ADVANCES IN MARINE STRUCTURES

MARSTRUCT.indb i 2/18/2011 5:39:02 PM

 PROCEEDINGS OF THE 3RD INTERNATIONAL CONFERENCE ON MARINE STRUCTURES –
MARSTRUCT 2011, HAMBURG, GERMANY, 28–30 MARCH 2011

Advances in Marine Structures

Editors
C. Guedes Soares
Instituto Superior Técnico, Technical University of Lisbon, Portugal

W. Fricke
Hamburg University of Technology (TUHH), Germany

MARSTRUCT.indb iii 2/18/2011 5:39:04 PM

 CRC Press/Balkema is an imprint of the Taylor & Francis Group, an informa business

© 2011 Taylor & Francis Group, London, UK

Typeset by Vikatan Publishing Solutions (P) Ltd., Chennai, India

Printed and bound in Great Britain by Antony Rowe (a CPI Group Company), Chippenham, Wiltshire

All rights reserved. No part of this publication or the information contained herein may be reproduced,
stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, by
photocopying, recording or otherwise, without written prior permission from the publisher.

Although all care is taken to ensure integrity and the quality of this publication and the information
herein, no responsibility is assumed by the publishers nor the author for any damage to the property or
persons as a result of operation or use of this publication and/or the information contained herein.

Published by: CRC Press/Balkema

P.O. Box 447, 2300 AK Leiden, The Netherlands
e-mail: Pub.NL@taylorandfrancis.com
www.crcpress.com – www.taylorandfrancis.co.uk – www.balkema.nl

ISBN: 978-0-415-67771-4 (Hbk + CD-ROM)

ISBN: 978-0-203-80811-5 (eBook)

MARSTRUCT.indb iv 2/18/2011 5:39:05 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Table of contents

Preface xi
Organisation xiii

Methods and tools for loads and load effects

Influence of raised invar edges on sloshing impact pressures—numerical investigations 3
S. Brizzolara, D. Villa, T. Gazzola, N. Tryaskin, N. Moirod, J. De Lauzon & L. Diebold
Modal approach to fluid structure interaction applied to a ship in waves 9
C. Cabos, B. Dilba, M. Krömer & A. Schwenkenberg
Sensors location and data processing algorithms of an optical fibers hull strength
monitoring system 19
A. Grasso, A. Vergine, D. Dimou, M. Samuelides, N. Tsouvalis & A. Ferrari
A hydroelastic investigation into the dynamic response characteristics
of bulk carriers 33
L. Kaydıhan, B. Uğurlu & A. Ergin
Numerical prediction of slamming loads on a rigid wedge subjected to water entry using
an explicit finite element method 41
H. Luo, S. Wang & C. Guedes Soares
On estimation of extreme ship response using upcrossing spectrum 49
W. Mao & I. Rychlik
Utilization of a whole ship finite element analysis from wave loads to structural strength
at real sea state 59
Y. Ogawa & M. Oka
Environmental and operational uncertainties in long-term prediction of slamming loads
of containerships 67
J. Parunov, M. Ćorak & I. Senjanović
Efficient calculation of fluid structure interaction in ship vibration 75
M. Wilken, A. Menk, H. Voss & C. Cabos

Methods and tools for strength assessment

Ultimate strength
Buckling analysis of composite delaminated ship plates under shearing 85
E.F. Beznea & I. Chirica
Shear buckling analysis of the composite plates with cut-outs 93
E.F. Beznea & I. Chirica
Robust ultimate strength formulation for stiffened plates subjected to combined loads 99
S.-R. Cho, H.-S. Kim, J.-B. Koo, H.-M. Doh & Y.-K. Chon

MARSTRUCT.indb v 2/18/2011 5:39:05 PM

 Rapid analysis techniques for ultimate strength predictions of aluminum structures 109
M.D. Collette
A revisit on design and analysis of stiffened shell structures for offshore applications 119
P.K. Das, K.K. Subin & P.C. Pretheesh
Finite elements modeling of delaminations in composite laminates 133
M. Gaiotti, C.M. Rizzo, K. Branner & P. Berring
Shakedown of welding-induced residual stress and effect on stiffened plate strength and behaviour 141
L.G. Gannon, N.G. Pegg, M.J. Smith & Y. Liu
Geometrical effects on strength and deformability of corroded steel plates 151
M.R. Islam & Y. Sumi
Methods to cope with up heaval buckling of high temperature offshore pipelines
in Mexico’s Bay of Campeche 161
J. Ochoa Z., J.E. Iturriaga F. & S. Meléndez P.
Benchmark study on use of ALPS/ULSAP method to determine plate and stiffened
panel ultimate strength 169
J.K. Paik, S.J. Kim, D.H. Kim, D.C. Kim, P.A. Frieze, M. Abbattista, M. Vallascas & O.F. Hughes
Modified Paik-Mansour formula for ultimate strength calculations of ship hulls 187
J.K. Paik, D.K. Kim, D.H. Park, H.B. Kim, A.E. Mansour & J.B. Caldwell
Development of advanced designed formulation to estimate the buckling/ultimate strength
of curved plates 203
J.-S. Park, M.-S. Chun & Y.-S. Suh
Ultimate strength assessment of ageing steel plates subjected to random non-uniform
corrosion wastage 213
J.E. Silva, Y. Garbatov & C. Guedes Soares
Comparison of numerical results with experiments on ultimate strength of short stiffened panels 221
M. Xu & C. Guedes Soares
Numerical study of the effect of geometry and boundary conditions on the collapse
behaviour of short stiffened panels 229
M. Xu & C. Guedes Soares
Hydro-elastoplasticity approach to ship’s hull girder collapse behavior in waves 239
W. Xu, K. Iijima & M. Fujikubo
A study on the dynamic buckling strength of containership’s bow structures subjected
to bow flare impact force 249
S.H. Yang, H.L. Chien, C.M. Chou, K.C. Tseng & Y.J. Lee
Ultimate strength of aluminum Y-stiffened panels 257
M.R. Zareei

Fatigue strength
Fatigue of high-speed aluminium ships: A master curve formulation 267
J.H. den Besten & R.H.M. Huijsmans
Stress and strain-based approaches for fatigue life evaluation of complex structural details 277
M. Biot & L. Moro
Different finite element refinement strategies for the computation of the strain energy density
in a welded joint 289
C. Fischer, A. Düster & W. Fricke
Round-robin on local stress determination and fatigue assessment of load-carrying
fillet-welded joints 295
W. Fricke, M. Codda, O. Feltz, Y. Garbatov, H. Remes, G. Risso, C. Rizzo & J. Romanoff
Fatigue estimation of a ship structural detail 303
S. Giuglea, A. Chirica, I. Chirica & E.F. Beznea
vi

MARSTRUCT.indb vi 2/18/2011 5:39:05 PM

 Adjoint design sensitivity analysis of crack propagation using molecular-continuum
multiscale approach 311
H.-L. Jang, M.-G. Kim & S. Cho
A study of design loads for fatigue strength utilizing direct calculation under real
operational conditions 317
M. Oka, T. Takami, Y. Ogawa & K. Takagi
The estimation of stress range distribution due to wide banded random loading obtained
by rain-flow counting method 325
J.B. Park, K.S. Kim & J. Choung
Three-dimensional fracture mechanics analyses of surface cracks
at welded joints in ship structure 335
S. Tanaka, H. Okada & S. Okazawa

Collision and impact strength

Ship hull composite plates analysis under blast loads 343
I. Chirica, D. Boazu, E.F. Beznea & A. Chirica
An investigation of a Suezmax tanker grounding accident 351
S. Ehlers, D. Polić, A. Klanac & M. Schröder
Research on anti-collision capability for double-hull design for the column of semi-submersible
by numerical simulation 357
Z. Hu, G. Chen & J. Yang
Effect of crushable bow on the overall crashworthiness in ship collision 365
Y.T. Huang, K.P. Wu, H.L. Chien, C.M. Chou, K.C. Tseng, C.F. Hung & C.L. Chang
A study on dynamic grounding of ships 373
T.-H. Nguyen, J. Amdahl, L. Garrè & B.J. Leira
Influence of bulbous bow structures on their collision behaviour 381
M. Schöttelndreyer, I. Tautz, J.M. Kubiczek, W. Fricke & E. Lehmann
Influence of ship motions in the numerical prediction of ship collision damage 391
K. Tabri & J. Broekhuijsen
Numerical prediction of impact loads in rectangular panels 399
R. Villavicencio & C. Guedes Soares
Influence of the neutral axis displacement on the residual strength of a damaged
tanker double bottom structure 411
R. Villavicencio, C. Guedes Soares, Z. Liu & J. Amdahl
Numerical simulation of laterally impacted clamped circular steel plates 419
R. Villavicencio, L. Sutherland & C. Guedes Soares
Study on the residual ultimate longitudinal strength of hull girder of a bulk carrier
against a sagging moment after ship collision 429
Y. Yamada & Y. Ogawa
Comparison of resistance to penetration of stiffened panels with T- and Y-stiffeners 437
M.R. Zareei
Ultimate load calculation during stranding 447
B. Zipfel & E. Lehmann

Dynamic analysis

Methods for hull structure strength analysis and ships service life evaluation, under
extreme hydroelastic wave loads, for a large oil-tanker 459
L. Domnisoru & A. Chirica
Dynamic analysis in the marine environment considering FSI—Ship-like structure case 473
A. Martínez Cimadevilla
vii

MARSTRUCT.indb vii 2/18/2011 5:39:05 PM

 Experimental analysis of structures
An experimental study on fatigue crack propagation life of T-joint fillet specimen considering
residual stress under storm loading 487
S.H. Kim, K.S. Kim, J.H. Lee, C.H. Yu & W.H. You
Experimental study on slamming load of the trimaran 495
S. Peng, H. Sun, J. Yue & W. Wu
Failures mode analysis of corroded steel structures subjected to compressive load 503
S. Saad-Eldeen, Y. Garbatov & C. Guedes Soares
Fatigue tests of butt welds and plates edges of 80 mm thick plates 511
H. von Selle, O. Doerk, J.K. Kang & J.H. Kim
Study on the ultimate strength test of high speed trimaran 521
W. Wu, S. Xu, W. Liu, X. Li & J. Yue
An experimental study on mechanical, fatigue and crack propagation properties
of IMO type B CCS materials at room and cryogenic temperatures 527
C.H. Yoo, K.S. Kim, J. Choung, C.S. Shim, J.K. Kang, D.H. Kim, Y.S. Suh, Y.L. Shim,
H.S. Urm, M.S. Kim & G.B. An

Materials and fabrication of structures

A study on laser assisted friction stir welding of C-Mn steel plates 539
P. Biswas & N.R. Mandal
Steel-concrete-steel sandwich structures in ship and offshore engineering 549
T.J. Grafton & J.R. Weitzenböck
Welding of large gaps of block joints in ships using different techniques 559
S. Zacke & W. Fricke

Methods and tools for structural design and optimisation

Scantling optimization of ship structures considering fatigue at the early design stage 569
A. Amrane & P. Rigo
Underwater noise emissions: Another challenge for ship design 581
M. André, T. Gaggero & E. Rizzuto
Normative framework for noise emissions from ships: Present situation and future trends 593
A. Badino, D. Borelli, T. Gaggero, E. Rizzuto & C. Schenone
Methods and criteria to manage airborne outdoor ship noise 603
M. Biot & L. Moro
Shape design optimization of fluid-structure interactions using SPH and geometrically
exact interfaces 611
H.-S. Kim, M.-G. Kim & S. Cho
Finite element-based shape optimization of an asymmetric steel sandwich panel joint 619
D. Polić, D. Frank, A. Klanac & S. Ehlers
Hull/superstructure-interaction in optimized passenger ships 625
H. Remes, J. Romanoff, P. Varsta, J. Jelovica, A. Klanac, A. Niemelä, S. Bralic & H. Naar
Optimization of monopile offshore wind structures 633
A. Thiry, F. Bair, L. Buldgen, G. Raboni & P. Rigo
Isogeometric shape optimization of design-dependent structures 643
M. Yoon, B.-Y. Koo & S. Cho

viii

MARSTRUCT.indb viii 2/18/2011 5:39:06 PM

 Structural reliability, safety and environmental protection
Combination of primary loading effects under various wave scatter diagrams 651
N.-Z. Chen, G. Wang, C. Guedes Soares & A.P. Teixeira
Uncertainty assessment of the ultimate strength of a stiffened panel 659
Y. Garbatov, M. Tekgoz & C. Guedes Soares
Safety analyses for bulk carriers using metamodels of still water loads 669
P. Georgiev
Reliability assessment of intact and damaged bulk carriers 679
A.W. Hussein & C. Guedes Soares
Structural optimization of the hold frame of a bulk carrier considering lifecycle risk 691
Y. Kawamura & M. Miyazaki
A study on long-term prediction of corrosion wastage 699
T. Matsukura, Y. Kawamura & E. Khoo
Optimization of composite maritime structures—effects of uncertainties on design
criteria and limits 707
L. Sanchez, J.W. Ringsberg & E. Johnson
Rule development for container stowage on deck 715
V. Wolf, I. Darie & H. Rathje
Reliability analysis of marine structural components using statistical data of steel strength 723
B. Yu & D.G. Karr

Author index 731

ix

MARSTRUCT.indb ix 2/18/2011 5:39:06 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Preface

This book collects the papers presented at the 3rd International Conference on Marine Structures,
MARSTRUCT 2011, which was held in Hamburg, Germany 28 to 30 March. This Conference follows
up from the initial ones that were held in Glasgow, Scotland and in Lisbon, Portugal, respectively in 2007
and 2009. These conferences aim at bringing together researchers and industrial participants specially
concerned with structural analysis and design of marine structures. Despite the availability of several con-
ferences about ships and offshore structures, it was felt that there was still no conference series specially
dedicated to marine structures, which would be the niche for these conferences.
The initial impetus and support has been given by the Network of Excellence on Marine Structures
(MARSTRUCT), which was funded by the European Union from 2005 to 2010, bringing together
33 European research groups from Universities, research institutions, classification societies and indus-
trial companies that are dedicated to research in the area of marine structures. With the end of this
EU project, a new organisation was created to maintain the cooperation ties among the groups that work
in this general area. The MARSTRUCT Virtual Institute was created in 2010, with the aim of being an
association of research groups interested in cooperating in the field of marine structures. It started with
the same members as the participants in the Network of Excellence but in the near future it is open to
accept the membership of other European groups that have the same aims.
The Virtual Institute is organised in the following Technical Committees:
• Methods and tools for establishing loads and load effects
• Methods and tools for strength assessment
• Experimental analysis of structures
• Materials and fabrication of structures
• Methods and tools for structural design and optimisation
• Structural reliability, safety and environmental protection
which in turn are divided in subcommittees. The aim is to promote the exchange of information and the
cooperation in these subject areas. This can take the form of promoting comparative or benchmark stud-
ies in various subjects, promoting joint research activities and joint research projects, organising short
specialised courses, workshops and conferences.
In particular the Virtual Institute will take the responsibility of organising the MARSTRUCT Confer-
ence biannually in different countries, starting from the present one that is already organised in a coop-
eration between the MARSTRUCT Virtual Institute and the Hamburg University of Technology in a
scheme that is planned to be continued in the future: the Virtual Institute will be responsible for the prepa-
ration of the technical programme and processing of the papers and the host country organization will be
responsible for the conference organization and management.
Despite being organised in Europe, this Conference is not meant to be restricted to European attendees
and a serious effort has been made to involve in the planning of the Conference participants from other
continents that could ensure a wider participation, which is slowly happening.
The conference reflects the advances that have been made in the last years within its domain including
the full range of methods and modelling procedures for the structural assessment of marine structures.
Various assessment methods are incorporated in the methods used to analyze and design efficient ship
structures, as well as in the methods of structural reliability to be used to ensure the safety and environ-
mental behaviour of the ships. This book deals also with some aspects of fabrication of ship structures.
This book includes almost 80 papers, which are organised into the themes that correspond to the Virtual
Institute Technical Committees, as listed above. The papers were accepted after a review process, based on
the full text of the papers. Thanks are due to the Technical Programme Committee and to the Advisory

xi

MARSTRUCT.indb xi 2/18/2011 5:39:06 PM

 Committee who had most of the responsibility for reviewing the papers and to the additional anonymous
reviewers who helped the authors deliver better papers by providing them with constructive comments.
We hope that this process contributed to a consistently good level of the papers included in the book.

Carlos Guedes Soares,

Wolfgang Fricke

xii

MARSTRUCT.indb xii 2/18/2011 5:39:06 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Organisation

Conference Chairmen
Prof. Wolfgang Fricke, Hamburg University of Technology (TUHH), Germany
Prof. Carlos Guedes Soares, IST, Technical University of Lisbon, Portugal

Technical Programme Committee

Prof. N. Barltrop, University of Strathclyde, UK
Prof. I. Chirica, University ‘Dunarea de Jos’ of Galati, Romania
Dr. M. Codda, CETENA, Italy
Prof. P. Das, University of Strathclyde, Glasgow, UK
Prof. R. Dow, University of Newcastle upon Tyne, England–UK
Prof. Y. Garbatov, IST, Technical University of Lisbon, Portugal
Prof. J. M. Gordo, IST, Technical University of Lisbon, Portugal
Dr. B. Hayman, Det Norske Veritas, Norway
Prof. A. Incecik, University of Strathclyde, UK
Prof. T. Jastrzebski, West Pomeranian University of Technology, Poland
Prof. B.J. Leira, Norwegian University of Science and Technology, Norway
Dr. S. Malenica, Bureau Veritas, France
Prof. T. Moan, Norwegian University of Science and Technology, Norway
Prof. U. Nielsen, Technical University of Denmark, Denmark
Prof. J. Ringsberg, Chalmers University of Technology, Sweden
Prof. P. Rigo, University of Liège, Belgium
Prof. E. Rizzuto, University of Genova, Italy
Prof. J. Romanoff, Aalto University, Finland
Prof. M. Samuelidis, National Technical University of Athens, Greece
Prof. R.A. Shenoi, University of Southampton, England–UK
Prof. M. Taczala, West Pomeranian University of Technology, Poland
Prof. P. Temarel, University of Southampton, UK
Prof. P. Varsta, Aalto University, Finland
Dr. A. Vredeveldt, TNO, The Netherlands

Advisory Committee
Prof. F. Brennan, Cranfield University, UK
Prof. A. Campanile, University of Naples, Italy
Prof. G. Chen, Shanghai Jiaotong University, P.R. China
Dr. F. Cheng, Lloyd’s Register, UK
Prof. S.-R. Cho, University of Ulsan, Korea
Prof. Y.S. Choo, Nat. Univ. Singapore, Singapore
Prof. W.C. Cui, CSSRC, P.R. China
Prof. C. Daley, Memorial University, Canada
Dr. M. Dogliani, Registro Italiano Navale, Italy
Prof. A. Ergin, ITU, Turkey
Prof. S. Estefen, COPPE/UFRJ, Brazil
Prof. M. Fujikubo, Osaka University, Japan
Prof. T. Fukasawa, Osaka Prefecture University, Japan
Prof. C.-F. Hung, National Taiwan University, Taiwan ROC

xiii

MARSTRUCT.indb xiii 2/18/2011 5:39:06 PM

 Prof. D. Karr, University of Michigan, Ann Arbor, Michigan, USA
Prof. H.W. Leheta, Alexandria University, Egypt
Prof. J.K. Paik, Pusan National University, Korea
Dr. N.G. Pegg, DND, Canada
Prof. M. Salas, University Austral of Chile, Chile
Dr. I. Senjanović, University of Zagreb, Croatia
Dr. R. Skjong, Det Norske Veritas, Norway
Prof. Y. Sumi, Yokohama National University, Japan
Dr. O. Valle Molina, Mexican Inst of Petroleum, Mexico
Dr. P. Videiro, Petrobras, Brazil
Dr. G. Wang, American Bureau of Shipping, USA
Dr. X. Wang, American Bureau of Shipping, USA
Prof. W. Wu, Wuhan University of Technology, P.R. China

Local Organizing Committee

Olav Feltz, Hamburg University of Technology (TUHH), Germany
Sonja Zacke, Hamburg University of Technology (TUHH), Germany

Technical Programme Secretariat

Maria de Fátima Pina, IST, Technical University of Lisbon, Portugal

xiv

MARSTRUCT.indb xiv 2/18/2011 5:39:06 PM

 Methods and tools for loads and load effects

MARSTRUCT.indb 1 2/18/2011 5:39:06 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Influence of raised invar edges on sloshing impact pressures—

numerical investigations

Stefano Brizzolara & Diego Villa

Marine CFD Group—Faculty of Engineering, University of Genova, Italy

Thomas Gazzola, Nikita Tryaskin, Nicolas Moirod, Jérôme De Lauzon & Louis Diebold
Bureau Veritas, Marine Division, Research Department, Neuilly-Sur-Seine Cdx—France

ABSTRACT: This paper presents a numerical investigation of the influence of raised invar edges on
wet drop tests pressures. The aim of this study is to evaluate the capabilities of the CFD software Open-
FOAM ([6]) to deal with fluid structure impact problems in the context of sloshing inside membrane
tanks. More precisely, the objective is to evaluate OpenFOAM capabilities for water drop tests simulations
for a smooth wedge on one hand and a wedge equipped with invar edges (like these which equip the Cargo
Containment System NO96 produced by GTT) on the other hand, both falling into calm water without
inclination. The obtained numerical results are then compared to Wagner’s solution for the smooth wedge
and to experimental measurements for the wedge with edges presented in ([2]). Numerical simulations are
in very good agreement with Wagner’s solution and experimental results ([2]) showing the OpenFOAM
capability to deal with fluid structure impact problems in the context of sloshing inside membrane tanks.
The agreement between numerical and experimental results confirms also that raised invar edges tend to
enhance the magnitude of sloshing pressures. This confirmation emphasizes the importance of consider-
ing the physics of invar edge effects in defining the design pressure to be used in assessing the integrity of
membrane LNG tanks.

1 INTRODUCTION In order to better understand the influence of

In order to manage the risk of failure due to slosh-
ing, adequate assessment of sloshing loads and
structural capacities are required. Even if the state
of the art of sloshing model tests has improved a
lot these last few years—by including more physics
such as reproduction of realistic irregular 6 degrees
of freedom motion for the tank, introduction of a
special ullage gas mixture in order to respect the
LNG gas/fluid ratio density in the model tank—
some physical phenomena are still not reproduced
in actual sloshing model tests.
Indeed, sloshing model tests are carried out using
tanks with smooth walls. However at prototype
scale, the two widely-used Cargo Containment
Systems (CCS) both have raised elements, corru-
gations in the case of MarkIII and raised invar
edges in the case of NO96. The MarkIII primary
membrane includes a square pattern of corruga-
tion cells formed by the crossing rows of larger
and smaller corrugations, both with spacings of
roughly 340 mm. In the case of NO96 system,
parallel rows of raised invar edges are present with
spacings of roughly 500 mm and contain the weld
used to join the sheets of invar that make up the
primary membrane.
these raised elements on sloshing pressures, differ-
ent studies have been carried out. Due to complex-
ity of sloshing model tests using raised edges, most
of these studies consisted in performing wedge
drop tests widely used in the industry as a means
to investigate fluid impact problems. The main
conclusion of these studies was that corrugations
significantly reduced the magnitude of impact
pressures by factors at least 2 when compared to
pressures measured on smooth wedge. The main
explanation for such reduction was the trapped air
cushioning effect.
However, a recent study (0) based on drop tests
and 2D sloshing tests showed that raised elements
effect on sloshing pressures is complex and may
not lead necessarily to lower pressures when com-
pared to the smooth cases.
The purpose of the present paper is first to
check/confirm the findings presented in (0) and
then to evaluate OpenFOAM capability to deal
with complex fluid structure impact problems by
comparing our numerical results to the experi-
mental results (0) for the wet drop tests with &
without invar edges. The calculations presented
in this paper concern only drop tests for wedges
with invar edges like these which equip NO96 CCS

MARSTRUCT.indb 3 2/18/2011 5:39:06 PM

 but not with corrugations like those which equip
MarkIII CCS.
Present work is complementary to the long vali-
dation study performed on different CFD methods
(unsteady free surface commercial RANSE solv-
ers and proprietary SPH method) with regards to
impact load of different ship like section shapes,
performed within MARSTRUCT activities and
already published in different occasions (0).
2 WEDGE DROP TESTS SETUP
2.1 Equations for the impact problem
The simulation of the impact of the falling wedge
in calm water is assumed uncoupled. The body is
rigid, and its position and velocity is prescribed
and imposed.
This impact problem can be solved in several
equivalent ways. It is possible to impose a flow at
a specific velocity that comes upward and hits a
static structure, or to let the structure go down in a
calm flow. The second method was chosen, because
it will offer more possibilities for structure motions
in future studies.
The solver used for this study is interDyMFoam
from OpenFOAM (0). The interDyMFoam is a
solver for two incompressible, isothermal immis-
cible fluids using a VOF (volume of fluid) phase-
fraction based interface capturing approach, in a
moving mesh domain. This solver solve the well
know Navier-Stokes equations, recombined for a
moving mesh. Following are presented the equa-
tion in the punctual versions and write for a mov-
ing control volume:
⎧⎪∇ ⋅ U = 0
⎨ ∂ρU + ρU ⋅ ∇ = ∇p + μΔU + f
(1)
⎩⎪ ∂t
⎧ ∇⋅
⎪∫
= ∫ ∇ ⋅UG dV
⎪ ⎛ ∂ρU ⎞ length of the wedge = 1000 mm
⎨∫ ⎜⎝ + ρU ⋅ ∇U ⎟ dV
⎠
(2)
⎪ ∂ t
⎪ = ∫ (∇ +
⎩ + ) dV + ∫ ( ⋅∇ ) dV
where UG represent the Velocity of the mesh.
Basically a new virtual flux is added to the equa-
tion to take in account the moving of the mesh.
That solver use a PISO loop at each time step
to ensure the pressure-velocity coupling. In par-
ticular, but an addition correction to the standard
PISO algorithm has been sued to better predict all
the no-linear terms of N-S equations. The time
marching is performed with a Eulerian implicit
approach, that increases the stability of the solu-
tion. The linearized system has been solved with
a GAMG (Generalized geometric-algebraic multi-
grid) for the Poisson equation and with a Smooth
solver (GaussSeidel as smoother) for the N-S equa-
tion. This solver use a VoF method to track the
interface between the two fluids, so it solves the
classical VoF equation, where alpha represents
the fraction of fluid present in each cell, and all
the physical quantities representing of the fluids
characteristics are weighted by this fraction.
∂α
+ α ⋅∇ = 0
∂t (3)
⎧ρ = αρ1 + ( − ) ρ2
⎨ μ = αμ + − μ
1 ( ) 2
⎩
To solve the VoF equation, that is a standard
fully convective equation, the interDyFoam use
the MULES method (multidimensional universal
limited explicit solver), that is a new generic solver
for simply convective equations with a bounded
range of free variable values.
2.2 Setup of the numerical simulations
The wedges configurations are the same one as the
ones described in (0). For sake of simplicity, only
the scale 1:6 is here presented in this section. For
the 2 other scales, one should refer to (0).
The set-up of the numerical simulations is the
following:
height of the wedge = 100 mm

Figure 1. Drop test scheme. Figure 2. Wedge configuration for scale 1:6.

MARSTRUCT.indb 4 2/18/2011 5:39:07 PM

 width of the wedge = 278 mm (however 2D
simulations were carried out)
dead rise angle = 10° (to ensure validity of
Wagner’s solution)
drop height = 0.6 m
On the contrary to experiments presented in
(0) where one side of the wedge is smooth and the
other side equipped with raised edges, two con-
figurations of the wedge were simulated. The first
one without invar edge (smooth case) and the sec-
ond one with invar edges scaled at 1:6. For both
configurations, a symmetry plan was considered as
shown in Fig. 3 hereafter. In the case of the wedge
with invar edges, the first invar edge is located at
230 mm away from apex of the wedge and the
distance between the two invar edges is 83 mm
like in (0). The locations of the pressure sensors
are chosen accordingly to (0) and are reminded
below. The dimensions are given with respect to
the apex of the wedge.
Three prescribed constant velocities of the
wedge equal to 2.80, 3.43 & 3.96 m.s−1 were sim-
ulated corresponding to the theoretical free fall
velocities obtained for the 3 drop heights of h = 0.4,
0.6 & 0.8 m. Thus variation of the wedge velocity
during impact is not taken into account.
For the two other scales (1:20 & 1:35), the 3 drop
heights were scaled from the scale 1:6.
Thus in our numerical simulations, the flow is
going from the pressure sensor n°1 to the pressure
sensor n°18.
3 WEDGE DROP TESTS RESULTS
3.1 Results for the smooth wedge (without invar
edge)
For sake of simplicity, only results at scale 1:6 are
here presented. Our numerical procedure displaces
the mesh at V = 2.80, 3.43 & 3.96 m.s−1. The mesh
used for this simulation is figured out below. The
total number of cells used to define this mesh is
equal to 290,000 cells.
For the case of the wedge without invar edge at
scale 1:6 and drop height h = 0.8 m, the pressure
time histories calculated at pressure sensor loca-
tion (10–18) are presented in Figure 4.
In order to validate this CFD numerical solution
obtained with OpenFOAM, this one is compared
with the reference Wagner’s analytical solution
for the smooth wedge without inclination which
is valid for the studied configuration (0, 0 & 0).
Comparison with experiments presented in (0) is
not possible since no results for the smooth case
are provided in (0). Comparisons of the maximum
peak impact pressures for all the pressure sensors
from #10 to #18 are figured out below for scale 1:6
and the 3 dropping heights. There is an excellent
agreement between the Wagner’s analytical solu-
tion and the CFD solution for the maximum peak
impact pressures. The difference for the maximum
peak impact pressure between the CFD solution
(OpenFOAM) and the Wagner’s analytical solu-
tion does not exceed 3% for the pressure sensors
of interest #10–#18. OpenFOAM and Wagner
pressure time histories for the pressure sensor #18
which presents the maximum error (of 3%) in com-
parison with Wagner’s solution is figured hereafter
on Fig. 5.
The difference between the CFD solution (blue)
and the Wagner’s analytical solution (red) arises
when the separation point reaches the end of the
wedge as figured below.

700000
p10
600000 p11
p12
p13
500000
p14
Figure 3. Mesh used for the smooth wedge simulation 400000
p15

(290,000 cells), simulation scale 1:6.

pressure (Pa)

p16
p17
300000
p18

Table 1. Pressure sensors location. 200000

N°Sensor Before 1st edge (mm) Between edges (mm) N°Sensor 100000

1 151.611 234.611 10 0
2 160.833 243.833 11
3 170.056 253.056 12 –100000

4 179.278 262.278 13 0 0.005 0.01 0.015 0.02 0.025

time (s)
5 188.500 271.500 14
6 197.722 280.722 15
7 206.944 289.944 16 Figure 4. Pressure time histories calculated by Open-
8 216.167 299.167 17 FOAM for the smooth wedge, pressure sensors 10–18 for
9 225.389 308.389 18
scale 1:6 and h = 0.8 m.

MARSTRUCT.indb 5 2/18/2011 5:39:11 PM

 3.2 Results for the wedge with invar edges Wagner s18
OpenFOAM s18
600000
For the sake of simplicity, only results at scale 1:6
are here presented. Our numerical procedure dis- 500000

places the mesh at V = 2.80, 3.43 & 3.96 m.s−1. The

400000
mesh used for this simulation is figured out below.
The total number of cells used to define this mesh 300000

is equal to 290,000 cells.

200000
For the case of the wedge without invar edge at
scale 1:6 and drop height h = 0.8 m, the pressure 100000

time histories calculated at pressure sensor loca- 0

tion (1–18) are represented in Figures 8. 0.008 0.01 0.012 0.014 0.016 0.018

One can notice that the time pressure histories

signals are noisy. This is the reason why it was Figure 5. OpenFOAM calculations & Wagner pressure
decided to apply a low-pass filter (2.5 kHz, 8th time histories for pressure sensor 18 for scale 1:6 and
order Butterworth low pass filter) to these signals. h = 0.8 m.
However, to ensure that this low-pass filter does
not affect the impact peak pressure of interest, this
low-pass filter is applied on the time pressure his-
tories obtained for the smooth wedge which was
validated in the previous section. For instance,

Table 2. Comparison of the maximum peak impact

pressures obtained by OpenFOAM calculations with the
Wagner’s solution.

1:6, h = 0.4 m OpenFOAM Wagner (OF/Wagner)

Sensor 10 310608 311000 0.999

Sensor 11 311117 311000 1.000
Sensor 12 314067 311000 1.010
Sensor 13 315277 311000 1.014
Sensor 14 319110 311000 1.026
Figure 6. Classical Wagner’s analytical solution is no
Sensor 15 319133 311000 1.026
more valid when the separation point reaches the end of
Sensor 16 317646 311000 1.021
the wedge.
Sensor 17 318834 311000 1.025
Sensor 18 319689 311000 1.028
1:6, h = 0.6 m OpenFOAM Wagner (OF/Wagner) the difference between the raw and filtered time
Sensor 10 466241 467000 0.998 pressure history for the pressure sensor #3 for scale
Sensor 11 465773 467000 0.997 1:6 and h = 0.6 m is figured out below.
Sensor 12 470233 467000 1.007 The low-pass filter has a limited influence
Sensor 13 472968 467000 1.013 (experiments post-processing procedure is not
Sensor 14 480360 467000 1.029 available) on the impact pressure time history for
Sensor 15 481153 467000 1.030 the case of the smooth wedge (without invar edge).
Sensor 16 478298 467000 1.024 Therefore, this low-pass filter was also applied to
Sensor 17 479665 467000 1.027 time pressure histories for all the pressure sensors
Sensor 18 480882 467000 1.030 for the drop test simulation of the wedge with
1:6, h = 0.8 m OpenFOAM Wagner (OF/Wagner) invar edges. These filtered time pressure histories
are figured out below.
Sensor 10 623481 622000 1.002 The Figure 11 depicts the time pressure histo-
Sensor 11 623824 622000 1.003
ries obtained in (0) for one case which is not speci-
Sensor 12 628800 622000 1.011
fied (so the y-axis legend is not given). However, it
Sensor 13 632826 622000 1.017
is possible to compare both results (Figure 10 &
Sensor 14 642751 622000 1.033
Figure 11) qualitatively. Doing so, one can notice
Sensor 15 643088 622000 1.034
that:
Sensor 16 638552 622000 1.027
Sensor 17 640465 622000 1.030
The first pressure sensors (before the first edge),
Sensor 18 642697 622000 1.033
pressures are similar to these ones obtained for the
smooth wedge. This can be explained by the fact

MARSTRUCT.indb 6 2/18/2011 5:39:13 PM

 Figure 9. Comparison between raw and filtered (low-
pass filter 2.5 kHz) time pressure histories calculated
by OpenFOAM for pressure sensor #3 for the smooth
wedge.

p1 p10
1.2e+006 p2 p11
p3 p12
p4 p13
1e+006
p5 p14
p6 p15
800000
pressure (Pa)

p7 p16
p8 p17
600000 p9 p18

400000

200000

Figure 7. Mesh used for the drop test simulation of the

0.005 0.006 0.007 0.008 0.009 0.01
wedge with invar edges (290,000 cells). time (s)

Figure 10. Filtered pressure time histories calculated

by OpenFOAM for all pressure sensors. Drop test sim-
1.8e+006
ulation of the wedge with invar edges for scale 1:6 and
1.6e+006
p1
p2
p10
p11
h = 0.8 m.
p3 p12
1.4e+006
p4 p13
1.2e+006 p5 p14
p6 p15
pressure (Pa)

1e+006 p7 p16
p8 p17
800000
p9 p18
600000

400000

200000

–200000
0 0.005 0.01 0.015 0.02 0.025

time (s)

p1 p10
1.6e+006 p2 p11

1.4e+006
p3 p12
Figure 11. Experimental drop test with invar edges (0).
p4 p13
1.2e+006 p5 p14
p6 p15
1e+006
that the first edge does not have any influence on
pressure (Pa)

p7 p16
p8 p17
800000

600000
p9 p18
upstream pressure sensors.
400000
OpenFOAM well predicts the over peak impact
200000
pressure (in comparison with smooth wedge) before
0 the first edge (see sensor #8 & #9).
-200000 OpenFOAM well predicts the under peak impact
0.004 0.005 0.006 0.007 0.008 0.009 0.01 0.011

time (s)
pressure (in comparison with smooth wedge) just
after the first edge (see sensor #10 & #11).
Figure 8. Pressure time histories for all pressure sensors OpenFOAM well predicts the over peak impact
calculated by OpenFOAM. Drop test simulation of the pressure (in comparison with smooth wedge)
wedge with invar edges (zoom for the right figure). before the second edge (see sensor #17 & #18).

MARSTRUCT.indb 7 2/18/2011 5:39:15 PM

 2.5
2
Ratio (invar/smooth)
1.5
1
0.5
0
9 10 11 12 13 14 15 16
Number of the Pressure Sensor
Figure 12. OpenFOAM calculations (symbols) and
experimental (0) (black line) maximum impact pres-
sures ratios between wedge with invar edges and the
smooth one.
3.3 Results for the ratio (invar case/Wagner)
All these qualitative observations can be quanti-
fied by comparing the ratio between the filtered
maximum pressures for the wedge with invar edges
and the Wagner’s solution as it is done in (0).
The agreement between OpenFOAM and Exxon
experiments (0) curves is satisfactory. Furthermore,
precise information on the experimental procedure
and its post-processing lack to perform clearer
quantitative conclusions. Hence, we can conclude
that OpenFOAM well predicts qualitatively and
quantitatively the peak impact pressures for the
drop test simulation of a wedge equipped with
invar edges.
4 CONCLUSIONS
The aim of this study was to check/confirm the
findings presented in (0) from one hand and to
evaluate the capabilities of the CFD software
OpenFOAM to deal with fluid structure impact
problems in the context of sloshing inside mem-
brane tanks from the other hand.
More precisely, the objective was to evaluate
OpenFOAM capabilities for drop tests simulations
for a smooth wedge on one hand and a wedge
equipped with invar edges (like these which equip
the Cargo Containment System NO96 produced
by GTT) on the other hand, both falling into calm
water without inclination.
In section 3-1, the drop test simulation of the
smooth wedge was performed and the obtained
time pressure histories were then compared to the
Wagner’s analytical solution (valid for this studied
configuration). The agreement between the two
solutions is excellent and thus OpenFOAM drop
test simulation for smooth wedge is validated.
MARSTRUCT.indb 8
In section 3-2, the drop test simulation of the
1:6, 0.4m
wedge equipped with invar edges was performed
1:6, 0.6m
1:6, 0.8m
and the obtained time pressure histories were
1:20, 0.12m
1:20, 0.18m
analysed. As these time pressure histories signals
1:20, 0.24m
1:35, 0.069m
are noisy, it was decided to filter these signals with
1:35, 0.086m
1:35, 0.103m
a low-pass filter (2.5 kHz, 8th order Butterworth).
Experimental
In order to quantify the influence of this low-pass
17 18 19
filtering on peak impact pressures of interest, this
filter was applied on the time pressure histories
obtained for the smooth wedge. It was then shown
that this low-pass filtering has a limited influence
on peak impact pressures of interest.
Then, the time pressure histories were com-
pared qualitatively and quantitatively with the
experimental ones obtained in (0). The agreement
is very good considering the complexity of such
simulation.
As a conclusion, it can be said that OpenFOAM
can be used to simulate drop tests simulations not
only for smooth wedge but also for wedge equipped
with invar edges which represent a much more
complicated case. The next step is to perform drop
tests simulations for wedges equipped with corru-
gations (like these which equip the Cargo Contain-
ment System MarkIII produced by GTT).
REFERENCES
Gazzola, T. “Contribution au problème d’impact non
linéaire: le problème de Wagner couplé”, Ecole Cen-
trale Paris 2007 (written in English).
He, H., Kuo, J.F., Rinehart, A. & Yung, T.W. “Influ-
ence of Raised Invar Edges on Sloshing Impact
Pressures”, 1st Sloshing Dynamics Symposium,
ISOPE 2009, Vol. 3, www.isope.org
Molin, B. “Hydrodynamique des Structures Offshore”,
Editions Technip.
OpenFOAM, v.1.7.1, www.openfoam.com
Viviani, M., Brizzolara, S. & Savio, L. “Evaluation
of slamming loads using smoothed particle hydro-
dynamics and Reynolds-averaged Navier–Stokes
methods”. Journal of engineering for the maritime
environment, 2009, 223:17–31, ISSN: 1475-0902, doi:
10.1243/14750902 JEME131.
Wagner, H. “Über Stoss und Gleitvorgänge an der Ober-
fläche von Flüssigkeite ”, ZAMM, Vol. 12, 193–215.
2/18/2011 5:39:22 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Modal approach to fluid structure interaction applied

to a ship in waves

Christian Cabos, Boris Dilba, Matthias Krömer & Anne Schwenkenberg

Germanischer Lloyd, Hamburg, Germany

ABSTRACT: We present a modal approach to the coupled computation of the fluid flow and the
motion and elastic deformation of a floating body. For the fluid part, i.e. the transient viscous free sur-
face flow around the structure, the open souce code OpenFOAM is adopted. For the structure part, i.e.
the fully non-linear six degree of freedom rigid body motion and the linear elastic deformations due to
the forces exerted by the fluid on the structure, an in-house code has been developed and linked to the
OpenFOAM solver. An algorithm maps fluid forces to the finite element model nodes of the structure
and maps the structure motion/deformations to the vertices of the fluid grid. In our modal approach,
the structure deformation is represented by a linear combination of a small number of precomputed
modes (in the current study, the eigenmodes of the structure), resulting in an ODE for the “modal coef-
ficients”. This approach proves to be computationally efficient: on the one hand, the size of the equation
system is much smaller than the full finite element equations, on the other hand, the restriction to low-
frequency eigenmodes suppresses high-frequency oscillations encountered in direct FE computations,
allowing for larger time step sizes and in general a more stable coupling. Also, suitable selection of the
modes allows for focussing on “critical” structure deformations e.g. for fatigue assessment. To this end,
deformations corresponding to the most relevant load cases according to classification rules can be used
as modes, too.

1 INTRODUCTION global strength analysis can be chosen. Integrating

Hydroelastic effects, i.e. the consideration of the
elasticity of a ship when computing behaviour in
waves, are important in particular when analysing
slamming events. Here, dynamic amplification of
the lowest global elastic eigenmodes of the vessel
(whipping) can occur. Since in this case fluid flow
and elastic deformation of the ship significantly
influence each other, the governing equations need
to be solved in a coupled fashion.
Because the global whipping deformations are
small compared to the ship length, linear elastic-
ity suffices to describe the deformation of the ves-
sel. On the other hand, translations and rotations
of the rigid ship can be significant; the rigid body
motion of the ship is therefore handled non-linearly
in this paper.
The elastic deformation of a ship and its result-
ing stresses can be represented through a finite
element model of the vessel. Different approaches
are possible concerning the mesh density. On the
coarsest scale, the ship can be represented by a few
beam elements, see (Oberhagemann, el Moctar, and
Schellin 2008). In order to also assess the torsional
deformations and e.g. local stress concentrations,
a global finite element as commonly used for
such a model can however lead to high numerical
effort on the one hand because of the large size of
the structural system of equations but on the other
hand since high mesh density can lead to unneces-
sarily small time steps in the time integration. For
the latter reasons, the elastic deformation of the
vessel is here approximated through a few defor-
mation modes of the global finite element model
only. The structural equations are described in
chapter 2 of this paper.
For analysing slamming events for a ship in
waves, non-linearities in the computation of the
fluid flow should be considered. In particular
computation in the time domain has advantages in
this case. For this reason, in this paper the Finite
Volume method is applied to solve the Reynolds
Averaged Navier Stokes Equations, see chapter 3.
For the solution of the coupled equation sys-
tems, the partitioned approach is used, i.e. separate
solvers for the fluid and the structure are called in
an iterative fashion, (see (Matthies, Niekamp, and
Steindorf 2006) and the references therein for an
introduction).
Mesh densities of the water near the ship and
of the wet part of the shell need not agree. This
is due to the differing typical wave lengths in both

MARSTRUCT.indb 9 2/18/2011 5:39:22 PM

 media. For this reason the meshes of fluid and
structure will not be compatible at their interface
in general. As a result a procedure is introduced to
transfer pressures to the hull and displacements to
the water between differing meshes. The coupling
procedure and the employed mapping algorithm
are described in section 4.
2 STRUCTURAL PART
Assuming linear elasticity the 3D FEM discretiza-
tion of the vessel leads to the well known equation
of motion
Mü + Cu· + Ku = F(t) (1)
in an inertial reference frame. Here M, C and K
represent the mass, damping and stiffness matrix,
F is the external load vector and u is the vector of
displacements. Its first and second derivative rep-
resent the velocity and acceleration, respectively.
In this paper, the structural displacements of the
hull will be approximated by a linear combination
of vectors φi (called modes in the following)
n
u( x t ) ∑ φ( x )i ri (t ) = Φ( x )r(t ). (2)
i =1
Since the ship will perform large rigid body
motions but only small elastic deformations, a
body fixed reference frame for solving the struc-
tural equations of motion will be chosen such that
the modal coefficients r1, … , r6 for the rigid body
modes are zero in the body fixed system. Due to
the movement of the body fixed coordinate sys-
tem, additional inertial forces will result. In the
following sections, the terms “inertial”, “global”
resp. “body fixed”, “local”, “moving” will be used
synonymously.
To derive the equation of motion in this mov-
ing coordinate system we first consider the iner-
tia forces acting on a structure as a consequence
of the moving reference frame. For this, Ü is the
translatory acceleration of the local coordinate
system in local coordinates and Ω and Ω  represent
the angular velocity and acceleration of the coor-
dinate system about its origin in local coordinates.
According to (Argyris and Mlejnek 1988) the iner-
tial forces are then given by
Mr
R ( H + Z )r + Gr + M (6)
in modal coordinates. Here the symbols R, H, Z,
G, and M denote
R ∫V ρ Φ
T  × x ) + Ω × (Ω × x )⎤ dV ,
⎡⎣U + (Ω ⎦
(7)

The first six modes φi, i = 1, …, 6 will be assumed the angular acceleration stiffness
to represent the rigid body motion of the ship. The
time dependent modal coefficients ri, i = 1, …, n are H ∫V ρ
T
( )dV , (8)
assembled in the vector
the centripetal forces
r = [ri, …, rn]T (3)

and the modes into the matrix Zr = ∫ ρ ΦT [ Ω × (Ω × Φ )]dV r, (9)

V

Φ = [φ1, …, φn]. (4) the Coriolis forces

A good choice for representing the global defor- Gr = 2∫ ρ ΦT (Ω × Φ )dV r, (10)
mation of the ship typically are the lowest eigen- V
modes which result from the modal decomposition
of the homogeneous, undamped and unsupported and the mass matrix
system
∫V ρ (11)
T
M dV .
Mü + Ku = 0. (5)
All terms of the equation of motion in the mov-
This leads to the “dry free vibration” mode ing coordinate system have been given above for
shapes φi, i =1‥n and the corresponding circular completeness. In ship applications, however, the
eigenfrequencies ωi, i = 1‥n. Note that since the forces resulting from the terms containing G, H,
ship is assumed to be unsupported ωi = 0, i = 1‥6 and Z are typically negligibly small whereas R
corresponding to six rigid body modes. While con- results in a significant contribution.
ceptually similar to “generalized” or “principal Now adding the stiffness
coodinates” in classical mechanics, we prefer the
term “modal coefficients” in order to stress the
∫V ρ (12)
T
K dV
approximation by a small number of modes.

10

MARSTRUCT.indb 10 2/18/2011 5:39:22 PM

 and damping C and decomposing the external The dry modes are computed from the
forces into modes leads to the equation of motion generalized eigenproblem
in modal coordinates in the moving coordinate
system (K − ω2M)φ

 ) Z (Ω )]r by requiring
Mr + [C G (Ω)]
)]r [ K + H (Ω
(13)
= ∫ φT FdA
d R( ,  , U ). det (K − ω2M) = 0,
A

If the modes used for the projection are nor- resulting in one rigid body eigenvalue/eigenvector
malized and orthogonal with respect to the mass
matrix, M becomes the unit matrix. In case moreo- ω 12 0 φ1 (1/ 2 )(1, 1)T
ver global vibration eigenmodes are chosen, K is
the diagonal matrix of the squared circular eigen- and one elastic eigenvalue/eigenvector
frequencies ω i2. Damping is then typically chosen
as applied for a global ship vibration analysis, i.e. ω 22 2k φ2 (1/ 2 )( 1, 1)T .
as a percentage of critical damping applied to the
respective mode (see e.g. (Asmussen, Menzel, and The modes
Mumm 2001)).
Introducing the abbreviations 1 ⎛1 −1⎞
Φ = (φ1, φ2 ) =
C* (Ω) = C + G(Ω), (14) 2 m ⎜⎝1 1⎟⎠

*
(Ω , Ω) 
K +H Ω ( ) Z (Ω ) , (15) are ortho-normal with respect to the mass
matrix, i.e.
 , Ω, U )  , U )
R* (t, Ω ∫A φ (16)
T
F (t )dA
d R(Ω, Ω φiT Mφ j = δ iijj ,

equation (13) reads where δij denotes the Kronecker delta. Substitut-
ing u = Φr and multiplying by ΦT, the equation of
Mr + C ( )r + K (  , Ω))r = R (t, Ω, Ω
* *  , U ).
* (17) motion (1) becomes the modal equation of motion
Note that the projected matrices can be com-
ΦT MΦr + ΦT KΦr
Φ = ΦT F ,
puted efficiently before the time integration, i.e. no
matrix vector products with the full system matri-
ces are required during time integration. with the modal mass matrix
Solution of equation (17) requires a nonlinear
M = ΦTMΦ = diag (1,1)
solver, because of the constraints ri = 0, i = 1, …, 6
and the fact that Ü, Ω, and Ω  are unknown for
and the modal stiffness matrix
the specific timestep. The solution procedure is
described in more detail in section 2.4.
K = ΦT KΦ = diag ( 1,
2
2 ).
2

2.1 Example: Two-mass oscillator Considering the inertia force due to rigid body
As a very simple example, consider two nodes, each acceleration Ü, i.e. x = u + (1,1)TU, where u denotes
of mass m, connected by a spring of stiffness k, the nodal degrees of freedom of the finite element
resulting in a two degree-of-freedom (each node’s model and U denotes the rigid body translation,
translation in the direction of the spring) system the nodal force vector becomes
represented by the mass matrix
F  F − M( , )T U.
⎛m 0⎞
=⎜ With r = (r1, r2)T etc., the final modal equations
⎝ 0 m⎟⎠ of motion in the body fixed reference frame are:
and the stiffness matrix 2m r1 1 2 2 mU,
− 2mU (18)

⎛ k −k⎞ 2k 1 (19)
K=⎜ . 
r2 + r2 = ( F2 F1 ).
⎝ −k k ⎟⎠ m 2m

11

MARSTRUCT.indb 11 2/18/2011 5:39:25 PM

 Because in the body fixed reference frame the ⎛A ⎞
modal coefficient r1 referring to the rigid body ⎜ elas − rbm ⎟ r 7..n
⎜ A ⎟ t
motion is zero at all times, according to eq. (18), ⎝ elas ⎠
the rigid body acceleration is ⎛R ⎞ ⎛M ⎞ ⎛C ⎞
elas − rbm ⎟ 7..n l − rbm
b ⎟ 7..n
= ⎜⎜
rb
bm ⎟
+ ⎜⎜  Δt
⎟ r t −Δ
+ ⎜⎜
eelas
 .
⎝ Relas ⎠
⎟
⎝ Mela l s ⎠ ⎝ Ceelas ⎟⎠ r t −ΔΔt
F F2 (25)
U = 1 ,
2m
Here matrices with subscript “elas−rbm”
i.e. rigid body motion is driven by the sum of the describe the influence of the elastic deformation
nodal forces. On the other hand, according to on the rigid body motion. Taking into account that
eq. (19), the modal coefficient r2 referring to elastic the mass matrix term Melas–rbm is equal to zero and
deformation is subject to a forced oscillation driven Melas is the identity matrix due to the normalized
by the difference of the nodal forces. eigenmode vectors and time independence, inte-
grating (13) with Newmark’s method is equivalent
to solve
2.2 Time domain integration
7..n
Newmark’s integration scheme (implicit and of Aelas − rbm
m rt = Rrbm Celas − rbm r 7t ..nΔt (26)
second order accuracy by canonical choice of
parameters) is applied to (17). Introducing the
and
abbreviations
7..n
Aelas rt7..n = Relas e asrt Δt .
rt 7..Δnt + Celas (27)
(  ) a1C ∗ ( ∗  ), (20)
t t 0M t) K ( t t

rt Δt = a1rt Δt + a4 rt Δt + a5 rt Δt , (21) In order to solve these equations the kinetic vari-
ables Ω, Ω and U have to be known. Furthermore
 = a0rt + a2 rt the time-variant rotation matrix S transforming
rt Δt Δt Δt + a3 rt Δt (22)
local to global coordinates is required due to the
fact that the hydrodynamic forces are determined
leads to the temporal discretization of the equation in the global reference frame; hence there is the
of motion need to transform into the body fixed coordinate
system. Determining the kinetic variables is a non-
A(  )r = R∗ (  U ) linear problem which can be solved by exploit-
t t t t t t t
∗
+ M 
rt 
t + C ( t )r t Δt , (23) ing equation (26). Here we take into account that
the rigid body motion is affected by the elastic
deformation represented by the coupling terms
where a0, …, a5 are the Newmark parameters that denoted by subscript “elas−rbm” in equation (26).
depend on the time step size Δt. According to this, the solution procedure will be
As a result taking the rigid body modes nested.
into account equation (23) can be subdivided
according to rigid body motion and elastic
deformations 2.3 Determining the rotation matrix
A popular method to determine the rotation matrix
⎛ A Aelas − rbm ⎞ in body dynamics is the successive execution of
⎜ rbm ⎟r
⎜ AT Aelas ⎟⎠ t three elementary rotations where the rotation
⎝ elas − rbm
⎛R ⎞ ⎛ M matrix is described by three parameters. The most
Meelas − rbm ⎞
= ⎜⎜
rb
bm ⎟
+ ⎜⎜ T
rrbm ⎟  popular parametrizations are the Euler-Angles

R
⎝ elas ⎠
⎟
⎝ M elas − rbm Meelas ⎟⎠ r t Δt and the Cardan-Angles. Here, a three parameter
⎛ C Ceelas − rbm ⎞ approach is applied with elementary rotations
+ ⎜⎜ T
rbm ⎟ 
 . (24)
⎝ Celas − rbm Celas ⎟⎠ r t Δt about z, y' and x" axes corresponding to yaw, pitch
and roll motion. The primes denote the axes of the
one time rotated system and the two time rotated
To enforce suppression of rigid body motion system, respectively. Summarizing the three rota-
in the local reference frame the modal coefficients tion angles in the vector of generalized coordinates
r1..6 corresponding to the rigid body modes have to α [ψ , θ , φ ]T the rotation matrix S reads
be 0. Dropping the first six columns of the matrices
reduces equation (24) to S(α) = Sz(α1)Sy'(α2)Sx"(α3). (28)

12

MARSTRUCT.indb 12 2/18/2011 5:39:31 PM

 The generalized coordinates and the angular 3. solve (33) to build the start residuum
 , Ω are related by
velocity and acceleration Ω 4. start newton iteration until convergence, i.e.
solve (33) for Ü, α
Ω = JR ( )α , (29) 5. integrate Ü in order to determine U and U
6. integrate α  ,
 in order to determine Ω a d Ω
 = J ( )  + K ( ,  )α ,
Ω (30) build new rotation matrix S
R R

where JR(α) and K R ( ) denote the Jacobian

matrix of rotation and the derivate of 3 FLUID PART
∂( R ) ∂α and α , α  are the first and second time
derivates of the generalized coordinate vector α. The open source Computational Fluid Dynam-
Note that Ω and Ω  are stated in global coordinates ics (CFD) solver OpenFOAM, which is based
at this point. Integrating α  in time with the trap- on the Finite Volume Method, is adopted for the
ezoidal rule leads to the velocity of the generalized fluid part. OpenFOAM (www.openfoam.com) fea-
coordinate vector tures the computation of the free surface between
two media (here, air and water) according to the
Δt Volume Of Fluid (VOF) method. This method is
α t = α t −Δt + t
(α Δt + α
t ). (31) based on the solution of a transport equation for
2
the volume fraction of one medium, where the con-
Finally integrating the velocity vector leads to vective term is discretized in a manner that on the
the generalized coordinate vector one hand maintains a sharp interface between vol-
ume fraction one and zero and on the other hand
guarantees that the volume fraction is bounded
Δt
αt = αt −Δt + (α t t + α t ). (32) between one and zero. The remaining transport
2 equations (mass conservation, momentum conser-
vation, etc.) are solved for an effective fluid whose
properties (density, viscosity) result from an vol-
2.4 Solution procedure ume fraction weighted average of the properties of
The solution of (13) in a body fixed reference the individual media.
frame leads to the time stepping scheme given by These transport equations are solved in a
(26) and (27). When Ω and Ω  are substituted by α sequentiell manner, i.e. each transport equation
according to (29) and (30) and moreover (31) and is solved for the variable which is governed by the
(32) are used in (26) and (27), the result of (26) is particular equation while the other variables are
a nonlinear system for the 6 unknown kinetic vari- kept unchanged. These sequence is repeated in
ables Ü and α each time step as outer iterations (in contrast to
the inner iterations denoting the iterative solution
7..n of sparse linear equation systems that result from
F (U t t) =Rrbm (U t t ) + Celas rbm ( t )r t Δt spatial discretization of the transport equation)
7..n
− Aela
l s rbm ( t )rt until convergence is achieved.
= 0. (33) Coupling between fluid flow and structure
motion/deformation can be explicit (once per
To solve this nonlinear system of equations time step) or implicit (once per outer iteration).
Newton’s method is applied in a classical way where Implicit coupling is achieved by simply solving for
the Jacobian matrix is built up in every iteration the modal co-efficients and adjusting the bound-
step. Damping (halve the correction) is applied if ary vertices of the CFD grid in each outer iteration
the minimizing functional doesn’t decrease. Due to until convergence is achieved even considering the
the use of the iterative solver the coupling terms can boundary movement.
be considered easily by embracing them when vary- In order to generate waves, velocities at the cor-
ing the variables to establish the Jacobian matrix. responding inlet boundaries and the elevation of the
This means that (27) can be treated as a common free surface is given according to Stokes wave the-
linear system once the kinetic variables Ü and α  ory of 2nd order. Whereas a hydrostatic pressure is
are known. Taking the last solution as initial guess, assumed at the outlet boundaries. Wave reflections
quadratic convergence can be obtained. The nested at the boundaries are avoided by introducing damp-
solution procedure can be summarized as follows: ing zones near to the corresponding boundaries. The
hydrodynamic forces acting at the interface between
1. set start values: Ut0 Utn−Δt, α
t0 = α
tn−Δt fluid and structure are obtained by integrating pres-
2. solve (27) for rt0,7..n sure and shear stress over the common surface.

13

MARSTRUCT.indb 13 2/18/2011 5:39:35 PM

 Technically, OpenFOAM provides an extensive
class library, that is used to compose (in a very
high-level way) a main program that is tailored for
a specific flow situation and has to be compiled
and linked against the library in order to build the
solver executable. This software architecture allows
for easy embedding of the structure solver that has
been implemented as an in-house library.
Prior to the current implementation with Open-
FOAM, the now discontinued commercial CFD
system COMET (www.cd-adapco.com), which is
in methology and features similar to OpenFOAM,
has been used for a prototype implementation of
the presented methodology. The numerical exam-
ple in section 5 was computed with the COMET
implementation.
4 COUPLING OF FLUID
AND STRUCTURE SOLVER
The coupling of fluid and structure solver imposes
information exchange between the FVM and
FEM at the interface. To maintain kinematic and
dynamic consistency at the interface the Dirichlet
υ( ξ (t, ξ ))) u( ξ ) ξ Γ 0 ,t ∈[ 0 ] (34)
and Neumann
σ f (t )n σ S (t y )n y Γ t ∈[ 0 T ] (35)
boundary conditions have to be fulfilled. Gener-
ally the Dirichlet boundary condition (34) will be
gained by determining the structural displacements
which will be derived and adopted in the CFD sim-
ulation. Whereas integrating the left hand side of
the Neumann boundary conditions (35) over the
fluid interface results in the hydrodynamic forces
acting on the structure. The organization of the
information exchange at the boundary influences
the accuracy of the results and affects the stabil-
ity of the coupled simulation as well. Keeping in
mind the non-matching grids at the interface and
the need to interpolate variables from one domain
into the other these conditions can’t be fulfilled
exactly.
In the partioned fluid structure interaction
approach two major coupling strategies are
applied namely the weak and strong coupling.
They differ in the information exchange frequency
per time step which influences the fulfillment of
kinematic and dynamic consistency at the inter-
face (34) and (35). Weak (explict) coupling has
only one boundary condition exchange per time
step whereas strong (implicit) coupling has several
exchanges a time step depending on the number of
iteration steps per time step performed for solving
14
MARSTRUCT.indb 14
the Navier Stokes equations. In this work only
strong coupling with underrelaxation is applied
due to the permanent failure in the kinematic
consistency made when using weak coupling. In
order to exchange information at the interface the
circumstances of non-matching grids have to be
considered. Strategies to map variables between the
domains are described in the following subsection,
see (Eisen and Cabos 2007) for more detail.
4.1 Mapping of hydrodynamic forces
and structural displacements
The CFD solver yields a pressure and velocity
field at the interface on the cell faces. Integrating
pressures and stresses over these faces results in a
force field where the forces are composed accord-
ing to
fi = piai + fi,shear. (36)
Here, αi is the normal vector of the cell face
scaled by its area and pi is the corresponding
pressure. pi as well as αi are located at the face
center xi. The task is to determine FE nodes and
distribute the pressure forces onto those nodes
such that the sum of output forces equals the FE
input forces while conserving the pressure induced
moment acting on the FE model.
On the other hand fluid cells at the interface have
to be adapted according to the structure deforma-
tion. Due to the fact that the hydrodynamic forces
are mapped according to the face center, a direct
inversion of this mapping isn’t applicable to deter-
mine the face defining node displacements at the
interface. Furthermore a fluid mesh smoothing
strategy is needed to maintain grid quality and
prevent cell deterioration.
4.1.1 Mapping of hydrodynamic forces
Since typically the fluid mesh is finer than the FE
mesh, only single finite elements will receive forces
related to a particular force result point. The con-
dition to conserve hydrodynamic force fi at location
xi when mapping on nodes xi,n of an FE element
reads as follows
fi ∑ fi ,n (37)
n
and for the moment
∑ fi ,n xi ,n − xi ) = 0. (38)
n
The task is to find appropriate weights wi,n that
fi,n = wi,n fi (39)
2/18/2011 5:39:40 PM
 holds and equations (37) and (38) are fulfilled. These
weights are determined by the FE element shape
functions. All nodes of an FE element will receive
forces with the same direction parallel to fi.
4.1.2 Mapping of structural displacements
In order to adapt the interface according to the
structural deformation, displacements of FE
nodes have to be mapped onto corresponding fluid
cell nodes at the interface. For this the mapping
between hydrodynamic forces and structure loads
can be exploited. First determine the displace-
ments of the force result points located at the face
center by using the weights wi,n
upres ∑ wi ,nui ,n . (40)
i
Here n denotes the structure element which is
subject for the mapping of the force result point.
Every fluid node at the interface will be adjusted
according to the weighted adjoining pressure result
node displacements
ufluid = ∑ wj′ u j pres , (41)
j
where the factors w ′j are determined according to
the distance to the fluid node.
4.2 Fluid mesh smoothing
After the boundary vertices of the fluid grid have
been moved according to the structure motion/
deformation, the interior fluid grid has to be
smoothed in order to prevent cell deterioration
and to maintain grid quality. While this has been
a major task (both conceptually and computation-
ally) in the past involving e.g. Finite Element com-
putation of an auxiliary spring system representing
the fluid grid, OpenFOAM now provides a robust
and efficient grid smoothing algorithm.
4.3 The partitioned algorithm
Only strong (implicit) coupling is applied is the
presented method which leads to the loop in
Figure(1) when integrating the structure solver
in the iterative solution procedure of the Navier
Stokes equation. This loop is cycled as many times
as outer iterations per time step are performed
for the fluid solution. If the maximum number of
outer iterations is reached it is advanced to the next
time step by setting new boundary conditions for
wave generation. To stabilize the whole solution
process in case of implicit (strong) coupling under
relaxation is applied.
15
MARSTRUCT.indb 15
Figure 1. Iterative loop.
5 NUMERICAL EXAMPLE
The aforementioned procedure has been applied
to a large container vessel (L = 294 m) at Froude
number υ / gL = 0.17 in both head (180°) and
oblique (140°) second order stokes waves of 335 m
length and 8.3 m amplitude. The following figures
show some results. In Fig. 2, the fluid-structure
boundary regions of the CFD grid and the finite
element model at the bow area of the ship are dis-
played. Obivously, the grids do not match, e.g. at the
bow, quadrilateral cell faces in the CFD grid cor-
respond to a somewhat coarser triangulation in the
FE model. For this reason, the mapping routines
described in section 4 have been applied in order to
transfer forces and deformations between the CFD
and FE representations of the vessel. In Fig. 3, the
elastic modes that are used to describe the linear
elastic deformation of the vessel are displayed.
These are the lowest elastic eigenmodes that cor-
respond to torsion (mode 7), horizontal bending
(8), vertical bending (9), and a higher hull girder
vibration mode (10). The displayed graphs indicate
that for wave lengths not smaller than the vessel’s
length, these modes suffice to capture the glo-
bal deformation in oblique seas, reducing the
number of degrees for the computation of the
elastic deformation from several thousands for
the full finite element equations to only four for
the modal approach. To verify this, a convergence
analysis comparing these results with an approxi-
mation using a higher number of eigenmodes,
could be performed. In Fig. 4, the vessel’s rigid
body motion due to head waves is displayed both
with and without the effect of structural elastic-
ity. It is demonstrated, that (for head waves) the
influence of elasticity on rigid body motion is very
small. In Fig. 5, temporal variation of the elastic
modal coefficients and, for reference, the pitch
angle is displayed for head waves. It can be seen
2/18/2011 5:39:42 PM
 Figure 2. Comparison of CFD boundary regions (left)
and finite element model (right) at the bow area of the
vessel. As resolution differs between grids, sophisticated
mapping for forces and displacements has to be applied.
Figure 3. Dry modes considered in simulation. The
lowest elastic eigenmodes refer to torsion, horizontal
bending, vertical bending, and finally a higher hull girder
vibration mode.
Figure 4. Comparison of rigid body motion in head
waves with and without consideration of elastic defor-
mation. Rigid body motion is hardly affected by elastic
deformation.
that vertical bending is the only mode that is sig-
nificantly excited. This is the expected behavior for
head waves. Also, this modal coefficient closely fol-
lows the pitch angle as the driving mechanism for
vertical bending. On the other hand, for oblique
waves, other modes become significant. In Fig. 6,
the modal coefficients for the elastic modes are
16
MARSTRUCT.indb 16
Figure 5. Pitch angle (scaled) b and elastic modal coef-
ficients r7, …, r10 in head waves. Vertical bending (mode 9)
is clearly dominant.
Figure 6. Elastic modal coefficients r7, …, r10, velocities,
and accelerations in oblique waves. Besides vertical bend-
ing (mode 9), torsion (mode 7) and horizontal bending
(mode 8) are significantly excited, whereas higher hull
girder vibration (mode 10) is not significant.
displayed for oblique sea. Here, torsion and and
horizonzal bending are clearly present, whereas
higher hull girder vibration remains negligible.
6 CONCLUSIONS
A method for the solution of the fluid structure
interaction problem in ship design has been pre-
sented. It is characterized by the approximation of
the structural deformation by a number of modes.
This serves three purposes,
• the significant reduction of the number of
degrees of freedom of the distretized structural
equation system,
• the suppression of high frequency contents of
the structural deformation which might lead to
overly small limits on the time step size, and
2/18/2011 5:39:44 PM
 • the possibility to precompute fluid mesh
deformations for each mode and therefore to
avoid solving artificial elastic problems for the
fluid mesh in each time step.
As a result of this modal approach, the equation
of motion for the ship must be solved in the moving
coordinate system. This leads to full system matri-
ces which depend on the ship translational accel-
eration and rotational velocity and acceleration.
The additional computational effort for these small
nonlinear sets of equations is nevertheless small,
since only few modes are applied. The rigid body
motion is also a result of this set of equations.
Compared to earlier approaches relying on a
beam approximation of the ship, torsional elastic
deformations can be considered. This is important
especially for large container vessels due to their
comparatively low torsional stiffness.
To improve the proposed method, future work
should focus on practical procedures for choosing
those modes, which are sufficient to describe the
most important effects of the structural deforma-
tion. In particular, the choice of global modes has
the natural limit of neglecting stresses due to local
effects. For this purpose, local deformation modes
of the particular structural elements could be
added to the set of global analysis modes.
17
MARSTRUCT.indb 17
REFERENCES
Argyris, J. & Mlejnek, H.-P. (1988). Die Methode der
Finiten Elemente, Band III - Einführung in die Dynamik.
Braunschweig: Friedr. Vieweg & Sohn.
Asmussen, I., Menzel, W. & Mumm, H. (2001). GL
Technology, Ship Vibration. Technical report, Ger-
manischer Lloyd.
Eisen, H. & Cabos, C. (2007). Efficient generation of cfd-
based loads for the fem-analysis of ship structures. In
International Conference on Computer Applications in
Shipbuilding, Portsmouth.
Matthies, H.G., Niekamp, R. & Steindorf, J. (2006).
Algorithms for strong coupling procedures. Computer
Methods in Applied Mechanics and Engineering
195(17–18), 2028–2049.
Oberhagemann, J., el Moctar, O. & Schellin, T. (2008).
Fluid-structure coupling to assess whipping effects on
global loads of a large containership. In 27th Sympo-
sium on Naval Hydrodynamics, Seoul.
2/18/2011 5:39:50 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Sensors location and data processing algorithms of an optical fibers

hull strength monitoring system

A. Grasso & A. Vergine

RINA Services S.p.A., Genova, Italy

D. Dimou, M. Samuelides & N. Tsouvalis

National Technical University of Athens, Athens, Greece

A. Ferrari
D’Appolonia S.p.A., Genova, Italy

ABSTRACT: The paper presents the design of two prototypes of a hull monitoring system which
employs temperature compensated laser based optical sensors. The prototypes have been developed for
a double side bulk carrier and an ice class tug boat. A set of finite element calculations has been carried
out for defining the optimum sensor’s installation areas in each ship. Preliminary tests on the algorithms
developed for the software onboard and a first analysis of a set of data measured on the tug boat are
reported.

1 INTRODUCTION The algorithms developed for the onboard

The paper presents the design of two prototypes of
a hull monitoring system that have been developed
within the MOSES project “Innovative Continuum
Multiplex Optical Sensors hull stress monitoring
system, supporting shipping safety and Enhanc-
ing the control capability over structural Ship
integrity”.
Two vessels, within the fleets of owners that
are partners of the project (Premuda and PKL
AS), have been selected for testing the prototypes:
a double side bulk carrier and an ice class tug boat.
Considering the characteristics of each vessel,
it has been decided to design the systems for moni-
toring primarily:
− Hull girder bending moments along the ship
length and local stresses in selected hot spots for
the bulk carrier;
− Stresses in the bow area due to ice pressure loads
and stresses on the forward winch deck support
structure for the tug.
A set of finite element calculations has been car-
ried out for defining the optimum sensor’s installa-
tion areas in each ship and results are presented.
Stress distributions in these regions have been ana-
lyzed in order to identify the best locations for the
installation of strain measuring devices, as well as
their proper gage length.
structural health assessment are described. Par-
ticularly, on the bulk carrier the software cor-
relates measured strains with hull girder bending
moments, while on the tug it estimates maximum
stress levels on selected areas of the hull structure
when the tug operates as an icebreaker. FE analy-
ses have been used for deriving the correlation
between the measured strains and the maximum
stress for the tug and for a preliminary test of the
algorithm employed on the bulk carrier.
The prototypes have been installed on the ves-
sels and preliminary results related to the tug boat
are presented.
1.1 The MOSES project
More than 4500 total ship losses were recorded
worldwide in the period from 1994 to 2002: a sig-
nificant percentage of such incidents are caused by
structural problems. Hull stresses due to loading,
or stresses imparted by wave and adverse weather
conditions constitute primary source of risk to
all types of ships with a large hull. Bulk carriers,
large oil tankers (Eliopoulou & Papanikolaou
2007, Papanikolaou et al. 2007), container ships,
LNG carriers and RoRo are vessel types particu-
larly subject to such type of risks. Due to concerns
about safety at sea, International Maritime Organi-
zation (IMO) has issued in 1994 a recommendation

19

MARSTRUCT.indb 19 2/18/2011 5:39:50 PM

 on bulk carriers more than 20,000 dwt to be fitted
with a hull stress monitoring system (IMO 1994).
The concept of MOSES project—Innovative
continuum Multiplex Optical SEnsors hull Stress
monitoring system—is to apply highly knowledge-
based methods to achieve control of loads in the
whole extension of the ship hull, using tempera-
ture compensated laser based optical sensors.
The employment of optical-fibre based archi-
tectures in hull stress monitoring systems was still
considered beyond commercial state of the art for
conventional metal ships in 1997 (Slaughter et al.
1997). The first systems adopting this technology
have been installed in Europe at the beginning of
this decade (Sagvolden et al. 2002) and nowadays
their commercial employment is growing. The
advantages of this technology rely mainly on:
− high mechanical, chemical and temperature resi-
stance of optic sensors (ideal for ship structures
and long term permanent structural health
monitoring);
− immunity of the system to electromagnetic and
radio frequency interference;
− intrinsic safety in hazardous areas, as sensors do
not employ electrical components;
− accuracy and quality of the signal;
− capability of being employed on long distances.
Sensors had to meet the stringent requirements
of the working environment, in terms of rugged-
ness, reliability, response accuracy, resistance to
electromagnetic interference and multiplex connec-
tion capability. Their development, not described
in the present paper, has been based on laser signal
transmission with optical fibres, exploiting Fibre
Bragg Grating (FBG) or Fabry-Perot interferom-
eter diffraction effect, a technology already applied
in other fields (Majumder et al. 2008), like civil
engineering (Li et al. 2004, Maaskant et al. 1997)
and composite materials structural monitoring
(Silva-Muñoz & Lopez-Anido 2009, de Oliveira
et al. 2008).
The project objectives involved sensor develop-
ment, FEM structural calculations and data con-
ditioning, to grant the applicability to the widest
types of ships. The new system, developed for an
operational employment, aims at providing more
accurate and reliable data for assessing hull struc-
tural condition, identifying damages at early stage
through direct readings or through analysis which
correlates the reading to the most severely stressed
areas. The design should allow achieving this goal
minimizing the number of sensors installed.
In heavy weather, the information provided by
the system should help the ship Master taking deci-
sions regarding the optimum speed, heading and
ballast condition to avoid excessive wave-induced
bending or, eventually, local overloads. Similarly,
20
MARSTRUCT.indb 20
during cargo operations, the officer in charge
could have a complete understanding of the actual
values of still water bending moments, which could
be compared with the ones expected. Moreover,
fatigue cycles on selected structural details can
be monitored, gathering information about the
fatigue status of the vessel. Benefits are hence
associated with immediate increase in safety of the
ship, during loading as well as in shipping condi-
tions, reduction of casualties, extended service life
of vessel, and targeted maintenance directed to
overcome the damages detected by the monitoring
system.
2 MONITORING SYSTEM INSTALLED
ON THE BULK CARRIER
2.1 Expected features
Considering the characteristic of the vessel, it has
been decided to design the system for monitoring:
− Hull girder bending moments along the ship
length;
− Local stresses in selected hot spots.
Sensors for estimating hull girder shear forces
have been also installed in two sections.
This prototype of the system aims mainly at
monitoring and storing the global bending moments
and informing the ship’s Master if their values
exceed a predefined threshold. Actually, informa-
tion on shear forces and hot spot stresses are to be
stored in order to be analyzed ashore for estimating
the condition of the structure under surveillance.
The monitoring software has been also linked
to the loading instrument, in order to provide,
if requested during loading and unloading, an
instantaneous comparison between the expected
and measured values of vertical bending moment.
2.2 Preliminary finite element analysis
A preliminary analysis has been carried out consid-
ering a set of load cases derived from four mutu-
ally exclusive load combinations referred to three
loading conditions, which are:
− Full load (departure), with the holds fully filled;
− Full load (departure), with γCargo = 3 t/m3;
− Ballast (departure).
The combinations of loads are provided in
RINA Rules for structural analysis (RINA 2010)
and are generally employed for structural element
analyses which do not require complete ship mod-
els. These load cases take into account also wave
induced hull girder loads, inertial loads due to ship
accelerations and static pressures induced by an
2/18/2011 5:39:50 PM
 equivalent wave, supposing the ship in both upright
and inclined conditions.
Only three cargo holds have been modeled, with
the center one actually analyzed. The three dimen-
sional model is assumed to be fixed at its aft end,
while at its fore end rigid constraint conditions
have been applied to all nodes located on longitu-
dinal members, in such a way that the transverse
section remains plane after deformation. Hull
girder loads have been forced at a selected section
of the central hold, applying “adjustment loads”
to the fore section of the model. Particularly, the
total vertical bending moment has been forced, for
each loading condition, in the section experiencing
the maximum value of still water bending moment.
Figure 1 shows distributions of vertical bending
moments due to steel weight, local loads (which Figure 2. Model mesh.
include seawater pressure) and adjustment loads
for one of the cases analyzed. The plot also reports
the bottom result to be the most stressed parts.
the total bending moment distribution (both the
Concentration effects are found at the hatch cor-
one estimated before the calculation and the one
ners, which appear to be a very important zone to
actually obtained by FE analysis), which accom-
be monitored. On the bottom, due to combination
plish to the condition above mentioned, referred to
of global and local loads, the part corresponding
“target” in the figure.
to the center of the hold presents higher values of
The structure of the ship has been modeled
stresses, compared both to the values found in the
employing plate and beam elements. The principal
surrounding elements and also predicted using an
dimensions of the elements correspond transver-
hull girder section theory. Nevertheless, this zone
sally to the spacing of the longitudinal ordinary
appears to be less critical.
stiffeners and longitudinally to the ordinary frame
The following hot spots have been also found to
spacing. Altogether, the model contains 50041
be of interest for the monitoring system:
nodes, 33178 beam and 60544 plate elements.
Figure 2 shows the mesh of the model. − Connection of the inner bottom with the hopper
Linear static analyses have been performed tank sloping plate;
employing Leonardo Hull 3D, a pre and post − Connection of the hopper tank sloping plate
processor developed by RINA, and the solver with the inner side;
MSC-NASTRAN 2008. − Connection of the inner bottom with the trans-
As expected, the higher stresses appear to be verse bulkhead lower stool;
mainly related to global loads (vertical bend- − Connection of the transverse corrugated bulk-
ing moments) and hence the deck and, less, heads with the topside tank;
− Ends of longitudinal hatch coamings.

2.3 Location of sensors

4
x 106

Considering the aims of the system and on the

2
basis of global loads distributions, FE analyses
Vertical bending moment [kN m]

0
and known issues, a preliminary list of points of
interest for measuring strains was defined, which
–2 Target included sensors on deck and bottom for bending
moments, sensors on the side for shear stresses and
–4
Steel weight sensors at the hot spots reported in the previous
–6
Local loads
Adjustment loads
section. From this preliminary list, a subset of loca-
VBM predicted tions has been selected, with the aim of optimizing
–8
VBM calculated
the number of sensors considering the constraints
90 100 110 120 130 140
Longitudinal location [m]
150 160 170
related to junction box specifications. Moreover,
there was the need for developing a system actu-
Figure 1. Example of load distributions and compari- ally affordable in the market. The following layout
son of the predicted total bending moment with the result appeared to be a good balance of technical and
of FE analysis. economical requirements.

21

MARSTRUCT.indb 21 2/18/2011 5:39:50 PM

 Figure 3 shows the sensors locations for
measuring hull girder bending moments. Four
sensors, which evaluate longitudinal strains on the
deck and on the bottom, are provided in each of
the sections highlighted in Figure 4.
Figure 5 shows the sensors locations for meas- Figure 6. Location of the sections in which shear
uring hull girder shear stresses. Two rosettes of stresses are evaluated.
three sensors each, with relative orientation of 45°,
are provided in each of the sections highlighted in
Figure 6.
Figures 7 to 10 show the sensors placement
for measuring local stresses; at each point of
measure, two sensors are mounted for measuring
longitudinal and transversal/vertical strains.
Temperature sensors are also provided for com-
pensation.

Figure 7. Location of sensors for monitoring local

Figure 3. Location of sensors for evaluating hull girder stresses in cargo holds.
bending moments in the section.

Figure 8. Location of the holds monitored for local

stresses.
Figure 4. Location of the sections in which bending
moments are evaluated.

2.4 Analysis of stress distributions near

Figure 5. Location of sensors for evaluating hull girder
shear stresses.
the locations selected for measuring local
stresses
Detailed FE analyses have been performed for
evaluating stress distributions in the areas selected
for monitoring local stresses (Figures 7 to 10). This
kind of analysis can provide useful information
for identifying the best length of each sensor and
a guideline for their installation in order to meas-
ure picks of stress.
Fine meshes have been obtained refining the
coarse one near each hot spot and the solutions
of the previous calculations have been forced, in
terms of displacements, as boundary conditions
on the refined models.

22

MARSTRUCT.indb 22 2/18/2011 5:39:52 PM


Figure 9. Location of sensors for monitoring local
stresses at hatch corners.
Figure 10. Location of the hatches monitored for local
stresses.
As described in Section 2.2, linear static analyses
have been performed employing Leonardo Hull 3D
and MSC-NASTRAN 2008. All the structures,
including stiffeners, have been described employing
quadrilateral and triangular plate elements. Shells
around each location have been modeled by ele-
ments of about 2 cm of edge, dimension which has
been progressively increased moving away from the
hot spot zone. In the outer region, elements with
edges of about 20 cm have been used. Welding has
not been taken into account.
Figure 11 shows an example of a refined mesh.
Two loading conditions have been analyzed:
− Full load (departure);
− Ballast (departure).
The combinations of loads are the same emp-
loyed for the preliminary analysis (Section 2.2).
Results showed that gradients of stress may be
great in the locations analyzed, with the exception
of the hatch corners. Small strain gauges (of about
2 cm of length) appear hence to be the preferable
solution for most of these hot spots, even if sensors
up to 10 cm, if installed properly, can be generally
employed, providing equally useful information on
local stresses. For the hatch corners, stress distri-
butions seem to allow the employment of strain
gauges with a length between 10 and 20 cm.
23
MARSTRUCT.indb 23
Figure 11. Example of fine mesh—Connection of the
inner bottom with the transverse bulkhead lower stool.
The final decision on the dimensions of the sen-
sors was taken however considering also technical
issues related to the sensors characteristics and
their capability to be properly protected from the
environment. Actually, small strain gauges resulted
to be less suitable for an employment in harsh envi-
ronments and hence have been avoided.
2.5 Calibration of the system
A calibration and zeroing procedure is required
prior to the employment of the system. The evalu-
ation of the “zero stress”, necessary for referring
the measurements to their actual absolute values,
has to be done in calm water in a well known
(reference) loading condition, comparing the aver-
age strain measured by each sensor during a proper
time interval with expected values provided by
means of FE analyses. Ballast condition has been
preferred, in order to reduce uncertainties related
to the loading. Temperatures measured during this
procedure are also stored as reference values.
2.6 Description of the algorithm for the software
onboard
When the monitoring software, which evaluates
onboard vertical and horizontal bending moments,
is turned on, a “set-up” subroutine loads all the
relevant data and arranges the connection with the
output of the reading unit, in order to read strains
and temperatures at its proper sampling frequency.
After each reading, stresses are calculated and
referred to the zero value derived during the cali-
bration procedure. Strains due to temperature var-
iations are also estimated for each sensor and the
stresses are compensated consequently, in order to
discard the effects related to temperature gradients
in the evaluation of bending moments.
2/18/2011 5:39:56 PM
 Subsequently, the monitoring software calculates
vertical and horizontal bending moments (MV and
MH) applied to each cross-section as follows:

4
MV ∑ KV iσ i (1)
i =1
4
MH ∑ K H iσ i (2)
i =1

where σi represents the compensated stress meas-

ured by the i-th sensor and KV and KH are coef-
ficients depending on the characteristic of the
cross-section and on the position of the sensor
in the section. The two coefficients are derived in
order to average the results obtained from the sen-
sors on the deck and on the bottom.
Results are also provided to the crew in terms of
a ratio between the measured bending moment and
the maximum allowed for the section considered.
Every time the monitoring task highlights a critical
situation (i.e. actual values exceed the allowed limit
values) the monitoring software raises an alarm,
both visual and sonorous, that could be of different
importance (e.g. one if a threshold is approached
but not reached, the other if a threshold is reached
or exceeded).
The Monitoring Software is governed through a
Graphical User Interface that allows an operator to
interact with the workstation and is also connected
to the Loading Instrument, allowing the compari-
son of the measured bending moments with the
expected values during loading and unloading
operations.
2.7 Preliminary tests of the algorithm
for the software onboard
A preliminary set of tests for ascertaining the reli-
ability of the algorithm has been performed using
FE analyses. A transversal ring, including the rel-
evant section in hold 5 (Figure 4), has been refined
and analyzed as already reported in Section 2.4.
The structures have been described employing
plate and beam elements. The reference dimension
of the elements was about 18 cm. Figure 12 shows
the mesh employed for this calculation.
For each combination of loads analyzed, stresses
predicted by FEM have been used as input of
the algorithm, for estimating the related bending
moments. These results have been then compared
with the bending moments obtained by FE analyses.
Figures 13 and 14 show the comparison of vertical
and horizontal bending moments respectively. In
the plots FL indicates full load condition, while
BL the ballast condition; the other letter represents
the load case, as reported in RINA rules.
Figure 12. Fine mesh for testing the algorithm.
Results appear to be in good agreement, with
errors related to maximum values of bending
moments satisfactory for the application.
The load cases employed for these preliminary
tests represent of course standard conditions for
structural checks of the ship and not simulations
of realistic situations. Nevertheless they should
give a satisfactory idea of the reliability of the pro-
cedure in severe weather conditions.
The tests have also shown that the order of mag-
nitude of the errors appears to be generally con-
stant and hence the percentage increases for small
values of bending moments. The system could
hence be employed satisfactorily for detecting haz-
ardous situations due to extreme loads, but cannot
be used efficiently, for instance, for the evaluation
of fatigue accumulated due to global loads.
The effect on the prediction of bending moments
varying the longitudinal location of the sensors
between two frames has been also analyzed.
In order to check the stability of the algorithm,
further tests have been performed, introducing
systematically, one sensor per time, an error cor-
responding to 10% of the stress value. The results
have shown that the influence of this error in the
prediction of higher values of bending moments is
generally satisfactorily low (lower than 2% of the
maximum allowable vertical bending moment).
3 MONITORING SYSTEM INSTALLED
ON THE TUG BOAT
3.1 Expected features
The main objective of the Tug boat hull sensor
monitoring system is to provide the master and
crew with accurate information about the stress
levels on selected areas of the hull structure when

24

MARSTRUCT.indb 24 2/18/2011 5:39:57 PM

 6.0E+06
FEM increased stress levels due to ice loading for the
5.0E+06 Estimated
4.0E+06
positioning of strain measuring devices. For the
3.0E+06 forward winch loading, a shell model of the main
2.0E+06 deck between frames 36 and 55 was used, includ-
VBM [kN*m]

1.0E+06
ing all support and stiffening members of the deck.
0.0E+00

–1.0E+06
Again, the aim of the analysis was to identify areas
–2.0E+06
with increased load levels during winch operation
–3.0E+06 for sensor placement.
–4.0E+06 Figure 15 presents the bow section area of the
–5.0E+06
FL-A FL-A FL-B FL-C FL-D BL-A BL-A BL-B BL-C BL-D
vessel modeled for the ice-loading study. All hull
Crest Trough Crest
Load Case
Trough
bottom and side plating including their respective
transverse and longitudinal reinforcing members
Figure 13. Comparison of vertical bending moments. between frames 46 and 59 have been included in
the FE model, and the platform deck (at 3200 mm
from baseline) and bulkheads at frames 50 and 55
2.0E+06
FEM have been fully modeled as well (Figure 16). The
Estimated
1.5E+06
structures at the fore and aft ends (Fr. 46 & 59)
1.0E+06 and the main deck itself have not been modeled
5.0E+05 directly, but they were taken into account by using
HBM [kN*m]

0.0E+00 appropriate boundary conditions (Figure 17).

–5.0E+05
The ice loading and seawater hydrostatic pressure
have been applied as normal and hydrostatic pres-
–1.0E+06
sure respectively, on all hull plate surfaces that
–1.5E+06
are actually loaded in working conditions. Two
–2.0E+06
FL-A FL-A FL-B FL-C FL-D BL-A BL-A BL-B BL-C BL-D
ice loading scenarios have been investigated, each
Crest Trough Crest Trough
Load Case
with a different width zone of application, extend-
ing from the waterline to a depth of about 50 and
Figure 14. Comparison of horizontal bending moments. 90 cm respectively, while the pressure is uniform
with a nominal value of 0.3 MPa (Figures 18 and

the tug operates as an icebreaker. The ratio between

the worst actual and maximum permissible stress
in the hull shell structure is to be continuously dis-
played, while appropriate warning and alarm trig-
gers are implemented in the system.
Strains from each sensor and the worst stress
ratios are to be stored so that post-processing anal-
ysis may be performed at later times, if needed.
Sensors for monitoring deck structure winch
loading have also been installed in the winch sup-
port members.
Figure 15. Modeled area, ice loading.
3.2 Preliminary finite element analysis
A preliminary analysis has been carried out with
two separate studies, one to investigate the ice load
response of the hull shell and another for the for-
ward winch deck support structure. The ice-load
study used a three-dimensional shell model of the
bow area below the main deck between frames
46 and 59, which is the area mainly affected by
ice pressure loads when the vessel acts as an ice-
breaker. The decision to limit the modeled part
of the vessel was based on the fact that longitu-
dinal strength is more than adequate in this type
of ships, because of their short length and the
generally increased scantlings of their structures.
The aim of the analysis was to identify areas with Figure 16. Bottom and side plating elements.

25

MARSTRUCT.indb 25 2/18/2011 5:39:59 PM

 Figure 17. Bow section model (Internal tank walls
omitted for clarity).
Figure 18. Load scenario 1, ice load zone ∼50 cm deep,
port side view.
19 respectively). The plate structures have been
modeled with 5-DOF shell elements, while the
HP hull side stiffeners and floor flanges have been
modeled using 3-D beam 6-DOF elements.
In the case of winch loading study, all main deck
transverse and longitudinal reinforcing members
between frames 36 and 55 have been included in
the FE model (Figures 20 and 21). The hull sides
and the structure at the model’s fore and aft ends
(Fr. 36 & 55) have not been modeled directly, but
they were taken into account by using appropriate
boundary conditions. The winch load of 65 tons
has been applied on the winch base positions on the
main deck while the model was fixed on the longi-
tudinal and vertical axes at its aft end on frame 36.
Boundary conditions preventing both vertical dis-
placement and rotation around the longitudinal
Figure 19. Load scenario 2, ice load zone ∼90 cm deep,
axis were applied on all nodes along the hull sides, port side view.
port and starboard. Along frame 55, only the verti-
cal displacement was restrained. The structure has
been modeled with about 80000 5-DOF shell ele-
ments, whereas 3500 6-DOF beam elements were
used to model the bulbs of the HP stiffeners.
The analysis was performed with version 23.1
of the general purpose code Autodesk ALGOR,
using the linear elastic analysis sparse solver.
It is noted that the magnitude of the applied
ice loading, which equals 0.3 MPa, is a “nominal”
static load. This loading condition gives the stress
pattern expected rather than the absolute values of
stresses under ice conditions. Actual measurements
of ice loads vary significantly. In way of example
and in order to get a feeling of what a 0.3 MPa Figure 20. Main Deck Fr. 36–Fr. 55.
uniform pressure represents, it is mentioned that
peak values of dynamic ice pressures according
to Kujala et al. (2007) were measured between members, given the fact that the affected structural
0.63 MPa and 0.84 MPa. part is made of high strength steel with a yield
Bearing the above in mind it is observed that point greater than 315 MPa. The winch support
for a static pressure of 0.3 MPa the side hull stress structure stresses are also low, because of the
levels are low and not critical, because of the increased plating thickness and the strong reinforc-
increased plating thickness and the reinforcing ing members.

26

MARSTRUCT.indb 26 2/18/2011 5:40:01 PM

 system includes sensors placed only longitudi-
nally according to the reference system shown in
Figure 23. To overcome the lack of strains data
in the transverse direction which are required for
accurate stress calculations, a FE analysis has been
performed, the aim of which was to derive relations
and mathematical expressions, linking the existing

Figure 21. Deck reinforcing members.

3.3 Location of sensors

The system installed on the tug boat performs
constant evaluation of strains in selected panels
and stiffeners of the bow structure, at the area just
below the water-line. The decision on sensor place-
ment has been made based on the results of the
FE analysis: Figure 22 shows the planning for the Figure 22. Hull sensor installation, port side.
placement of sensors on the port side of the bow
between frames 51 and 53, while a second set of
sensors was similarly installed on starboard side.
In addition, sensors on the deck winch supporting
structure can provide information about the tow-
ing operational status of the vessel, as a function
of time.
Figure 23 shows the reference system for the
general case of sensors placement on panels:
aspect ratio a/b is to be taken with “a” on the
indicated transverse axis of each panel, and sen-
sors are to be installed in any of the two positions
shown as “middle” and “quarter”. Both are on the
b/2 transverse line, at 50% and 75% of the trans-
verse length “a”. Each sensor is to be positioned
along the indicated longitudinal axis (x) of the
respective panel.
Sensors placement on stiffeners is to follow the ref- Figure 23. Hull sensors installation on panels.
erence system shown in Figure 24. Sensors are to
be positioned on top of the stiffener flange and at
the middle of the stiffener span, along the stiffener
longitudinal axis.
Regarding the winch loading sensors, they are
placed on the flange and web of the main deck
support girders, under the winch base (Figure 25).
Temperature sensors are also provided for
compensation.

3.4 Stress and strain calculations on panels

In the loading case under study, which is lat-
eral pressure acting on the hull panels from the
ice, the developed stress state in the panels is
2-dimensional. Because of the limited number
of available channels/sensors for installation, the Figure 24. Hull sensors installation on stiffeners.

27

MARSTRUCT.indb 27 2/18/2011 5:40:03 PM

 Figure 25. Location of sensors for evaluating winch
loading.
Figure 26. Test panel example, aspect ratio 2,
b = 600 mm.
longitudinal strains to the missing transverse ones
and, additionally, linking the longitudinal stresses
at the sensor location to the maximum von Mises
stress developed elsewhere on the panel. All calcula-
tions were carried out for flat plates with fixed ends
and having various aspect ratios and thicknesses.
Three separate lateral pressure load conditions due
to ice were taken into account:
− The ice thickness is either equal to or greater
than the panel dimension α, thus covering the
full span of the side panel in the transverse
direction of the vessel.
− The ice thickness is approximately equal to 2/3
of the panel dimension α, thus partly loading
the side panel, and
− The ice thickness is approximately equal to 1/3
of the panel dimension α, again partly loading
the side panel. Figure 27. Strain ratio εyy/εxx, middle position.
Trend line for aspect ratio less than 2.5.
In all cases, ice loading was taken as uniformly
distributed. Taking in addition into account dif-
ferent locations of the partial ice lateral pressure,
similar vessels. The magnitude of panel length b
six load cases were defined in total. The proposed
is also based on typical frame spacing values of
algorithm performs all calculations for every one
similar vessels. The analysis performed was linear
of the above load conditions and finally selects the
static, with uniform pressure applying totally or
highest von Mises stress value to compare with the
partially on panels with all four sides fixed. The
permissible stress.
analysis was performed only for steel panels.
The FE analysis was performed in a series of
For each load case and panel geometry, the val-
flat rectangular test panels (Figure 26), with the
ues of x-strain (εxx), y-strain (εyy) and x-stress (σxx)
following parameters:
were calculated for both the “middle” and “quar-
− Aspect ratio a/b (b is the longitudinal side, paral- ter” positions, along with the maximum von Mises
lel to the sensor axis) ranging from 1 to 4. Actual stress σVM of the whole panel. The results were
values examined were 1, 1.25, 1.5, 1.75, 2, 2.5, then tabulated and the following strain and stress
3 and 4. ratios were plotted against the aspect ratio values
− Plate thickness: Three values have been taken for both “middle” and “quarter” positions:
into account: 12 mm, 15 mm and 18 mm.
− Panel length b: Three values have been taken − εyy / εxx
into account: 500 mm, 600 mm and 700 mm. − σxx / σVM
The panel thickness values have been selected The analysis showed that both ratios are unaf-
based on actual thickness measurements found in fected by the thickness value and the value of the

28

MARSTRUCT.indb 28 2/18/2011 5:40:05 PM


Figure 28. Strain ratio εyy / εxx, quarter position.
Figure 29. Stress ratio σxx / σVMmax, middle position.
Figure 30. Stress ratio σxx / σVMmax, quarter position.
panel length b (frame spacing). As an example,
the plots for the fully pressured panel load case
are presented in Figures 27 to 30. For each plot,
an exponential or polynomial trend line has been
fitted and this equation is used to define the strain
or stress ratio (measured of strain or stress over the
anticipated maximum value) for any given aspect
ratio. The same procedure was followed for all load
cases in order to define a full set of mathematical
expressions for the strain and stress ratios.
29
MARSTRUCT.indb 29
3.5 Description of the algorithm for the software
onboard
During the setup phase, each sensor is to be
uniquely identified and correlated to a panel/
stiffener position, and also to the side of the hull it
is placed on (port or starboard). When setting up
the system, the user has to specify geometric and
material details as the panel’s aspect ratio and the
sensor’s position (“middle” or “quarter”, for panel
sensors only), the material Modulus of Elasticity,
Poisson’s ratio and thermal expansion coefficient,
and also the maximum permissible stress values for
each panel/stiffener.
The system must be calibrated and zeroed in the
harbor condition; this is the zero strain and tem-
perature state for all calculations. The only exter-
nal load acting on the hull in harbor conditions is
hydrostatic pressure, and for the panels where the
sensors are installed, expected ice load pressure
will be 50–60 times greater than this hydrostatic
sea pressure. Strains due to hydrostatic pressure
are very low and thus can be safely ignored.
At the end of the setup phase all data is stored,
and in the sequence retrieved whenever the system
is turned on; the setup procedure will need to be
repeated only to modify the input data, for exam-
ple the permissible stress values. The system reads
the setup data and then starts the measurement
phase, reading and storing the measured strains
from each sensor. In the sequence, a temperature
compensation calculation is carried out for the
strains measurements. Having the final value of
strains, a series of calculations provides the respec-
tive stress at each position, σi, and, additionally,
the corresponding maximum expected stress in
each panel and stiffener.
In the sequence, the system passes into the dis-
play and evaluation of results phase. In this phase,
the system calculates the ratio between the calcu-
lated maximum stress and the permissible stress,
for each monitored panel and stiffener and the
worst value of this ratio is then displayed to the
crew and stored for post processing. A warning is
to be provided when the ratio is between 0.8 and 1.
Values greater than 1 shall trigger an alarm.
3.6 Verification tests of the algorithm
for the software onboard
For verification purposes, the finite element model
of the Tug bow described in section 3.2 was used
to compare the numerical results of the ice loading
analysis with the stress values calculated using
the formulas derived from the analysis described
in section 3.4. Figure 31 shows one typical test
panel between the platform deck and the horizon-
tal girder on the hull side. The aspect ratio for all
2/18/2011 5:40:07 PM

Figure 31. Tug bow FE model, typical test panel.
neighboring panels is about 2.5, and two of sec-
tion 3.4 analysis load cases were used to evaluate
the results.
The numerical strains calculated from the FE
model at the “middle” and “quarter” positions of
the panel shown in Figure 26 were used as input
in the corresponding mathematical expressions of
section 3.4, thus resulting in the analytical calcula-
tion of the maximum panel von Mises stress. This
value was then compared to the maximum von
Mises stress calculated by the FE analysis and the
difference was found to be between 2.5 and 6%.
This difference is very small and fully acceptable
for the purposes of the present study and may be
justified because of the curvature of the actual
bow panels and the hydrostatic sea pressure on the
hull sides, which were both not considered in the
flat rectangular panel analysis.
4 PRELIMINARY RESULTS
The system installed in the tug boat is currently
under test. At present, the ship operators have
provided for evaluation purposes only one set of
runs recorded during a week in the beginning of
February 2010, when the tug operated in 20 cm
thick floating ice.
Sample plots of strain data for some 24-hour
periods verified the presence of increased strain
levels for those time frames the Tug was actually
operational. Figure 32 shows a 3-hour strain plot
of two strain sensors (P1P is on a port side plate
and P1S is on the symmetric panel on starboard
side, see Figure 22) where the Tug was operational
between 09:35 am and 11:00 am. The scale between
two horizontal grid lines is 10 με, which illustrates
that the structure is not heavily stressed and the
30
MARSTRUCT.indb 30
Figure 32. Sample strain plot, plate sensors port &
starboard.
stress level is low under the specific ice breaking
operational conditions.
Data measured on the bulk carrier are not yet
available.
5 CONCLUSIONS
The paper describes a hull stress monitoring sys-
tem which employs temperature compensated
laser based optical sensors, developed within the
MOSES project. The design of two prototypes are
described, one installed on a double side bulk car-
rier and the other on an ice class tug boat. The eval-
uation of hull girder bending moments along the
ship length and local stresses in selected hot spots
has been selected as main features for the system
on the bulk carrier. The measurement of stresses in
the bow area due to ice pressure loads and stresses
on the forward winch deck support structure have
been considered instead more important for the
tug boat.
The optimum sensor’s installation areas have
been selected by means of FE analyses, taking into
account also constraints related to junction box
specifications and looking for a good balance of
technical and economical requirements.
The paper describes also the algorithms for
the software provided onboard. Particularly, on
the bulk carrier the software correlates measured
strains with hull girder bending moments, while
on the tug it estimates maximum stress levels on
selected areas of the hull structure when the tug
operates as an icebreaker.
Preliminary tests on the software, performed
employing FE analyses, show that the two proto-
types should be capable of providing results with
accuracy fully acceptable for the purposes of each
system.
At the present time, no measured data are avail-
able for the bulk carrier, while initial readings from
the tug boat operating in the gulf of Tallinn have
been obtained and currently investigated. The ini-
tial readings indicate that the structure of the tug
does not suffer high stresses under normal ice
breaking operating conditions. In any case, further
analyses for both the vessels are expected in the
future on a wider set of data.
2/18/2011 5:40:08 PM
 ACKNOWLEDGMENTS
The research work for this paper has been partly
funded by the European Commission under the
Framework Programme 7, Grant Agreement 22083.
The partners of the consortium are D’Appolonia
S.p.A., Italy; Smartec BV, Switzerland; Pegaso
Systems S.r.l., Italy; PKL AS, Estonia; HENER
Henryka Weber, Poland; Premuda, Italy; RINA
Services S.p.A., Italy; National Technical University
of Athens, Greece.
REFERENCES
De Oliveira, R., Ramos, C.A. & Marques, A.T. 2008.
Health monitoring of composite structures by embed-
ded FBG and interferometric Fabry-Pérot sensors.
Computers & Structures, Volume 86, Issues 3–5,
pp. 340–346.
Eliopoulou, E. & Papanikolaou, A. 2007. Casualty Anal-
ysis of Large Tankers. Journal of Marine Science and
Technology 12: 240–250.
International Maritime Organization (IMO). 1994.
Recommendations for the Fitting of Hull Stress Mon-
itoring Systems.
Kujala et al. 2007. Maximum ice-induced loads on
ships in the Baltic Sea. Proceedings of PRADS 2007,
pp. 1278–1286.
Li, H.N., Li, D.S. & Song, G.B. 2004. Recent applica-
tions of fiber optic sensors to health monitoring in
civil engineering. Engineering Structures, Volume 26,
Issue 11, pp. 1647–1657.
31
MARSTRUCT.indb 31
Maaskant, R., Alavie, T., Measures, R.M., Tadros, G.,
Rizkalla, S.H. & Guha-Thakurta, A. 1997. Fiber-
optic Bragg grating sensors for bridge monitoring.
Cement and Concrete Composites, Volume 19, Issue 1,
pp. 21–33.
Majumder, M., Gangopadhyay, T.K., Chakraborty, A.K.,
Dasgupta, K. & Bhattacharya, D.K. 2008. Fibre Bragg
gratings in structural health monitoring—Present sta-
tus and applications. Elsevier Sensors and Actuators A:
Physical, Volume 147, Issue 1, pp. 150–164.
Papanikolaou, A., Eliopoulou, E., Alissafaki, A.,
Mikelis, N., Aksu, S. & Delautre, S. 2007. Casualty
Analysis of AFRAMAX Tankers. Journal of Engi-
neering for the Maritime Environment, Proceedings
of the Institution of Mechanical Engineers, Part M,
Volume 221, pp. 47–66.
RINA S.p.A. 2010. Rules for the Classification of Ships,
Part B, Vol. 2, pp. 33–35.
Sagvolden, G., Pran, K., Vines, L., Torkildsen, H.E. &
Wang, G. 2002. Fiber Optic System for Ship Hull
Monitoring. 15th IEEE Optical Fiber Sensors Confer-
ence, 1, pp. 435–438.
Silva-Muñoz, R.A. & Lopez-Anido, R.A. 2009. Struc-
tural health monitoring of marine composite struc-
tural joints using embedded fiber Bragg grating strain
sensors. Composite Structures, Volume 89, Issue 2,
pp. 224–234.
Slaughter, S.B., Cheung, M.C., Sucharski, D. &
Cowper, B. 1997. State of the art in hull response
monitoring systems. Report SSC-401, Ship Structure
Committee.
2/18/2011 5:40:09 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

A hydroelastic investigation into the dynamic response

characteristics of bulk carriers

L. Kaydıhan
Delta Marine, Istanbul, Turkey

B. Uğurlu & A. Ergin

Faculty of Naval Architecture and Ocean Engineering, Istanbul Technical University, Maslak,
Istanbul, Turkey

ABSTRACT: This paper presents a hydroelastic investigation into the dynamic response characteristics
of a group of bulkers with different load carrying capacities, e.g., two handysize vessels with carry-
ing capacities of 20 000 and 32 000 dwt, respectively; one handymax vessel with a carrying capacity of
53 000 dwt; one panamax vessel with a carrying capacity of 76 000 dwt; two capesize type vessels with
carrying capacities of 140 000 and 180 000 dwt, respectively. For all the bulk carriers adopted in this study,
detailed three dimensional finite element structural models are prepared, separately, by using commercial
finite element software. The calculations are carried out for two different loading conditions, namely, fully
loaded and ballast conditions. The dry and wet frequencies are computed by using the finite element, and
they are compared with those calculated by using a higher-order 3-D hydroelasticity theory.

1 INTRODUCTION forces in conjunction with 2D beam idealization of

Bulk carriers, container vessels and tankers are the
three largest groups of vessels within the merchant
fleet. However, bulk carriers comprise approxi-
mately 40% of the world merchant fleet. In 2006,
seven bulk carriers over 10 000 dwt were identified
as having suffered total loss together with the loss
of thirty seven lives, according to a report sub-
mitted to the maritime safety committee of IMO
by INTERCARGO (2007). One of those losses
was directly attributable to structural failure with
twenty six seafarers reported as lost from that one
incident. Once again the bulk carrier industry was
concerned by the heavy loss of life associated with
the total loss of a ship due to a catastrophic struc-
tural failure. Therefore, considerable effort has
been made to understand the wave induced struc-
tural response behavior of bulk carriers.
Bishop et al. (1985, 1991) used a two dimensional
hydroelasticity theory to investigate structural fail-
ures experienced by bulk carriers such as the Onom-
ichi Maru and OBO MV Derbyshire, respectively.
In their studies, they investigated the steady-state
and transient response behaviors of the ship struc-
tures. Hirdaris et al. (2003, 2006) applied 2D and 3D
hydroelasticity theories to predict and compare the
dynamic behavior of a bulk carrier in waves based
on OBO MV Derbyshire. They employed strip the-
ory for calculating 2D fluid-structure interaction
the ship structure, and also 3D boundary element
method with a pulsating source distribution over the
mean wetted surface for 3D fluid-structure interac-
tion effects. They reported differences between 2D
and 3D fluid-structure interaction models, for the
anti-symmetric case, and mostly on the evaluated
wave loads. These differences are attributed to the
inability of the employed Timoshenko beam the-
ory to model the non-prismatic features of the bulk
carrier and open-deck structures realistically. Very
recently, Tian et al. (2009) used a three dimensional
hydroelasticity theory in the analysis of a large bulk
carrier of 180 000 dwt traveling at its design speed in
regular and irregular head waves. The analysis was
repeated without the forward speed effect, and it
was observed that the forward speed effect had cer-
tain influence on the springing induced bending
moment.
This paper presents a hydroelastic investigation
into the dynamic response characteristics of a group
of bulkers with different load carrying capacities,
e.g., two handysize vessels with carrying capacities
of 20 000 and 32 000 dwt, respectively; one hand-
ymax vessel with a carrying capacity of 53 000
dwt; one panamax vessel of 76 000 dwt carrying
capacity; two capesize type vessels with capacities
of 140 000 and 180 000 dwt, respectively. For all
the bulk carriers adopted in this study, detailed
three dimensional finite element structural models

33

MARSTRUCT.indb 33 2/18/2011 5:40:09 PM

 are prepared, separately, by using the commercial
finite element software, i.e., Abaqus (2008).
The three dimensional structural models consist
entirely of the shell finite elements, representing
all major external and internal structural compo-
nents. The surrounding fluid domain is discretized
by using three dimensional fluid finite elements
available in the commercial software. Both sym-
metric and antisymmetric response characteristics
are obtained in terms of the dry and wet frequen-
cies with the associated mode shapes. The calcu-
lated dry and wet frequency values are presented as
a function of characteristic parameter of the bulk
carriers, in order to reflect overall response behav-
iors of the ship structure, for fully loaded and bal-
last conditions.
In a further study, a higher-order 3-D hydroe-
lasticity method was employed for calculating the
fluid-structure interaction effect in terms of the
generalized added mass and hydrodynamic damp-
ing coefficients. The wet resonance frequency val-
ues were calculated and they are compared with
those obtained from the finite element analysis for
the bulk carrier of 140 000 dwt carrying capacity.
The response behaviors in terms of principal coor-
dinates are also presented for the 32 000 dwt bulk
carrier excited by regular head waves. For the wet
resonance frequency calculations, an infinite fre-
quency limit condition is assumed on the free sur-
face. It is to say that the fluid-structure interaction
forces are associated with the inertia effect of the
fluid, independent of the frequency of vibration,
and that the hydrodynamic damping is zero.
2 MATHEMATICAL MODEL
2.1 Generalized equation of motion
The equation of motion describing the response of
a 3D discretized (finite element) structure to exter-
nal excitation may be written as Ergin (1997)
. form
MÜ + CVU + KU = P (1)
where M, CV, K denote the mass, structural damp-
ing .and stiffness matrices, respectively. The vectors
U, U and Ü represent the structural displacements,
velocities and accelerations, respectively, and the
column vector P denotes the external forces.
In an in vacuo analysis, the structure is assumed
to vibrate in the absence of any structural damp-
ing and external forces reducing equation (1) to the
form
MÜ + KU = 0. (2)
The trial solution is expressed in the form
of U = d eiωt and substituted in equation (2).
34
MARSTRUCT.indb 34
By canceling the common factor eiωt one obtains
the equation
(−ω2M + K)d = 0. (3)
This equation describes the simple harmonic
oscillations of the free undamped structure, and
the in vacuo normal modes, d, and natural frequen-
cies, ω, are obtained from this equation.
The distortions of the structure may be
expressed as the sum of the deflections in the
normal modes as
U = D p(t), (4)
where D is the modal matrix whose columns are
the in vacuo, undamped mode vectors, d, of the
structure and p is the principal coordinates vector.
By substituting equation (4) into equation (1) and
pre-multiplying by DT, the following generalized
equation in terms of the principal coordinates of
the structure is obtained:
 + bp(
 t) (5)
ap(t) cp((t ) Q( )
Here a, b, c denote the generalized mass, damp-
ing and stiffness matrices, respectively, and are
defined as
a = DTMD, b = DTCVD, c = DTKD, Q = DTP (6)
It should be noted that the generalized mass,
a, and stiffness, c, matrices are diagonal. The
generalized force matrix Q(t) represents the fluid
structure interaction and all other external forces
(e.g., wave forces, etc.), and may be expressed as
Q(t)) 
(Ap(t)  ) + Cp( ))
Bp( (t ), (7)
where A, B and C are the generalized added mass,
fluid damping, and fluid stiffness matrices, respec-
tively, and Ξ(t) denotes the generalized external
forces. Thus, equation (5) may be rewritten in the
( 
A)p(t) (b  t)
B)p( (c C)p( ) Ξ( ) (8)
2.2 Formulation of fluid problem
The fluid is assumed ideal, i.e., inviscid and incom-
pressible, and its motion is irrotational, so that the
fluid velocity vector associated with the unsteady
flow, v, can be defined as the gradient of a veloc-
ity potential function Φ as v(x, t) = ∇Φ, where
x = (x, y, z)T and t denote the position vector and
time, respectively. In general, Φ satisfies the Laplace
equation, ∇2Φ = 0, throughout the fluid domain,
an appropriate free surface boundary condition,
the kinematic boundary condition on the wetted
2/18/2011 5:40:10 PM
 surface of the structure, and a far-field radiation
condition. For time-harmonic flows without for-
ward speed, the linearized form of the free surface
boundary condition states g∂Φ/∂z + ∂2Φ/∂t2 = 0 on
the free surface, where g is the gravitational accel-
eration and z denotes the vertical axis.
For an elastic structure in contact with fluid
medium, the principal coordinates describing
the vibratory response of the structure may be
expressed as (Uğurlu & Ergin, 2006, 2008)
p(t) = p0eiωt,
where p0 represents the response amplitude vector.
The velocity potential function due to the vibra-
tion of the structure in the rth in vacuo vibrational
mode may also be written in terms of principal
coordinates as (Ergin & Temarel, 2002)
Φr ( , t) r( ) p0 r iω t
, r 1, 2,.., nm .
Here, nm represents the number of modes of
interest, and p0r is the amplitude for the rth prin-
cipal coordinate.
The kinematic boundary condition for the rth
modal vibration of a structure in contact with fluid
can be expressed as
∂Φ r ⎛ ∂u r ⎞ ⋅ n,
∂n = ⎝
∂t ⎠
where n is the unit normal vector on the wetted
surface pointing out of the fluid domain, and ur
denotes the displacement response of the struc-
ture in the rth principal coordinate that may be
written as
ur(x, t) = ur(x)preiωt.
Here, u is the modal displacement vector of the
median surface of the elastic structure obtained
from the in vacuo analysis.
2.3 Numerical evaluation of perturbation
potential
The boundary value problem for the perturbation
potential φ can be expressed by the boundary inte-
gral equation
c( ξ ) ( ξ ) ∫∫ {G( , ξ)q
ξ ) q( )
ξ) G n ( , ξ ) φ ( )} dS
SW
Here, ξ = (ξ, η, ζ) and x respectively represent
the source and field points on the mean wetted
surface of the structure, Sw, q = ∂φ/∂n refers to the
flux, the free term c(ξ) identifies the fraction of
MARSTRUCT.indb 35
φ(ξ) lying inside the potential domain of interest,
and G(x, ξ) stands for the Green function for time-
harmonic flows without forward speed, which can
be given in the form
4π G 1/ r + 1/ r′ H. (14)
Here, r and r ′ denote the distance between the
field and source points, and the field point and free
surface image of the source point, respectively, and
H represents the contained free-surface effects.
(9) For the solution of equation (13) with the bound-
ary condition (11), the mean wetted surface can be
idealized by using boundary elements, over which
the variations of the potential function and its
flux are described in terms of shape functions and
nodal values as
ne ne
(10) φe ∑ N ej φ ej qe = ∑ N ej q ej . (15)
j =1 j =1
Here, ne and N ej represent the number of nodal
points assigned to the eth element and the shape
function adopted for the distribution of the potential
function, respectively, and they together determine
the imposed approximation order for the potential
and flux distributions over the wetted surface.
In this study a higher order, linear, representation
is preferred by adopting four nodded quadrilateral
(11)
and triangular boundary elements.
2.4 Generalized fluid structure interaction
forces
Using the Bernoulli’s equation and neglecting the
second-order terms, the dynamic fluid pressure on
the elastic structure due to the oscillation of the
(12) elastic structure can be written as
P = −ρf (∂Φ / ∂t), (16)
where ρf is the fluid density.
The kth generalized fluid structure interaction
force component due to rth modal vibration of
the elastic structure is defined as (Ergin et al, 2007;
Uğurlu & Ergin, 2009)
Zkr ρf ∫∫ k (i ω )φr dS. (17)
SW
d . For time-harmonic free-surface flows without
(13) forward speed, the radiation potential is complex
in general, and Zkr may be expressed in terms of
the generalized added mass coefficient, Akr, and
hydrodynamic damping coefficient, Bkr, that are
in phase with the acceleration and velocity, respec-
tively. Namely,
35
2/18/2011 5:40:11 PM
 ships are already built and they are in service. The
Akr (ρ f 2
) [ ∫ (i ) k r dS ], (18a)
others are either under construction in Turkish
SW
shipyards or at the final stage of their design. Six
bulkers with different load carrying capacities
Bkr ( ρf )I [ ∫ (i ) k r dS ]. (18b) were investigated: two handsize type (20 000 and
SW
32 000 dwt), one handymax type (53 000 dwt),
one panamax type (76 000 dwt) and two capsize
2.5 Wave excitation and resonance frequencies type (140 000 and 180 000 dwt). The handsize and
handymax type bulkers under investigation have
For a floating structure the main excitation source
double hull side constructions.
is the ambient waves, generally composed of
For the each bulk carrier adopted in this study, a
two components that are related to the incident
detailed three dimensional finite element structural
wave system and its disturbance due to the scat-
model was prepared by using a commercial finite
tering effect of the body. The incident wave and
element software, i.e., Abaqus (2008). The three
diffraction potential fields, respectively denoted
dimensional structural model consists entirely of
by Φi and Φd, are connected through the relation
the shell finite elements, representing all major
∂(Φi + Φd)/∂n = 0 on the wetted surface of the
external and internal structural components. Shell
structure, which may invoke another boundary
element S4 was selected for modeling the primary
integral equation in the form of equation (13) for
and secondary members of the structural model.
the diffraction potential distribution. Considering
The shell element adopted for the calculations can
a generalized excitation in the form of Ξ(t) = Ξ0eiωt,
carry bending, membrane and shear loads. This is a
which may be caused by sinusoidal waves of fre-
general purpose shell element with four nodes, and
quency ω, the diffraction problem can be avoided
three nodes for triangle elements. The four nodded
by using the Haskind relations, and the resulting
shell element is mainly adopted for the structural
generalized wave forces for a stationary structure
model, and the three nodded triangle shell element
may be defined as
is used when it becomes necessary. At each node
of the shell elements, six degrees of freedom are
0r f ∫∫ (φi r φr i ) u r ⋅ n dS , (19) assigned: three translations and three rotations.
SW The triangle elements were used in the areas of
large curvature, such as the bilge and side shells in
where Ξ0r represents the force component associ- the vicinities of bow and stern. A limited number
ated with the rth modal vibration, and φi is the of beam elements was also adopted for the cal-
amplitude of the incident wave potential. culations. Beam element B31 available in Abaqus
The equation of motion for a harmonically (2008) was used, and it is a Timoshenko beam and
excited elastic structure may be written as allows for the transverse shear deformation. C3D
linear 3D solid elements were also used for mod-
Dp0 = Ξ0, (20) eling the main engine, etc. This element can have
where eight or six nodes, and each node has three degrees
of freedom (three translations). The three dimen-
D = −ω2(a + A) +(c + C) + iω(b + B). (21) sional models of the bulk carriers were generated
with sufficient detail to model the decks, inner and
The solution of equation (20) can be obtained outer bottom, machinery spaces, cargo spaces, etc.
from All the primary structural members were mod-
eled by using shell elements, and for the second-
adj
dD aadj (d D)*
dj D (det ary structural members such as stiffeners, shell and
p0 = Ξ0 = 2
Ξ0 . (22) beam elements were adopted. In order to obtain
det D det D an accurate idealization of the dynamic behavior
of the vessels, a close agreement was obtained
Here, the asterisk denotes a complex conjugate between the finite element models and designed
expression. From the calculated principal coordi- vessels for the vertical and horizontal centers of
nates, the resonance frequencies of the coupled gravity and total weights.
fluid-structure system can be deduced. The investigations were performed for two dif-
ferent loading conditions, i.e., fully loaded and
ballast conditions. Depending on the loading con-
3 NUMERICAL STUDY dition (fully loaded and ballast conditions), the
cargo loads, ballast loads, fuels, hatch covers, etc
The bulkers adopted in this study were designed are idealized by using the mass elements available
by Delta Marine Turkey, and some of the adopted in the finite element program. The main engines,

36

MARSTRUCT.indb 36 2/18/2011 5:40:14 PM

 however, were idealized by using 3D solid finite Table 1. Comparison of the dry natural and wet
elements, whose length, weight and centers of grav- resonance frequencies for the 140 000 dwt bulk in water
ity (vertical and horizontal) coincide, respectively, for fully loaded condition (Hz).
with those provided by the engine manufacturer.
FEM FEM BEM
The shaft systems and propellers were modeled, Mode (dry) (wet) (wet)
respectively, by using beam elements and mass
elements. 2 Node VB 0.923 0.656 0.674
For the handysize bulkers with 20 000 dwt and 2 Node HB 1.312 1.142 1.148
32 000 dwt load carrying capacities, respectively, 1 Node T 2.339 1.797 1.793
135 700 and 24 462 structural finite elements were 3 Node VB 1.789 1.295 1.330
adopted for the structural analysis. 25 546 struc- 3 Node HB 2.590 2.245 2.266
tural finite elements were used for the handymax 4 Node VB 2.540 1.886 1.935
bulker of 53 000 dwt load carrying capacity, and 4 Node HB 3.634 3.249 3.093
70 609 finite elements for the panamax type bulker 5 Node VB 3.183 2.420 2.486
with the load carrying capacity of 76 000 dwt. On 2 Node T 5.598 3.837 3.681
the other, for the finite element models of the cap-
size bulkers with the load carrying capacities of
140 000 and 180 000 dwt, respectively, 64 810 and Table 2. Comparison of the dry natural and wet res-
78 370 structural finite elements were distributed. onance frequencies for the 140 000 dwt bulk carrier in
It is believed that the finite element idealizations water for ballast condition (Hz).
used in the modeling are adequate to obtain the
FEM FEM BEM
dynamic response behavior of the bulk carriers in Mode (dry) (wet) (wet)
vertical bending and coupled horizontal bending
and torsion. Figure 1 shows the finite element ide- 2 Node VB 1.260 0.848 0.827
alization of the 32 000 dwt bulk carrier. 2 Node HB 1.857 1.699 1.600
The surrounding fluid domain is discretized by 1 Node T 2.413 1.875 1.873
using three dimensional acoustic fluid finite ele- 3 Node VB 2.543 1.629 1.639
ments available in the commercial software. The 3 Node HB 3.710 3.391 3.389
calculations are carried out for the fully loaded 4 Node VB 3.750 2.348 2.398
and ballast conditions. The results of the finite ele- 4 Node HB 5.322 5.064 5.180
ment calculations are presented for the 140 000 dwt 5 Node VB 4.716 2.985 3.042
bulker in Table 1 and Table 2, respectively for the 2 Node T 5.893 3.881 3.704
fully loaded and ballast conditions. The modes of
the ship hull are identified by the most dominant
vibrational shape. Table 3. Comparison of the dry natural and wet reso-
The amplitudes of principal coordinates for nance frequencies for the 32 000 dwt bulk carrier in water
the first three mode shapes are presented for the for fully loaded condition (Hz).
full loaded bulk carrier of 32 000 dwt capacity in
FEM FEM BEM
Figure 2 for the head waves. The dry natural and Mode (dry) (wet) (wet)

2 Node VB 1.285 0.875 0.903

2 Node HB 1.553 1.337 1.343
1 Node T 1.833 1.510 1.547
3 Node VB 2.919 1.960 2.018
3 Node HB 3.263 2.832 2.833
4 Node VB 4.539 3.055 3.211
4 Node HB 5.539 4.687 4.777
5 Node VB 6.576 4.135 4.391
2 Node T 6.214 5.117 5.115

wet resonance frequency values are presented in

Figure 1. 32 000 dwt capacity bulk carrier finite ele-
ment model in ballast condition.
Table 3 for the fully loaded case. It can be seen
from Table 3 that there is a very good agreement
between the finite element and boundary element
calculations. The principal coordinates presented
in the figures are associated with the excitation
by regular sinusoidal waves of 1 m amplitude.

37

MARSTRUCT.indb 37 2/18/2011 5:40:16 PM

 The resonance frequencies can be estimated by
observing the major peaks in these figures.
For instance, the first wet resonance frequency
is 0.903 Hz (2-noded vertical bending mode) and
the major peak occurs at the vicinity of this fre-
quency value. Due to the coupling of these modes
with other principal modes, each principal coordi-
nate generally has humps occurring in the data set
near these frequencies. The effect of the irregular
frequencies is observed in the principal coordinate
data presented. It should be noted that the princi-
pal coordinates presented in Figure 2 represent the
first three elastic principal modes.
Figure 2. Amplitudes of the first three components
of the principal coordinate vector for elastic modes and
head waves.
38
MARSTRUCT.indb 38
Figures 3–4 and Figures 5–6 present the wet
resonance frequency curves, respectively, for
the fully loaded and ballast conditions. It should
be noted that these curves were drawn by using
the resonance frequency data obtained from the
finite element calculations of six bulk carriers
with different load carrying capacities. The figures
only show the vertical bending dominated modes
(Figures 3–4) and torsional vibration dominated
modes (Figures 5–6). The resonace frequency val-
ues are presented as a function of I / L3 , where
I is the second moment of area for the mid-ship
section, Δ is the displacement and L overall length.
The parameter I / ΔL3 represents the bending and
torsional mechanical properties of the midship of
bulk carriers and a similar representation was also
adopted by Todd (**). The freqeuncy curves shown
in these figures can be used to predict the resonance
Figure 3. Wet resonance frequency curves of fully
loaded bulk carriers.
Key: X – 2-node vertical bending; – 3-node vertical
bending; – 4-node vertical bending; – 5-node vertical
bending.
Figure 4. Wet resonance frequency curves of bulk car-
riers in ballast condition.
Key: X – 2-node vertical bending; – 3-node vertical
bending; – 4-node vertical bending; – 5-node vertical
bending.
2/18/2011 5:40:17 PM

Figure 5. Wet resonance frequency curves of bulk car-
riers for fully loaded condition.
Key: – 1-node torsion; – 2-node torsion.
Figure 6. Wet resonance frequency curves of bulk carriers
in ballast condition.
Key: – 1-node torsion; – 2-node torsion.
frequencies of bulk carriers in preliminary design
stage. The frequency values increase with increas-
ing parameter value, except for the 1-noded tor-
sional mode shape.
4 CONCLUSIONS
In this paper, a hydroelastic investigation was car-
ried out into the dynamic response characteristics
of a group of bulkers with different load carrying
capacities, e.g., two handysize vessels with carrying
capacities of 20 000 and 32 000 dwt, respectively;
one handymax vessel with a carrying capacity of
53 000 dwt; one panamax vessel of 76 000 dwt car-
rying capacity; two capesize type vessels with capac-
ities of 140 000 and 180 000 dwt, respectively. For
all the bulk carriers adopted in this study, detailed
39
MARSTRUCT.indb 39
three dimensional finite element structural models
were prepared, separately, by using the commercial
finite element software. In a first group of calcula-
tions, the wet resonance frequencies and associated
mode shapes of the bulkers were calculated for
the fully loaded and ballast conditions by using
the finite element method. The surrounding fluid
domain was discretized by using three dimensional
fluid finite elements. Both symmetric and antisym-
metric response characteristics were obtained in
terms of the dry and wet frequencies with the asso-
ciated mode shapes, and they are presented for the
bulk carrier of 140 000 dwt capacity in Tables. The
calculated dry and wet frequency values for six dif-
ferent bulk carriers are also presented as a function
of the characteristic parameter, in order to reflect
overall response behaviors of the ship structures,
for fully loaded and ballast conditions. These
frequency curves could be used by the designers at
the preliminary design stage of the bulk carriers.
In a second group of calculations, a higher-
order 3-D hydroelasticity method was employed
for calculating the fluid-structure interaction effect
in terms of the generalized added mass and hydro-
dynamic damping coefficients. The wet frequency
values were calculated and they are compared with
those obtained from the finite element analysis for
the bulk carrier of 140 000 dwt carrying capacity.
As can be seen from Tables 1–2, there is a very
good comparison with those obtained from the
finite element analysis.
The frequency dependent response behaviors in
terms of principal coordinates were calculated by
using the higher order boundary element method,
and presented for the 32 000 dwt bulk carrier
excited by regular head waves and beam waves.
The resonance behaviors might be observed from
the peaks of the principal coordinate response
amplitudes.
REFERENCES
ABAQUS 2008. Theory Manual. SIMULIA, U.S.A.
Bishop, R.E.D., Price, W.G. & Temarel, P. 1985. A hypoth-
esis concerning the disastrous failure of the Onomichi-
Maru, Transactions of RINA 127: 169–186.
Bishop, R.E.D., Price, W.G. & Temarel, P. 1991. A theory
on the loss of MV Derbyshire. Transactions of RINA
133: 389–453.
Ergin, A. 1997. The response behavior of a submerged
cylindrical shell using the doubly asymptotic approxi-
mation method (DAA). Computers and Structures
62: 1025–1034.
Ergin, A., Kaydihan, L. & Uğurlu, B. 2007. Hydroelas-
tic analysis of a 1900 TEU container ship using finite
element and boundary element methods. Proceedings
of the International Conference of Asian-Pacific
Technical Exchange and Advisory Meeting on Marine
Structures, Yokohama, Japan.
2/18/2011 5:40:21 PM
 Ergin, A. & Temarel, P. 2002. Free vibration of a partially
liquid-filled and submerged, horizontal cylindrical
shell. Journal of Sound and Vibration 254: 951–965.
Hirdaris, S.E., Miao, S.H., Price, W.G. & Temarel, P.
2006. The influence of structural modelling on the
dynamic behaviour of a bulker in waves. Proceeding
of the 4th International Conference on Hydroelasticity
in Marine Technology, China, 25–33.
Hirdaris, S.E., Price, W.G. & Temarel, P. 2003. Two- and
three-dimensional hydroelastic modelling of a bulker
in regular waves. Marine Structures 16: 627–658.
INTERCARGO. 2007. Bulk carrier casualty report,
London.
Tian, C., Wu, Y.S. & Chen, Y.Q. 2009. Numerical pre-
dictions on the hydroelastic responses of a large bulk
carrier in waves. Proceedings of the 5th International
Conference on Hydroelasticity in Marine Technology,
England.
40
MARSTRUCT.indb 40
Uğurlu, B. & Ergin, A. 2006. A hydroelasticity method
for vibrating structures containing and/or submerged
in flowing fluid. Journal of Sound and Vibration
290: 572–596.
Uğurlu, B. & Ergin, A. 2008. A hydroelastic investigation
of circular cylindrical shell containing flowing fluid
with different end conditions. Journal of Sound and
Vibration 318: 1291–1312.
Uğurlu, B. & Ergin, A. 2009. Using higher-order bound-
ary elements in hydroelasticity analysis of surface
piercing structures. Proceedings of the 5th International
Conference on Hydroelasticity in Marine Technology,
England.
2/18/2011 5:40:23 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Numerical prediction of slamming loads on a rigid wedge

subjected to water entry using an explicit finite element method

Hanbing Luo, Shan Wang & C. Guedes Soares

Centre for Marine Technology and Engineering (CENTEC), Technical University of Lisbon,
Instituto Superior Técnico, Lisboa, Portugal

ABSTRACT: An explicit finite element code is applied to study the impact loads on one two-dimensional
rigid wedge subjected to water entry. This wedge with deadrise angle 30 degree impacting the calm water
is modeled. The numerical results are compared and validated against published experimental slamming
force, pressure distributions at different time instances, and pressure histories at different points on the
wetted surface, obtaining very good comparisons. A convergence study for parameters, such as mesh
density and penalty factor, is carried out. The computational efficiency and accuracy of the results is
discussed.

1 INTRODUCTION Wagner type of formulation eventually adjusted by

When a ship travels in rough seas, it will impact
water because of large vertical relative motions
between the ship and the wave surface. This
hydrodynamic impact phenomenon is defined as
slamming. The impulsive pressure loads induced
by slamming will affect the ships structures both
locally and globally. In rough seas, this impact force
is so large that many ships have reported local struc-
tural damages due to the slamming loads, especially
in heading waves with high forward speed. For
example, the tragedy of MV Estonia in the Baltic
Sea on 28 September 1994, one of the deadliest
marine disasters of 20th century, was initialized by
the break of the bow door due to the severe slam-
ming, which the Ro/Ro ferry experienced. In this
paper, the local slamming problem will be studied
and a numerical model will be developed to repro-
duce the conditions used in an early experimental
study with which results it will be compared.
There is a considerable amount of research
conducted on slamming by experimental, ana-
lytical, and numerical simulation methods since
Von Karman (1929) and Wagner (1932). Ship
slamming depends on the relative motion, body
geometry, water surface profile, air cushion,
hydroelasticity of structures, compressibility of
water, and others. This makes ship slamming such
a complicated physical process that it is difficult to
model in all aspects. To develop a model of ship
slamming, it is necessary to build upon accurate
predictions of slamming forces by arbitrary shapes
such as ship sections on water.
Much work has been done by modeling the
slamming forces in different ship sections by a
correction factors calibrated with experiments as
reviewed for example in Ramos and Guedes Soares
(1998). However, improvements are required on
this type of approach based on better predictions
of slam loads.
Most of earlier research work focused on 2D
and simple geometry section problems, e.g. Zhao
and Faltinsen (1993), Engle and Lewis (2003),
Wu, et al. (2004), Sun and Faltinsen (2009). Not
much research has been carried out on 3D slam-
ming problems, except the hemisphere, or conical
shapes, e.g. Faltinsen et al. (2004), Peseux, et al.
(2005), Korobkin, et al. (2006).
Several good reviews have been published dur-
ing past years. Faltinsen (2004) reviewed practical
slamming problems for ships and offshore struc-
tures, including water entry on an initially calm
free surface, wetdeck slamming, green water and
sloshing. Xu (2009) presented a review of theoreti-
cal and numerical simulation techniques on hydro-
dynamic impact of ships.
With the development of computing technology
and capability, codes based on explicit Finite Ele-
ment Method (FEM) began to be applied to predict
local slamming loads. Bereznitzki (2001) analyzed
hydroelastic problems using the MS-Dytran code.
Stenius et al. (2006) studied modeling techniques for
rigid wedge impact problems using the LS-DYNA
code. Several parameters that influence the conver-
gence of simulation, such as mesh density and con-
tact stiffness were discussed. Aquelet et al. (2006)
discussed the influence of penalty factor on the
damping effect. Luo et al. (2010) used MS-Dytran
to study the impact of one stiffened panel, showing
that an explicit code has the potential to predict

41

MARSTRUCT.indb 41 2/18/2011 5:40:23 PM

 water entry impact problems for both rigid and
elastic structures.
Alexandru, et al. (2007) carried out a compari-
son of simulation of 2D slamming problems, using
Boundary Element Methods (BEMs), Computa-
tional Fluid Dynamics (FLUENT and FLOW-3D
codes), Smooth Particle Hydrodynamics (SPH),
and Explicit FEM (LS-DYNA code). Fairly good
agreements were achieved. Some differences in time
domain after the peak value for slamming pres-
sure on one wedge with dead rise angle 25 degree
were observed. There were even much differences
between the pressures predicted with each other on
an impacting rigid bow section. Nevertheless, the
results obtained are encouraging overall, but more
validation work still needs to be carried out on
tuning parameters in order to get better numerical
results.
In this paper, the explicit FEM code LS-DYNA
is applied to study the fundamental water entry
problem on initially calm free surface of 2D rigid
wedge. The predicted results are compared with the
experimental results Zhao, et al. (1996). The main
purpose of this paper is to carry out a convergence
study on the parameters that influence the simula-
tion efficiency and accuracy of the results.
2 EXPLICIT FEM CODE PROGRAM
SETUP
The commercial explicit FEM code LS-DYNA
(version 971, 2007) is used with double precision
for the numerical simulation. It is based on explicit
time integration. The Arbitrary Lagrangian—
Eulerian (ALE) algorithm is chosen in this paper.
Water and air are modelled as Multi-material
Eulerian mesh. Then the free surface can be mod-
eled by Volume of Fluid method. The structure
is modeled by the Lagrangian mesh with rigid
material.
The penalty coupling method is applied for the
interaction of Eulerian fluids and the Lagrangian
solid, which is different from the penalty based
contact coupling method used by Stenius et al.
(2006). The penalty factor used here is defined to
simulate the coupling effect between the fluids and
the structure. The coupling effect is limited to the
normal direction of the solid surface. So the sliding
is allowed in the tangent direction. Usually, com-
pression coupling direction option is selected for
rigid body impact. By combining the ALE solver
and the Eulerian-Lagrangian penalty coupling
algorithm, LS-DYNA has the capability to simu-
late the slamming problems.
The critical time step size is the minimum time
value that the sound travels through all elements
(solid and shell mesh). The time step used in the
42
MARSTRUCT.indb 42
simulation can be set by one scale factor based on
the critical one. Pressure sensors are located on the
center of each shell elements at the coupling sur-
face to obtain the pressure signals.
3 DESCRIPTION OF THE WATER ENTRY
PROBLEM
The drop test of one wedge section was carried out
in MARINTEK by Zhao et al. (1996) as shown
in Figure 1. Table 1 presents the main data for
the test section. Results of measured pressure and
slamming force will be adopted to compare with
predicted numerical results in this paper.
Figure 2 shows the wedge model setup
in LS-DYNA. There is one element along the z direc-
tion. All nodes are constrained for displacement in
y direction in order to simulate the two-dimensional
water entry in the plane x-y. Symmetry boundary
condition is applied on the surfaces where x = 0.
Only half of the wedge, water and air are needed
to be modeled. So the simulation CPU time will be
decreased dramatically. Nonreflecting boundary
conditions are added on the external surfaces of
air and water domain, except the symmetric sur-
faces. The size of air and water domain in x and y
direction is selected to try to eliminate the effects
of limited boundaries as much as possible. The size
of water domain is 1250 mm * 700 mm in plane
x-y which is about five times of the wedge’s size,
and that of air domain is 1250 mm * 200 mm.
Figure 1. Locations of gauges in wedge section from
Zhao et al. (1996).
Table 1. Main particulars of the test section.
Breadth of section 0.50 m
Length of measuring section 0.20 m
Length of each dummy section 0.40 m
Total length 1.00 m
Dead rise angle 30°
Total weight 241 kg
Weight of measuring section 14.5 kg
2/18/2011 5:40:23 PM
 Table 2. Three models with different mesh densities.

Parameters Model 1 Model 2 Model 3

Mesh size 5 mm 2.5 mm 1.25 mm

Number
of elements
(Fluids + 5725 + 29500 + 127000 +
Structures) 118 215 430
Number of
elements on
impacting
wedge surface 58 115 230
Figure 2. 2-D water entry model in LS-DYNA. CPU time * 56 m 9 h 49 m 96 h 54 m
*
Note: It was run on one PC with 2.4 GHz processor and
Initial impact velocity above calm water surface 2 Gigabytes of memory.
is −6.15 m/s in vertical y-direction. Gravity is not
considered.
From considerations of computational effi-
ciency, it is not a good choice to mesh uniformly
the total fluids (air and water) domains. Only the
domain near the wedge and also where the wedge
will pass through are meshed uniformly with finer
meshes. This mesh density will be studied to obtain
a compromise between CPU time and simulation
results. The domain far away the wedge is moder-
ately expanding toward the boundaries.

4 ANALYSIS OF PREDICTED RESULTS

FOR 2D RIGID WEDGE
Figure 3. Part of meshes in air and water domain of
4.1 Slamming force Model 2.

Different mesh densities are chosen to simulate

the water entry problem. The mesh size for air and
water is chosen as 5 mm, 2.5 mm, 1.25 mm respec-
tively, and mesh size for wedge is set the same as
that for fluids. Table 2 lists the main parameters for
the three models. The predicted accelerations are
proportional to the slamming force on the measur-
ing section in the test. The instant when the vertex
of 2D wedge touches the element on the water sur-
face is set as 0.0s. Figure 3 shows part of meshes in
air and water domains near the vertex of the wedge
from model 2.
Figure 4. compares the predicted vertical slam-
ming forces with experimental results. It shows Figure 4. Comparison of experimental slamming forces
that: and predicted results with different mesh sizes.
• The predicted slamming forces agree satis-
factorily with the experimental one in the time the drop test. But this difference is not so large
domain, from the beginning of slamming, to the as up to 20%, which was predicted by BEMs of
flow separation on the knuckle at about time Zhao et al. (1996).
0.0158s, and to the later stage of water entry in
the figure. The predicted results are about 5% • When the mesh size is 5 mm (Model 1), there are
larger than the experimental ones in the middle some high frequent oscillations on the curve
of water entry before flow separation. Maybe it because of numerical noise. Especially at
is mainly due to the three-dimensional effect on the beginning of impact, the peaks are very

43

MARSTRUCT.indb 43 2/18/2011 5:40:24 PM

 obvious. The difference between predicted and
experimental slamming force after flow separa-
tion is too big.
• When the mesh size is 2.5 mm (Model 2), the
result is better except the small oscillations at the
beginning. Further it can simulate the separa-
tion very well at the knuckle. The slamming force
reduces quickly after flow separation, and then
decays slowly. The CPU time is also acceptable.
• Finally, when the mesh size is 1.25 mm (Model 3),
no obvious oscillations are observed in the curve.
It shows best consistency in this case, but it will
take about 4 days to run even one simulation in (a) t1 = 0.00435s
a normal personal computer.

4.2 Pressure distributions at different

time instances
Non dimensional pressure distributions on the
wetted surface of the wedge are presented at three
time instances, and are compared with experimental
results in Figure 5. V(t) is the drop velocity from
the test. The time instance in Figure 4 is the time in
the test. Y is the vertical coordinate on the wedge
surface, yk is vertical coordinate of the vertex, and
yD is the draft of the wedge. (b) t2 = 0.0158s
• The predicted pressure results agree well with
the measured pressures. The measured pressures
are smaller than the predicted ones in general,
especially after the initial stage as shown in
Figures 5 (b) and (c). It is mainly because of
three-dimensional effects in the test then.
• When a much finer mesh is used, the predicted
results are much smoother. The pressures from
Model 3 are the best. It shows good consist-
ency with experimental results. Only the value
of pressure P2 in Figure 5 (a) from the experi-
ment is larger than the predicted results from
LS-DYNA. The reason for this difference is not (c) t3 = 0.0202s
clear, maybe mesh density at the initial stage here
is not finer enough, or maybe due to experimen- Figure 5. Comparison of pressure distributions at
tal errors, because this value of pressure P2 from different time instances.
the experiment is also larger than the predicted
results by BEMs in Zhao et al. (1996).
• In the initial stage, Figure 5 (a) shows that results
Model 2 near the vertex of wedge is set much finer,
from both Model 1 and 2 are not good, while in
for example, 1.25 mm, then the predicted results
Figures 5 (b) and (c), results from both Model 1
at the initial stage in Figure 4 and Figure 5 (a) will
and 2 are better. It may be explained as that the
become better and acceptable comparing with
mesh density near the vertex of wedge in model
those from Model 3. Not so much CPU time is
1 and 2 is not finer enough to describe the slam-
needed as Model 3. It may be a good solution from
ming pressure in this case.
the view point of computational efficiency.
Comparing with slamming force results in
Figure 4, it may be concluded that the mesh den-
4.3 Pressure histories at different points
sity in the water and air domains near the ver-
tex of wedge is most important. In other words, Figure 6 shows a comparison of pressure histories
if the mesh density in water and air domain of at different points on the wetted surface of the

44

MARSTRUCT.indb 44 2/18/2011 5:40:28 PM

 wedge from P1 to P5. Because the drop velocity
varies during water entry process, non dimensional
pressure is not applied here. The predicted pressure
results from model 2 and 3 are shown, but those
from model 1 are not included because the results
are not good due to too coarse meshes.
• The predicted results show good agreement with
the experimental ones in general. Usually, they
are a little larger than the experimental ones,
(a) P1 depth = 12.5 mm except at point P2 in Figure 6 (b), the reason was
explained in 4.2 before.
• The peak values of pressure histories from
model 3 with finer mesh 1.25 mm are larger
than those from model 2, while the pressure dis-
tributions after peak are almost same. There is
one obvious impulse in pressure curves before
the pressure peak from model 2 in Figures 6
(b), (c) and (d). Maybe these impulses are due
to numerical noise, because when finer mesh
1.25 mm is used, the phenomena disappear as
(b) P2 depth = 37.5 mm shown in Figure 6.
• The peak value of pressure P1 is smaller than that
of P2, P3, and P4, and the shape of P1’s peak is
not as good as others. Perhaps it is because the
mesh density near the vertex of wedge and also
P1 is still not fine enough in model 3.
In order to capture the pressure peak phenom-
ena correctly and also to eliminate the numerical
noise in the curves, much finer mesh is needed in
this case, for example 1.25 mm. For the case of
(c) P3 depth = 62.5 mm
slamming force, mesh size 2.5 mm maybe is enough
as shown in Figure 4 before.

4.4 Water jet and pressure contour in water

at different time instances
Figure 7 presents both water jet and pressure con-
tour phenomena from time 0.00435s, 0.0158s, to
0.0202s, corresponding to pressure distributions
described in Figure 5. Half of rigid wedge and part
of the water domain near the wedge is shown. The
(d) P4 depth = 87.5 mm vertex of the rigid wedge of model 3 will touch
the calm water at time 0.0036 s in the LS-DYNA
simulation, which corresponds to the time 0.0s in
the drop test. Three coupling points option is cho-
sen for each coupled Lagrangian element, and no
fluid leakage is observed on the coupling wedge
surface.
• Before flow separation, as shown in Figures 7 (a)
and (b), the maximum pressure appears in the
inner domain, or the up-rise which was described
(e) P5 depth = 131.8 mm in Wagner (1932) theory. The slamming pressure
will decrease in the outer domain in the water,
Figure 6. Comparison of pressure histories at different or in the jet flow along the wedge surface. The
positions. pressure in the jet flow is so small that it can

45

MARSTRUCT.indb 45 2/18/2011 5:40:29 PM

 on the knuckle. The pressure is almost the same
along the coupling wetted surface along wedge,
but it is much smaller comparing with that
before flow separation as shown in Figures 7 (a)
and (b). It is also observed from pressure distri-
butions in Figure 5. That is the reason why the
slamming force will reduce very much compar-
ing with that before flow separation. It is con-
sistent with what is presented in slamming force
curve in Figure 4.

(a) t1 = 0.00435s 4.5 Penalty factor

The penalty factor (PFAC) is one parameter for
scaling the estimated stiffness of the interacting cou-
pling system. It is applied to compute the coupling
force to be distributed on the structure and fluids.
Simulations are run for three cases when PFAC are
set as 0.5, 0.1, and 0.01 for model 3. Figures 8 and
9 shows comparison of slamming force and pres-
sure distributions with different values of PFAC
respectively. Little difference is observed. In this
paper the default value 0.1 is used.

(b) t2 = 0.0158s

Figure 8. Comparison of slamming forces with differ-

(c) t3 = 0.0202s ent values of PFAC.

Figure 7. Predicted water jet flow and pressure contour

in water by LS-DYNA.

be ignored. That’s the reason why the Wagner

theory can explain the water impact almost cor-
rectly, and still be widely used today.
• The position of maximum pressure on the wet-
ted surface of wedge will move from the vertex
to the knuckle during the water entry process.
The peak value of slamming pressure does not
decrease much before flow separation compar-
ing from Figures 7 (a) and (b), so slamming force
will increase gradually when the wetted surface
is increased as shown in Figure 4.
• Figure 7 (c) shows the water jet and pressure Figure 9. Comparison of pressure distributions with
contour in the final stage after flow separation different values of PFAC at t2 = 0.0158s.

46

MARSTRUCT.indb 46 2/18/2011 5:40:30 PM

 4.6 Time step
The critical time step size is the minimum time value
that the sound travels through any elements in the
model. One scale factor TSSFAC can be adopted
to compute the time step used in simulations. The
time step calculated should not be larger than the
critical one, otherwise negative volume errors will
appear. But if the time step is set to one value that
is too small, then the simulation CPU time will
increase correspondently. The critical time step size
can be approximated firstly before the simulation,
in order to set one sale factor to obtain one appro-
priate time step.
5 CONCLUSIONS
Numerical simulation of slamming loads on a
two-dimensional rigid wedge with the deadrise
angle 30 degree is carried out by the explicit FEM
code LS-DYNA in order to validate experimental
results. The Arbitrary Eulerian-Lagrangian solver
and penalty coupling algorithm is used.
Slamming force, pressure distributions at dif-
ferent time instances, and also pressure histories
at different points on wetted surface of wedge are
predicted, and the results are compared with the
experimental ones published by Zhao et al. (1996).
It shows good agreement between each other.
The predicted 2D slamming force correctly
describes the impact process from the initial stage,
flow separation, to the final one, and the values
are a little larger than the 3D experimental ones.
So are the pressure values, especially after the ini-
tial stage of impact. This is mainly because of the
three-dimensional effects during water entry.
A convergence study is carried out. Different
mesh sizes, 1.25 mm, 2.5 mm and 5 mm, are com-
pared. The finer the mesh size is, the better results
can be obtained. But the CPU time will increase
dramatically once the mesh density is increased.
The compromise between mesh density and CPU
time should depend on what will be predicted.
In order to capture the peak of slamming
pressure, much finer mesh is needed, for exam-
ple, 1.25 mm in this case. The value of the pres-
sure peak predicted by the coarser mesh model is
smaller than that by the finer mesh. For example,
the maximum non-dimensional pressure coeffi-
cient from model 3 at the time 0.0158s before the
flow separation is 7.32, while that from model 2
is only 6.65, which is about 9% smaller. The peak
value of pressure P1 predicted in model 3 is smaller
than that of P2, P3, and P4, and the shape of P1’s
peak is not as good as others. Perhaps it is because
the mesh size 1.25 mm near the vertex of wedge
and also P1 is still not finer enough. Both models
47
MARSTRUCT.indb 47
with mesh size 1.25 mm and 2.5 mm can predict
pressure histories after the peak very well. How-
ever, the coarser mesh, such as 2.5 mm for this
case, is enough to predict the slamming force.
It is found that the mesh density is the most
important parameter influencing the predicted
accuracy. Small differences are observed when the
penalty factor is modified. One appropriate value
for time step scale should be set to reduce CPU
time and also to prevent the negative volume errors
during the simulation.
The water jet flow and pressure contours in
the water domain are presented during different
stages of the impact. It simulates the phenomena
correctly in general from model 3 with 1.25 mm
mesh size. The peak pressure appears in the inner
domain before the flow separation, while the pres-
sure along wedge surface is very low after the flow
separation. The value of pressure predicted in the
jet flow is so small that it may be ignored.
The modelling techniques described here can be
adopted to model wedges of different types, and to
study other hydrodynamic impact problems.
ACKNOWLEDGEMENTS
The work has been performed in the scope of the
project EXTREME SEAS—Design for Ship Safety
in Extreme Seas, (www.mar.ist.utl.pt/extremeseas),
which has been partially financed by the EU under
contract SCP8-GA-2009-234175.
REFERENCES
Alexandru, I., Brizzolara, S., Viviani, M., Couty, N.,
Donner, R., Hermundstad, O., Kukkanen, T.,
Malenica, S. & Termarel, P. (2007). Comparison of
experimental and numerical impact loads on ship-like
sections. Advancements in Marine Structures, Guedes
Soares, C, and Das, P.K., (Eds), Taylor & Francis,
UK, 339–349.
Aquelet, N., Souli, M. & Olovsson, L. 2006.
Euler–Lagrange coupling with damping effects:
Application to slamming problems. Computer Methods
in Applied Mechanics and Engineering. 195, 110–132.
Bereznitski, A. 2001. Slamming: the Role of Hydroelastic-
ity. International Shipbuilding Progress. 48, 333–351.
Engle, A. & Lewis, R. 2003. A comparison of hydro-
dynamic impacts prediction methods with two-
dimensional drop test data. Marine Structures. 16, 2,
175–182.
Faltinsen, O.M. & Chen, Z.M. 2005. A generalized
Wagner method for three-dimensional slamming,
Journal of Ship Research, 49, 4, 279–287.
Faltinsen, O.M., Landrini, M. & Greco, M. 2004.
Slamming in marine application. Journal of Engineer-
ing Mathematics. 48, 187–217.
2/18/2011 5:40:34 PM
 Korobkin, A.A. & Scolan, Y.-M. 2006. Three-dimensional
theory of water impact. Part 2. Linearized Wagner
problem. Journal of Fluid Mechanics. 549, 343–373.
LS-DYNA, Keyword User’s Manual, Livermore Soft-
ware Technology Corporation, Version 971, May
2007.
Luo, H.B., Hu, J.J. & Guedes Soares C. 2010. Numerical
simulation of hydroelastic response of flat stiffened
panels under slamming loads. Proceedings of the 29th
International Conference on Ocean, Offshore and Arctic
Engineering (OMAE´10), 6–11 June, 2010, Shanghai,
China, ASME, Paper OMAE2010-20027.
Peseux, B., Gornet, L. & Donguy, B. 2005. Hydrody-
namic impact: Numerical and experimental investiga-
tions. Journal of Fluids and Structures, 21, 277–303.
Ramos, J. & Guedes Soares, C. (1998). Vibratory response
of ship hulls to wave impact loads. International Ship-
building Progress. 45 (441): 71–87.
Stenius, I., Rosn, A. & Kuttenkeuler, J. 2006. Explicit
FE-modeling of fluid-structure interaction in hull-
water impacts. International Shipbuilding Progress.
53, 1031–121.
Sun, H. & Faltinsen, O.M. 2009. Water entry of a bow
flare section with a roll angle. Journal of Marine
Science and Technology. 14, 69–79.
48
MARSTRUCT.indb 48
Von Karman, T. 1929. The impact on seaplane floats
during landing. National Advisory Committee for
Aeronatics. Techinical note No. 321, 309–313.
Wagner, H. 1932. Uber Stoss- und Gleitvergange an der
Oberflache von Flussigkeiten. Zeitschrift fuer Ange-
wandte Mathematik und Mechanik, 12, 193–215.
Wu G.X., Sun H. & He Y.S. 2004. Numerical simulation
and experimental study of water entry of a wedge in
free fall motion. Journal of Fluids and Structures, 19,
3, 277–289.
Xu, G.D. & Duan, W.Y. 2009. Review of prediction
techniques on hydrodynamic impact of ships. Journal
of Marine Science and Applications. 8, 204–210.
Zhao, R. & Faltinsen, O.M. 1993. Water Entry of Two-
Dimensional Bodies. Journal of Fluid Mechanics, 246,
593–612.
Zhao, R., Faltinsen, O.M. & Aarsnes, J.V. 1996. Water
entry of arbitray two-dimensional sections with and
without flow separation. Proc. 21st Symposium on
Naval Hydrodynamics. 408–423.
2/18/2011 5:40:34 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

On estimation of extreme ship response using upcrossing

spectrum

Wengang Mao & Igor Rychlik

Department of Mathematical Sciences, Chalmers University of Technology, Gothenburg, Sweden

ABSTRACT: In this paper a simple method is proposed to estimate extreme ship response, defined as
the return values of the responses. Real ship responses are often non-Gaussian, hence a transformation,
defined by the cubic Hermite polynomials, of a Gaussian process is employed to model the responses.
The transformation is a function of the standard deviation, skewness, kurtosis and zero up-crossing fre-
quency of a response. The parameters vary with changing sea conditions and operation conditions and
are easily available from measured responses. In the case when measurements are not available the param-
eters are estimated by an empirical model from the significant wave height and operation conditions.
The model is derived from the measurements, but can be also estimated by means of a theoretical analysis.
The proposed method is compared with the typical engineering approach to estimate the return values of
a response. The full-scale measurements of a 2800TEU container ship during the first six months of 2008
are used in the comparisons.

1 INTRODUCTION xT could happen is very low, some importance

Let M be the maximum value of the response X,
say, during one year. The extreme response is often
characterized by the most probable value, i.e. the
location of a mode of M, or by the return value.
More precisely, let T = 10, 20 or 100 years, then the
return value of response xT is the 1/T quantile of
M, viz. a solution of the following equation,
1 It employs the fact that the distribution of local
P (M xT ) = (1)
T
If long records of ship responses are available,
assuming stationary shipping and ergodicity of
the responses, xT could be estimated by one of
standard statistical procedures. For example, the
Gumbel or Generalized Extreme Value distribu-
tion could be used to fit the distribution of the
observed maxima of blocks of recorded response.
Another popular method is the Peak Over Thresh-
old (POT), for more details see Coles [1].
If there are no data available, then the distribu-
tion of M in Eq. (1) has to be estimated in other
ways, using the available information. Often, one
can assume that shipping is known, so that the long
term distribution of sea states can be determined.
Then response at a given sea state can be simulated
by means of dedicated numerical software. Next
suitable statistical method could be used to estimate
the distribution of M, see e.g. Naess et al. [2]. Since
frequency of sea states when responses exceeding
sampling methodology is often employed to limit
the computation effort. For example, the envi-
ronmental contour line, see Winterstein et al. [3],
is a popular way to limit the number of sea states
needed to be considered.
In the engineering community method, des-
cribed below, is often used. The methods is
originally derived for Gaussian responses but
can be used for non-Gaussian responses as well.
maxima Xm in a stationary Gaussian response X is
known, viz. the so-called Rice’s distribution char-
acterized by few spectral moments. Then the long
term cumulative distribution function (cdf) of Xm
in one year is estimated and used to approximate
the distribution of yearly maximum M. Finally xT
is estimated by solving
1
P (X m xT ) = , (2)
n T
where n is the expected number of local maxima in
one year. The long term cdf of Xm is often approxi-
mated by a Weibull distribution, see DNV [4].
Here the approach to approximate the long term
cdf of Xm (even for non Gaussian responses) by
the Weibull distribution and then solving Eq. (2)
to estimate the return response xT will be called
“Method 1”.
This paper focuses on an alternative method for
extreme response prediction, the so called Rice’s

49

MARSTRUCT.indb 49 2/18/2011 5:40:34 PM

 method, in which the distribution of M is bounded
as follows
P (M x) E ( N1+ ( x )) + P(
P (X (0) x ), (3)
where N1+ ( x ) denote the number of upcrossings
of level x by the response X in one year, while
E(N) is the expected value of N, see Cramer
and Leadbetter [5] for more detailed discussion.
Neglecting the term P(X(0) >x) and combining
Eq. (1) and Eq. (3) gives the Rice’s estimate of
the return response xT. In the following section
we shall slightly reformulate the Rice’s estimate of
xT to demonstrate its connection with the narrow-
band approximation of long term cdf of Xm. Here,
this method will be referred to as “Method 2”.
Method 2 requires the computation (estima-
tion) of E(N N1+ ( x )) . Typically, developing a model
for ship response variability involves: (a) using
some well established models for variability of
encountered waves; (b) mathematical description
of ship wave interaction to compute wave loading;
(c) a model for structure properties to compute
structural stresses. In the simplest case, this pro-
cedure leads to Gaussian model for the response,
under stationary sea conditions, and the expected
number of upcrossing during a sea state is given
by an analytical formula, see Rice [6]. Next, given
a shipping, the long term distribution of sea states
enables evaluation of E ( N1+ ( x )), see Eq. (11).
However, due to the complexity of interaction
between ship structure and encountered waves,
the real ship responses are non-Gaussian, particu-
larly under large sea states. In such an eventuality,
the Gaussian model can lead to large prediction
errors, e.g., severe underestimation (50%) of xT
was reported in Mao et al. [7]. There are several
numerical approaches proposed in the literature
to numerically compute Rice’s formula for non-
Gaussian responses, see some recent references
Naess and Karlsen [8], Butler et al. [9] or Galtier
et al. [10]. In this paper, the 4 moments Hermite
transformation, proposed in Winterstein et al. [11],
is used to model the non-Gaussian ship responses.
The transformation requires only the knowledge
of variance, skewness, kurtosis and mean level
upcrossing frequency of the responsse. The advan-
tage of the method is that the expected number of
crossings of any level x, during a sea state, is given
by an explicit analytical formula, see Eq. (16).
2 EXTREME ESTIMATION
BY METHOD 1 & 2
The variability of ship responses is mainly caused
by the encountered waves, which can be modeled
50
MARSTRUCT.indb 50
as “locally” stationary processes (sea states). A sea
state (usually from 20 minutes to several hours) is
generated using, for example, a linear (Gaussian)
or a Stokes (quadratic) wave model. Most often, it
is defined by a vector of parameters; say W, whose
elements could be the significant wave height Hs,
wave period Tp, etc. The encountered wave envi-
ronments are described by a sequence of sea states
Wi, i = 1, …, K, where K is the average number of
sea states encountered during a long term period.
The distribution of W is called long term distribu-
tion of sea states. The probability density function
(pdf) of the distribution will be denoted by f(W).
However it is not obvious how to define (estimate)
the pdf f(W). Loosely speaking long term pdf of
sea state parameters describes the variability of
encountered sea state W, which depends on ship-
ping routes. However it will also depend on the goal
for the computations. We will not discuss this sub-
ject deeply just point out nature of the problem.
In this paper, ship responses X are estimates
of structural stresses and clearly depend on the
encountered sea states. The mean value of stresses
is assumed to be a constant and without loss
of generality, is set to be zero, for simplicity of
expressions.
2.1 Review of method 1
The long term cdf of response local maxima Xm
is defined as the limiting value of the ratio of the
number of local maxima with height below x and
the total number of local maxima as the observa-
tion period tends to infinity. Since one has only
finite length measurements, the long term cdf of
Xm has to be estimated practically. Often, when
long records of measurements are available, the
Weibull distribution is used to estimate the long
term cdf of Xm, viz.
⎡ ⎛ x −θ⎞k⎤
FX m ( x ) ≈ 1 − exp ⎢ ⎥. (4)
⎢⎣ ⎝ λ ⎠ ⎥⎦
Suppose that the observed number of local
maxima during t (unit: year) is n(t), then expected
number of local maxima during a year, n, is esti-
mated by
n(t )
n= , (5)
t
while the return value xT is obtained by solving
Eq. (2) (T often equals to 20, 50 or 100 years).
However, the long term Weibull cdf in Eq. (4) is
very sensitive to the value of location parameter θ,
which is hard to estimate when measuring cam-
paign length t is not very long. Furthermore the
2/18/2011 5:40:35 PM
 effect of high frequency vibration of ship structures
also makes estimate of n, Eq. 5, very uncertain.
This makes practical use of the original version of
“Method 1” somewhat difficult. In the following,
the upcrossing spectrums of ship responses will be
used to approximate the long term cdf.
2.2 Narrow-band approximation of FXm(x)
Let the long records of measured stresses be divided
into a series of stationary parts (sea states). For
each sea state W, expected number of upcrossing
of level x in unit time by the stationary response is
denoted by μ+(x|W). The short term cdf of Xm at
the sea state W can then be bounded by
μ + (x |W )
1 − FX m ( x | W ) ≤ , (6)
μ + (0 | W )
see Rychlik and Leadbetter [12] for a proof. The
upcrossing frequency μ+(x|W) need to be estimated
or computed from the model.
In Figure 1 (a), empirical cdf of Xm is compared
with the approximated bound when μ+(x|W) is
Figure 1. (a): short term distribution FXm(x|W),
bounded by Eq. (6) (solid line) and empirical distribution
(dotted line) from full-scale measurements at a stationary
sea state; (b): long term (t = 1 month) empirical cdf FXm(x)
(dotted line) and the bound given in (9) (solid line).
51
MARSTRUCT.indb 51
estimated from the measured stresses. One can see
the typical behavior that the empirical cdf of Xm is
close to the bound in the tails defined in Eq. (6).
The bound is often known as the narrow-band
approximation.
However the narrow-band method is most use-
ful when μ+(x|W) can be computed from a model
for the responses. For example when ship response
at W is Gaussian, then the upcrossing spectrum can
be analytically evaluated by Rice’s formula and one
derives that the cdf of Xm is bounded by Rayleigh
cdf. The long term cdf of Xm during period t, is
given by
FX m ( x ) = ∫ FX m ( x | W f (W )dW
d , (7)
where f(W) is the pdf of the vector of parameters
defining the sea states encountered in the period t.
Further the fraction n(t)/t in Eq. (5) is approxi-
mated by
n(t)/t ≈ Ûμ+(0|W)f(W)dW . (8)
Combining Eqs. (6–8) gives the narrow-band
estimation of the return value.
2.3 Examples
In the following, ship response from half year’s
full-scale measurements of a 2800TEU container
vessel is taken to compare different estimates of
the long term cdf of Xm. For the region there are
well established models (Weibull) long term pdf of
encountered significant wave height f(W). How-
ever the measuring period is rather short and we
do not expect that the variability of the encoun-
tered significant wave heights is well described by
the model. Hence the empirical cdf of Xm could dif-
fer from that estimated by Eq. (7). The alternative
could be to use the encountered seas to estimate
the long term pdf of W. This would still require
numerical computations of the integral in Eq. (7).
Since in this section we only wish to compare dif-
ferent means to estimate the long term cdf of Xm,
a related method, not requiring the integration,
will be used instead in the following.
As before, let Nt+ ( x ) denote the number of up-
crossings of level x by the response in a period of
length t, then the empirical distribution of encoun-
tered response maximums Xm satisfies
Nt+ ( x )
FX m ( x ) ≥ 1 − . (9)
Nt+ ( )
Again the bound is close to 1–FXm(x) for high
values of x as can be seen in Figure 1(b) where one
presents the empirical cdf of encountered local
2/18/2011 5:40:37 PM
 maximums and the bound in Eq. (9) for measured
stresses in t = 1 month. It is shown that the bound
using upcrossing spectrums as Eq. (9) can well
model the empirical cdf of high levels (at the tail
area).
We turn next to comparisons with the Weibull
fit to the empirical cdf. In Figures 2 and 3, the
logarithmic scale on Y-axis is chosen to check
the agreement of the exceedance probability, i.e.
1–FXm(x), at very high levels.
Figure 2 presents the exceedance probability
obtained from the full-scale measurements. In the
figure one can see that the fitted Weibull distribu-
tion slightly underestimate the extreme response
at the same probability level, while there is a trend
Figure 2. Long term empirical cdf FXm(x) of measured
ship response (solid line); the bound proposed in Eq. (9)
(dashed line) and fitted Weibull distribution (dash-dotted
line).
Figure 3. Long term empirical cdf FXm(x) of simulated
Gaussian response for each sea state, where the response
spectrums are obtained from measurements.
52
MARSTRUCT.indb 52
that the bound given in Eq. (9) converges to the
empirical distribution at high levels.
If ship response at each sea state is assumed to
be Gaussian, the Gaussian response can be simu-
lated using the response spectrum from measure-
ments. Next the same comparisons, as presented
in Figure 2, were performed for the Gaussian
responses and presented in Figure 3. Now, one can
see a very good agreement between the empirical
distribution, fitted Weibull distribution and the
bound from Eq. (9).
2.4 Comparison of methods 1, 2
As derived in the introduction Method 2 proposes
to estimate the return value of response xT, viz. as
a solution of the following equation
1
E ( N1+ ( xT )) = , (10)
T
+
where E ( N1 ( x )) is the expected number of
upcrossings of the level x by the responses in
one year. Obviously the expected value has to be
estimated.
Let assume that one can choose a minimal
length of the stationary periods (sea states) Δt,
e.g. 20 or 30 minutes. Furthermore, one estimate
that in average the ship will encounter K sea states
during a year, i.e. the sailing time is KΔt. Then
the expected number of upcrossings at level x,
E ( N1+ ( x )), is given by
E ( N1+ ( x )) K ⋅ Δt ∫ μ + ( x | W f (W )dW
d , (11)
where μ+(x|W) is the upcrossing frequency of level
x at a sea state W, and Δt is the duration time of a
sea state and assumed to be 1800 seconds. Finally,
the return value of the response is a solution of the
equation
K ⋅ ∆t∫μ+(xT|W)f(W)dW = 1/T . (12)
Next we shall compare the narrow-band estimate
xT of Eqs. (2, 6–8), with the Method 2 estimate,
i.e. solution to Eq. (12). It is easy to see that if the
zero-crossing frequency μ+(0|W) is independent of
sea state, then the two methods would give identi-
cal estimates. Since only severe sea states are essen-
tial for estimation and for those seas the heading
angle 0 thus approximately μ+(0|W) is constant.
Remark: Suppose sea states are characterized by
a single parameter, the significant wave height W,
say. Further let each sea state be experienced by a
ship for the same period of time Δt, e.g. 30 minutes.
If one plans to employ Method 2, i.e. Eq. (3), to
estimate the return response and use the standard
2/18/2011 5:40:40 PM
 decoupling argument, leading to. Eq. (11), then
long term pdf f(W) of sea states is the limit, as sail-
ing time increases, of the normalized histogram of
encountered significant wave height. In contrary, if
one plans to employ Method 1, i.e. Eq. (2), to esti-
mate the return response and use the decoupling
argument, e.g. Eq. (7), then long term pdf f(W)
of significant wave height is biased by the intensity
of the maximums. More precisely, it is the limit, as
sailing time increases, of the normalized (to have
integral equal to one) histogram of encountered
significant wave height scaled by the intensity of
local maximums of the responses. Note that the
two long term pdf are equal if the intensity of local
maximums does not depend on the sea state.
3 THE UPCROSSING INTENSITY
In order to use Method 2 as Eqs. (11–12), upcross-
ing intensity μ+(x|W) has to be estimated for at
least severe sea state. Then the long term pdf of
sea states for the particular shipping needs also be
determined first. In this section we shortly review
computations of μ+(x|W) .
If the joint pdf of ship responses X(t) (zero
mean stresses) and its derivative X (t ) under a sea
state W is known, then the upcrossing frequencies
can be computed by Rice’s formula viz.
∞ denoted by Θ = (m,σx, X ,α3,α4) with skewness
μ + ( |W ) = ∫ ⋅f ( x,, |W ) .
0 X ( ), (t )
However, the joint pdf f(⋅) is often not known
or hard to compute, e.g., for quadratic responses
see Butler et al. [9], Naess [13] . In our previous
work Mao et al. [14], the so-called Laplace Moving
Average (LMA) is shown to be able to model the
non-Gaussian ship responses. It requires knowl-
edge of response power spectrums, skewness and
kurtosis of the stresses. A limitation of the Laplace
model, similar to the second order Stokes Waves, is
that the pdf fX (t ),X (t ) x, z |W ) is not available in an
analytical form (the pdf is defined in the frequency
domain by its characteristic function and has to be
computed using numerical methods).
We will next present cases when the crossing
intensity is given by an explicit analytical for-
mula.
3.1 Up-crossing intensity of Gaussian processes
If ship response at a sea state W is Gaussian, the
up-crossing intensity is computed by Rice’s viz.
1 λ2 ⎛ x2 ⎞
μ + ( |W ) = exp ⎜ − ⎟,
2π λ0 ⎝ λ0 ⎠
MARSTRUCT.indb 53
where λ0, λ2 are the zero-order and second-order
spectral moments of responses X(t). The first two
terms on the right hand side of Eq. (13) are often
referred to as the zero up-crossing frequency
1 λ2
fz μ+ W)= . (14)
2π λ0
However, ship responses are often non-Gaussian
as the real environmental loads, e.g. ocean waves,
show considerable non-Gaussian features, such
as a skewed marginal distribution with heavy
tails. Further, the non-linear interaction between
ships and wave loads can no longer be neglected
for extremely large sea states. Gaussian assump-
tion of ship responses largely underestimates the
extreme values based on the investigation of full-
scale measurements. Hence, an alternative method
is needed to model the expected up-crossings for
the non-Gaussian responses.
3.2 Winterstein’s transformed Gaussian
processes
Winterstein’s transformation is defined by cubic
Hermite polynomials with quadratic terms in
the Hermite model. Providing the stochastic
parameters of response X(t) at a sea state W,
α 3 E [ X (t )3 ] / σ X3 , and kurtosis α 4 E [ X (t )4 ] / σ X4 ,
the transformed Gaussian process is defined by
X(t) = G(u(t)) = m + σX ⋅
[c1H1(u(t))] + c2H2(u(t)) + c3H3(u(t))], (15)
where, Hi are Hermite polynomials and u(t) is a
standard Gaussian process (in what follows, the
mean stress m = 0). The coefficients of the Hermite
polynomials are chosen so that the first 4 moments
of X(t) could match that of the transformed
Gaussian process, more details see Winterstein
et al. [11].
Let G−1 be the inverse function of G, and then
u(t ) = G 1(X (t )). Hence,
⎛ G −1( )2 ⎞
μ + ( |W ) = fz (W )exp ⎜ − ,
⎝ 2 ⎟⎠
1 σ G 1
(X )
fz W ) = , (16)
2π σ G 1
(X )
(13)
Note that fz(W) is defined for the Gaussian
process u(t).
53
2/18/2011 5:40:44 PM
 3.3 Estimation of parameters in transformed
Gaussian model
The parameters of Θ in the transformed Gaussian
model are functions of the encountered sea
state W. The parameters of the non-Gaussian ship
responses were widely studied in Sikora [15] and
Mansour and Wasson [16], and more recently in
Jensen and Mansour [17]. The estimations pre-
sented in these papers are efficient for applica-
tions at ship’s conceptual design stage, since the
parameters can be derived from the ship’s main
dimensions, encountered waves, and operational
parameters.
As is known, wave environments are generally
described by the significant wave height Hs and
wave period Tp. Due to the difficulty in determin-
ing the strongly uncertain parameter Tp, its long Figure 4. Kurtosis of measured responses for both
term conditional distribution given Hs, for exam- mid-section and after-section.
ple DNV [18], can be used to simplify the estima-
tions. Hence, in the following, the sea conditions
are characterized by a single parameter—the
significant wave height Hs. In the previous work
by the present authors, Mao et al. [7], the standard
deviation of response, σX(W), is estimated from Hs
by an explicit relation terms of ship speed U and
heading angles β viz. σX (W) = C(β; U)⋅Hs. The
zero up-crossing frequency, fz(W), is approximated
by the encountered wave frequency.
Finally, only the relation between skewness (and
kurtosis) and the encountered significant wave
height should be further established. Skewness and
kurtosis are measures of non-Gaussianity of the
responses. It is well known that the effects of non-
linear interactions between ship and waves are no
longer negligible for large sea states. Therefore, we
expect skewness and kurtosis to depend mostly on Figure 5. Skewness of wave induced responses and the
the encountered significant wave height, see also whole response inducing high frequency vibration such as
springing and whipping, computed from observations and
Jensen and Mansour [17]. by the formula proposed in Jensen and Mansour [17].
The following investigation is based on the pre-
viously used full scale-measurements. The meas-
urements contain both winter and spring voyages, E(Nt+(x)) was not very sensitive for small variations
so that they can be used to represent the variability of the kurtosis.
of longer term wave environments. For the extreme In order to compute the values of skewness of
response prediction, only stresses under heavy seas responses at all different sea states Wi, the relation
are of interest, here sea states with significant wave between α3 and Hs should also be established. When
height Hs above 4 meters are considered. measurements of ship responses are available, the
The values of kurtosis α 4 for sea states with relation can be easily regressed by, for example,
H s ≥ 4 m, are presented in Figure 4. It shows that a least square method. Alternatively, a numerical
there is no significant trend between kurtosis and analysis is usually used to get the responses when
Hs. Similar conclusion is also derived in Mansour no measurements are available. As is known that
and Wasson [16]. Here, the kurtosis is assumed to the high frequency responses, such as whipping
be 3.4. For some pairs of parameters ( 3 ,α 4 ), the and springing, are very important for the extreme
cubic Hermite polynomials in Eqs. (15–16) does response analysis, so that ship hull should be mod-
always remain monotone. One can resolve this eled as a flexible body for a numerical analysis.
problem by using alternative values of kurtosis. However, this makes the computation extremely
This approach is motivated by an observation time consuming and expensive. Based on the
that the computed expected upcrossing spectrum investigation of full-scale measurements, Figure 5

54

MARSTRUCT.indb 54 2/18/2011 5:40:48 PM

 tells us that only using wave induced responses
are good enough to compute the skewness of real
ship repsonses, since the skewness of wave induced
responses is almost identical with that of the whole
responses. This conclusion is also consistent with
that reported in Jensen and Mansour [17]. Hence,
ship hull is then modeled as a rigid body and even
some commercial software is able to conveniently
compute the wave induced responses.

3.4 Equivalency of the methods 1 and 2

As we have mentioned above, if fz (W) is relatively
constant for severe sea states then the narrow band
estimation (Method 1) of the return response is
equivalent to the Rice’s approach (Method 2). This
is also what we assume in order to be able to com-
pare the methods, see the Remark in Section 2.
In practice the long term cdf of response
maxima (or long term narrow band cdf) is often
approximated by means of Weibull distribution. If
the tails of the long term cdf are well approximated
by the Weibull tails then the Methods 1 and 2
would still be equivalent. In order to facilitate the
comparisons between the two methods we propose
to approximate the expected number of upcrossing
during one year of level x as follows

⎛ ⎛ x⎞k⎞
( N1+ ( x )) ( N1+ (0 ))exp ⎜
E(N
E(
E (N ⎟, (17)
⎝ ⎝ λ⎠ ⎠
where E(N1+(0)) is estimated by means of Eq. (11)
or from the actual measured responses.
The methods will be compared using the time
history from the full-scale measurements at two
places of the 2800TEU container ship. The meas-
ured places are located at the 1/4 ship length for-
ward of after perpendicular (denoted as after
section), and amidships (denoted by mid section),
respectively. The detailed information can be
referred to Storhaug et al. [19].
Firstly, assuming the ship response under each
sea state to be Gaussian processes, the correspond-
ing upcrossing spectrums are computed by Rice’s
formula as Eq. (13). The upcrossing spectrum at all
sea states are then integrated as Eq. (11) to get the
long term upcrossing spectrum in time t. Secondly,
instead of Gaussian assumption, the upcrossing
spectrum at each sea states is estimated by the trans-
formed Gaussian approach as Eqs. (15–16) (also
refer to the Method 2 in this paper). In addition,
when the time history of ship responses is avail-
able, the fitted Weibull distribution is then used to
approximate the long term upcrossings as Eq. (17).
Figure 6 presents the upcrossings computed by the
above 3 approaches, together with the observed
Figure 6. Upcrossing spectrum in time t = 0.5 year
estimated by Eqs. (15–16) (referred to Method 2), by
fitted Weibull distribution as Eq. (17) (referred to as
Method 1), and computed by Eq. (9) assuming Gaussian
response, together with the observed upcrossing for both
after section (a) and mid section (b).
upcrossings. It can be seen that the Gaussian
assumption of ship response largely underesti-
mates the upcrossing spectrum at high levels x. For
the real non-Gaussian ship response, fitted Weibull
approach (Method 1) and transformed Gaussian
approach (Method 2) give almost identical results,
and both are close to the observed upcrossings.
Further, the upcrossing spectrums computed by
Method 1 and 2 converge when extrapolating to
even higher levels. Therefore, both methods are
able to estimate the extreme response for the given
set of data.
4 EXAMPLE OF EXTREME PREDICTION
In this section, the Method 2 will be applied for
the extreme response prediction using the full-
scale measurements through an practical example.

55

MARSTRUCT.indb 55 2/18/2011 5:40:52 PM

 The relation between α3 and Hs is obtained from
the measurements by a linear regression analy-
sis. Method 1 is also applied and its result is
taken as a validation for Method 2. Further, in
Method 2 the significant wave height is a very
important parameter in computing the upcrossing
spectrums, two different sources of encountered
waves are used for comparison.
4.1 Estimation of parameters from measurements
The standard deviation of ship responses σX,
is determined through its relation with Hs, i.e.,
σX = C(β; U)Hs. The values of C(β; U) are com-
puted by a linear strip theory. For the 2800TEU
container ship, its value in terms of heading angle
β under service ship speed U = 10 m/s is given in Mao
et al. [20]. Zero upcrossing frequency of response,
fz, is approximated by the encountered wave fre-
quency. Again, kurtosis is assumed to be 3.4 in
Figure 7 Linear regression of skewness as a function
of significant wave heights. (a): Results for After-section;
(b): Results for Mid-section, also including the skew-
ness computed by the closed expression in Jensen and
Mansour [17].
56
MARSTRUCT.indb 56
the example. In order to make the transformation
remain montone at high levels, we have increase the
value of kurtosis in some cases with large skewness.
It is also checked that the increase will not affect the
computation of upcrossing spectrum.
Skewness of responses under large sea states
against the encountered significant wave height
Hs is plotted in Figure 7. It shows that the value
of skewness will increase with the encountered Hs.
The linear regression method gives the value of
skewness as a function of Hs for both after section,
α 3 _ aftf , and midsection, α 3 _ mid , as follows
α3 _ f 0 11 0 35 α 3 _ mid
id 0.063H s − 0.39.
(18)
In the current study, the skewness models in
terms of other parameters, e.g., heading angle,
are also tested by a linear regression method. But
the more complex models do not explain the vari-
ability of the skewness any better than the simple
regression model as Eq. (18).
Note that for the midsection of the ship, a
semi-analytically closed formula in terms of sig-
nificant wave height and operational profiles (ship
speed and heading angle) was proposed to esti-
mate the skewness of ship responses in Jensen and
Mansour [17]. But as shown in Figure 7(b), there
is a big gap of skewness computed by these two
approaches. In particular for the full-scale meas-
urements, there are a lot of sea states with negative
skewness, and some of them are Gaussian even for
very high sea states. However, in order to check if
the model can be applicable for extreme response
predictions, the skewness regressed from the full-
scale measurements will be used for the following
study.
4.2 Encountered waves
For the computation of upcrossing spectrums,
besides the relation between the stochastic param-
eters Θ and wave environments W (mainly char-
acterized by significant wave height Hs), one also
needs to know the expected number of encountered
sea states K and the long term distribution of Hs.
The first quantity is related to the expected sail-
ing time while the second depends on the ship-
ping. Since the available stress data are measured
in North Atlantic, this region will be considered in
what follows.
The variability of sea environments, here Hs,
has been extensively studied and many databases
are available. (Note that it is not always clear that
the distribution is adequate for the studied prob-
lem; see the Remark in Section 2.) In the following,
the measurements of Hs from the onboard radar
installed on the above 2800TEU container ship
2/18/2011 5:40:55 PM
 are used for the extreme estimation. Further, the
distribution of Hs recommended by DNV [18] is
also used as an input of Method 2, and compared
with the onboard measurements.
The distribution of Hs measured onboard con-
tains much more moderate seas than DNV [18]
recommended, but the probability of high Hs from
DNV is larger than that from the onboard obser-
vations. For extreme analysis, large Hs are of more
interest to estimate the extreme values. Hence, we
expect that using DNV recommended Hs will give
larger values of extreme responses than that using
observed Hs.
The difference of Hs obtained from above
2 approaches could be a consequence of the rout-
ing plan system installed in the measured ship.
However, the difference may be just caused by sta-
tistical errors, since the distribution of observed
Hs is obtained from only half year’s measurements
Figure 8. Results of 100-year stress using the six
months’ full-scale measurements. The expected num-
bers of upcrossings are computed by Method 1 Eq. (17)
(dotted lines), and Method 2 with onboard measured
Hs (dashed lines) and Hs recommended by DNV [18]
(dash-dotted lines). Solid lines represent the observed
upcrossings. Horizontal dash-dotted lines represent the
expected number of upcrossings related to the 100-year
stress. (a): Results for After-section; (b): Results for
Mid-section.
57
MARSTRUCT.indb 57
while the DNV [18] recommended Hs is collected
over many years.
4.3 Results of extreme values
In the following, the so-called 100-year response xT
(T = 100 years) will be estimated by both Method
1 and 2 on the basis of six-month (t = 0.5 year)
full-scale measurements. In this case the expected
number of upcrossings by the level of 100-year
response is equal to t/T = 0.05. If the onboard
observed Hs are used as the input of Method 2, it
is denoted as Method 2(a). While if Method 2 uses
DNV [18] recommended Hs, it is then denoted as
Method 2(b).
Figure 8 presents the expected number of
up-crossings and the estimates of x100 for both after
section and mid section. The expected numbers of
up-crossings computed by Method 1 (Eq. (17))
and Method 2(a) are very close to each other and
converge fast to the observed upcrossings at high
levels. For the after section, the value of x100 is
about 210 Mpa and 230 Mpa estimated by Method
1 and Method 2(a), respectively. For the midsec-
tion, the two methods give almost identical results
of x100, 350 Mpa.
The expected number of upcrossings computed
by Method 2(b) significantly deviates from the
other two methods, in particular for the midsection.
Method 2(b) overestimates more than 40% of the
100-year than Method 1 and Method 2(a). It is due
to that the distribution of Hs used in Method 2(b)
is quite difference from the measurements. Hence,
for extreme prediction during ship’s design stage, it
is extremely important to describe the encountered
waves accurately along its operation period.
5 CONCLUSIONS
This paper presented a simple approach for the
prediction of extreme response, e.g. 100-year stress
x100. In the method, Winterstein’s transformed
Gaussian approach is used to model the non-
Gaussian ship responses. The expected numbers
of upcrossings by the real ship responses are then
computed by Rice’s formula from the transformed
Gaussian processes. The computed upcrossings
are easily applied to estimate the values of extreme
responses. The accuracy of this method is vali-
dated by the typical Weibull fitting method, on the
basis of full-scale measurements of a 2800TEU
container ship.
Parameters of the transformed Gaussian model,
i.e., standard deviation and skewness of stationary
ship responses, are derived as a function of encoun-
tered significant wave height. The relation between
skewness and Hs can be directly computed using
2/18/2011 5:40:58 PM
 only wave induced responses. It can be achieved
through a simple nonlinear numerical analysis
assuming as a rigid ship body.
Finally, the proposed method is conveniently
applicable for extreme estimation with limited
information, mainly encountered significant wave
height Hs. However, due to the strongly relation
between the encountered Hs and estimates of
extreme responses, a correct distribution of
encountered Hs should always be initially deter-
mined for applications.
ACKNOWLEDGEMENTS
The authors acknowledge the support from the EU
project SEAMOCS (Applied Stochastic Models for
Ocean Engineering, Climate and Safe Transporta-
tion) and Gothenburg Stochastic Center and the
Swedish foundation for Strategic Research through
GMMC, Gothenburg Mathematical Modeling
Center. Also many thanks to DNV, crews, manage-
ment company and owner for providing data.
REFERENCES
[1] Coles, S. An Introduction to Statistical Modeling of
Extreme Values. Springer-Verlag, 2001.
[2] Naess, A., Gaidai, O. & Haver, S. Efficient estima-
tion of extreme response of drag dominated off-
shore structures by Monte Carlo simulation, Ocean
Eng. Vol. 34 (16), 2007, pp. 2188–2197.
[3] Winterstein, S.R., Ude, T.C., Cornell, C.A.,
Bjerager, P. & Haver, S. Environmental Param-
eters for Extreme Response: Inverse FORM with
Omission Factors, Proceedings, ICOSSAR-1993,
Innsbruck, Austria.
[4] DNV. Postprocessor for Statistical Response Calcu-
lations. December 15th, 2007.
[5] Cramer, H. & Leadbetter, M.R. Stationary and
Related Stochastic Process: Sample Function Prop-
erties and Their Applications. Wiley, 1967 (Republi-
cation by Dover 2004).
[6] Rice, S.O. The mathematical analysis of Random
Noise. Bell syst Tech J, 1944 & 1945; 23 & 24,
pp. 282–332 and pp. 46–156.
[7] Mao, W., Ringsberg, J., Rychlik, I. & Storhaug, G.
Development of a Fatigue Model Useful in Ship
Routing Design. To be appear in the Journal of Ship
Research.
58
MARSTRUCT.indb 58
[8] Naess, A. & Karlsen, H. Numerical Calculation of
the Level Crossing Rate of Second Order Stochas-
tic Volterra Systems. Probab. Eng. Mech. 2004; 19,
pp. 155–160.
[9] Butler, R., Machado, U. & Rychlik, I. Distribution
of wave crests in non-Gaussian sea. Applied Ocean
Research, 2009; 31, pp. 57–64.
[10] Galtier, T., Gupta, S. & Rychlik, I. Crossings of
Second-order Response Processes Subjected to
LMA Loadings. Journal of Probability and Statis-
tics 2010; Volume 2010, Article ID 752452, 22 pages,
doi:10.1155/2010/752452.
[11] Winterstein, S.R., Ude, T.C. & Marthinsen, T.
Volterra models of ocean structures: extreme and
fatigue reliability. Journal of Engineering Mechan-
ics, ASCE, 1994; 120(6), pp. 1369–1385.
[12] Rychlik, I. & Leadbetter, M.R. Analysis of ocean
waves by crossing and oscillation intensities. Inter-
national Journal of Offshore and Polar Engineering,
2000; 10, pp. 282–289.
[13] Naess, A. Statistical analysis of second-order
responses of marine structures. Journal of Ship
Research, 1985; 29(4), pp. 270–284.
[14] Mao, W., Li, Z., Galtier, T., Ringsberg, J.W. &
Rychlik, I. Estimation of wave loading induced
fatigue accumulation and extreme response of a
container ship in severe seas. In: Proceedings of the
OMAE2010, Shanghai, China, 2010. (Reference:
OMAE2010-20125).
[15] Sikora, J.P. Cumulative Lifetime Loadings for Naval
Ships. Symp. On Hydroelasticity and Unsteady
Fluid Loading on Naval Structures, 1998; Anaheim
CA, Novermber 15–20.
[16] Mansour, A.E. & Wasson, J.P. Charts for estimating
Nonlinear Hogging and Sagging Bending Moments.
Journal of Ship Research, 1995; 39(3), pp. 240–249.
[17 Jensen, J.J. & Mansour, A.E. Estimation of Ship
Long-term wave induced bending moment using
closed-form expressions. Trans, RINA, 2002;
pp. 41–55.
[18] DNV-RP-C205. Environmental conditions and
environmental loads, Recommended Practice. April,
2007.
[19] Storhaug, G., Moe, E. & Piedras Lopes, T.A. Whip-
ping Measurements Onboard a Midsize Container
Vessel Operating in the North Atlantic. Interna-
tional Symposium on Ship Design and Construc-
tion, 28–29th of November, 2007, Shanghai, China,
pp. 55–70.
[20] Mao, W., Rychlik, I. & Storhaug, G. Safety Index of
Fatigue Failure for Ship Structure Details. Journal
of Ship Research, Vol. 54(3), pp.197–208.
2/18/2011 5:40:59 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Utilization of a whole ship finite element analysis from wave

loads to structural strength at real sea state

Yoshitaka Ogawa & Masayoshi Oka

National Maritime Research Institute, Tokyo, Japan

ABSTRACT: The authors developed a whole ship finite element analysis system from wave loads to
structural strength at real sea state. In the present study, methodology for the rational analysis of struc-
tural strength by means of such a whole ship analysis particularly from the viewpoint of loads is dis-
cussed. Consequently, for the rational evaluation of strength in waves, the effect of operation particularly
on wave loads should be considered. It is verified that the evaluation without the effect of operation may
overestimate the stress induced by waves.

1 INTRODUCTION clarified that adequate stress can be evaluated by

It is known that recent some kinds of ships get
longer and faster. As a result, it is difficult for such
ships to apply the empirical background for the
determination of design loads, structural require-
ments and so forth while those for conventional
ships are adjusted by means of the empirical back-
ground.
In addition to this, the transparency of technical
background of design loads and structural require-
ments will be strongly required owing to the adop-
tion of IMO/GBS because such transparencies are
clearly drafted in the functional requirement of
the GBS.
Based on this background, authors developed
a whole ship finite element analysis system from
wave loads to structural strength at real sea state.
In the present study, methodology for the rational
analysis of structural strength by means of such a
whole ship analysis particularly from the viewpoint
of loads is discussed.
Firstly, the whole ship finite element analy-
sis system is developed by the combination with
the computation of nonlinear wave loads. In this
system, time-domain nonlinear strip method is
used for the robust and rational computation of
dynamic pressure. A basic function of the present
system and verification of wave loads, which is one
of dominant factors for evaluation of structural
strength, is indicated.
Secondly, time-domain finite element analysis is
conducted. In this computation, statistic finite ele-
ment analysis by means of the wave pressure distri-
bution in each time step is carried out. An adequate
time steps in one wave encounter period is exam-
ined through the computation in regular wave. It is
the computation with sufficient number of time
steps in an encounter period. It is also clarified that
high stress close to the yield stress can be evaluated
by the present computation although the present
method takes the nonlinear effect of wave loads
into account.
It is considered that there may be certain dis-
crepancies between the computed wave loads and
real situation because such a computation basically
doesn’t take the effect of operation into account.
For the further investigation, finally, the effect
of operation, that is the effects of speed reduc-
tion, change course and operational limitation
(limitation of wave height), on the fatigue strength
was examined. The sensitivity of such effects is
discussed for the rational analysis of structural
strength.
2 WHOLE SHIP FINITE ELEMENT
ANALYSIS
2.1 Computation of wave pressure
on hull surface
Ship motions and wave pressure are computed by
means of the time domain simulation program,
developed by the National Maritime Research
Institute of Japan (Ogawa, 2005). The program,
namely NMRIW (Nonlinear Motion in Regu-
lar and Irregular Waves), is based on a nonlinear
strip method (Bishop et al., 1977; Jensen, 1979;
Yamamoto, 1980; Fujino, 1983; ISSC, 2000).
The NMRIW was developed reflecting the lat-
est results of a seakeeping and manoeuvring
study (Hamamoto, 1993). Forces due to linear
and nonlinear potential flow are combined with

59

MARSTRUCT.indb 59 2/18/2011 5:40:59 PM

 manoeuvring forces and viscous drag forces. It is
well known that it is difficult to compute nonlinear
wave loads in bow and quartering seas by means
of the existing time domain computation method.
Using the present method, wave loads in bow and
quartering seas can be estimated rationally.
Ship motion components, Xj ( j = 1, 2 ..., 6), are
determined from a set of 6 differential equations of
motion with its origin at the centre of gravity. Here
j = 1, 2, 3, 4, 5, 6 refer to surge, sway, heave, roll,
pitch and yaw modes, respectively. With respect to
rotations, a right-handed convention is used. The
equations of motion for a ship traveling with for-
ward speed U are described as
(M )  waves with random phase angles. Irregular wave
ij A ij X j Biijj X j Cij X j
= FW
j FVj i j = 1, 2, 3, 4, 5, 6 (1)
where X  and X  denote motion acceleration and
j j
velocity, respectively. Mij and Aij denote mass and
added mass respectively. Bij denotes damping.
Cij denotes the restoring coefficient. The index,
i, denotes the direction of the fluid force. FWj denotes
the wave exciting force. FVj denotes the excitation
force due to viscous effects.
In terms of global flexible modes, the modal
superposition approach is applied in accordance
with formulation by Yamamoto et al. (Yamamoto,
1980). Equations of motion including flexible
modes are solved in the time domain by means
of a 4th-order Runge-Kutta scheme. The Froude-
Krylov force, which has considerable effect on the
nonlinearity of ship motions, is estimated by the
integration of the hydrostatic and hydrodynamic
wave pressure along the instantaneous wetted sur-
face of the hull at each time step.
With respect to the sectional wave radiation force
and potential value at each time step, the integral
equation method is utilized. Source and doublet are
distributed at the origin of each section to avoid the
irregular frequency, in accordance with Ohmatsu’s
method (Ohmatsu, 1975). The sectional diffraction
force, in the present method, is computed by solv-
ing the Helmholtz equation at each time step.
Wave impact load due to slamming is com-
puted by means of the displacement potential
approach (Takagi, 2007; Ogawa, 2009) in terms of
the instantaneous wetted surface at each time step.
The viscous effect of roll damping due to ship hull
and bilge keels is estimated using various empiri-
cal formulae. The propeller thrust is described by
means of the propeller characteristics. The hull
resistance is a function of the instantaneous speed
and draft. Lateral force and yaw moment due to
rudder is considered to keep a target course o for
the ship in the simulation. The rudder is controlled
by the PID control as follows:
60
MARSTRUCT.indb 60
t
de
δ = − K P e + K I ∫ edt + K D (2)
dt
0
e =ψ ψ0
where δ denotes the rudder angle. e denotes the
deviation of the ship course. ψ and ψ0 denote the
instantaneous course and target course respec-
tively. KP, KI and KD denote the proportional gain,
integral gain and differential gain, respectively.
NMRIW can compute the wave pressure in both
regular and irregular waves. The sea surface and
wave kinematics are described based on the linear
wave theory. The sea surface of irregular waves is
described by the linear superposition of regular
was realized by the sum of 200 components of
waves in accordance with the Jonswap spectrum.
To obtain the stable results, 10 times simulations
with the duration of 3600 seconds of ship scale were
carried out in each condition. The combination of
phase angle of each wave component was varied
in each simulation. Figure 1 shows the example of
the spectrum of incident wave. It is found that the
planed spectrum, which is shown as “Base”, is the
same as the computed spectrum, which is derived
from the time history of computed incident wave
and is shown as “Cal.”. It is found that present
computation method can adequately realize the
wave spectrum, which is the basis for the utiliza-
tion of direct computation for the clarification of
statistical value of ship response in waves.
Figure 2 shows the example of the time history
of vertical bending moment at midship and S.S. 7.5.
It is also found that two-node vibration owing to
the whipping occurs with natural frequency of
the present container ship. It is clarified that the
present method can execute the robust computa-
tion of wave loads including whipping vibration.
Present computation method has been verified
through the comparison with experiments of many
kinds of ship (Ogawa, 2005 & 2007). In addition
7
Spectrum (T02 = 8sec., H = 5m)
Base
6
Cal.
5
4
S(ω)
3
2
1
0
0 0.5 1 1.5
ω
2 2.5 3 3.5
Figure 1. The example of spectrum of incident
wave (significant wave height:5 m, mean wave period:
8.0 sec.).
2/18/2011 5:41:00 PM
 Wave vertical bending moment (Head seas, Hw = 9m, Fn = 0.219) Wave torsional moment (wave height :6m)
400000
S.S. 5 0.0012 Cal. (Fn = 0.11)
300000 S.S. 7.5 Cal. (Fn = 0.164)
Wave vertical bending moment

0.001
Exp. (Fn = 0.11)
200000
Exp. (Fn = 0.164)
0.0008
100000
(ton-m)

Tx/ρgζL B
0.0006

2
0
300 320 340 360 380 400 0.0004
-100000
0.0002
-200000
0
-300000 0 0.5 1 1.5
Time (sec.)
λ/L

Figure 2. Time history of the wave vertical bend- Figure 4. The wave torsional moment response ampli-
ing moment of the post-panamax container ship at tude operator of the mega container ship at S.S.5.5
S.S.7.5 and S.S.5 (χ = 180 deg., Fn = 0.219, Hw = 9 m, (χ = 150 deg., wave height: Hw = 6 m).
λ/L = 1.0).

Wave vertical bending moment (wave height :6m)

Cal. (Head seas, Fn = 0.11)
Cal. (Head seas, Fn = 0.164)
0.03 Cal. (Bow seas, Fn = 0.164)
Exp. (Head seas, Fn = 0.11)
Exp. (Head seas, Fn = 0.164)
0.025 Exp. (Bow seas, Fn = 0.164)

0.02
MV/ρgζL2B

0.015

0.01

0.005

0
0 0.5 1 1.5
λ/L

Figure 3. The wave vertical bending moment response

amplitude operator of the mega container ship at S.S.5.5
with various ship speed and headings (wave height: Figure 5. Example of wave pressure distribution on
Hw = 6 m). finite element model by means of the present computa-
tion (GUI of NMRIW).

to the previous results, present method is further

verified through the comparison with experiments experiments and computations, it is found that the
of the large container ship. torsional moment in the stern quartering seas was
Details of the experiments have already been significant in the stern quartering seas compared
published (Oka, 2009). Figure 3 shows example of with that in bow quartering seas.
validation for the wave vertical bending moment in It is found that the present method can explain
head and bow seas, for the ship travelling at various the measured wave torsional moments, which were
speeds. The fundamental frequency component, measured accurately using the newly developed
Mv, is divided by ρgςBL2. Here, ρ and g describe backbone model (Oka, 2009). It is confirmed that
the density of fluid and acceleration of gravity, the present method, which can estimate roll motion
respectively. ς, B and L describe wave amplitude, in the stern quartering seas adequately, estimates
ship breadth and ship length, respectively. not only the vertical bending moment but also the
It is found that the computed wave bending wave torsional moment in various wave condition.
moment of the mega container ship is in ample
agreement with experiments. It is also found that 2.2 Whole ship finite element analysis
ship speed and wave encounter angle have effect on
the bending moment. It is clarified that the present The present system computes a time history of
method can explain the effect of ship speed and stress in regular and irregular waves. In this com-
wave encounter angle on wave load quantitatively. putation, static finite element analysis by inputting
Figure 4 also shows example of validation for the computed wave pressure distribution, which is
the wave torsional moment in bow seas, at vari- shown in Figure 5, in each time steps are carried
ous ship speeds. The fundamental frequency com- out. By means of the present system, outer shell
ponent, Tx, is divided by ρgςBL2. Based on both of a subject ship can be extracted automatically.

61

MARSTRUCT.indb 61 2/18/2011 5:41:02 PM

 Figure 6. Example of wave pressure distribution
on finite element model by means of the present
computation.

Figure 6 shows an example of the wave pressure Figure 7. The whole ship finite element analysis in the
distribution on finite element model. It is found regular head seas (Stress distribution under the hogging
condition) (N = 12, Head seas, wave height:12 m, wave
that wave pressure is loaded only on the outer length ratio to ship length:0.8.).
shell.
The NASTRAN software is used for the present
finite element analysis. Prior to the computa-
tion, density of container was tuned to adjust the
weight, center of gravity and radius of inertia of a
finite element model. Dynamic forces and inertia
of a finite element model is adjusted by means of
the inertia relief function of the NASTRAN.

3 EXAMINATION OF RESULTS
OF FNITE ELEMENT ANALYSIS

3.1 Examination of time step of time-domain

analysis
Firstly, an adequate time step in one wave encoun-
ter period is examined through the computation in
regular wave. In the present study, the computed
stress of the large container ship is examined to
clarify the adequate step with the variation of time
step “N” from 12 to 120. Figures 7 and 8 are exam-
ples of the stress distribution under the hogging
condition in the case of N = 12 and 40. It is con-
firmed that stress distribution in the case of N = 40
are same as that in the case of greater number of
step. It is clarified that stress distribution converges
at N = 40 in various wave period. It is found that
stress amplitude in waves is 10% different between
in the case of N = 12 and in the case of N = 40.
It is confirmed that the present analysis system
with sufficient number of time step can evaluate
the stress amplitude in waves adequately.
3.2 Consideration of the computed stress
In the meanwhile, it is also clarified that high
stress close to the yield stress can be evaluated
by the present computation although the present
method takes the nonlinear effect of wave loads
into account. It is considered that the validity of
Figure 8. The whole ship finite element analysis in the
regular head seas (Stress distribution under the hogging
condition) (N = 40, Head seas, wave height:12 m, wave
length ratio to ship length:0.8.).
inputs of wave pressure should be clarified. It is
supposed that there is certain discrepancy between
computed wave loads and real situation.
Figure 9 shows the example of the long term
prediction of wave vertical bending moment of a
large container ship. Response amplitude operator,
which is basis for these long term values, is com-
puted by means of the linear strip method. It is
well known that probability of occurrence of once
in 25 years correspond to 10–8. In the meanwhile,
probability of occurrence of the design wave load,
determined by the IACS UR-S11 (IACS, 2006), of
this container ship corresponds to 10−6. It is found
that there is certain discrepancy in wave loads
related to those two probabilities.
For the consideration of design wave loads for
existing ship, that discrepancy is adjusted based on

62

MARSTRUCT.indb 62 2/18/2011 5:41:05 PM

 Long term prediction of vertical bending moment (S.S.5, Fn = 0.18) Vertical Bending Moment (Fn = 0.182, T02 = 15sec. H1/3 = 10m)
(Post-panamax container ship)
0.002 0.06 Midship
? = 180deg
S.S. 7 1/2
? = 90deg
0.0016 0.04
? = 0deg
All Headings
0.02
0.0012
Mv/?gBL3

Mv/ρgζBL2
0
0.0008 600 1100 1600 2100 2600
-0.02
0.0004
-0.04
0
-8 -7 -6 -5 -4 -3 -2 -1 0 -0.06
log10Q t(sec.)

Vertical Bending Moment (Fn =0.182, T02 = 15sec. H1/3 = 10m)

Figure 9. Long term prediction of vertical bending 0.06
Midship
moment of a large container ship computed by means of S.S. 7 1/2
0.04
linear strip method.
0.02

Mv/ρgζBL2
the voyage records. However, it is considered that it 0
1400 1410 1420 1430 1440 1450 1460 1470 1480
is difficult for new type of ship to compensate such -0.02
a discrepancy based on a voyage re-cord. There-
-0.04
fore, it is important to evaluate wave loads ration-
ally and to assess the effect of operation on it by -0.06
t(sec.)
means of a direct computation.
Figure 10. The example of time history of wave ver-
tical bending moment (χ = 180 deg., significant wave
4 EVALUATION OF THE EFFECT height:10 m, mean wave period:15.0 sec.) (Below: magni-
OF OPERATION ON WAVE LOADS fication of above time history of 2000 second within one
hour duration).
4.1 The effect of the change of ship speed
on the short term probability of wave loads
Sagging (T02 = 15sec., H1/3 = 10m)

Figure 10 shows the example of the computed time 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
1 Without speed reduction (Fn = 0.182)
history of vertical bending moment. It is found With speed reduction
that the hull girder vibration owing to the whip- Wave component only (without hull girder vibration)
ping occurs frequently in such a severe sea state. 0.1
It is confirmed that two-node vibration occurs with
natural frequency of the present container ship.
P

It is also found that the large bending moment can 0.01

be induced due to whipping in the sever sea state.
Figure 11 shows the relation of speed reduc-
tion with the probability of wave vertical bending 0.001
Mv/ρgζBL2
moment. This probability is derived from the histo-
gram of the computed sagging moment. The value
Figure 11. The relation of ship speed with the prob-
of horizontal axis denotes the non-dimensional ability of occurrence of wave vertical bending moment
value divided by ρgςBL2 (in this case, ς is the at midship (χ = 180 deg., significant wave height:10 m,
significant wave amplitude). The solid line in mean wave period:15.0 sec.).
Fig. 11 shows the probability of occurrence with-
out speed reduction. This means that computation
is carried out with constant ship speed. For the the probability of occurrence of wave loads. It is
evaluation of the effect of hull girder vibration, clarified that it is important to consider the effect
wave loads taking only rigid motion into account hull girder vibration on the statistical characteris-
is computed in the same irregular waves. This com- tics of wave loads.
putation derives the wave loads without hull girder However, in the real navigation, ship speed
vibration. The probability of wave loads com- is reduced owing to the nominal speed loss and
puted without hull girder vibration is also shown deliberate speed loss. Particularly, those speed
in Fig. 11 as the small dotted line. loss become significant in rough seas because ship
Although the number of peak value is differ- motion and wave loads become large. In addi-
ent owing to the hull girder vibration, it is found tion to this, in the real navigation, ship course are
that hull girder vibration has a certain effect on changed to avoid severe condition particularly in

63

MARSTRUCT.indb 63 2/18/2011 5:41:07 PM

 rough seas. In the meanwhile, usual computations
for the evaluation of statistical characteristics of
ship motion and wave loads have been generally
conducted under the assumption that ship speed
and course are constant. Therefore, it is inadequate
to assess the effect of hull girder vibration on wave
loads quantitatively based on the computation and
the model experiments with constant ship speed
and course. This implies that computation and
model experiments taking the effect of operation
should be conducted.
Using the present computation method, the
probability of occurrence with speed reduction is
computed. It is shown as the dotted line in Fig. 11.
In this computation, ship speed was reduced to
two third of initial ship speed when the pitching
motion becomes large in the computation.
It is found that the probability with speed reduc-
tion becomes smaller than that without speed
reduction because the whipping induced hull girder
vibration is significantly reduced owing to the
speed reduction. It is clarified that the evaluation
without the effect of operation may overestimate
the effect of hull girder vibration on statistical
characteristics of wave loads quantitatively.
Figure 12 shows the example of the computed
probability of occurrence of wave vertical bending
moment with various ship speeds. This probabil-
ity is derived from the histogram of the computed
sagging and hogging moment separately. It is
found that the occurrence probability of sagging
moment is slightly different from that of hogging
moment because the effect of hull girder vibration
and nonlinearity of ship motion are not significant
in 5 m wave height. In the meanwhile, it is clari-
fied that the effect of ship forward speed on the
probability is significant. In the real navigation,
criteria for the speed loss and the course change
are based on not only hull girder vibration but also
V.B.M. (T02 = 15sec., H1/3 = 5m)
1 0
0 0.005 0.01 0.015 0.02
0.1
P
0.01
0.001
Mv/ρgζBL^2
Figure 12. The effect of ship speed on the probability
of occurrence of wave vertical bending moment at mid-
ship (χ = 180 deg., significant wave height:5 m, mean
wave period:15.0 sec.).
MARSTRUCT.indb 64
large motion and acceleration. This implies that
the statistical characteristics of wave loads should
be also evaluated in various wave conditions that
hull girder vibration doesn’t occur.
Based on the results in Figures 11 and 12, it is
verified that the setting of ship speed has much
effect on the probability. Therefore, it is important
to compute wave loads taking the effect of speed
reduction into account.
4.2 The effect of course change on the short
term probability of wave loads
Figure 13 shows the relation of course change
with the probability of wave vertical bending
moment. The solid line in Fig. 13 shows the prob-
ability of occurrence without course change. This
means that computation is carried out with con-
stant headings. In the meanwhile, the dotted line
shows the probability of occurrence with course
change. In this computation, heading angle was
varied from head seas (180 deg.) to bow quartering
seas (170 deg.) when the pitching motion becomes
large in the computation. It is found that the prob-
ability with course change becomes smaller than
that without course change because the whipping
induced hull girder vibration is certainly reduced
owing to the course change. It is clarified that the
direct computation can explain the effect of course
change on the probability rationally.
In addition to them, probability of wave loads
at bow seas (170 deg.) computed without course
change is also computed. This is shown as the
small dotted line in Fig. 13. Although the resulting
heading is same as bow seas (170 deg.), it is found
that the probability at bow seas is different from
that with course change in head seas. It is clari-
fied the evaluation of statistical characteristics of
wave loads with constant course may be different
Sagging (T02 = 15sec., H1/3 = 10m)
0.005 0.01 0.015 0.02 0.025 0.03 0.035
0.025 0.03 0.035 1 Without Course change (Head seas only)
With Course change
Sagging(Fn = 0.182) Bow seas (170 deg. ) without course change
Hogging(Fn = 0.182)
Sagging(Fn = 0.12) 0.1
Hogging(Fn = 0.12)
0.01
0.001
Mv/ρgζBL2
Figure 13. The relation of wave encounter angle with
the probability of occurrence of wave vertical bend-
ing moment at midship (χ = 180 deg., significant wave
height:10 m, mean wave period:15.0 sec.).
64
2/18/2011 5:41:09 PM
 from the real situation. Therefore, it is important
to compute wave loads taking the effect of course
change into account.
In the meanwhile, it is considered that further
quantitative examination of the degree of speed
reduction is required in the future study. Particu-
larly, the relation of the criteria for the speed loss
and the course change with statistical character-
istics of wave loads should be examined further
although it is clarified that the direct computation
can explain the effect of speed reduction on the
probability rationally.
5 CONCLUSIONS
Through the whole ship finite element analysis
from load to structural strength at real sea state,
the effect of operation on wave loads and strength
is examined. Conclusions are as follows:
1. The direct computation by means of the present
method can explain wave loads in various
wave height, wave period and wave direction,
rationally.
2. The present analysis system with sufficient
number of time step can evaluate the stress
amplitude in waves adequately.
3. In the meanwhile, for the rational evaluation of
the stress in waves, the effect of operation on
wave loads should be considered. It is verified
that the evaluation without the effect of opera-
tion may overestimate the stress induced by
waves.
ACKNOWLEDGEMENTS
A part of the present study was supported by a
Grant-in Aid for Scientific Research of the Japan
Society for Promotion of Science (No. 20360400).
REFERENCES
Bishop, R.E.D., Price, W.G. & Tam, P.K.Y. 1977.
A unified dynamic analysis of ship response to waves,
Transaction of Royal Institute of Naval Architects,
Vol. 119, pp. 363–90.
Fujino, M. & Chiu, F. 1983. “Vertical Motions of High-
speed Boats in Head Sea and Wave Load”, J Soc
Naval Arch Japan, Vol. 154, pp. 151–163.
65
MARSTRUCT.indb 65
Hamamoto, M. & Kim, Y.S. 1993. “A New Coordinate
System and the Equations Describing Manoeuvring
Motion of Ship in Waves”, J Soc Naval Arch of Japan,
Vol. 173, pp. 209–220.
IACS, “Longitudinal strength standard”, 1993. IACS
S-11.
IACS, 2001. “Standard Wave Data”, IACS Recom-
mendation No. 34.
ISSC, 2000. “Extreme hull girder loading”, special task
committee VI. 1, 14th international ship and offshore
structures congress, Nagasaki, Japan, pp. 263–320.
Jensen, J.J. & Pedersen, P.T. 1979. “Wave-induced bend-
ing moments in ships—a quadratic theory”, Transac-
tion of Royal Institute of Naval Architects, Vol. 121,
pp. 151–165.
Ogawa, Y. et al., 2005. “The effect of a bow flare shape
on the water impact pressure”, International Journal
of Offshore and Polar Engineering (IJOPE), Vol. 16,
No. 2.
Ogawa, Y. 2007. “A study on nonlinear wave loads of a
large container carrier in rough seas”, 10th Interna-
tional Symposium on Practical Design of Ships and
other Floating Structures (PRADS), 1, pp. 132–140.
Ogawa, Y. & Takagi, K. 2009. “An evaluation of whip-
ping vibration utilizing the displacement potential
method”, Hydroelasticity 2009, pp. 213–222.
Ohmatsu, S. 1975. “On the Irregular Frequencies in
the Theory of Oscillating Bodies in a Free Surface”,
Papers of Ship Res Inst, Tokyo, Vol. 48, pp. 1–13.
Oka, M., Oka, S. & Ogawa, Y. 2009. “An experimental
study on wave loads of a large container ship and its
hydroelastic vibration”, Proc. 4th Int. Conf. Hydroe-
lasticity in Marine Technology.
Takagi, K. & Ogawa, Y. 2007. “Flow Models of the Flare
Slamming”, Proceedings of International Conference
on Violent Flows (VF-2007).
Yamamoto, Y., Fujino, M. & Fukasawa, T. 1980. “Motion
and longitudinal strength of a ship in head seas and the
effects of nonlinearities”, Naval Architects and Ocean
Engineering, Journal of Society of Naval Architects of
Japan, Vol. 18.
2/18/2011 5:41:10 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Environmental and operational uncertainties in long-term

prediction of slamming loads of containerships

J. Parunov, M. Ćorak & I. Senjanović

Faculty of Mechanical Engineering and Naval Architecture, University of Zagreb, Zagreb, Croatia

ABSTRACT: The purpose of the paper is to investigate long-term distribution of slamming loads of
containerships accounting for different types of environmental and operational uncertainties. Following
uncertainties are studied: the choice of the wave scatter diagram, the effect of the avoidance of heavy
weather, the effect of the maneuvering in heavy weather and the uncertainty of the method for predic-
tion of the long-term extreme slamming pressure. Long term distributions of bottom slamming pres-
sures are computed using different combinations of the aforementioned uncertainties. The purpose of
the study is the uncertainty assessment for the application in the reliability based design of ultra-large
containerships.

1 INTRODUCTION • the lifetime weighted sea method,

In the design and operation of ultra-large con-
tainerships important hydroelastic effects appear
in addition to the rigid body response. The most
relevant hydroelastic phenomenon concerning the
longitudinal strength of large containership is whip-
ping, the transient vibration of ship hull occurring
as a consequence of slamming. Such vibration may
considerably increase the extreme vertical wave
bending moments amidships and thus needs to be
considered in the ship structural design.
Estimation of design slamming parameters, as
the frequency of the slamming occurrence and
design slamming pressures is demanding task
depending on numerous uncertainties. The reason
for this is that slamming phenomenon is very sen-
sitive to the environmental conditions, ship speed
and heading angle. These parameters depend on
assumed shipping route, the ship master’s actions
to avoid heavy weather and on the maneuvering in
heavy weather. These assumptions can not be set
with large confidence that makes design slamming
loads quite uncertain.
This paper aims to quantify influence of men-
tioned uncertain parameters on the long-term
extreme slamming pressures. The effect of the fol-
lowing uncertainties is studied:
• design wave environment,
• avoidance of heavy weather,
• maneuvering in heavy weather.
Furthermore, two methods for calculating long-
term extreme slamming pressures are compared:
• the long-term prediction via Poisson outcrossing
method.
Two types of slamming loads appear in contain-
erships: the bow flare slamming and the bottom
slamming. Although both types of the slamming
are important and can excite hull-girder vibration,
only the bottom slamming is considered in the
present study.
It is assumed that the bottom slamming pres-
sure psl is proportional to the square of the relative
velocity v of the bottom and the wave surface:
1
psl ρk
kv 2 (1)
2
Critical (threshold) value of the impact velocity
is assumed according to Ochi and Motter (1973).
v0 0 093 gL (2)
It is worth mentioning that the threshold veloc-
ity given by Equation (2) is used by Jensen et al.
(2008) in recently published study of the wave
induced hull girder loads on containerships. The
associated threshold pressure takes the form:
1
p0 kv02
k (3)
2
The formulation of the problem is set in such
a way to avoid assumption on the method for
calculating slamming pressure coefficient k. The
approach is elaborated in Section 2.

67

MARSTRUCT.indb 67 2/18/2011 5:41:11 PM

 Table 1. Main particulars of containerships. where
Ship particular Flokstra S175 1
λ= (6)
Lpp m 270 175 ρkσ r2
B m 32.2 25.4
T m 10.85 9.5 Slamming pressure may be calculated using
Tballast m – 7.0 pressure coefficient k = 1. This is possible, as pres-
CB – 0.598 0.572 sure as well as 1/λ in the exponential part of the
Δfull tonnes 57499 24742 Equation (5) is proportional to k and therefore k
Δballast tonnes – 17157 can be cancelled. Actual extreme pressures can be
Vn knots 24.5 22.15 obtained by simply multiplying calculated unitary
pressure values by actual value of the pressure
coefficient k.
Two well-known containerships are analyzed in Long-term distribution of slamming pressure psl
the present study: can be modeled as series of short term sea states
where each heading angle β have the same probabil-
• Flokstra containership,
ity of occurrence. Probability distribution of slam-
• S175 containership.
ming pressure in short term is given by Equation (5)
Results of model tests of two ships are well doc- and the long-term distribution is given as:
umented and available in the references Flokstra
(1974) and Wu et al. (2002). Main particulars of nβ
Δβ
the ships are specified in Table 1. Fp ( psl ) = ∑
The hydrodynamic assessment is performed by i =1 2 ⋅π
⎛ nH ⎞
( ) r (T β ) ⋅ p (H )
the linear strip theory. Although more sophisticated nT

3D hydrodynamic tools are available nowadays, ⋅⎜ ∑F i j ,k sl Sj ,,TZk ⎟

Zk βi Zk i Sj
for the purpose of comparative study performed ⎝ jk ⎠
herein, the linear strip theory is considered to be (7)
a convenient tool. The hydrodynamic strip models where p(HS,TZ) is the probability of occurrence
of the containerships and their validations are pre- of sea state, while the relative number of slam-
sented by Parunov & Ćorak (2010). ming appearances in each short-term sea state is
The present paper is organized in such a way that given as:
methods for long-term slamming pressure calcula-
tion are described firstly. After that, the effect of υ ssl
different environmental and operational uncertain- r ( TZ , β ) = *
(8)
ties on the long-term extreme slamming pressures υ ssl
is studied. The number of slams in 20 years is also
estimated. Finally, paper ends with conclusions where υsl is the slamming frequency in each indi-
summarizing obtained results. vidual short-term sea state, given as:

1 σV
υ sl = ⋅ ⋅ Psslam (9)
2 CALCULATION METHODS OF LONG- 2π σ M
TERM EXTREME BOTTOM SLAMMING
PRESSURE while the mean slamming frequency in all sea states
is given as:
2.1 The lifetime weighted sea method for long- nβ
term distribution of slamming pressure Δβ
υ sl = ∑
*

Impact pressure psl in the short-term sea state is a i =1 ⋅ π

2
⎛ nH nT ⎞
( ) ( )
random variable which follows the probability law
given as truncated probability density function in ⋅ ⎜ ∑ υ sl i j ,k HS j TZk , βi ⋅ p HS j TZk ⎟
the exponential form (Ochi and Motter 1973): ⎝ jk ⎠
(10)

f ( psl ) = λ ⋅ e ⎣
(
⎡ − λ ⋅ p0 psp ⎤
⎦) (4) Model tests and full-scale trials show that nec-
essary and sufficient conditions for occurrence of
psp slam impact are the bottom emergence, i.e. rela-
(
F psl HS TZ , β = ) ∫ f ( p) dp = 1 − e ⎡⎣−λ ⋅ ( p
sl p )⎤⎦ tive bow motion being larger than local draught d,
and impact velocity being higher than the critical
p0
(5) velocity v0.

68

MARSTRUCT.indb 68 2/18/2011 5:41:12 PM

 Relative bow motion and velocity are statistically ln(- ln( q ) b
independent random processes which follow pext +e a (17)
Rayleigh distribution (Senjanović et al., 2003).
Based on that fact, probability of slamming can
be written as: 2.2 Long-term extreme slamming pressure via
Poisson outcrossing method
⎛ ⎛ d2 v2 ⎞⎞
⎜ −⎜ + 02 ⎟⎟ The extreme distribution Fext(pext|HS,TZ,β;tsea), of
Pslam (M > d V > v ) ⎝ ⎝ 2⋅σ M 2⋅σV
2
⎠⎠ the slamming pressure over the stationary period
e (11)
of time of a single sea state tsea, conditional on the
sea state defined by the three parameters (HS, TZ, β)
where σ M2 and
σV2 are variances of the relative is obtained via the Poisson outcrossing approxima-
motion and the relative velocity, respectively, in a tion. The appropriate value for tsea depends on the
short term sea state. These are different for each geographical location and is considered to vary
mean zero crossing period TZ, heading β, ship between 1800 and 43200s. Here a value of 10800s
speed and loading condition. Short term sea states (3 hours) is assumed.
are modeled by Pierson–Moskowitz wave spec- First step is evaluation of the mean outcrossing
trum, while σ M σV
2 and 2 are given as:
rate υp, for the value of the extreme slamming pres-
sure pext, based on the exponential function:
∞
σM
2
= ∫ SM ( x ) dω (12)
⎡ − λ ⋅( pext p )⎤⎦
0 υp ( ext S Z β ) υ sll e ⎣ (18)
∞
σV2 = ∫ SV ( x ) dω (13)
where υsl is the slamming frequency in the short
0
term sea state calculated according to (9). Distri-
After establishing long-term distribution in dis- bution of extreme slamming pressures pext over the
crete form by applying Equation (7), the theoreti- sea state duration is than given as:
cal three–parameter Weibull distribution may be
fitted:
( −υ ( p ) )
⎛ p −ε ⎞
α
(
Fext pext HS TZ , β ;tsea ) e p eext
xt HS TZ ,β tsea
(19)
− ⎜ sl
⎝ ϑ ⎟⎠
FWeibulll ( psl ) = 1 − e (14)
By unconditioning extreme value distribution
in the short term conditional on sea state and
where ϑ, α and ε = p0 are the Weibull scale heading angle, one obtains extreme distribution
parameter, shape parameter and location of slamming pressure for all sea states for the
parameter respectively. Probability FWeibull given assumed sea state duration tsea:
by Equation (14) represents the probability that
nβ
the slamming pressure is less than psl in one, ran- Δβ
domly chosen wave cycle. The Weibull 3P distribu- Fext ( pext ;tsea ) = ∑
i =1 ⋅ π
2
tion may be presented in the linear scale, and then
⎛ nH nT ⎞
Weibull parameters may be calculated by the least
square method as:
⋅ ⎜ ∑ Fi
⎝ jk
j ,k ( pext HS TZ , β;ttsea ) (
p HS j ,TZk ⎟
⎠
)
⎛ b⎞ (20)
−
⎝ a⎠
α , ϑ =e (15)
Finally, the extreme distribution of pext over
where a and b are parameters of the straight line the long-term return period (e.g. TC = 20 years) is
fitted. For the long-term return period (e.g. TC = 20 obtained by the order statistics:
years), the probability of exceeding the most prob-
Fext ( pext ;T
TC ) xt ( pext tsea ) ⎤
nsea
able extreme slamming pressure is given as: ⎡⎣ Feext (21)
⎦
1 1
q= = (16) where the assumption is made that, in the long
N sl T ⋅ υ * term, the extremes reached in different sea states
C sl
are independent random variables and where the
The most probable extreme impact pressure for number of stationary sea states in the period TC,
given return period can then be calculated as: nsea, is given by TC /tsea.

69

MARSTRUCT.indb 69 2/18/2011 5:41:14 PM

 Extreme distribution obtained numerically by The course changes in heavy weather are mainly
described procedure may be fitted by the theoreti- to avoid the ship capsizing or excessive ship roll-
cal Gumbel distribution. Probability that the slam- ing amplitudes that may interfere with normal
ming pressure remains less than a given value pext working activities on board. Consequently, the
over a long time period TC, is then given as: probability of head seas is much higher in heavy
weather than in normal sea conditions. How-
⎛ p − p* ⎞
ever, this is valid only for smaller ships (less than
− ⎜ ext ⎟
⎜⎝ λ ⎟⎠ about 200 m in length). For large ships, the course
FGumbel ( pext ) = e −e (22) changes in heavy weather are not so frequent.
The explanation for this finding could be that the
masters of larwge ships feel safe even in rather
where p*and λ are Gumbel distribution parameters. rough seas.
If Gumbel distribution is presented in the linear Another important maneuver in rough seas is the
scale, then Gumbel parameters can be calculated speed reduction. This action is not dependent on
by the least square method as: the ship size. Speed could be reduced due to techni-
cal reasons, such as the added resistance of wind
1 and waves, change of submerged part of the hull,
λ , p* λ (23) change of wake field and loss of thrust. Very large
a
motion amplitudes, velocities and accelerations,
slamming, green seas, overload of the main engine
where a and b are parameters of the fitted straight could be the reason for voluntary speed reduction.
line. Finally, the most probable extreme impact The effect of the speed reduction in heavy
pressure over the long-term return period is equal weather is assessed by analyzing three different
to the Gumbel parameterp*, i.e. pext = p*. ship speeds:
• Nominal ship speed (Vn)
• Reduced speed –80% of the nominal speed
3 RESULTS OF THE LONG-TERM
• Zero speed.
ASSESSMENT OF EXTREME BOTTOM
SLAMMING PRESSURE

In order to assess the influence of the shipping Table 2. The most probable bottom slamming pressures
route on the extreme bottom slamming pressures, in 20 years according to the lifetime weighted sea
long-term calculations are performed for following method (k = 1).
scatter diagrams:
The most probable extreme slamming pressure
• Standard IACS North Atlantic scatter diagram
Floxtra S175
(IACS)
• North Pacific scatter diagram (N-P) Wave scatter
• Scatter diagram for shipping route North diagram Ship speed FL FL BL
Europe–Far East through Suez Chanel (Suez).
IACS vn 237 242 278
More details on the wave zones considered and 0.8 vn 201 207 236
the resulting extreme slamming pressures are pre- zero 60 40 61
sented in Section 4. N-P vn 179 223 275
The next issue covered in the paper is the effect 0.8 vn 152 191 236
of the heavy weather avoidance. The problem is zero 49 37 64
approached by the truncating probability density Suez vn 170 203 257
function of a significant wave height, as explained 0.8 vn 142 172 221
in Section 5. The truncation of the scatter diagrams zero 48 39 69
results in two additional scatter diagrams for which
long-term calculations are performed: N-P modified vn 170 214 267
0.8 vn 145 185 230
• Modified N-P scatter diagram, zero 56 47 79
• Modified Suez scatter diagram. Suez modified vn 147 199 254
0.8 vn 128 175 223
Two main maneuver that master can undertake zero 59 53 89
when the ship is in heavy seas are course changing
from beam seas to head or following seas and vol- * FL – full load condition
untary speed reduction (Guedes Soares 1990). BL – ballast condition

70

MARSTRUCT.indb 70 2/18/2011 5:41:19 PM

 Table 3. The most probable bottom slamming pressures 1.40

in 20 years according to the Poisson outcrossing method 1.30

(k = 1, short-term duration of sea states −3 hours). N-P/N-P modified (full load)

1.20 SUEZ/SUEZ modified (full load)

The most probable extreme slamming pressure
1.10

Floxtra S175 1.00

Wave scatter diagram Ship speed FL FL BL 0.90

0.80
IACS vn 277 327 423
0.8 vn 227 265 337 0.70

zero 58 38 60
Figure 2. Influence of the avoidance of the heavy weather
N-P vn 181 248 333 on the slamming pressure for full load condition.
0.8 vn 167 208 273
zero 48 36 64
Suez vn 169 217 293 10.0

0.8 vn 149 183 241 9.0

zero 47 38 69 8.0

N-P modified vn 173 241 330 7.0

0.8 vn 155 204 275
6.0
zero 54 45 79
5.0
Suez modified vn 147 225 308
4.0
0.8 vn 133 194 262
zero 55 50 90 3.0 vn/zero (full load)

2.0

Figure 3. Influence of the ship speed on the slamming

1.80 pressure for full load condition.

1.60

1.10
1.40 1.05

1.00
1.20
0.95

0.90
1.00
IACS/N-P (full load) 0.85

IACS/SUEZ (full load) 0.80

0.80
0.75
LongTerm/Poisson extreme - vn
Figure 1. Influence of the shipping route on the extreme 0.70
LongTerm/Poisson extreme - 0.8 vn
slamming pressure. 0.65 LongTerm/Poisson extreme - zero
0.60

Results are elaborated in Section 6 of the paper.
Furthermore, for each scatter diagram, calcula-
tions are performed for two long-term calculation
methods:
• The lifetime weighted sea method
• Poisson outcrossing method.
Detailed results of that analysis are presented in
Section 7.
Summary results of the long-term slamming
pressure analysis for two ships are presented in
Tables 2 and 3 for the lifetime weighted sea method
and Poisson outcrossing method respectively.
Also, these results are graphically presented in
Figures 1–3 and Figure 4 for full load condition
Figure 4. Influence of the calculation method on the
slamming pressure (3 hours assumed duration of short-
term seas states in Poisson outcrossing methods).
and all load conditions respectively. In all cases,
unitary pressures (k = 1) are calculated at section at
the distance of 10% from the fore perpendicular.
4 CHOICE OF THE SHIPPING ROUTE
One of the main issues analyzed within the present
study is the influence of the shipping route on the
extreme slamming pressure. The North Atlantic
environment, proposed by IACS for calculation of

71

MARSTRUCT.indb 71 2/18/2011 5:41:20 PM

 extreme global wave loads, is certainly not the most
frequent trading route for containerships, although
some of containerships are sailing on that route as
well. However, the North Atlantic is considered as
the most severe environment and as such can cover
all situations that the ship may encounter during
her lifetime, including the situation that the ship
permanently sails in the North Atlantic (Guedes
Soares 1996).
The idea of the present research is to study
differences in the extreme slamming pressure if
some other shipping route is selected, aiming to
model more realistically the operational features of
containerships.
Two important containership routes considered
in the present study are the North Pacific route
between Yokohama and San Francisco, and the
Europe-Asia route connecting European ports
Hamburg and Rotterdam with Shangai through
the Suez Channel.
The scatter diagram for the North Pacific trad-
ing route is obtained by combining wave zones 13,
20, 22 & 29 from Global Wave Statistics (GWS)
(Hogben et al. 1986).
The scatter diagram for the Europe–Asia trading
route passing through Suez Channel is obtained by
combining wave zones 11, 16, 17, 25, 26, 27, 29, 40,
41, 50, 60, 61 & 62 from GWS.
For both mentioned scatter diagrams, it is
assumed that the ship spends equal time in each
of the zones, i.e. equal probability of occurrence is
attached to each of them.
The diagram showing the influence of the ship-
ping route is presented in Figure 1. The figure
presents the most probable extreme pressures of
N-P and Suez scatter diagrams relative to IACS
diagram.
The results in Figure 1 indicate that extreme
slamming pressures for IACS design wave environ-
ment are considerably larger than slamming pres-
sures calculated for other wave environments. Only
in the case of ships speed equal to zero, this effect
is less pronounced. This is seen as ratios approxi-
mately equal to 1 in the Figure 1. Maximum ratio
between extreme pressures for IACS and N-P
environment is about 1.5, while the maximum ratio
between extreme pressures for IACS and Suez
environment is about 1.6.
5 INFLUENCE OF THE AVOIDANCE
OF HEAVY WEATHER
Avoidance of heavy weather is assessed by com-
paring long-term extreme pressures calculated for
truncated and original Suez and N-P wave scatter
diagrams. Scatter diagrams are truncated at signifi-
cant wave heights of 10.5 m. Truncation procedure
72
MARSTRUCT.indb 72
is described by Parunov & Ćorak (2010) and it is
not repeated herein.
Ratios of the most probable extreme slamming
pressures between original and truncated scatter
diagrams are presented in Figure 2.
In the most cases the effect of truncation is to
reduce extreme pressures up to about 10% and 30%
for N-P and Suez wave environments respectively.
Values lower than one indicates that larger extreme
pressures are obtained for truncated than for the
original scatter diagram. Numerical inaccuracies in
fitting theoretical distributions used in calculation
of the most probable slamming pressures could
be the reason for these discrepancies within the
low values of pressures. These values are typically
achieved for zero speed, but since those pressures
are relatively low, such results may be disregarded.
6 INFLUENCE OF THE SHIP SPEED
The influence of the ship speed on the long-
term extreme slamming pressures is presented
in Figure 3. The figure represents ratios of extreme
slamming pressures calculated for nominal ship
speed and extreme slamming pressures calculated
for zero speed case.
It may be seen from Figure 3 that the ship speed
has crucial effect on the bottom pressures. By
reducing speed from design speed to zero, extreme
bottom pressure is decreased several times. As
speed reduction in heavy weather is decision of
ship master, it is evident that such maneuvering
may be critical for extreme slamming loading of
containerships.
7 INFLUENCE OF THE METHOD
FOR CALCULATION OF THE MOST
PROBABLE EXTREME SLAMMING
PRESSURE
Two methods for calculation of long-term extreme
slamming pressure are similar, but not exactly
the same. The main difference is that in Poisson
outcrossing method, duration of individual short
term sea states needs to be specified, while that is
not required in the lifetime weighted sea method.
Physically, by modifying duration of individual
short-term sea states, one determines level of cor-
relation between successive sea states. Therefore,
there is an additional uncertainty due to the type
of the long-term calculation method employed.
In Figure 4, comparison between two methods
is presented with assumed duration of short-term
sea states in Poisson outcrossing method equal
to 3 hours. For zero speed, the lifetime weighted
sea method is slightly conservative. However,
2/18/2011 5:41:23 PM
 Table 4. Number of slams in 20 years. Large dispersion of results is present when
assessing influence of the heavy weather avoid-
Floxtra S175 ance. In average, the effect of truncation of the
scatter diagram is to reduce extreme pressures up
Wave scatter
diagram Ship speed FL FL BL to about 30%.
As expected, there is a huge influence of the ship
IACS vn 25991 58262 371453 speed on the extreme slamming pressures. Reduc-
0.8 vn 17492 37471 244470 tion of the ship speed from nominal speed to zero
Zero 360 136 2883 speed causes decrease of slamming pressures in
average 4–5 times.
N-P vn 7697 28002 203907
Method for calculating long-term extreme slam-
0.8 vn 5678 20052 146297
ming pressure is also one potential source of uncer-
zero 178 144 3758
tainty. For zero speed case, dispersion of results is
Suez vn 3963 15906 128593 relatively low and both studied methods, i.e. the
0.8 vn 2964 11793 95839 lifetime weighted sea method and the Poisson
zero 111 117 3395 outcrossing method, lead to the similar extreme
N-P modified vn 7010 27206 211755 values. Increased ship speed, however, causes that
0.8 vn 5104 19387 151425 the method of Poisson outcrossing predicts larger
zero 172 177 4082 extreme slamming pressures, but also uncertainty
Suez modified vn 2364 12066 108599 of results increases considerably.
0.8 vn 1828 9207 82106 Impulsive forces acting on the ship bottom are
zero 125 190 3991 proportional to the pressures calculated in this
study. Furthermore, whipping bending moments
are also proportional to the impulsive forces.
by increasing ship speed, the lifetime weighted Therefore, calculated uncertainties will have direct
sea method tends to produce lower extreme values consequence on uncertainties of whipping bend-
comparing to Poisson outcrossing method. ing moments. Furthermore, whipping bending
moments are to be combined with rigid hull ver-
tical wave bending moments for the assessment
8 NUMBER OF SLAMS of the hull-girder strength of the ultra-large con-
tainerships. Because of large uncertainties in both
As a result of the present analysis, number of wave-induced and whipping bending moments,
slams in ship’s lifetime is calculated by Equation reliability-based methods are to be used in design
(16) and presented in Table 4. These values can be of these ships.
useful in design, for example, they can be used to
assess number of transient vibration loading proc-
esses reducing the hull-girder fatigue life of the ACKNOWLEDGEMENTS
containership.
The investigation is funded by EU FP7 Project
“Tools for Ultra Large Container Ships” (TULCS),
9 CONCLUSIONS what is gratefully acknowledged.

The paper deals with environmental and operational

uncertainties in calculation of long-term extreme NOMENCLATURE
slamming pressure. Uncertainty in calculation
of the pressure coefficient k is excluded from the
analysis in a way that k was taken as unity. Actual psl Slamming pressure
extreme pressure may be calculated by multiplying ρ Sea density
obtained results by actual value of k according to
some of available methods. k Sllamming pressure coefficient
The influence of the choice of the shipping
route is analyzed in the paper firstly. In average, v Relative bow velocity
adopting Suez route, extreme slamming pressures v0 Critical (treshold) velocity
are reduced by factor of 1.5 compared to the
North Atlantic IACS scatter diagram. If the North g Gravity constant
Pacific shipping route is adopted instead of IACS L Ship lenght
diagram, then average reduction of extreme slam-
ming pressures would be 1.3. p0 Threshold pressure

73

MARSTRUCT.indb 73 2/18/2011 5:41:23 PM

 f psl ) Probability density function
n Fext ( pext ;tsea ) Unconditional extreme
of the impact pressuree distribution of the
(
F psl HS TZ , β ) Short-term probability
slamming pressure for tsea
distribution of the slamming
pressure Fext ( pext ;TC ) Extreme distribution of the
slamming pressure forr TC
FP ( psl ) Long-term probability
distribution of the slamming nsea Number of stationary sea
states in TC
pressure
FGumbel Gumbel distributio
u n
p( HS TZ ) Probabiility
l of occurrence of
sea
e state p ,λ
*
Parameters of the Gumbel
distribution
r (TZ , β ) Relative number of slamming
a
apperances in sh
hort-term sea
state
υ sl Slamming ffrequency in each REFERENCES
short term sea state
* Flokstra, C. 1974. Comparison of Ship Motion Theories
υ sl Average slamming frequency with Experiments for a Container Ship, International
in all sea states Shipbuilding Progress, 21, 168–189.
σM
2
Variance of the relative Guedes Soares, C. 1990. Effect of Heavy Weather
bow motion Manoeuvring on the Wave-Induced Vertical Bending
Moments in Ship Structures, Journal of Ship Research,
σV2 Variance of the relative Vol. 34, No. 1, 60–68.
bow velocity Guedes Soares, C. 1996. On the Definition of Rule
Requirements for Wave Induced Vertical Bending
Pslam Probability of slamming Moments. Marine Structures 9. 409–425.
Hogben, N., Dacunha, N.M.C. & Olliver, G.F. 1986.
SM (ω , x ) Response spectrum of the
t Global Wave Statistics. British Maritime Technology
relative bow motion Ltd. Felltham.
Juncher Jensen, J. et al. 2008. Wave induced extreme hull
SV (ω , x ) Response
p spectrum
p of the girder loads on containerships. Transactions SNAME
relative bow velocity 116, 128–152.
Ochi, M.K. & Motter, L.E. 1973. Prediction of slamming
FWeibulll Three-parameter Weibull characteristics and hull responses for ship design.
distribution Transactions SNAME. Vol. 81.
Parunov, J. & Ćorak, M. 2010. Influence of environ-
ϑ ,α , ε Parameters of the Weibull mental and operational uncertainties on vertical wave
distribution bending moments of containerships. Proceedings of:
The William Froude Conference—“Advances in Theo-
q P obability of exceeding
Pr retical and Applied Hydrodynamics, Past and Future”,
the most Portsmouth, 24–25 November 2010. UK., 201–207.
N sl Number of slams Senjanović, I., Tomašević, S. & Parunov, J. 2003. Ship
Slamming and Whipping in Rough Sea. Brodogradnja
TC Long-term return period 51, 45–56.
Wu, M.K. & Hermundstad, O.E. 2002. Time-domain
pext Random variable extreme Simulation of Wave-induced Nonlinear Motions and
Loads and its Applications in Ship Design, Marine
Fext Probability distribution of the Structures 15, 561–59.
extreme slamming pressure
υ p ( pext HS TZ , β ) Mean outcrossing rate of the
slamming pressure in
short-term sea stateiod

74

MARSTRUCT.indb 74 2/18/2011 5:41:24 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Efficient calculation of fluid structure interaction

in ship vibration

M. Wilken
Germanischer Lloyd SE, Hamburg, Germany

A. Menk
Robert Bosch GmbH, CR/APJ3, Stuttgart-Schwieberdingen, Germany

H. Voss
Hamburg University of Technology, Hamburg, Germany

C. Cabos
Germanischer Lloyd SE, Hamburg, Germany

ABSTRACT: Simulating global ship vibration can be split into three steps: firstly, the computation
of the dry elastic vibration of the ship structure, secondly determination of the hydrodynamic pressures
caused by a given time harmonic velocity distribution on the outer shell and thirdly, the solution of the
coupled vibration problem by considering the interaction of fluid and structure. In this paper various
approaches for the solution of the third problem for large models are compared and discussed. They are
based on reduction methods for the hydrodynamic mass matrix and make use of fast solution methods
for the exterior fluid problem for given velocity distributions of the shell. A numerical example is used to
assess the accuracy and the speed of the solution procedures.

1 INTRODUCTION

Forces induced by engines and propellers excite ship

vibrations which despite of their small amplitudes
can affect human comfort and may cause fatigue
damages. In order to predict ship vibrations it is
indispensable to account for the effect of the sur-
rounding water because the hydrodynamic forces
acting on the ship’s hull can considerably reduce
the natural frequencies of the dry ship and there- Figure 1. Illustration of the effect of water on the
fore can significantly affect the vibration response. vibrating outer shell.

1.2 Equations describing the hydrodynamic

1.1 The effect of water on a vibrating structure
The hydrodynamic influence of the water on the
vibrations of a ship can be modeled as an addi-
tional mass distribution on the outer shell. The
acceleration of the structure causes the fluid near
the interface to accelerate which in turn exerts an
opposing force on the ship’s hull (see Figure 1). The
additional force which is needed to accelerate the
surrounding fluid can be interpreted by Newton’s
law as an additional mass distributed on the ship’s
hull. That mass is often called hydrodynamic mass
or added mass.
mass effect
Since the structural displacements in ship vibration
are small compared to the dimensions of the ship,
the ship and the surrounding fluid can be modeled
by a set of linear PDEs. Moreover the flow of the
water around the ship’s hull is assumed to be invis-
cid and irrotational.
Hence, the velocity field of the fluid is the
gradient of a velocity potential—which due to
mass conservation satisfies the Laplace equation
Δp = 0 (1)

75

MARSTRUCT.indb 75 2/18/2011 5:41:25 PM

 Γf Γf
outer shell
Γkk
Γ
Ω
Ω
Figure 2. Problem description.
Assuming an infinitely wide and deep fluid
domain, boundary conditions have to be speci-
fied at the free water surface and at the submerged
ship’s hull (Figure 2). The exact boundary con-
dition at the free surface is nonlinear (Newman
1977), but for frequencies above 1 Hz and small
displacements it can be linearized to yield a pres-
sure release condition which takes the form p = 0
on Γf . At the submerged ship surface the fluid
velocity normal to this surface must be equal to the
normal velocity of the structure,
∂p  
= ρω 2u T n on Γ k (2)
∂n
where p denotes the pressure at the outer shell,
 with the number of fluid panels.
ρ the fluid density, ω the excitation frequency, u

the displacement of the structure and n is the out-
ward normal.
2 CONSIDERATION OF
HYDRODYNAMIC MASS EFFECTS IN
GLOBAL SHIP VIBRATION ANALYSIS
2.1 Standard procedures
2.1.1 Full hydrodynamic mass matrix method
To account for the surrounding water, an FE
model of the ship can be complemented by an FE
discretization of the water to solve the Laplace
equation with coupling boundary condition (2)
(Arman et al., 1979). This causes considerable
additional cost since only a bounded region of the
fluid domain can be modeled this way and suitable
boundary conditions on the outer boundary have
to be specified or the remaining unbounded region
of the water has to be discretized by semi infinite
elements. If Boundary Element (BE) methods are
used in combination with a special fundamental
solution, an unbounded fluid domain can be mod-
eled, but only the submerged ship hull has to be dis-
cretized. Thus the problems previously mentioned
are avoided. The mesh can simply be generated
from the FE mesh of the ship’s hull. Today this is a
standard approach if three dimensional effects have
to be included in the analysis (Cabos et al., 2003).
76
MARSTRUCT.indb 76
An advantage of this method is the fact that the
discretization of the coupled fluid-solid problem
has the same dimension as the FE model of the
dry ship alone in case the free surface boundary
condition is handled using the method of images
(Wilken et al., 2009). However, as a drawback the
part of the mass matrix corresponding to the wet
hull of the ship is fully populated.
To determine the eigenfrequencies of the
coupled problem, the following eigenvalue prob-
lem must be solved:
wet (M S
K S u − Ωwet MH ) u = 0
2 (3)
where KS, MS are the stiffness and structural
mass matrix of a FE model of the ship which are
large-scaled and sparse. The hydrodynamic mass
matrix MH models the impact of the surrounding
water on the ship. Only rows and columns of MH
corresponding to wet degrees of freedom contain
entries which are different from zero. However, the
total number of non-zero entries in the coupled
system is still increased considerably. Computing
the complete hydrodynamic mass matrix in this
way leads to a cubic scaling of required computa-
tion time and a quadratic scaling of memory usage
2.1.2 Lewis approach
Consideration of the effect of the surrounding
water for the computation of global ship vibra-
tions dates back to the first half of the 20th cen-
tury. Regarding a ship as a slender body, Lewis
(Lewis 1929) showed that the inertia of the water
can be approximately be accounted for by analyz-
ing the two-dimensional flow around ship cross
sections.
The hydrodynamic mass of a cylindrical cross
section was generalized to more complex shapes
by introducing reduction coefficients. Assuming
that a ship is a slender body, Lewis succeeded in
determining the hydrodynamic mass affecting
vertical bending vibrations of a ship (Figure 3).
Since the hydrodynamic mass derived with this
method depends on the particular bending mode
of the ship, it is typically valid only for a specific
range of frequencies around the corresponding
eigenfrequency of this mode. Wendel (Wendel
1950) and Landweber (Landweber 1957) extended
Lewis’ work by considering also horizontal and
rotational acceleration of ship cross sections. Grim
(Grim 1953, 1960) examined the reduction coef-
ficients for higher modes. The Lewis method is
most appropriate for a Finite Element analysis, if
the ship is modeled by several beam elements. For
the three dimensional analysis of ship vibrations
based on a Finite Element model, the use of the
2/18/2011 5:41:26 PM
 application to vibrating ship structures as shown in
the next sections.

λ 2.2.2 Projection approach

Neglecting the influence of the surrounding water
one obtains the so called dry eigenvalue problem
for the ship structure

K S u − Ωdry 2MS u = 0 (4)

that can be solved efficiently due to the sparse

x structure of the matrices.
Figure 3. Lewis assumption of deformation of the ship
structure over the ship length.
so called Lewis Method leads to the problem that
only the total force onto a ship cross section (or
better its total hydrodynamic mass) can be com-
puted. The actual distribution of the hydrodynamic
mass over the contour of the section is determined
with a heuristic approach. In a three dimensional
vibration analysis this can lead to hydrodynamic
forces having components tangential to the shell
surface. Despite these approximations and short-
comings, the Lewis approach has proven to yield
good results in the low frequency range having a
very good performance in terms of CPU time and
memory consumption.
2.2 Advanced approaches using fast BEM
2.2.1 Fast multipole method
The standard boundary element method com-
putes the pressure on a fluid panel caused by
each other vibrating panel. This is done in an
exact manner with no restriction on the shape of
the immersed hull. The so called fast multipole
method takes advantage of the concrete hull shape
of the vibrating structure based on the following
advisement:
Solving the Laplace equation (1) for vibrat-
ing point sources close to each other, the far field
result is a pressure field that decreases with one to
the square of the distance to these points. It can
therefore be idealized in the far field as a pressure
While the eigenvectors of (3) and (4) exhibit
similar characteristics, it is found that the eigen-
modes of these problems are quite close to each
other. This observation suggests to project the
eigenproblem (3) onto a space spanned by a
selected subset of nsel eigenvectors Udry of (4).
This is done by assuming the wet eigenvectors as
linear combination of the dry eigenvectors and
multiplying the resulting equation from the left
side with UTdry yielding a new eigenvalue problem
of a smaller dimension nsel:
T
( wet (M S
ry K S − Ωwet
U dr 2
)
M H ) U drryw = 0 (5)
The required subset of dry eigenvectors may
be choosen as eigenvectors with eigenfrequencies
smaller than an appropriate multiple of the upper
frequency bound of interest. Instead of explic-
itly calculating the hydrodynamic mass matrix
it suffices to evaluate the matrix vector product.
This can be done by a standard boundary ele-
ment method but also more efficiently by the fast
multipole boundary element method.
Solving this eigenvalue problem yields eigen-
vectors w defining linear combination factors to
be used for approximating the wet eigenvectors
from the dry eigenvectors. This approach can be
improved by projecting onto approximated wet
modes (“semi wet modes”) resulting from an eigen-
value problem with an approximate hydrodynamic
mass matrix:
(K S semi (MS + MH semi )) usemi = 0 (6)

field caused by a single vibrating point source. This

effect is exploited by the fast multipole method Using these semi wet modes for projection
yielding a fast and memory saving procedure for yields
computing the resulting pressure field of a vibrat-
ing structure (Wilken et al., 2009).
T
U se ( wet (M S
mi K S − Ωwet
2
)
M H ) U semi w = 0 (7)
Considering a typical hull form, it is obvious
that the far field approximation above could be respectively a new eigenvalue problem
applied for the majority of pairs of panels since
only a small fraction of them is close to each
other. For this reason the fast multipole bound-
⎡Ω2semi
⎣ Ωwet
2
I( T
Ussemi
ary element method seems to be well suited for its (
× MH ) ⎦ )
M H semi U semi ⎤ w = 0 (8)

77

MARSTRUCT.indb 77 2/18/2011 5:41:28 PM

 Depending on the number of semi wet modes
used for projection, the precision of the wet modes
can be influenced.
In the present work the semi wet modes are
calculated using the Lewis approach described in
section 2.1.2. The solution of the forced vibration
problem is then easily obtained by modal superpo-
sition using the derived wet modes from (8).
2.2.3 Modal hydrodynamic mass matrix
approach
The eigenvalue problem (3) can be solved by the
shift-and invert Lanczos method (Bai et al., 2000).
Using this method requires the solution of the lin-
ear system
KSx − σ(MS + MH)x = b (9)
for x in every iteration step. This can be done by
a direct method which requires the explicit form
of the matrix MH (i.e., nH solves of the BE sys-
tem) and which, due to the structure of the sys-
tem matrix A: = KS − σ(MS + MH), is very time and
memory consuming. Solving (9) by an iterative
solver like MINRES (Saad 2003), every Lanczos
step requires a suitable Krylov subspace. Hence,
one has to apply the system matrix A to a couple of
vectors, and each of these multiplications demands
the solution of one BE system. In the following we
derive a reduction method which is much more
efficient from a computational point of view. The
hydrodynamic mass matrix MH is symmetric and
positive definite. We take advantage of the spectral
decomposition
hH
MH ∑ λi xi xiT (10)
i 3.1 Model
of MH, where the eigenvalues λi λi ≥ … ≥ λ nH are
ordered by magnitude. Then the best approxima-
tion to MH by a matrix of rank n with respect to
the spectral norm is the truncation of (10)
n

M ∑ λi xi xiT (11)
H
i
The Lanczos method is favorable for computing
an approximation to M  since it does not employ
H
the explicit form of the matrix MH but only matrix-
vector products and according to the Kaniel-Paige
theorem (Golub et al., 1996) it converges first to
extreme eigenvalues and in particular to the largest
ones which are better separated than the smallest
ones. Notice that the Lanczos process for com-
puting M  can even be accelerated by replacing
H
the solution of the BE approximation by the fast
multipole approach (Wilken et al., 2009).
78
MARSTRUCT.indb 78
Having determined the approximation M 
H
to MH, we have to solve the reduced eigenvalue
problem
 2 M
KS u = Ω wet
wet S ( MH u) (12)
Tackling it by the shift-and-invert Lanczos
method one has solve a linear system
KS x − (M S MH x) b (13)
in every iteration step where σ is a preselected shift.
Since KS and MS are sparse the LU factorization
of KS − σMS can be determined efficiently and
since the rank of M  is quite small compared to
H
the dimension of the problem it is inexpensive to
employ the Sherman-Morrison-Woodbury formula
(Golub et al., 1996) for solving problem (13).With
this approach the shift-and-invert Lanczos method
for the reduced problem (12) essentially requires
the following work: To initialize one provides those
vectors and matrices which are independent of the
right hand side b when solving (13) by the Sher-
man-Morrison-Woodbury formula. To this end a
slightly larger number than n solutions of the BE
system are necessary in a Lanczos process for com-
puting M  . Additionally n solutions of linear sys-
H
tems (Ks − σMS)wj = xj and n 2 scalar products of
length n are required in this preprocessing phase.
Thereafter, every iteration step requires one solve
of a linear system of dimension n, and n scalar
products of length n.
3 NUMERICAL EXAMPLE
An FE model of a typical container vessel of 250 m
length and 32 m breadth having 35262 degrees of
freedom was investigated to assess the accuracy and
the speed of the described techniques. This model
is capable to compute global vibration responses
Figure 4. Container Ship of 252.2 m length and 32.2 m
breadth.
2/18/2011 5:41:30 PM
 Figure 5. BEM mesh with 1490 boundary elements.
to a frequency up to 20 Hz. A draft of 10.8 m was
used in the calculations leading to 1490 wetted
elements of the outer shell with 4554 degrees of
freedom.
3.2 Computations
For the assessment of the various methods, the
eigenvalues and the forced vibration response
caused by a typical propeller induced excitation
were calculated. The following approaches for the
solution of the vibrational fluid-structure interac-
tion problem were performed
1. Usage of the explicit hydrodynamic mass matrix
computed by a standard BEM method herein-
after referred as “FULL_HYM” method.
2. Usage of a diagonal hydrodynamic mass matrix
approximation hereinafter referred as “LEWIS_
HYM” approach.
3. Projection of vibration equation (3) onto a
set of semi wet modes hereinafter referred as
“PROJECTION” approach. The semi wet modes
were taken from an eigenvalue computation
approximating the hydrodynamic mass matrix
according to Lewis. This Lewis approximation
yields 338 eigenvectors below 20 Hz (88 modes
more than the reference FULL_HYM method)
which all were used for the projection.
4. Usage of a modal approximation of the hydro-
dynamic mass matrix hereinafter referred as
“MODAL_HYM” approach. Due to com-
parability reason with the PROJECTION
approach the same number of (Fast-) BEM
applications were performed leading to a modal
hydrodynamic mass matrix approximation of
rank 338.
3.3 Comparison
In the following, the modes, i.e. the pairs of
eigenvalue and eigenvector, and forced vibration
results (i.e. velocity at dedicated locations) of the
FULL_HYM method serves as reference values
for the comparison of the various approaches.
79
MARSTRUCT.indb 79
Figure 6. Forced vibration configuration.
The FULL_HYM method yields 250 modes below
20 Hz.
3.3.1 Precision
For comparing the eigenvectors of the different
approaches the modal assurance criterion (MAC),
see e.g. Allemang (1980), will be used
(v )
T 2
i ref M H v j appr
Cij =
MAC (14)
(viTref M H vi ref
e )(
vTj ,appr M H v j ,appr )
where vi,ref denotes the i-th eigenvector computed
by the FULL_HYM method, MH denotes the
full hydrodynamic mass matrix and vj,appr denotes
the j-th mode computed by the approaches
2–4. A MAC value near to 1 indicates that the
approached eigenvector vj,appr is quite similar to the
reference eigenvector vi,ref .
This criterion can be subsequently used to com-
pare the eigenvalues of the most similar eigenvec-
tors computed by the FULL_HYM method and
the particular approach:
E ri =
relErr
Ωi ,appr
Ω j ref
{j M C )}
(MAC i j
(15)
As it can seen from Figure 7 the eigenvectors
computed according to LEWIS matching the
reference eigenvectors only for the first 100 modes.
The differences in eigenfrequencies is below 5% in
the lower frequency range and goes up to 20%
in the higher frequency range to 20 Hz.
Relative differences in eigenvalues of the PRO-
JECTION and MODAL_HYM approach are of
the same magnitude, i.e. below 5% over the total
frequency range from 0 Hz to 20 Hz. Also the
eigenvectors of these 2 different approaches are
of the same similarity compared to the reference
eigenvectors.
2/18/2011 5:41:35 PM
 Figure 7. Similarity of LEWIS_HYM eigenvectors and
reference eigenvectors according to MAC. Figure 10. Relative error of MODAL_HYM eigenval-
ues and reference eigenvalues ordered by MAC value.

Figure 8. Relative error of LEWIS_HYM eigenvalues

and reference eigenvalues ordered by MAC value. Figure 11. Similarity of PROJECTION eigenvectors
and reference eigenvectors according to MAC.

Figure 9. Similarity of MODAL_HYM eigenvectors Figure 12. Relative error of PROJECTION eigenvalues
and reference eigenvectors according to MAC. and reference eigenvalues ordered by MAC value.

80

MARSTRUCT.indb 80 2/18/2011 5:41:38 PM

 In Figure 13 typical frequency response curves
caused by the single and double propeller blade
passage excitations computed by the different
approaches are shown. The response curves com-
puted according to LEWIS gives the qualitative
characteristics in the low frequency range but devi-
ates in the higher frequency range considerable
from the reference curve. Application of the PRO-
JECTION approach yields the best accordance
with the reference velocity whereupon the velocity
results of the MODAL_HYM approach are only a
little less inaccurate.
3.3.2 Run time
All computations were performed on a 64-bit linux
computer Quad-Core AMD Opteron(tm) Proces-
sor 2356 with 32 GB RAM and a clock rate of
2.3 GHz.
The FULL_HYM method as the most accu-
rate procedure is also the most time consuming
method with a total computation time of ∼1.8 h
and the most inaccurate approach (LEWIS) is the
fasted with ∼2 min CPU time. The PROJECTION
and the MODAL_HYM approach having total
computation time of the same order of magnitude
(∼15 min and ∼20 min) only the distribution
between pre computation time and mode compu-
tation differ considerably: The pre computation
time of the PROJECTION approach, i.e. the semi
wet mode computation time according to Lewis
is much less than the pre computation time of
the MODAL_HYM approach where the modal
approximation of the hydrodynamic mass matrix
has to be computed.
4 CONCLUSIONS
The presented PROJECTION approach com-
bines the fast and robust LEWIS method with
an advanced fast boundary element technique
yielding very accurate eigenfrequencies and accu-
rate forced vibration results within small com-
putation times. The MODAL_HYM approach
exhibits only slightly worse characteristics in preci-
sion and run time. Both approaches require user
experience: the PROJECTION approach in case
of selecting the number of eigenvectors used for
projection and the MODAL_HYM approach in
case of number of modes needed for approxima-
tion the hydrodynamic mass matrix.
A particular advantage of the proposed meth-
ods is that they scale very well. The effort to com-
pute the hydrodynamic mass effect is dominated
by evaluations of the hydrodynamic mass opera-
tor. Through application of the fast multipole
method, the cost for this application grows approx-
imately like N log2(N) for large numbers N of wet
panels.

REFERENCES

Allemang, R.J. 1980. Investigation of Some Multiple

Input/Output Frequency Response Function Experi-
mental Modal Analysis Techniques. Doctor of Philoso-
phy Dissertation, University of Cincinnati, Department
of Mechanical Engineering, pp. 141–214.
Armand, J.-L. & Orsero, P. 1979. A method for evaluat-
ing the hydrodynamic added mass in ship hull vibra-
Figure 13. Typical response curves of the different
tions. SNAME Transactions, 87:99–120.
approaches.
Bai, Z., Demmel, J., Dongarra, J., Ruhe, A. & van der
Vorst, H.A. 2000. Templates for the Solution of Alge-
braic Eigenvalue Problems: A Practical Guide. SIAM,
Pre Computation Mode Computation Total
Philadelphia.
Approach Item CPU Time [s] CPU Time [s] CPU Time [s] Cabos, C. & Ihlenburg, F. 2003. Vibrational Analysis of
FULL_HYM full hydro mass 6060 553 6613
Ships with Coupled Finite and Boundary Elements.
LEWIS_HYM Lewis hydro mass 2 126 128 Journal of Computational Acoustics, 11(1):91–114.
Golub, G.H. & Van Loan, C.F. 1996. Matrix Computa-
MODAL_HYM Modal hydro mass 787 409 1196
tions. The John Hopkins University Press, Baltimore
PROJECTION Lewis modes 126 744 870 and London, 3rd edition.
Grim, O. 1953. Berechnung der durch Schwingungen
eines Schiffskörpers erzeugten hydrodynamischen
Figure 14. Comparison of run time. Kräfte. STG Jahrbuch.

81

MARSTRUCT.indb 81 2/18/2011 5:41:45 PM

 Grim, O. 1960. Elastische Querschwingungen des
Schiffskörpers. Schiffstechnik, 7(35):1–3.
Landweber, L. 1957. Mass of Lewis Forms Oscillating in
a Free Surface. Proceedings: Symposium on the Behav-
iour of Ships in a Seaway, Wageningen.
Lewis, F.M. 1929. The Inertia of the Water Surrounding
a Vibrating Ship. Transactions of the SNAME, 37.
Newman, J.N. 1977. Marine Hydrodynamics. The MIT
Press, Cambridge, Mass.
Saad, Y. 2003. Iterative Methods for Sparse Linear Sys-
tems. SIAM, Philadelphia, 2nd edition.
82
MARSTRUCT.indb 82
Wendel, K. 1950. Hydrodynamische Massen und
hydrodynamische Massenträgheitsmomente 14. Jahr-
buch der STG, vol. 44, 207–255.
Wilken, M., Of, G., Cabos, C. & Steinbach, O. 2009. Effi-
cient calculation of the effect of water on ship vibra-
tion. C. Guedes Soares and P.K. Das, Analysis and
Design of Marine Structures, pages 93–101, London,
Taylor & Francis.
2/18/2011 5:41:47 PM
 Methods and tools for strength assessment

Ultimate strength

MARSTRUCT.indb 83 2/18/2011 5:41:47 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Buckling analysis of composite delaminated ship plates under

shearing

E.F. Beznea & I. Chirica

University “Dunarea de Jos” of Galati, Romania

ABSTRACT: In this paper postbuckling behaviour and estimation of global buckling ultimate strength
of the delaminated rectangular plates are presented. The influence of the position and geometry of
elliptical delamination on the changes in the buckling behaviour of ship deck plates made of composite
materials is considering. The composite plates models were analyzed using COSMOS/M and database
is prepared for different diameter ratios and position of delamination. A delamination model, describ-
ing delaminating mode, by using COSMOS/M soft package, is applied, so that the damaged part of the
structures and the undamaged part have been represented by layered shell elements. The influence of the
position and the ellipse’s diameters ratio of delaminated zone on the critical buckling force is investigated.
The applied methods have been improved in the Marstruct Project.

1 INTRODUCTION whole structure (collapse of the stiffeners and the

Laminated composite panels, which are aniso-
tropic, are gaining popularity in structural applica-
tions such as ship hulls, decks, ship and offshore
superstructures. These panels are becoming incr-
easingly used in structural marine applications due
to their high specific stiffness and specific strength
(Altenbach, Altenbach & Kissing 2001). The use
of laminated composites provides flexibility to
tailor different properties of the structural ele-
ments to achieve the stiffness and strength char-
acteristics. These panels, unfortunately, have one
important characteristic connected to big sensitiv-
ity on geometrical and mechanical imperfections
(different dimensions comparative with the design
ones). Another kind of imperfections is about
material (Adams, Carlsson & Pipes 2003; Jones
1999). Taking into account that fabrication tech-
nologies of composite materials are hand made
based, the probabilistic occurrence of defects is
quite too high.
These defects are of following types: directions
of fibers are different of the designed ones, varia-
tions in thickness, inclusions and initial transversal
deformations (Thurley & Marshall 1995).
Ship structure plates are subjected to any com-
bination of in plane, out of plane and shear loads
during application. Due to the geometry and gen-
eral load of the ship hull, buckling is one of the
most important failure criteria. Buckling failure
mode of a stiffened plate can further be subdivided
into global buckling, local skin buckling and stiff-
ener crippling. Global buckling is collapse of the
shell as one unit). Local plate buckling and stiffen-
ers crippling on the other hand are localized failure
modes involving local failure of only the skin in
the first case and the stiffener in the second case.
A grid stiffened panel will fail in any of these failure
modes depending on the stiffeners, plate thickness,
shell winding angle and type of applied load.
Understanding delamination is essential for pre-
venting catastrophic failures. Therefore, analysis
of delamination behavior from tests, modelling
delamination, analysis of structural performance
under delamination and preventing and mitigation
of delamination are the main aim of the research
team that performed the work.
In (Thurley & Marshall 2002) the buckling
behaviour of laminated panels with one stiffener,
subjected to compression by using a layer wise
finite element formulation, is presented.
In (Nemeth 1992; Nemeth 1997) Nemeth have
done some parametric studied based on ortho-
tropic plate theory and produced generic buckling
design charts in terms of useful nondimensional
parameters for unstiffened composite panels sub-
jected to different loadings.
In (Mallela & Upadhyay 2006) some parametric
studies on simply supported laminated composite
blade-stiffened planels subjected to in-plane shear
loading. Few important parameters influencing
the buckling behaviour are identified and guide-
lines are developed.
The aim of the work presented in this paper is
to analyze the influence of delamination on the
changes in the buckling behaviour of ship deck

85

MARSTRUCT.indb 85 2/18/2011 5:41:47 PM

 plates made of composite materials. This problem and
has been solved by using the finite element method,
in (Beznea 2008). An orthotropic delamination Ζk = Ζk−1 + tk/2. (4)
model, describing mixed mode delaminating, by
using COSMOS/M soft package, was applied. So, In equation (2), the coefficients Qijk have the fol-
the damaged part of the structures and the undam- lowing forms
aged part have been represented by well-known
finite elements (layered shell elements). The influ- k
Q11 ′ 4
C11c (′
2 C12 ′
2C33 ) ′ 4
c 2 s 2 C22 s ;
( ) ( )
ence of the position and the ellipse’s diameters k ′ ′ ′ ′
ratio of delaminated zone on the critical buckling Q12 C11 + C22 4C33 c 2 s 2 + C12 c4 + s4 ;
force was investigated.
If an initial delamination exists, this delamina-
k
Q22 ′ 4
= C11 ′
s + 2 C12 ( ′
+ 2C33 ′ 4
c 2 s 2 + C22 )
c ;
tion may close under the applied load. To prevent (
′
Q3k3 = C11 ′
11 C222 − 2
′
12
′
2C333 ) ′
c 2 s 2 + C33(c4 + s4 ; )
the two adjacent plies from penetrating, a numeri-
cal contact model is used.
k
Q13 (C ′ ′ ′
111 − 2C333 − C12 )c s + (
3 ′
C12 ′
− C22 ′
+ 2C33 )cs ;
3

k
Q23 = (C ′ ′
11 − 2C33 − C12
′
)cs + ( 3 ′
C12 ′
− C222 ′
+ 2C33 )c s.
3

s = sin θ ; c = cos θ .
2 SHEAR BUCKLING EQUATIONS
OF ORTHOTROPIC PLATES For the orthotropic plane material

When plates are subjected to the application large ′ El ′ Et

in-plane loads either compressive or shear, they C11 = ; C22 = ;
1 μ lt μtl 1 − μ lt μtl
buckle. The phenomenon of buckling is a non- μ lt Et μtl El
′ ′
linear one which is characterized by dispropor- C12 = = = C21 ;
tionate increase of the displacements associated 1 μ lt μtl 1 − μ lt μtl
′ ′ ′ ′ ′
with the small increments of the loads. The meth- C33 Gllt ; C13 C31 = C23 C32 = 0.
odology for determining and analyzing the buck-
ling behaviour of laminated composite plates is in where El, Et, Glt, μlt and μtl are the material
essence identical to that applied to isotropic plates. characteristics in the longitudinal (l ) and trans-
As in isotropic plates it involves the solution of versal (t) orthotropic directions, and θ is the angle
an eigen-value problem associated with a govern- between the x axis and l direction.
ing set of homogeneous differential equations The last two terms from equation (1) are the
and a prescribed set of homogeneous boundary measure of the orthotropic coupling, resulting
conditions. from the fact that the principal orthotropic axes
In the case of isotropic plates, of sizes axb, are not orthogonal with the plate geometry axes.
exact buckling solutions are available only for For a special orthotropic plate, D13 = D23 = 0.
a few combinations of loading and boundary This is a case which has received the most atten-
conditions. tion by the researchers.
The theory and differential equation of bending The problem of the stability of orthotropic
of anisotropic plates were established by Huber plates due to shear was apparently first examined
and the governing differential equation for shear by Bergmann and Reissner (Johns 1971), who
buckling of a general orthotropic plate is considered it an infinitely long in the x direction
and they also neglected the bending stiffness in that
D11∂ 4w ∂x 4 + 2 D33∂ 4w / ∂x 2∂y2 + D22∂ 4w y4 direction. The governing differential equation
used was
+ xy ∂ w / ∂
2
∂y + 13∂
4
w / ∂x 3∂y
+ 23∂
4
/∂ ∂y3 = 0 (1) 2 D33∂ 4w ∂x 2∂y2 + D22∂ 4w ∂y 4
+ 2N
N xy∂ 2w / ∂x∂y 0 (5)
where D11, D22, D33, D13, D23 are the orthotropic plate
stiffnesses, calculated according to the equation
In (Smith 1946) approximate solutions are
N obtained for the buckling of clamped edged
Dij ∑ Qijk ( zk zk )3 (2) finite plates, using the Rayleigh-Ritz method.
k =1 The same approximate deflection functions are
assumed as
The thickness and position of every ply can be
calculated from the equation
tk = zk−zk-1, (3)
w ∑ ∑ AmnX mYn (6)
m n

86

MARSTRUCT.indb 86 2/18/2011 5:41:47 PM

 where
Xm x 2 + x + (−
x 3 − 2x )m (x x ) − m1π sin mπx
Yn y 3 − 2yy 2 + y + ( − )n ( y y ) − n1π sin nπyy
In the upper equations x = x/a and y = y/b.
Analyses are made with m and n as variables and
it is shown that solutions only exist in two distinct
ranges, i.e. when (m + n) even and (m + n) odd.
The case when (m + n) is even gives the lower criti-
cal shear loads. That this may not necessarily be
true in general has been shown in isotropic panels
where m + n < 6.
3 NUMERICAL ANALYSIS ON BUCKLING
OF IN-PLANE SHEARING LOAD
PLATES
The finite element delamination analysis was car-
ried out using COSMOS/M finite element soft-
ware. There are several ways in which the panel
can be modeled for the delamination analysis.
For the present study, a 3-D model with 3-node
SHELL3L composite element of COSMOS/M is
used. The panel is divided into two sub-laminates
by a hypothetical plane containing the delamina-
tion. For this reason, the present finite element
model would be referred to as two sub-laminate
model. The two sub-laminates are modeled sepa-
rately using 3-node SHELL3L composite element,
and then joined face to face with appropriate
interfacial constraint conditions for the corre-
sponding nodes on the sub-laminates, depending
on whether the nodes lie in the delaminated or
undelaminated region.
The delamination model has been developed
by using the surface-to-surface contact option
(Fig. 1). In case of surface-to-surface contact, the
FE meshes of adjacent plies do no need to be iden-
Figure 1. Delamination model.
87
MARSTRUCT.indb 87
tically. The contact algorithm of COSMOS/M has
possibility to determine which node of the so-called
master surface is in contact with a given node on
the slave surface. Hence, the user can define the
interaction between the two surfaces.
In the analysis, the certain layers are intention-
ally not connected to each other in ellipse regions.
The condition is that the delaminated region does
not grow. In COSMOS/M these regions were
modeled by two layers of elements with coinci-
dent but separate nodes and section definitions to
model offsets from the common reference plane.
Thus their deformations are independent. At the
boundary of the delamination zones the nodes
of one row are connected to the corresponding
nodes of the regular region by master slave node
system.
The square plates (320 × 320 mm), clamped
on the all sides are made of 9 macro-layers hav-
ing the charactersitics presented in Table 1. The
material is E-glass/polyester having the material
characteristics:
Ex = 46 GPa, Ey = 13 GPa, Ez = 13 GPa, Gxy =
5 GPa, Gxz = 5 GPa, Gyz = 4.6 GPa,
RTx = 1.062 GPa, RTy = 0.031 GPa, RCy = 0.118 GPa,
R xy = 0.72 GPa. μxy = 0.3, μyz = 0.42, μxz = 0.3
The in-plane loading was applied as a uniform
shear pressure on the sides (Figure 2).
The ellipse’s diameters of the delamination area
placed in the middle of the plate are considered
from the condition of the same area for all cases.
In the parametric calculus, the following diameters
ratios were considered:
– Case 1 (Dx/Dy = 0.5): transversal diam-
eter Dy = 141 mm, longitudinal diameter
Dx = 70,5 mm;
– Case 2 (Dx/Dy = 1): transversal diam-
eter Dy = 100 mm; longitudinal diameter
Dx = 100 mm;
– Case 3 (Dx/Dy = 2): transversal diam-
eter Dy = 70.5 mm; longitudinal diameter
Dx = 141 mm.
The position of the delamination along the
thickness it’s been considered between two neigh-
bors macrolayers i and i + 1, (i = 1, 9). We have
considered all cases. Taking into account the thick-
ness symmetry of the plates, will be presented only
cases of position of delamination on one side of
symmetry axis.
For the material model two cases has been
considered:
– linear behaviour;
– nonlinear behaviour (Tsai-Wu failure criterion).
2/18/2011 5:41:51 PM
 Table 1. Plate lay-up.

Nr. Macro layer 1 2 3 4 5 6 7 8 9

Nr. Layers × angle 2 × 0° 1 × 45° 2 × 90° 1 × 45° 4 × 0° 1 × 45° 2 × 90° 1 × 45° 2 × 0°

Thickness [mm] 0.62 0.31 0.62 0.31 1.24 0.31 0.62 0.31 0.62

The solution procedure takes the following

Figure 2. Shear loaded plate.
Table 2. Buckling load range obtained in graphical
method.
Buckling load [MPa]
Position of delamination Min pcr Max pcr
Macro-layer 1 – Macro-layer 2 115.42 279.92
Macro-layer 2 – Macro-layer 3 119.61 284.35
Macro-layer 3 – Macro-layer 4 121.23 286.39
Macro-layer 4 – Macro-layer 5 123.71 289.97
For perfect plate (without delaminations), the
value of the buckling load is:
– linear calculus: pcr = 169.76 MPa;
– nonlinear calculus: pcr = 334.71 MPa.
To solving geometrically and material nonlinear
problems, the load is applied as a sequence of suffi-
ciently small increments so that the structure can be
assumed to respond linearly during each increment.
For each increment of load, increments of displace-
ments and corresponding increments of stress and
strain are computed. These incremental quantities
are used to compute various corrective stiffness
matrices (variously termed geometric, initial stress,
and initial strain matrices) which serve to take into
account the deformed geometry of the structure.
A subsequent increment of load is applied and the
process is continued until the desired number of
load increments has been applied. The net effect is
to solve a sequence of linear problems wherein the
stiffness properties are recomputed based on the
current geometry prior to each load increment.
mathematical form
(K + KI)i-1 Δdi = ΔQ (7)
where
K is the linear stiffness matrix,
KI is an incremental stiffness matrix based upon
displacements at load step i-1,
Δdi is the increment of displacement due to the
i-th load increment,
ΔQ is the increment of load applied.
The correct form of the incremental stiffness
matrix has been a point of some controversy.
The incremental approach is quite popular (this is
the procedure applied in this study), due to the ease
with which the procedure may be applied and the
almost guaranteed convergence if small enough
load increments are used.
In this paper, buckling and post-buckling analy-
sis has been performed for all types of panels.
Since the plates have an initial imperfection, as it
is seen in the Figures 1 and 2, the increasing of the
transversal deformation is starting from the begin-
ning, that is so named buckling is starting as the
in-plane load is starting to increase from 0.
The explanation is that due to the initial imper-
fection, the in-plane loading produces the shearing
in the plate and also bending in the area of imper-
fection. Therefore it is difficult to determine the
buckling load by numerical way.
This is why we have chosen the graphical method,
by drawing the asymptote to the curve in the area
where the slope is changing almost suddenly. The
intersection of the asymptote with the loading axis
can be considered as buckling load.
Anyway the asymptote is not an unique one and
we may determine the buckling loading in a range
of values.
So, as it is seen in the Figures 3–6, according
to the plotted asymptotes, for all cases of position
of delamination, the buckling loads of the plates
(having the diameters ratio presented above) are
placed in the domain
119 MPa < pcr < 290 MPa.
Also, according to these Figures, the buckling
load is decreasing since the diameters ratio of the
delamination is increasing.
In this paper, buckling and post-buckling analy-
sis has been performed for all types of panels.

88

MARSTRUCT.indb 88 2/18/2011 5:41:54 PM

 Figure 3. Shear buckling and postbuckling behaviour
of the plate having the delamination between macrolay-
ers 1 and 2.
The buckling load determination may use the
Tsai-Wu failure criterion in the case of the general
buckling does not occur till the first-ply failure
occurring. In this case, the buckling load is con-
sidered as the in-plane load corresponding to the
first-ply failure occurring.
The Tsai-Wu failure criterion provides the math-
ematical relation for strength under combined
stresses. Unlike the conventional isotropic materi-
als where one constant will suffice for failure stress
level and location, laminated composite materials
require more elaborate methods to establish failure
89
MARSTRUCT.indb 89
Figure 4. Shear buckling and postbuckling behaviour
of the plate having the delamination between macrolay-
ers 2 and 3.
stresses. The strength of the laminated composite
can be based on the strength of individual plies
within the laminate. In addition, the failure of plies
can be successive as the applied load increases.
There may be a first ply failure followed by other
ply failures until the last ply fails, denoting the ulti-
mate failure of the laminate. Progressive failure
description is therefore quite complex for lami-
nated composite structures. A simpler approach
for establishing failure consists of determining the
structural integrity which depends on the defini-
tion of an allowable stress field. This stress field is
2/18/2011 5:41:55 PM
 Figure 5. Shear buckling and postbuckling behaviour Figure 6. Shear buckling and postbuckling behaviour
of the plate having the delamination between macrolayers of the plate having the delamination between macrolay-
3 and 4. ers 4 and 5.

usually characterized by a set of allowable stresses commonly used for orthotropic materials with une-
in the material principal directions. qual tensile and compressive strengths. The failure
The failure criterion is used to calculate a fail- index according to this theory is computed using
ure index (F.I.) from the computed stresses and the following equation (Thurley & Marshall 1995)
user-supplied material strengths. A failure index
of 1 denotes the onset of failure, and a value less
than 1 denotes no failure. The failure indices are F.I. = F1 σ1 F2 ⋅ σ 2 + F1111 ⋅ σ12 (8)
computed for all layers in each element of your
+ F22 σ 22 F66 ⋅ σ 62 + 2FF1122 ⋅ σ1 ⋅ σ 2
model. During post processing, it is possible to
plot failure indices of the mesh for any layer. 1 1 1
The Tsai-Wu failure criterion (also known F1 = − ; F11 = T C ;
as the Tsai-Wu tensor polynomial theory) is R1T R1C R1 R1

90

MARSTRUCT.indb 90 2/18/2011 5:42:00 PM

 Table 3. Ultimate stress for delaminated plates [MPa].

Position Type of
of delamination degradation Dx/Dy = 0.5 Dx/Dy = 1 Dx/Dy = 2

Macro-layer 1 Tension 25 20 20
– Macro-layer 2 Compression 90 – –
Macro-layer 2 Tension 25 20 20
– Macro-layer 3 Compression 90 – –
Macro-layer 3 Tension 25 20 20
– Macro-layer 4 Compression 90 – –
Macro-layer 4 Tension 25 20 20
– Macro-layer 5 Compression 90 – –

1 1 1 4
F2 = T
− C ; F22 = T C ;
R2 R2 R2 R2
1 importance to the composite industry. It involves a
F66 = 2 . (9)
R12
The coefficient F12 which represents the
parameter of interaction between σ1 and σ2 is to be
obtained by a mechanical biaxial test. In the equa-
tions (9), the parameters RCi , RTi are the compres-
sive strength and tensile strength in the material in
longitudinal direction (i = 1) and trasversal direc-
tion (i = 2). The parameter R12 is in-plane shear
strength in the material 1–2 plane.
According to the Tsai-Wu failure criterion, the
failure of a lamina occurs if
F.I > 1 (10)
In COSMOS, nonlinear material is considered
as a material with nonlinear behaviour (the non-
linear material curve) or case of introducing the
material strength components for Failure criteria
using for composites. This latest case is the case
analyzed in the paper.
The failure index in calculated in each ply of each
element. In the ply where failure index is greater
than 1, the first-ply failure occurs, according to the
Tsai-Wu criterion. In the next steps, the tensile and
compressive properties of this element are reduced
by the failure index. If the buckling did not
appeared until the moment of the first-ply failure
occurring, the in-plane load corresponding to this
moment is considered as the buckling load.
In the nonlinear calculus, for the buckling load,
the graphical method and Tsai-Wu failure criterion
were used. The values obtained for buckling load
were placed in the range specified in each case.
In Table 3, the variations of the buckling load
corresponding to the fails in the tension and
compressive cases, versus diameters ratios are
presented.
CONCLUSIONS
Delamination is a phenomenon that is of critical
breakdown in the bond between the reinforcement
and the matrix material of the composite. Under-
standing delamination is essential for preventing
catastrophic failures. Due to the geometry and
general load of the ship deck, buckling is one of
the most important failure criteria.
The FEM based methodology was success-
fully developed for the investigation of buck-
ling problems of composite plates having central
delamination.
Two hypotheses regarding the type of material
modeling were used (linear and nonlinear). The
FEM model is robust in that it can be used to predict
the global buckling loads of composite plates either
on one side or both sides. Finite-elements analysis
was carried out to assess the reliability of the meth-
odology. The two-sublaminate model developed in
this work provides a convenient method to model
delaminated composite panels.
For the values of in-plane loads lower than
80 MPa, the displacement values are increasing
since the diameters ratio is decreasing. This trend is
due to the fact the transversal diameter is decreas-
ing (since the delaminated area remains constant).
Smaller transversal diameter means increasing of
shear stiffness.
In the case of the in-plane loading values
bigger than 80 MPa, the displacement values are
increasing since the diameters ratio is increasing.
This trend is due to the contact pressure between
the layers in contact in the delamination, which is
increasing since the loading force is increasing.
For an in-plane loading of about 130–300 MPa
in each case, a small instant jumping of transversal
displacement is observed.
This means that what is recover in plate stiffness
after the increasing of contact pression in the dela-
mination area, is lost due to the lamina damage
occurring.

91

MARSTRUCT.indb 91 2/18/2011 5:42:03 PM

 As it is seen in Figures 3–6 for the plates with
delamination having the diameter ratio less than 1,
the buckling load is almost double than the plates
with the delamination having diameters ratio big-
ger than 1.
The buckling load determination is too difficult
without applying a graphical method, or applying
the Tsai-Wu failure criterion in the case of the gen-
eral buckling not occurred till the first-ply failure
occurring.
The first two failures occur in the material of the
plate at values of buckling load that is not depend-
ing on the position of delamination.
The first failure occurring in an element is based
on the Tsai-Wu failure criterion, which provides the
mathematical relation for strength under combined
stresses was used.
The failure index is calculated in each ply of
each element. In the ply where failure index is
greater than 1, the first-ply failure occurs, accord-
ing to the Tsai-Wu criterion. In the next steps, the
tensile and compressive properties of this element
are reduced by the failure index. If the buckling
did not appeared until the moment of the first-ply
failure occurring, the in-plane load corresponding
to this moment is considered as the buckling load.
ACKNOWLEDGEMENTS
The work has been performed in the scope of
the Romanian Project PN2—IDEI, Code 512
(2009–2011).
REFERENCES
Adams, D.F., Carlsson, L.A. & Pipes, R.B. Experimental
Characterization of Advanced Composite materials,
Ed. Taylor & Francis Group, 2003.
Altenbach, H., Altenbach, J. & Kissing, W. Mechanics of
Composite Structural Elements, Ed. Springer, Berlin,
2001.
Beznea, E.F. Studies and researches on the buckling
behaviour of the composite panels, Doctoral Thesis,
University Dunarea de Jos of Galati, 2008.
Beznea, E.F., Chirica, I., Boazu, D., Chirica, R. &
Chirica, A. Buckling Analysis of Delaminated Ship
Deck Plates, Made of Composite Materials, Proceed-
ings of the 24-th DAS-2007: Danubia-Adria Sympo-
92
MARSTRUCT.indb 92
sium on Development in Experimental Mechanics,
19–22 sept. 2007, Sibiu, Romania, pp. 213–214.
Beznea, E.F., Chirica, I. & Chirica, R. Buckling behav-
iour of plates with central elliptical delamination,
Proceedings of MARSTRUCT 2009, The 2-nd Inter-
national Conference on Marine Structures, Lisbon,
Portugal, 16–18 March 2009, Analysis and Design of
Marine Structures - Guedes Soares & Das (eds), 2009
Taylor & Francis Group, London, pp. 429–434.
Chirica, I., Beznea, E.F. & Chirica, R. Placi compozite,
Editura Fundatiei Universitare Dunarea de Jos,
Galati, 2006.
Chirica, I., Beznea, E.F., Chirica, R., Boazu, D. &
Chirica, A. Buckling Behavior of the Delaminated
Ship Hull Panels, Proceedings of The 12-th Interna-
tional Maritime Association of the Mediterranean
Congress—IMAM, 2–6 sept. 2007, Varna, Bulgaria,
pp. 161–166, vol. 1-Maritime Transportation, Ed.
Taylor & Francis, ISBN 978-0-415-43725.
Huo, M.V., Harik, I.E. & Ren, W.X. Buckling behav-
iour of stiffened plates, Int. J. Solids Struct, 39, 30,
pp. 39–55, 2002.
Jones, R.M. Mechanics of Composite Materials, Ed.
Taylor & Francis Group, London, 1999.
Johns, D.J. Shear Buckling of Isotropic and Orthotropic
Plates, A Review, R. & M. No. 3677, 1971.
Mallela, K.U. & Upadhyay, A. Buckling of laminated
composite stiffened panels subjected to in-plane
shear: A parametric study, Thin-Walled Structures 44,
pp. 354–361, 2006.
Nemeth, M.P. Buckling behaviour of long symmetrically
laminated plates subjected to combined loadings,
TP 3195, NASA, 1992.
Nemeth, M.P. Buckling behaviour of long symmetrically
laminated plates subjected to shear and linearly vary-
ing axial edge loads, TP 3659, NASA, 1997.
Smith, R.C.T. The buckling of plywood plates in shear,
Australian C.S.I.R. Aero. Res.Labs. (Melbourne) Rep.
SM 51, 1946.
Thurley, G.J. & Marshall, I.H. Buckling and Postbuck-
ling of Composite Plates, Ed. Chapman & Hall,
London, 1995.
2/18/2011 5:42:04 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Shear buckling analysis of the composite plates with cut-outs

E.F. Beznea & I. Chirica

University “Dunarea de Jos” of Galati, Romania

ABSTRACT: Shear buckling and postbuckling behaviour of the square plates, made of composite
materials, with central cut-outs is treated in this paper. In the analysis, finite element method (FEM) was
applied to perform parametric studies on various plates based on the shape and position of the ellipti-
cal hole. This study addressed the effects of an elliptical/circular cutout on the buckling load of square
composite plates. The laminated composite plates were arranged as symmetric cross-ply. The cutouts
are either circular holes or elliptical holes. The shear-buckling strengths of the plates could be increased
considerably only under aspect ratios. The plate-buckling mode can be symmetrical or anti-symmetrical,
depending on the plate boundary conditions, aspect ratio, and the hole size. In this paper, the analysis
has been performed only for the plate clamped on sides. The results and illustrations provide important
information for the efficient design of ship structural panels made of composite materials, having cut-
outs. The aim of the work presented in this paper is to analyze the influence of cut-out on the changes
in the buckling behaviour of ship deck plates made of composite materials. For each diameters ratio
there are plotted variation of the transversal displacement of the point placed in the middle of the plate,
according to the pressure that has been applied. Buckling load determination for the general buckling of
the plate has been made by graphical method. The post-buckling calculus has been performed to explain
the complete behaviour of the plate.

1 INTRODUCTION The buckling of flat square plates with central

Laminated polymer composites are being used
in many advanced structural applications and in
the last decade are extensively used in ship hull
structure. Cut-outs are commonly used in ship
hull components as access ports for mechanical
and electrical systems, or simply to reduce weight.
Structural panels with cut-outs often experience
in-plane loads that are induced mechanically can
result in panel buckling. The buckling behavior of
the structural panels with cut-outs is interested by
naval architects in the structural design.
For an unperforated rectangular plate of finite
extent (i.e., with finite length and finite width)
under uniform shear loading on the sides, the
closed-form buckling solutions are easily obtained
because the prebuckling stress field is uniform
everywhere in the plate. When a finite rectangu-
lar plate is perforated with a central cutout (e.g.,
a circular or a square hole), however, the buckling
analysis becomes extremely cumbersome because
the cutout introduces a load-free boundary that
causes the stress field in the perforated plate to
be non-uniform. Hence, the closed-form buckling
solutions are practically unobtainable, and various
approximate methods had to be developed to ana-
lyze such perforated plates.
circular holes under in-plane edge shear has been
studied both theoretically and experimentally by
various authors. (Mallela & Upadhyay 2006) pre-
sented a parametric study on laminated composite
blade-stiffened panels subjected to in-plane shear.
They proposed some design charts that can be
used to selecting the optimum parameters for bet-
ter stiffener propositions.
The methods of theoretical analysis used by
most of the past investigators were the Rayleigh-
Ritz minimum energy method and the Timoshenko
method. However, except for Schlack (1964) and
Kawai and Ohtsubo (1968), the theoretical analy-
sis methods used do not allow the boundary and
loading conditions to be precisely defined for
larger hole sizes because the stress distributions
of the infinite perforated plate are used as the pre-
buckling stress solution for the finite perforated
plate. Thus, most of the earlier buckling solutions
are limited to small hole sizes, and are not fit for
studying the effects of different plate boundary
conditions on the buckling strengths of the finite
plates with arbitrarily sized holes using those
approximate solutions.
The objective of this paper is to describe the
results of the research that has been conducted
on the buckling and postbuckling behaviour of

93

MARSTRUCT.indb 93 2/18/2011 5:42:04 PM

 rectangular composite plates under shearing loads
that have a cutout.
2 PLATES CHARACTERISTICS
The square plates (320 × 320 mm), clamped on the
sides, are made of E-glass/epoxy having the mate-
rial characteristics:
Ex = 46 [GPa], Ey = 13 [GPa], Ez = 13 [GPa],
Gxy = 5 [GPa], Gxz = 5 [GPa], Gyz = 4.6 [GPa].
μxy = 0.3, μyz = 0.42, μxz = 0.3.
Strengths:
T
x Pa ], RCx
[GPa [GPa ] ,
RTy [GPa ], RCy [GPa ] ,
Rxy = 0.72 [Gpa].
The thickness of the plate is 4.96 mm. The thick-
ness of a layer is 0.31 mm.
Topological code of the plate is [02/45/902/45/02]s.
For the material behaviour model two cases have
been considered:
− linear behaviour;
− nonlinear behaviour (Tsai-Wu failure criterion).
The in-plane loading was applied as a uniform
shear pressure on the sides (Figure 1).
The ellipse’s diameters of the cut-out area
placed in the middle of the plate are considered
from the condition of the same area for all cases.
In the parametric calculus, the following diameters
ratios are considered:
– Case 1 (Dx/Dy = 0.5): transversal diameter
Dy = 141 mm, longitudinal diameter
Dx = 70.5 mm;
– Case 2 (Dx/Dy = 1): transversal diameter
Dy = 100 mm; longitudinal diameter
Dx = 100 mm;
– Case 3 (Dx/Dy = 2): transversal diam-
eter Dy = 70.5 mm; longitudinal diameter
Dx = 141 mm.
3 SHEAR BUCKLING ANALYSIS
The analysis was carried out using COSMOS/M
finite element software. For the present study, a
3-D model with 4-node SHELL4L layered com-
posite element of COSMOS/M is used.
For shear buckling analysis the uniform pres-
sure on sides was incremental applied.
3.1 FE linear buckling analysis
A mesh convergence study for numerical eigenvalue
buckling loads has been done.
The results of the study is presented in the
Figure 2. As it is seen, the optimum number of the
element on side is 16 for all plates with cut-out hav-
ing the analized diameters ratio.
In Table 1, the results of linear buckling calculus
for each diameters ration are presented.
For the plate without cut-out, the load buckling
from linear buckling calculus is
pcr = 176.73 [MPa].
In the Figures 3, 4 and 5, the deformed plate
after buckling is presented for all cases of diam-
eters ratio.
As it is seen, due to the diametric symmetry of
the shear loading, the buckled shape plate in the
cases 1 and 3 are similar.

Table 1. Buckling load obtained in linear calculus.

Dx/Dy 0.5 1 2
pcr [MPa] 154.52 158.87 160.159

Dx
x
Dy

y
Figure 1. Shear loaded plate with cut-out. Figure 2. Results obtained in the convergence study.

94

MARSTRUCT.indb 94 2/18/2011 5:42:04 PM


Figure 3. The deformed plate after buckling, for
Dx/Dy = 0.5.
Figure 4. The deformed plate after buckling, for
Dx/Dy = 1.
Figure 5. The deformed plate after buckling, for
Dx/Dy = 2.
The results of the linear analysis are presented
in Figure 6, where the variation of the buckling
load function of the ratio Dx/Dy is plotted.
As it is seen, the buckling load is increasing since
the ratio Dx/Dy is increasing.
Opposite, in the case of uniaxial compressive
buckling the variation is so as the buckling load
is decreasing since the ratio Dx/Dy is increasing
(Chirica, Beznea & Chirica 2009).
95
MARSTRUCT.indb 95
161
160
159
Pcr [MPa]
158
157
156
155
154
0 0.5 1 1.5 2 2.5
Dx/Dy
Figure 6. Variation of buckling load (linear calculus)
versus diameters ratio Dx/Dy.
3.2 FE buckling nonlinear calculus
To solving of geometrically and material nonlinear
problems, the load is applied as a sequence of suf-
ficiently small increments so that the structure can
be assumed to respond linearly during each incre-
ment. For each increment of load, increments of
displacements and corresponding increments of
stress and strain are computed. These incremental
quantities are used to compute various corrective
stiffness matrices (variously termed geometric,
initial stress, and initial strain matrices) which
serve to take into account the deformed geom-
etry of the structure. A subsequent increment of
load is applied and the process is continued until
the desired number of load increments has been
applied. The net effect is to solve a sequence of
linear problems wherein the stiffness properties are
recomputed based on the current geometry prior
to each load increment. The solution procedure
takes the following mathematical form
(K + KI)i−1Δdi = ΔQ (1)
where
K is the linear stiffness matrix,
KI is an incremental stiffness matrix based upon
displacements at load step i−1,
Δdi is the increment of displacement due to the
i–th load increment,
ΔQ is the increment of load applied.
The correct form of the incremental stiffness
matrix has been a point of some controversy. The
incremental approach is quite popular (this is the
procedure applied in this study). This is due to
the ease with which the procedure may be applied
and the almost guaranteed convergence if small
enough load increments is used.
Buckling and post-buckling analysis has been
performed for all types of panels.
In certain cases, the general buckling does not
occur till the occurring of the first-ply failure.
2/18/2011 5:42:06 PM
 In these cases, for the buckling load calculus the
Tsai-Wu failure criterion may be used.
The Tsai-Wu failure criterion provides the
mathematical relation for strength under combined
stresses. Unlike the conventional isotropic materials
where one constant will suffice for failure stress
level and location, laminated composite materials
require more elaborate methods to establish failure
stresses. The strength of the laminated composite
can be based on the strength of individual plies
within the laminate. In addition, the failure of plies
can be successive as the applied load increases.
There may be a first ply failure followed by other
ply failures until the last ply fails, denoting the ulti-
mate failure of the laminate. Progressive failure
description is therefore quite complex for lami-
nated composite structures. A simpler approach
for establishing failure consists of determining the
structural integrity which depends on the defini-
tion of an allowable stress field. This stress field is
usually characterized by a set of allowable stresses
in the material principal directions.
The failure criterion is used to calculate a failure
index (F.I.) from the computed stresses and user-
supplied material strengths. A failure index of 1
denotes the onset of failure, and a value less than
1 denotes no failure. The failure indices are com-
puted for all layers in each element of your model.
During post processing, it is possible to plot failure
indices of the mesh for any layer.
The Tsai-Wu failure criterion (also known as the
Tsai-Wu tensor polynomial theory) is commonly
used for orthotropic materials with unequal ten-
sile and compressive strengths. The failure index
according to this theory is computed using the fol-
lowing equation, (Altenbach & al, 2004).
F.I. F1 σ1 + F2 σ 2 F1111 σ12 (2)
+ F22 σ 22 F66 σ62 + 2F F1122 ⋅ σ1 σ 2
where
1 1 1
F1 = − ; F11 = T C ;
R1T R1C R1 R1
1 1 1 1
F2 = T − C ; F22 = T C ; F66 = 2 . (3)
R2 R2 R2 R2 R12
1
The coefficient F12 which represents the
parameter of interaction between σ1 and σ2 is
to be obtained by a mechanical biaxial test. In
the equations (3), the parameters RCi , RTi are
the compressive strength and tensile strength in
the material in longitudinal direction (i = 1) and
trasversal direction (i = 2). The parameter R12 is
in-plane shear strength in the material 1–2 plane.
96
MARSTRUCT.indb 96
According to the Tsai-Wu failure criterion, the
failure of a lamina occurs if
F.I. > 1. (4)
In COSMOS, nonlinear material is considered
as a material having the behaviour according to the
Tsai-Wu Failure criteria.
The failure index is calculated in each ply of
each element. In the ply where failure index is
greater than 1, the first-ply failure occurs. In the
next steps, the tensile and compressive properties
of this element are reduced by the failure index.
If the buckling did not appeared until the moment
of the first-ply failure occurring, the in-plane load
corresponding to this moment is considered as
being the buckling load.
In the Figure 7, the variations of the buckling
load corresponding to the fails in the tension cases
(Fail 1) and compressive cases (Fail 2), versus
diameters ratio are presented.
As it is seen, the value of the buckling load cor-
responding to the fail 1 (tension) has the same
value for all diameters ratios.
For the panels with elliptical central cut-out the
values of the buckling load is placed in the range.
150 [MPa] < pcr < 190 [MPa]
Using graphical method, the buckling load may
be estimated by drawing an asymptote to the curve
in the point where the slope is changing suddenly.
The postbuckling behaviour of the plate may be
explained according to the curves in the Figure 8,
from region drawn after buckling occurring.
Figure 7. Variation of buckling load, corresponding to
fail 1 and fail 2, versus diameters ratio.
2/18/2011 5:42:09 PM
 Figure 8. Variation of maximum transversal displacement versus in-pane load for each diameters ratio.
In the nonlinear calculus, for the buckling load,
the graphical method and Tsai-Wu failure criterion
were used. The values obtained for buckling load
were placed in the range specified in each case in
Figure 7.
4 CONCLUSIONS
In the paper, the results of the FEM based meth-
odology that was successfully developed for the
investigation of buckling problems of composite
plates with central elliptical cut-out is presented.
Two hypotheses regarding the type of material
modeling is used (linear and nonlinear).
The buckling behavior of plates with cen-
tral holes as presented in figure 8 is quite pecu-
liar because, under certain boundary conditions
(clamped edges) and cut-out aspect ratios, the
mechanical-buckling strengths of the perforated
plates, contrary to expectation, increase rather
than decrease as the hole sizes grow larger. The
conventional wisdom is that, as the hole sizes
increase, the plates lose more materials and
become weaker. Therefore, the buckling strengths
were expected to decrease as the hole sizes
increase. This was not the case. Such peculiar
buckling phenomenon of the perforated plates
may be explained as follows.
Certain conclusions after the shear buck-
ling analysis of the perforated plates may be
performed:
− the buckling load is increasing since the ratio
Dx/Dy is decreasing;
− the buckling load corresponding to an element
failing which is damaged by tension, has the
same value that is not depending on the ratio
Dx/Dy;
97
MARSTRUCT.indb 97
− as the ratio Dx/Dy is increasing, the value of
the buckling load corresponding to an element
failing which is damaged by compression is also
increasing.
When the hole size becomes considerably large
relative to the plate width, most of the compres-
sive load is carried by the narrow side strips of
material along the plate boundaries. As it is well
known, a stronger plate boundary condition (e.g.,
clampedrather than simply-supported bounda-
ries) increases the buckling strength, while the
higher stress concentration decreases the buckling
strength.
Thus, which effects become dominant will
determine the increase or decrease of the buckling
strengths of the perforated plates.
For the circular-hole cases, the narrow shear
side strips are under stress concentration, which
reduces the buckling strengths.
The unusual buckling characteristics of the per-
forated plates offer important applications in ship
structural panel design. Namely, by opening holes
of proper sizes in ship structural panels for weight
saving, their buckling strengths can be boosted
simultaneously.
The buckling load determination is too difficult
without applying a graphical method. In certain
cases, the using of Tsai-Wu criterion may predict
so named “buckling load”, if general buckling of
the plate does not occurred before first-ply failure
occurring.
The first failure occurring in an element is based
on the Tsai-Wu failure criterion, which provides
the mathematical relation for strength under com-
bined stresses may be used.
The lack of the criterion is referring to the antici-
pation of the real mode to occurring the cracking.
2/18/2011 5:42:11 PM
 Taking into account the mathematical
formulation, the Tsai-Wu failure criterion is easy
to be applied. Additionally, this criterion offers
advantages concerning the real prediction of the
strength at variable loadings. It is to remark that
by applying linear terms, it is possible to take into
account the differences between the tension and
compression strengths of the material.
ACKNOWLEDGEMENTS
The work has been performed in the scope of
the Romanian Project PN2—IDEI, Code 512
(2009–2011).
REFERENCES
Adams, D.F., Carlsson, L.A. & Pipes, R.B. 2003. Experi-
mental Characterization of Advanced Composite mate-
rials, Ed. Taylor & Francis Group.
Altenbach, H., Altenbach, J. & Kissing, W. 2004.
Mechanics of Composite Structural Elements, Ed.
Springer, Berlin.
Ambarcumyan, S.A. 1991. Theory of Anisotropic Plates:
Strength, Stability, and Vibrations, Hemispher Pub-
lishing, Washington.
Baba, B.O. & Baltaci, A. 2007. Buckling Characteristics
of Symmetrically and Antisymmetrically Laminated
Composite Plates with Central Cutout, Appl Compos
Mater (2007) 14, 265–276.
Chirica, I., Beznea, E.F. & Chirica, R. 2006. Placi com-
pozite (in Romanian). Edit. Fund. Univ. Dunarea de
Jos, Galati, ISBN (10) 973-627-337-7; ISBN (13) 978-
973-627-337-7.
Chirica, I., Beznea, E.F. & Chirica, R. 2009. Buckling
behaviour of the ship deck composite plates with cut-
outs, Proceedings of MARSTRUCT 2009, The 2-nd
International Conference on Marine Structures,
Lisbon, Portugal, 16–18 March 2009, Analysis and
Design of Marine Structures - Guedes Soares &
Das (eds), 2009 Taylor & Francis Group, London,
pp. 423–428, ISBN 978-0-415-54934-9.
Dash, S., Asha, A.V. & Sahu, S.K. 2004. Stability of
Laminated Composite Curved Panels with Cutout Using
Finite Element Method, International Conference on
Theoretical, Applied Computational and Experimen-
tal Mechanics (ICTACEM 2004) December 28–31,
2004, IIT, Kharagpur.
Engelstad, S.P., Reddy, J.N. & Knight, N.F. Jr. 1992. Post-
buckling Response and Failure Prediction of Graphite-
Epoxy Plates Loaded in Compression, AIAA Journal,
30(8), 2106–2113.
98
MARSTRUCT.indb 98
Hilburger, M.F. 2001. Nonlinear and Buckling Behavior
of Compression-loaded Composite Shells, Proceedings
of the 6th Annual Technical Conference of the
American Society for Composites, Virginia.
Kawai, T. & Ohtsubo, H. 1968. A Method of Solution for
the Complicated Buckling Problems of ElasticPlates
With Combined Use of Rayleigh-Ritz’s Procedure in
the Finite Element Method, AFFDLTR-68-150.
Khamseh, A.R. & Waas, A.M. 1992. Compression Failure
Mechanisms of Uni-Ply Composite Plates with a Circu-
lar Cutout, AIAA-92-2276-CP.
Komur, M.A., Sen, F., Atas, A. & Arslan, N. 2010.
Buckling analysis of laminated composite plates with
an elliptical/circular cutout using FEM, Advances in
Engineering Software, 41/2: 161–164.
Kremer, T. & Shurmann, H. 2007. Buckling of tension-
loaded thin-walled composite plates with cut-outs,
Composite Science and Technology.
Liu, Y., Jin, F. & Li, Q. 2005. A strength-based multiple
cutout optimization in composite, plates using fixed grid
finite element method, Composite Structures 73 (2006)
403–412.
Mallela, U. & Upadhyay, A. 2006. Buckling of laminated
composite stiffened panels subjected to in-plane
shear: a parametric study, Thin-Walled Structures,
(44), 354–361.
Nemeth, M.P. 1996. Buckling and postbuckling behavior
of laminated composite plates with a cutout, NASA
Technical Paper 3587.
Qablan, H.A., Katkhuda, H. & Dwairi, H. 2009.
Assessment of the Buckling Behavior of Square
Composite Plates withCircular Cutout Subjected to
In-Plane Shear, Jordan Journal of Civil Engineering,
Volume 3, No. 2, 2009.
Rezaeepazhand, J. & Jafari, M. 2008. Stress Analysis
of Composite Plates with Non-circular Cutout,
Key Engineering Materials Vols. 385–387 (2008)
pp. 365–368.
Rezaeepazhand, J. & Jafari, M. Stress Analysis of
Composite Plates with a Quasi-Square Cutout
Subjected to Uniaxial Tension, Journal of Reinforced
Plastics and Composites July 2010 29: 2015–2026.
Schlack, A.L., Jr. 1964. Elastic Stability of Pierced Square
Plates. Experimental Mechanics, June 1964: 167–172.
SRAC. 2001. Cosmos/M FEM program user guide.
Structural Research & Analysis Corporation.
(www.cosmosm.com)
Thurvey, G.J. & Marshall, I.H. 1995. Buckling and
Postbuckling of Composite Plates, Chapman & Hall,
London.
Yazici, M. 2009. Influence of Cut-Out Variables on
Buckling Behavior of Composite Plates, Journal of
Reinforced Plastics and Composites October 2009
Vol. 28 No. 19, 2325–2339.
2/18/2011 5:42:12 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Robust ultimate strength formulation for stiffened plates

subjected to combined loads

Sang-Rai Cho
University of Ulsan, Ulsan, Korea

Hyun-Soo Kim
Hyundai Heavy Industries, Ulsan, Korea

Jeong-Bon Koo
Samsung Heavy Industries, Geoje, Korea

Hyung-Min Doh & Young-Kee Chon

Korean Register of Shipping, Daejon, Korea

ABSTRACT: A robust ultimate strength formulation is proposed for stiffened plates subjected to com-
bined axial compression, transverse compression, shear force and lateral pressure loadings. Before the
formulation was derived, a simplified numerical method was developed to trace the structural behavior
of stiffened plates under combined loadings. A rigorous parametric study was, then, performed using the
developed numerical method to predict the ultimate strength of various stiffened plates under various
combinations of loadings. The formulation was derived through a regression study using the parametric
study results. The accuracy and reliability of the proposed formulation were compared with those of a
commercial package, ABAQUA and DNV PULS and with the experimental results.

1 INTRODUCTION The followings summarizes the factors that

Most ship structures and topside decks of offshore
structures are composed of stiffened plates, which
are normally fabricated by welding. Welded plates
contain initial shape imperfections and residual
stresses. Furthermore, ship structures are subjected
to various combined loadings. Therefore, it is dif-
ficult to develop analytical methods or procedures
for predicting the ultimate strengths of marine
stiffened plates with reasonable accuracy.
Accurate ultimate strength analyses of ships’
stiffened plates require nonlinear analyses using
commercial packages. Learning to operate and
perform the analyses with these packages is time
consuming. To overcome these shortcomings, Det
Norske Veritas (DNV) developed a simplified
analysis package called Panel Ultimate Limit State
(PULS) and requested the performance of ulti-
mate strength analyses for all the stiffened plates.
Therefore, we developed our own simplified analy-
sis program and developed design formulations
with which structural designers can easily predict
the ultimate strength of stiffened plates subjected
to combined loadings.
should be considered in ultimate strength
predictions:
a. Interaction between yielding and elastic
buckling
b. Combined loading effects
c. Interaction between buckling modes
d. Initial shape and material imperfections
The ultimate strengths of marine structural
elements can be predicted using analytical meth-
ods, numerical methods, experimental methods,
simplified analytical or numerical methods and
design formulations. However, in most cases, ana-
lytical methods are not appropriate when the above
effects are considered in the procedures. Numeri-
cal predictions using commercial computer codes
are popular among researchers. Nonlinear analy-
ses using commercial codes are necessary for ulti-
mate strength predictions, but they are difficult to
operate and are time consuming. Among others,
experimental methods seem most reliable to struc-
ture owners, but they are still the most expensive
and require a relatively longer time to perform.
Any simplified analytical or numerical methods

99

MARSTRUCT.indb 99 2/18/2011 5:42:12 PM

 are welcomed by structural designers, but it is not
easy to develop simplified methods that provide
acceptable levels of accuracy. However, due to the
shortcomings and disadvantages of the aforemen-
tioned four methods, in structural design, the ulti-
mate strengths of marine structural elements are
predicted using easy-to-use strength formulations.
Ultimate strength formulations can be derived
through direct regression analysis of test data if the
numbers of available test data is sufficiently large
and if the ranges of their geometric and material
parameters and reduced slenderness parameters
include those of the actual structural elements.
However, the strength formulations obtained via
direct regression analysis may not demonstrate
meanings related to the physical states of the struc-
tural elements. Therefore, it seems more sensible to
adopt a strength formula as the basis of the for-
mulation and to perform regression analysis for
the modification factors in the formula. Test data
have their own uncertainties, which may obstruct
the collection of meaningful regression results for
certain parameters. This kind of difficulty can
be overcome by employing theoretical analysis
methods.
In this study, a simplified numerical method was
developed and the predictions using this method
were substantiated with those of ABAQUS (2009)
and PULS. Rigorous parametric study was con-
ducted covering wide range of design parameters.
Adopting the generalized Merchant-Rankine for-
mula as the basis a design formulation was devel-
oped using the results of the parametric study.
2 SIMPLIFIED ANALYSIS METHOD
2.1 General
For the ultimate strength analysis of stiffened plates
with initial shape and material imperfections, the
Finite Element method, Finite Difference method
and Dynamic Relaxation (DR) technique (Day
1965; Frieze et al., 1978) can be employed. In this
study, however, the DR technique is adopted as the
basis of the developed analysis method, for which
the procedure is straightforward and can also be
employed in dynamic problems.
In this study the stiffened plate is divided into
flat triangular elements and membrane and bend-
ing deformations of the element are decoupled.
This concept was originally proposed by Chan
and Davis (1983) and has been employed in plate
collision problems (Cho et al., 1996) and ultimate
strength analyses under hydrostatic pressure (Han
1999; Kim 2001).
The membrane deformation is represented
with finite triangular elements of constant strain
100
MARSTRUCT.indb 100
Figure 1. Physical concept of the analysis method.
and the bending deformation is assumed to
be concentrated at the edges of triangles (see
Figure 1). In other words the triangular elements
remain flat throughout the analysis, and triangle
edges can be folded. The edge rotations between
the triangles are restrained by non-linear torsional
springs representing the actual bending stiff-
ness of the flat element. Only three translational
degrees of freedom are contained in the analysis.
The component of the resultant displacement at
each node parallel to the element plane causes the
in-plane deformation and internal forces. Using the
displacement component normal to the element
plane the hinge folding angle is calculated and then
the bending moment is obtained. The out-of-plane
internal forces are derived to balance the bending
moment.
2.2 Formulation
2.2.1 Equations of motion
In the ultimate strength analysis using the DR
technique, the applied loads or imposed displace-
ments are increased step by step and the responses
of the structure are dynamically traced. When
the structural behavior is elastic the equations of
motion can be written as follows:
[ { }n + [C ]{D }n + [K ]{D}n = {Rext }n

] D (1)
where [M] = mass matrix; [C] = damping matrix;
[K] = stiffness matrix; {D}n = nodal displacement
vector at the nth time step: {  }n = nodal velocity
vector at the nth time step; { }n = nodal accelera-
tion vector at the nth time step; and {Rext}n = nodal
external force vector at the nth time step.
For elasto-plastic dynamic analysis, however,
the second term on the left hand side of
Equation 1, [K]{D}n can be written as {Rint}n and
we write Equation 1 as follows:
2/18/2011 5:42:12 PM
 [M ]{D}n + [C ]{D}n + {R } {R }
n
ext
n
(2)
where {Rint}n = nodal internal force vector at the
nth time step, which can be obtained by superpos-
ing {Reint}n for all of the corresponding elements.
The nodal internal force vectors, {Rint}n are
represented in the global coordinate, which are
transformed from the element local coordinate.
The nodal internal force vectors of each element in
the local coordinate are composed of those paral-
lel or normal to the element plane. The in-plane
internal forces (which are parallel to the element
plane) can be obtained through volume integra-
tion of the in-plane stresses of the element. The
normal components of the internal forces can be
calculated using the folding angles of the element
edges. The details of the internal force calculation
are discussed later.
The temporal central difference method pro-
vides the equations for nodal velocity and accelera-
tion as follows:
earlier the nodal internal forces in the global
coordinate, {Rint}n can be obtained by superposing
the internal forces of each element on the global
coordinate, {Reint}n. Actually, the internal forces of
each element are calculated in the local member
coordinate and then transformed into the global
coordinate. Therefore, we first explain how to cal-
culate the internal forces in the local coordinate and
describe the transformation procedure later.
In elasto-plastic problems, the strain-stress rela-
tionship is nonlinear. Therefore, it is necessary
to express this relationship in incremental form.
The strain increments, {Δε} are easily calculated
when the strains at the (n−1)th and nth time steps
are known. The stress increments, {Δσ} can be
expressed using Hooke’s law as in Equation (5).
{Δσ } [ ]({Δε } − {Δε }) (5)
The plastic strain increments {Δεp} can be
obtained using the Prandtl-Reuss flow rule, and
this study adopts the von Mises yield criterion.

{D }n = 2Δ1 t ({D} {D} ) (3a) 1. In-plane internal forces

In developing the analysis method, the con-

{D}n = Δ1t ({D} )

stant strain flat triangular element is employed
{D}n + {D} (3b) to analyze the in-plane deformation. The nodal
displacements in the local coordinate can be
where Δt = calculation time interval. obtained via transformation of those in the
By substituting Equations 3a and 3b into the global coordinate. Herein, we explain the pro-
equations of motion, Equation 2 can be rewritten cedure for calculating the in-plane nodal inter-
as follows: nal forces in the local coordinate from the nodal
displacements in the local coordinate. For this
purpose, the strain of each element should first
⎛ 1
[C ]⎞⎠ {D}n +1
1
⎜⎝ 2 [M ] + be calculated from the nodal displacements and
Δt 2 Δt the stress components can then be calculated
{ } { }
= R
n
R
n
1
Δt
(
+ 2 [M ] 2 {D}n − {D} ) (4) using the elasto-plastic strain-stress relation-
ship. Finally, the in-plane nodal forces can be
1 obtained through volume integration of the
+
2Δt
[ ]{ }n−1 stresses over the element as follows:

2.2.2 Internal forces
The nodal displacements at the (n + 1)th time step
can be calculated using Equation 4 together with
the nodal displacements at the (n − 1)th and nth
time step and {Rint} and {Rext} of the nth time
step. Accordingly, the velocity and accelerations at
the (n + 1)th time step can then be obtained from
Equations 3a and 3b. We assume in this study
that the characteristics of the applied loading are
known. Therefore the problem to be solved is how
to calculate the internal force vectors, {Rint}n using
the nodal displacements at the nth time step.
For the simplicity of equations the subscript ‘n’
will be omitted when deriving the equations for the
internal forces, {Rint}n when the displacement vec-
tors at the nth time step are provided. As mentioned
∫ [B ] { o } d (vol
T
re v ) (6)
vol
where [B] is the strain-displacement matrix.
2. Normal internal forces
In order to satisfy the moment equilibrium
conditions two adjacent element normal forces
are required at node i and l due to the torsional
moment at the common edge j–k (see Figure 2).
These normal forces are then balanced by the
normal forces at node j and k. The first step in
deriving the relationship for the out-of-plane
internal force is to describe the folding angle in
terms of the nodal translational displacements.
The nodal internal forces for each element
should be calculated in the local coordinate and

101

MARSTRUCT.indb 101 2/18/2011 5:42:13 PM

 compression and shear force ultimate strength
analyses were also performed for two models pro-
vided by DNV and the results were also compared
with those of PULS and ABAQUS.

2.3.1 For axial compression

Figure 2. Bending between two elements with normal
internal forces.
the forces need to be transformed into the global
coordinate before summing for a node. The coor-
dinate transformation is necessary not only for
this purpose but also when calculating the element
strain from the nodal displacements in the global
coordinates.
3. Development of computer program: SPUSA
Following the formulation of the developed
analysis method, a computer program, denoted
as Stiffened Plates Ultimate Strength Analysis
(SPUSA) was developed and its calculation pro-
cedure is as follows:
The dimensions and material properties of the
ten axially compressed test models used for
the comparison study are summarized in Table 1.
The comparison results of the predictions not
only by SPUSA, but also by PULS are provided in
Table 2, along with the test data.
In Table 1, a, B and b are the length, whole
breadth and spacing of stiffeners of the model,
respectively; ns is the number of stiffeners; tp is
the thickness of the plate; hw and tw are the height
and thickness of the stiffener web, respectively;
wf : and tf : are the width and thickness of the stiff-
ener flange, respectively; E and σY are the Young’s
modulus and the mean yield stress of the material,
respectively; and FB, AB and TB denote a flat-bar,
angle-bar and tee-bar, respectively.
As seen in Table 2, the theoretical predictions
of SPUSA and PULS agree relatively well with
the test results. However, the initial shape imper-
fection levels of the small-scale test models should
Table 1. Dimensions and material properties of test
models under axial compression (Unit: mm, MPa).
Stiff.
Model a B b ns tp type

step 1: set-up boundary conditions S3F3 600 250 100 3 2.13 FB

step 2: set-up displacement increments S5F3 600 450 100 5 2.13 FB
step 3: calculate strain increments S3A3 600 250 100 3 2.13 AB
step 4: calculate stress increments S5A3 600 450 100 5 2.13 AB
step 5: calculate bending moments S3A100 600 260 100 3 1.86 AB
step 6: calculate internal forces C-1–2 1434 1197 239 4 5.80 FB
step 7: calculate new nodal displacements C-3–4 1152 960 192 4 6.00 FB
step 8: < repeat steps 3–7 > R 1700 1168 457 3 9.95 TB
FL1 577 635 136 5 4.93 FB
After trying with different values for the dis- FL2 577 635 136 5 4.93 FB
placement increment the value finally chosen was
10−5 times the models dimensions (a × b), which Model hw tw wf tf E σY
provided the calculation efficiency and consistency
of results. S3F3 50 2.13 – – 248000 332
S5F3 50 2.13 – – 248000 332
2.3 Substantiation of the developed analysis S3A3 40 2.13 15 2.13 235000 330
method S5A3 40 2.13 15 2.13 235000 330
S3A100 40 1.99 15 2.13 220000 316
To substantiate the developed analysis method, ten
stiffened models were collected from the literature, C-1–2 105.8 5.70 – – 205000 271
which were subjected to axial compression. The C-3–4 102 8.00 – – 205000 269
models were analyzed and their ultimate strengths R 136.1 7.36 28.6 15.9 205000 377
were predicted using SPUSA. These predictions FL1 63.5 3.02 – – 190000 321
were compared with the test results and those FL2 63.5 3.02 – – 190000 247
of PULS. For combined loadings of transverse

102

MARSTRUCT.indb 102 2/18/2011 5:42:16 PM

 Table 2. Comparison of results of SPUSA with those Table 3. Dimensions and material properties of
of PULS (Unit: MPa). analyzed models.

Exp. PULS SPUSA Stiffener Plate 12.0 mm /

Model Ref. (1) (2) (3) (1)/(2) (1)/(3) span 8800 mm thickness 14.5 mm
Stiffener Stiffener
S3F3 Cho &
spacing 890 mm type Tee-bar
Song
(2003) 230 260 270 0.88 0.85 No. of Stiffener
S5F3 Cho & stiffeners 6 height 700 mm
Song
Young’s Web
(2003) 213 259 255 0.82 0.84
modulus 208 GPa thickness 13 mm
S3A3 Cho &
Song Poisson ratio 0.3 Flange width 150 mm
(2003) 261 281 291 0.93 0.90
Yield stress 355 MPa Flange
S5A3 Cho & thickness 18 mm
Song
(2003) 215 280 270 0.77 0.80
S3A100 Cho &
Table 4. Analyzed results of SPUSA, PULS and
Song
ABAQUS (Unit: MPa).
(2003) 260 259 253 1.00 1.03
C-1–2 Fukumoto Load ABAQUS PULS SPUSA
et al. Model type (1) (2) (3) (1)/(2) (1)/(3)
(1974) 202 218 241 0.93 0.84
C-3–4 Fukumoto Model I Pure
et al. shear 204 200 184.3 1.02 1.10
(1974) 239.1 242 238 0.99 1.00 Pure
R Murray trans.
(1975) 271 296 276 0.92 0.98 comp. 62.0 55.0 61.5 1.13 1.01
FL1 Faulkner Model II Pure
(1977) 250.1 266 250 0.94 1.00 shear 204 200 186.7 1.02 1.09
FL2 Faulkner Pure
(1977) 172.4 186 186 0.93 0.93 trans.
comp. 71.5 69.0 76.0 1.04 0.94

be higher than those assumed in the theoretical
analyses. Therefore, further study seems necessary
to investigate the effects of imperfection level on
the ultimate strength of stiffened plates.
2.3.2 Shear force and transverse compression
For the case of combined transverse compression
and shear force loadings the ultimate strengths of
two models provided by DNV (Byklum 2003) were
analyzed using SPUSA and the results were com-
pared with those of PULS and ABAQUS. Table 3
provides the dimensions and material properties
of the two models. The models with thicknesses of
12.0 mm and 14.5 mm are denoted as Model-I and
Model-II, respectively.
The ultimate strength analysis results of the two
models subjected to pure shear force and pure
transverse compression are summarized in Table 4
together with those of ABAQUS and PULS. When
the pure transverse compression is applied to
Model-I, the SPUSA prediction is approximately
10% greater than that of PULS but is similar to
that of ABAQUS. Under shear force, the ultimate
strength predicted by SPUSA is approximately
10% less than those of both PULS and ABAQUS.
Similar to the trend of Model-I, the strength
predicted by SPUSA is 10% greater than that of
PULS for transverse compression, but is 10% less
for shear force.
3 PARAMETRIC STUDY
3.1 Stiffened plates for parametric study
Prior to deriving the ultimate strength formula-
tion proposed in this study, a rigorous parametric
study was performed. The stiffened plates consid-
ered in the parametric study were similar to those
provided in ISSC 2000 committee VI.2 (Yao et al.,
2000). The size of the local panel between stiffen-
ers was taken as:
a × b = 2,400 × 800; 4,000 × 800 (mm)
tp = 10; 13; 15; 20; 25 (mm)
Three types of stiffeners were considered;
a flat-bar, an angle-bar and a tee-bar, and three

103

MARSTRUCT.indb 103 2/18/2011 5:42:16 PM

 Table 5. Dimensions of stiffeners.

Type Size 1 Size 2 Size 3

Flat-bar 150 × 17 250 × 19 350 × 35

Angle-bar 150 × 90 × 9/12 250 × 90 × 10/15 400 × 100 × 12/17
Tee-bar 138 × 9 + 90 × 12 235 × 10 + 90 × 15 383 × 12 + 100 × 17

Notes: flat-bar (h × t); angle-bar (h × bf × tw/tf); tee-bar (h × tw + bf × tf)

Table 6. Imposed boundary conditions.

Location Constraints

x=0 Tx = 0, Ry = 0
x = a/2, 3a/2 Tz = 0
x = 2a Ty = 0, Ry = 0
y = 0, 4b Tz = 0

Figure 3. Stiffened plate model for axial compression

loading.

sizes were assumed for each type of stiffener, as indi-

cated in Table 5.
The initial shape imperfections assumed in the
study were determined using Equation 7 for plates
and Equation 8 for stiffeners.

w po = Ao si mπ x
sin π y
a πx
+ Be sin (7)
b a
wso Bo sin π x vso i πx
Co sin
, (8)
a a

where Ao = tp/200, Bo = a/1000, Co = a/1000.

Instead of using m in Equation 7 as the aspect
ratio of the plate, we assumed m = 1 to better rep-
resent reality. The residual stresses due to welding
were assumed as those provided by ISSC2000 com-
mittee VI.2.
The stiffened plate model considered in the par-
ametric study is depicted in Figure 3, which shows
one and two halves span model. The number of
stiffeners in the standard model was three as shown
in Figure 3.
The boundary conditions imposed for the
ultimate strength analyses are summarized in
Table 6 and the displacement loadings applied
in the analyses are depicted in Figure 4 for axial
compression, transverse compression and shear
force. The displacement loadings were linearly
Figure 4. Applied linearly varying displacement load-
ings along the boundaries.
varying along the boundaries as shown in the
figures.
In the analyses using ABAQUS, the element
size of the model was 100 mm × 100 mm for the
plate and 100 mm × 50 mm for the stiffener web.
The stiffener flange was divided into 2 elements at
one side. The finite element chosen for the anal-
yses was S4R which is a 4-node, quadrilateral,
stress/displacement shell element with reduced
integrations and a large-strain formulation.

104

MARSTRUCT.indb 104 2/18/2011 5:42:17 PM

 3.2 Axial and transverse compression loadings Table 7. Comparison of predictions by PULS, ABAQUS
and SPUSA for axial and transverse compression
As a part of the parametric study, computa- loadings. <axial compression>
tions were first conducted for single loadings
including axial compression and transverse com- (Xm)axial
pression using PULS, ABAQUS, and SPUSA. The
results of these computations were utilized not PULS/ PULS/ ABAQUS/
only for the further substantiation of the developed Model ABAQUS SPUSA SPUSA
analysis method but also as part of the raw data 1.034 0.959 0.929
F3 (10∼25)15
with which the strength formulation was derived.
F3 (10∼25) 25 0.954 0.945 0.989
The computation results are summarized
F3 (10∼25) 35 1.010 1.024 1.013
in Table 6. The stiffened plates are denoted as
L3 (10∼25) 15 1.003 1.037 1.035
‘Xijklm’, indicating that:
L3 (10∼25) 25 0.976 0.970 0.992
X = F for flat-bar, L for angle-bar, T for
L3 (10∼25) 35 0.993 0.989 0.995
tee-bar;
T3 (10∼25) 15 0.954 0.988 1.036
i = aspect ratio (a/b); jk = plate thickness; lm = 15
T3 (10∼25) 25 0.959 0.955 0.995
for a stiffener of size 1, 25 for a stiffener of size 2,
35 for a stiffener of size 3. T3 (10∼25) 35 1.021 0.998 0.977
For example, F31015 represents a stiffened plate mean 0.989 0.985 0.996
with an aspect ratio and plate thickness of 3 and COV 5.30% 6.38% 4.02%
10 mm, respectively, for which the stiffener is a flat-
bar of size 1. <transverse compression>
As can be seen in the table, 45 cases were ana-
(Xm)trans.
lyzed for axial compression, and the ratios of pre-
dictions are summarized. The prediction ratios PULS/ PULS/ ABAQUS/
(Xm) of PULS/ABAQUS, PULS/SPUSA, and Model ABAQUS SPUSA SPUSA
ABAQUS/ SPUSA provide means of 0.989 (5.30%
COV), 0.985 (6.38% COV) and 0.996 (4.02% F3 (10∼25)15 0.912 0.927 1.017
COV), respectively. The results for the transverse F3 (10∼25) 25 1.014 1.087 1.076
compression load are also presented in the table. F3 (10∼25) 35 1.292 1.410 1.090
Comparing with the accuracy of the predictions L3 (10∼25) 15 0.967 1.003 1.042
for axial compression larger uncertainties can be L3 (10∼25) 25 1.012 1.072 1.061
found for transverse compression. L3 (10∼25) 35 1.069 1.132 1.059
T3 (10∼25) 15 0.948 0.979 1.037
T3 (10∼25) 25 1.008 1.033 1.032
3.3 Combined loadings T3 (10∼25) 35 1.026 1.093 1.068
The parametric study continued for combined mean 1.028 1.082 1.054
axial compression, transverse compression, shear COV 12.01% 13.60% 6.85%
force and lateral pressure loadings using PULS,
ABAQUS and SPUSA. Some of the results
Further computations were performed in the
obtained using PULS are depicted in Figures 4(a)
parametric study by changing the aspect ratio and
and 4(b).
plate thicknesses of the stiffened plates.
As shown in the figures, the effects of transverse
compression for the case of bi-axial loadings can be
negligible if the transverse compressive stress is less
4 DERIVATION OF STRENGTH
than about 10% of the yield stress. If the transverse
FORMULATION
compression is greater than that, however, the axial
capacity may decrease dramatically. When com-
4.1 Basis of the derived formulation
bined with shear force the reduction in axial com-
pression capacity is negligible until the shear stress As mentioned earlier in this paper, in order to
is less than about 30% of the yield shear stress. derive a good ultimate strength formulation, we
The effects of lateral pressure on the ultimate must take into account not only the interaction
strength of stiffened plates can be found by com- between yielding and elastic buckling but also
paring Figures 4(a) and 4(b). When axial compres- the interaction between different buckling modes.
sion is dominant lateral pressure may strengthen Even under axial compression alone various buck-
the ultimate capacity. However, when transverse ling modes may be experienced depending on the
compression is dominant the opposite effects can geometries and material properties of stiffened
be expected. plates. The axial compression becomes much more

105

MARSTRUCT.indb 105 2/18/2011 5:42:19 PM

 complicated when it is combined with transverse Using the parametric study results for axial
compression and/or shear force. compression and lateral pressure loadings, the
In this study, the generalized Merchant-Rankine knock-down factors ρc and ρt were determined
formula (Odland & Faulkner 1981) is adopted as through regression analysis as Equations 10(a)
the basis of the proposed ultimate strength formu- and 10(b), respectively. The knock-down factor for
lation for stiffened plates. In the formula proposed overall buckling, ρoa was assumed to be 1.0.
by Odland and Faulkner, the von Mises yield cri-
terion is adopted to monitor the collapse of stocky ρc β −0.69 λc1 63λt 0.09 (10a)
structures while the elastic buckling stress sub-
jected to combined loading is estimated using the ρt β 0.47
47
λc 3 51
λt −1.14 (10b)
linear sum of each component.
All the parametric study results obtained using ρoa = 1.0 (10c)
SPUSA, ABAQUS and PULS were utilized in
the regression analyses to derive the knock-down where β p σY E is the plate slenderness
factors. ratio; λc σY σ ec is the slenderness ratio of the
stiffener for column buckling; and λt σY σ et is
the slenderness ratio of the stiffener for tripping.
4.2 Axial compression
The generalized Merchant-Rankine formula can be
4.3 Transverse compression alone
written as Equation 9 for axial compression incor-
porated with lateral pressure (Cho et al., 1998). The proposed formulation for transverse compres-
sion alone can be written as follows:
2 2
⎛ σ xa σ xa + σ xbs σ xa ⎞ ⎛ σ xa + σ xb ⎞ 2
⎜⎝ ρ σ + ρ σ + + ⎟⎠ = 1 ⎛⎛ σ ⎞ ⎞
2
ρoaσ eoax ⎟⎠ ⎜⎝ σY
2
⎛ σy ⎞
c ec t et ⎜⎜ y
⎟ ⎟ +⎜ =1 (11)
⎝ σY ⎟⎠
(9)
⎜⎝ ⎝ ρ yCbcσ ey ⎠ ⎟⎠
where σxa = applied axial compressive stress;
( )
2 2
σxbs = bending stress at the stiffener flange where σ ey π E ( ν ) ⎡⎣ (b a ) ⎤⎦ t p b is
2 2 2

due to Meq; σec = Euler column buckling the, local plate buckling stress under transverse
stress of stiffener including associate plating; compression; ρ y λc1 34 } is the knock-
σY = σYp Ap σYs As Apps , is the mean yield stress;
down factor for local plate buckling under trans-
σ et 1 I o (G
GJJ 4π 2 L2 ( ECw )), is the elastic verse compression;
tripping stress of a stiffener; σ xb = σY × M eq M p , Cbc p{ . β1 65 (b a ) } is the strengthen-
03
is the equivalent bending stress due to the ing coefficient of the boundary condition for local
end bending moment and lateral pressure; plate buckling.
M eq M e + pbl 2 16 ; J hsf tsf 3 hswtssww3 3 , is the
St. Venant torsion constant; I 0 I w + As es2 + I f , is
the moment of inertia of the stiffener; Iw = polar 4.4 Shear force alone
moment of inertia of stiffener web; If is the polar When pure shear force is applied the mean ultimate
moment of inertia of the stiffener flange; es = dis- strength approaches the yield stress. The formula-
tance between the stiffener centroid (plate excluded) tion for pure shear force derived through regres-
and its toe; Cw I f (h ( hw t f )2 , is the torsional sion analysis of the parametric study results is as
warping constant;
onst follows:
σ eoax = n π Dy ax B 2 [ Dx B 2 Dy L2 + 2 m 2 Dxxy n2
Dy m 4 L2 n 4 B 2 ], is the overall grillage buck- 2
ling stress; L,B are overall length and breadth, ⎛ τ xy ⎞
2
⎛ 3τ xy ⎞
respectively; ax is the average cross-section area per ⎜⎝ ρ τ ⎟⎠ + ⎜ σ ⎟ = 1 (12)
τ e ⎝ Y ⎠
unit width of plating and longitudinal stiffeners; Dx,
Dy are effective flexural rigidity per unit width of
where τ e = { 806 + 3.. (b′ a )2 }E (tt p b )2 is the
stiffeners with attached plating in the longitudinal
elastic shear buckling stress of local plate; and
(x) and transverse (y) directions, respectively; Dxy is
ρτ = 4.0 is the knock-down factor for plate shear
the twisting rigidity per unit width
buckling
ρc is the knock-down factor for column buckling
of stiffener; ρt is the knock-down factor for trip-
4.5 Combined loads
ping of stiffener; and ρoa is the knock-down factor
for overall buckling of stiffened plate under axial Finally, the formulation to predict the ultimate
compression. strength of stiffened plates subjected to combined

106

MARSTRUCT.indb 106 2/18/2011 5:42:19 PM

 axial compression, transverse compression, shear
force and lateral pressure is obtained as follows:

⎛ σ xa σ xa + σ xbs σ xa
⎜⎝ ρ σ + ρ σ +
ρoaσ eoax
c ec t et
2
τ xy ⎞
2
⎛ σy ⎞
+⎜ ⎟ + ⎟
⎝ ρ yCbcσ ey ⎠ ρτ τ e ⎟
⎠
2
⎛
(σ xa + σ xb )2 − (σ xa + σ xb )σ y + σ y2 + 3τ xy2 ⎞⎟
+⎜ =1
⎜ σY ⎟
⎝ ⎠
(13)

4.6 Accuracy of the proposed formulation

The proposed strength formulation was derived
for single loadings via regression analyses using
the results of the parametric study. Therefore it is

Figure 5(b). Ultimate strength of stiffened plates sub-

jected to combined axial compression, transverse com-
pression and shear force with lateral pressure loadings
(p = 0.10 MPa).

necessary to determine its accuracy for combined

loadings. Figure 5 compares the predictions for
combined axial compression and transverse load-
ings according to the proposed formulation with
those of PULS, ABAQUS and SPUSA. Fairly
good agreements were observed.

5 CONCLUSIONS

In this study an ultimate strength analysis pro-

Figure 5(a). Ultimate strength of stiffened plates sub-
jected to combined axial compression, transverse compres-
sion and shear force without lateral pressure loadings.
gram called SPUSA, is developed for analysis of
stiffened plates subjected to combined axial com-
pression, transverse compression, shear force and
lateral pressure. The developed method employs the
Dynamic Relaxation technique temporally and the
Finite Element method spatially. Relevant test data
and PULS and ABAQUS predictions are utilized
to substantiate the developed method. Good agree-
ments were obtained between the predictions.

107

MARSTRUCT.indb 107 2/18/2011 5:42:27 PM


Figure 6. Comparison of predictions by the pro-
posed formulation with those of PULS, ABAQUS and
SPUSA.
A rigorous parametric study was conducted
using SPUSA, PULS and ABAQUS. Adopting
the generalized Merchant-Rankine formula as
the basis and using the parametric study results, a
robust ultimate strength formulation was derived
for stiffened plates subjected to combined axial
108
MARSTRUCT.indb 108
compression, transverse compression, shear force
and lateral pressure loadings.
REFERENCES
ABAQUS. 2009. Standard User’s Manual, 6.9.2.
Byklum, E. 2003. ABAQUS analyses of stiffened panel
subjected to combination of shear load, transverse
compression and lateral pressure. Det Norske Veritas,
Report No. 2003-0285.
Chan, A.S.L. & Davis, G.A.O. 1983. A simplified finite
element model for the impact of thin shells. Struc-
tures under Shock and Impacts, Proc. 1st Intl Conf on
Structures under Shock and Impact, Bulson, P.S. (Ed),
Elsevier, Amsterdam: 361–380.
Cho, S-R. & Song, I-C. 2003. Experimental investiga-
tions on the ultimate and post-ultimate strength of
Stiffened plates under axial compression. Jour. of Ship
and Ocean Technology (SOTECH), 7(1): 1–12.
Cho, S-R., Choi, B-W. & Frieze, P.A. 1998 Ultimate
strength formulation for ship’s grillages under com-
bined loadings. Proc. 7th Intl Sym. Practical Design
of Ships and Mobile Units (PRADS), Oosterveld,
M.W.C. and Tan, S.G. (Eds), Elsvier, Amsterdam:
125–132.
Cho, S-R., Lee, S-B. & Kim, I-W. 1996. Experimental
and theoretical investigations on the collision strength
of plates. Proc. 10th Asian Technical Exchange and
Advisory Meeting on Marine Structures (TEAM X),
Pusan National University, Pusan: 287–305.
Day, A.S. 1965. An introduction to dynamic relaxation.
The Engineer, 219: 218–221.
Faulkner, D. 1977. Compression tests on welded
eccentrically stiffened plates panels. Steel Plated
Structures, Dowling, P.J. Harding, J.E. and Frieze, P.A.
(Eds), Crosby Lockwood Staples, London: 581–617.
Frieze, P.A., Hobbs, R.E. & Dowling, P.J. 1978. Appli-
cation of dynamic relaxation to the large deflec-
tion elasto-plastic analysis of plates. Computers &
Structures, 63 (2): 301–310.
Fukumoto, Y., Usami, T. & Okamoto, Y. 1974. Ultimate
compressive strength of stiffened plates. Proc. ASCE
Speciality Conf. on Metal Bridges, St Louis.
Han, D-W. 1999. A study on ultimate strength analysis
technique for ring-stiffened cylinders having initial
shape imperfections. M.Sc. Thesis, Dept. of Naval
Architecture and Ocean Engineering, Univ. of Ulsan
(in Korean).
Kim, S-M. 2001. Experimental and theoretical investiga-
tion on the ultimate strength of ring-stiffened cylin-
ders having initial shape and material imperfections.
M.Sc. Thesis, Dept. of Naval Architecture and Ocean
Engineering, Univ. of Ulsan (in Korean).
Murray, N.W. 1975. Analysis and design of stiffened
plates for collapse load. Structural Engineer, 53:
153–158.
Odland, J. & Faulkner, D. 1981. Buckling of curved steel
structures—design formulations. Integrity of Offshore
Structures, Faulkner, D. Cowling, M.J. and Frieze, P.A.
(Eds), Applied Science Publishers, London: 419–443.
Yao, T. et al. 2000. Report of ISCC committee VI.2—
Ultimate hull girder strength. 2: 321–391.
2/18/2011 5:42:30 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Rapid analysis techniques for ultimate strength predictions

of aluminum structures

M.D. Collette
University of Michigan, Ann Arbor, Michigan, USA

ABSTRACT: A series of rapid semi-analytical methods for predicting the collapse of aluminum
structures is presented, including methods for tensile and compressive limit states. The methods presented
have been designed to be extensible to a wide range of structural topologies, including both conventional
stiffened-panel topologies and more advanced extrusion topologies. Unlike existing steel ultimate strength
methodologies, particular attention is paid to capturing aluminum-specific response features, such as
alloy-dependent material stress-strain curve shapes and the weakening effect of fusion welds. The methods
are validated against finite element analysis and previously published experimental results.

1 INTRODUCTION for steel, there are several reasons why existing

The design of the current generation of aluminum
high-speed vessels for commercial and military
applications has created the need for improved
aluminum structural engineering tools. The
increased size of these vessels and their exposed
operating environments has resulted in a corre-
sponding increase in the wave-induced loads on
these vessels. As these vessels become larger and
take on increasingly challenging military missions,
the safety implications of structural failure have
grown as well. To effectively optimize and approve
lightweight structures for such vessels, naval archi-
tects require rapid ultimate strength methods for
the 5xxx and 6xxx-series aluminum alloys suitable
for use in the design process.
The industry has developed mature rapid ulti-
mate strength tools for steel vessel structures,
mostly based on Smith (1977) type approaches to
the ultimate collapse strength of hull girders under
global bending loads. In the Smith-type approach,
individual load-shortening and load-extension
curves are derived for the different plates and
stiffened panel elements in the structure, and then
these curves are used to approximate the moment-
curvature relation for the hull girder. While there is
reason for concern that aluminum structures may
not fully satisfy the interframe collapse assump-
tion inherent in the Smith method (Benson et al.,
2010), extending the Smith-type collapse approach
is a logical first step in developing ultimate strength
methods for the new generation of aluminum
vessels.
Although there are a large number of mature
rapid load-shortening prediction methods available
steel formulations are unlikely to be sufficient
to support the analysis of aluminum structures.
First among these reasons is the weakening effect
of fusion welding on marine aluminum alloys.
In the heat-affected zone (HAZ) near a weld bead,
the aluminum material is weakened compared to
surrounding material, with reductions in material
proof stress typically between 30% and 50%. This
material inhomogeneity significantly complicates
the structural response; when loaded in tension,
plastic strains tend to localize in the HAZ and final
failure often involve rupture in these HAZ at low
global strains. Likewise, the HAZ may also impact
the compressive strength of the panel.
Another key difference between aluminum and
steel is that aluminum has a much more rounded
stress-strain curve than steel with an elastic modu-
lus only 1/3 that of steel. Furthermore, the degree
of rounding varies by alloy type. The rounded curve
removes a clear yield stress and thus aluminum
alloys are typically rated in terms of their 0.2%
offset proof stress. The 5xxx-series aluminum
alloys also show a marked softening before the
proof stress is reached which reduces the buckling
strength of these alloys as the tangent modulus has
been significantly reduced at stress values below
the proof stress.
A final important difference between aluminum
and steel is that aluminum is easy to extrude, espe-
cially in the 6xxx-series. This opens up many dif-
ferent geometric alternatives for stiffened panels
beyond rolled profiles welded to flat plates, such
as hat-shape stiffeners or double-sided extru-
sions, examples of which are shown in Figure 1.
The ability to employ custom extrusions means

109

MARSTRUCT.indb 109 2/18/2011 5:42:31 PM


Figure 1. Extruded alternatives to conventional
stiffened panel arrangements in aluminum.
that the naval architect must now check local
buckling modes of the stiffening shape—a check
traditionally performed by the material producer
when rolling compact steel profiles for ships.
This paper presents a series of new methods
for rapidly assessing the tensile and compressive
strength of aluminum structures. Key goals of the
methods are to include alloy specific stress-strain
curve in the formulation, be applicable to a wide
range of extrusion topologies, and to use only rapid
semi-analytical formulations so that the method is
suitable for early-stage design and optimization
approaches. The work performed on compressive
analysis is presented first, including analysis of un-
stiffened plates and stiffened extruded structures
and panels in compression. This is followed by an
examination of tensile limit states. Finally, conclu-
sions and recommendations for future work are
presented.
2 COMPRESSIVE ANALYSIS
The compressive analysis of stiffened panels and
extrusions is developed on a plate-separation
model approach, where the response of individual
un-stiffened plate elements is determined initially,
and then the combined properties of assemblies of
plate elements is addressed. In the work to date,
only formulations for uni-axial compression have
been validated and are presented here, although
potential extensions to shear and bi-axial compres-
sion are possible at the plate level using existing
interaction equations.
2.1 Plate element formulation
The basis for this theory is Hopperstad et al.,’s
(1999) extension of Stowell’s (1948) unified buck-
ling theory. Stowell’s theory proposes that the uni-
axial buckling stress can be expressed as:
σBuckle = ησElastic (1)
where σelastic is the elastic buckling stress and η is
used to correct for plastic effects in stocky sections.
The classical elastic buckling stress for plates is:
2
π 2E ⎛t⎞
σ Elasti
l c =k ⎜⎝ ⎟⎠ (2)
(
12 1 − ν 2 ) b
110
MARSTRUCT.indb 110
where ν is the elastic Poisson’s ratio for the material,
taken as 0.3 for aluminum, t is the thickness of the
member, b is the breadth, E is the elastic modu-
lus, and k is the buckling geometry coefficient. For
the flat plates simply-supported on all four edges
investigated in this report, k was taken as 4. The
η factor also changes based on plate geometry.
Stowell (1948) proposed the following for simply-
supported flat plates:
ESEC ⎛ 1 1 1 3 ETAN ⎞
η= ⎜2 + 2 4 + 4 E ⎟ (3)
E ⎝ SEC ⎠
where ESEC and ETAN are the secant and tangent
modulus of the material’s stress-strain curve, thus
capturing the alloy-dependent stress-strain curve.
As both the secant and tangent modulus depend
on the instantaneous stress level, an iterative
approach is required to calculate the buckling stress
via Equation 2 and Equation 3. This approach has
been shown to be reasonably accurate at predicting
the initial buckling stress in both the elastic and
plastic regions; however, it does not include any
post-buckling strength.
Hopperstad et al. (1999) extended Stowell’s the-
ory so that the entire compressive load-shortening
curve of the plate in compression could be pre-
dicted, thus including any post-buckling strength
that may occur. The basis for the approach is
the effective width approach, where the nominal
width of the plate is reduced as buckling occurs to
account for the loss of stiffness and resisting force
associated with buckling. Thus, for any instantane-
ous average compressive stress, the effective width,
beff, can be found as:
σ avg
befff b (4)
σe
where σavg is the average axial compressive stress
across the plate and σe is the current edge stress,
which is equal to the stress obtained from the
material compressive stress-strain curve at the cur-
rent value of axial strain.
Hopperstad et al.,’s approach fundamentally
assumes that the effective width of the plate once
the edge stress of the plate has surpassed the
initial buckling stress can be found by calculating
the width of the plate that would first buckle at the
current edge stress value. Thus, the stress after
the onset of buckling can be evaluated as:
σ = ησ Elasti
l cσ e ≤ σ e (5)
where the buckling coefficient η is calculated now at
the current edge-stress value and the elastic bucking
2/18/2011 5:42:32 PM
 stress is the same one from Equation 2, calculated
with the original plate geometry. The inequal-
ity simply enforces an effective breath proportion
equal to or less than one. The ultimate strength of
the plate is now found numerically as the maximum
value on the stress-strain curve. Hopperstad et al.,
presented some promising results for outstands
where one long edge the plate was left free, and
noted that this approach also seems to work well
when compared to simply-supported data, using in
part the Mofflin (1983) plate test results, without
providing details.
The initial proposal by Hopperstad et al., does
not include the impact of initial out-of-plane
deformations (IOOPD), residual stresses, or HAZ
on plate strength. Extensions to include these were
formulated. For IOOPD, a simplified model was
developed to reduce the effective plate thickness
based on the assumptions that: the IOOPD will be
roughly equal to the plate thickness and can be rep-
resented by a single parameter and the reduction in
plate ultimate strength is related to the amount of
bending induced by the IOOPD. By investigating a
series of aluminum plates tested by Mofflin (1983)
with various levels of IOOPD, the following for-
mula was developed which showed a good fit to
the Mofflin data:
⎛ M σU ET σ 0.2 ⎞ t central region, tension block region, and HAZ
tefff t ⎜1 − k ⎟ , with 2 < tefff
⎝ M σ 0.2 ET σU ⎠
Mσu = σutδMAX
σ 0.2t 2
M 0.2 =
6 Interaction between compressive residual
where k is a scaling factor found to be roughly
0.2 for the Mofflin plate dataset, ET is the tangent
modulus of the material stress-strain curve calcu-
lated at that 0.2% proof stress or the ultimate stress
of the plate without considering IOOPD. The
moment from Equation 7 is the moment induced
at the maximum IOOPD by the ultimate stress
σu, calculated by Equation 5 without considering
IOOPD, and the moment from Equation 8 is that
required to reach the yield stress at the extreme
fiber of the plate.
With the effective thickness determined from
Equation 8, the load-shortening curve determined
for the un-deformed plates is then corrected by the
ultimate strength of the plate with reduced effec-
tive thickness:
σU _ IOOP
σ IOOP σ PERFECT
σU _ PERFECT
This correction includes the slight reduction in
stiffness observed in plates with large IOOPD.
MARSTRUCT.indb 111
Additionally, the impact of HAZ in the direction
of the applied loading was included, as were resid-
ual stress effects. Residual stresses were modeled
with the familiar tension-block assumption along
the plate’s edges parallel to the applied load, with
a balancing compressive stress in the mid-region
of the plate. Three different corrections were made
to the load-shortening curve to account for these
impacts. First, the load-shortening curve for the
central part of the plate that was loaded in com-
pression was adjusted at each strain value by add-
ing the elastic strain resulting from the residual
stress and then subtracting off the residual stress
from the resulting stress value:
σ PLATE _ R (ε a ) = σ PLATE (ε a + ε R ) − σ R− (10)
where εa is the applied stress, εR is the locked-in
compressive residual strain, and σR is the compres-
sive residual stress. The outer edges of the plate
were then modified in the subsequent two steps.
The tension block was assumed to unloaded elas-
tically at first, and then re-load along the HAZ
material stress-strain curve. The HAZ that extends
outside of the tension block was assumed to fol-
low the HAZ stress-strain curve in compression.
The overall resisting force in the plate was then
assembled by adding the contributions from the
t (6)
region. This approach is only valid for welds par-
allel to the direction of applied stress. Transverse
(7) welds are also important, however, a simplified
model for these types of welds has not yet been
(8) developed.
stresses and IOOPD has been observed in steel
plates, and simply combining the strength reduc-
tion from these sources is usually overly conserva-
tive (Guedes Soares 1988). A similar situation
was observed in aluminum, and a method for cor-
recting the observed IOOPD in the presence of
residual stresses for a reduced IOOPD for use in
Equation 6 was defined:
⎛ σ l c σ R− ⎞
δ IOOP _ REDUCED δ IOOP ⎜ Elasti ⎟ (11)
⎝ σ Elasti
l c ⎠
where σElastic is the elastic buckling stress from
Equation 2, and σR is the compressive residual
stress.
(9) 2.2 Plate element verification
Three different sources of experimental validation
data are available for simply-supported aluminum
plates. The total available database consisted of
111
2/18/2011 5:42:34 PM
 130 different plate specimens. For this comparison
study, the well-known Ramberg-Osgood curve
was used to represent the stress-strain curve of the
aluminum alloy:
n alloy, although a smaller number of 2024-T3 and
σ ⎛ σ ⎞
ε= + 0.002 ⎜ (12)
E ⎝ σ 0.2 ⎟⎠
Each of the test programs noted here used
measured compressive stress-strain properties.
Using compressive stress-strain properties in place
of tensile stress-strain properties is important for
aluminum, especially for the 5xxx-series alloys in a
work-hardened condition as there may be large dif-
ferences between the tensile and compressive proof
stress, often exceeding 10%.
The first test program is that of Mofflin
(1983) which tested 76 aluminum plates in uni-
axial compression at the University of Cambridge.
The plates were made from 5083 and 6082 alloys.
The 6082 plates were in a “TF” temper, which
is roughly equivalent to a modern T6 temper,
while the 5083 were in a mill finish condition; a
loose specification used at the time with considera-
ble variability in proof stress. All of the plates were
nominally 6 mm thick, tested at a single aspect
ratio of four to one, and tested with simple sup-
ports on the unloaded edges that did not constrain
the plate from pulling in. Initial imperfections were
intentionally induced, although not in the plate’s
lowest buckling mode, and welds were simulated
along the longitudinal edges by making TIG passes
without depositing weld metal.
The second test program was a series of 6061
and 5456 plates tested under uni-axial compres-
sion at the David Taylor Model Basin in the 1960s
(Conley et al., 1963). The 6061 plates were in the T6
temper, while the 5456 plates were tested in the H24
and H321 tempers. Eight un-welded 6061 plates and
16 un-welded 5456 plates were used in the current
comparison. The plates were all 914 mm long, with
widths of 305 mm and 457 mm. The compressive
proof stress of the plates was measured, but the knee
exponent, n, in the Ramberg-Osgood relation was
not measured. For the purposes of the current vali-
dation study, it was assumed that the 6061-T6 alloys
had a knee exponent, n, of 25 while both tempers of
the 5456 alloy had an exponent of 16. The experi-
mental program addressed quite slender plates, with
b/t ranging from 48 to 144. IOOPD were noted to
be as-supplied, and believed to be typical of what is
present in the shipyard, but were not measured.
The third and final test program is a series of
aerospace alloy plates tested by NACA (Anderson
and Anderson 1956). While the plates were tested
over a wide range of slenderness, including very
stocky plates, the base material was normally quite
112
MARSTRUCT.indb 112
thin (1.6 mm) so the initial imperfections and
manufacturing residual stresses in these plates may
be different from thicker marine plates. 58 differ-
ent plate tests are included in the current validation
effort. The plates were primarily of the 7075-T6
2014-T6 plates were also included. A compressive
stress strain curve was presented for each alloy. The
plates covered b/t ranges from 15 to 60. The plate
length is not specified, but was selected so at least
five buckles could form in the direction of applied
load. IOOPD are thought to be small.
Initially, only Hopperstad et al.,’s extension to
the Stowell theory was compared to the unwelded
plates in the dataset, without corrections for
IOOPD. The results are shown in Figure 2.
Defining the bias of the prediction as the
predicted strength divided by the experimental
strength, the mean bias for the predicted ultimate
strength was 1.03 with a Coefficient Of Variation
(COV) of the bias of 8%. The initial buckling
stress, measured only in the DTMB and NACA
data sets was predicted with a mean bias of 1.03
and COV of 12%. Notably, the bias in the formula
was independent of the alloy type, indicating that
the effect of alloy-dependent material stress-strain
curve has been adequately captured.
IOOPD, residual stresses, and HAZ corrections
were applied progressively to Mofflin plates that
Predicted Failure Stess/0.2% Proof Stress
1.5
Perfect Prediction
Mofflin Plates with large IOOP
Mofflin Plates with small IOOP
DTMB Plate-All Alloys
1 NACA Plates - All Alloys
0.5
0
0 0.2 0.4 0.6 0.8 1 1.2
Experimental Failure Stess/0.2% Proof Stress
Figure 2. Comparison of ultimate strengths from
Equation 5 against all unwelded plate data.
1.5
Perfect Prediction
Predicted Failure Stess/0.2% Proof Stress
Stowell, No OOP
Stowell, With OOP
Stowell, With OOP and Weld
1
0.5
0
0 0.2 0.4 0.6 0.8 1
Experimental Failure Stess/0.2% Proof Stress
1.2 1.4
Figure 3. Corrections for IOOPD and residual stress
from welding on Mofflin plate test data.
2/18/2011 5:42:37 PM
 had both measured IOOPD and simulated welds.
The results are shown in Figure 3, indicating that
the proposed corrections largely remove the error
arising from these sources.
2.3 Panels and extrusions formulation
Load-shortening response is required at the panel
level, as well as the plate level, so that grillage and
hull-girder collapse modes can be investigated.
Typically, stiffeners and attached plate or extru-
sion section can be idealized as a column sup-
ported by the transverse frames on each end of the
panel. Using this column idealization, an elastic
column buckling parameter, λ, can be defined for
the stiffened panel roughly analogous to the plate
slenderness parameter β ⋅ λ depends on the length
of column, l, and the cross-sectional area, A, and
area moment of inertia, I.
b σ 0.2
β=
t E (13)
l σ 0.2 I
λ= , r=
πr E A plates, except in the Faulkner et al., approach, the
An approximate load-shortening methodology
is presented in this section. This is based upon the
classical panel column buckling approach devel-
oped and presented by Faulkner et al. (1973) for
steel panels, which was further extended to pre-
dict the entire load-shortening curve by Gordo
and Guedes Soares (1993). In this approach, the
strength of a steel plate and stiffener combination
in compression can be determined as:
σC σCB ⎛ AS bet ⎞
= (14)
σY ⎜⎝ AS bt ⎟⎠
σY
where σC is the column failure stress, σY is the
yield stress of the steel, σCB is the column buckling
stress, b is the plate width, be is the effective plate
width at failure, t is the plate thickness, and AS is
the stiffener cross-sectional area. The second term
in this expression assumes that the plate between
stiffeners will buckle before the overall panel fails,
and thus an area reduction factor for this must be
included. In this approach, the column buckling
stress is calculated using the Johnson-Ostenfeld
interaction approach to transition from elastic
buckling to in-elastic buckling as:
σ CB 1 σY
= 1− , σ E ≥ 0 5σY
σY 4 σE (15)
σ CB σ E
= , σ E < 0 5σY
σY σY
113
MARSTRUCT.indb 113
where σE is the Euler column buckling stress that is
distinct from the edge stress, σe, used above with a
lower-case e subscript. σE is found as:
π2 I F ′
σE = 2
E EEFF (16)
a AS bet
where a is the panel length between support, and
IEFF′ is a tangent effective moment of inertia, cal-
culated assuming the plate between stiffeners has a
tangent effective width, be′, related to the instanta-
neous compressive modulus of the plate account-
ing for its buckling failure. This is different than the
effective width, be, discussed previously. Faulkner
et al., gives formulations so effective width and
tangent effective width can be predicted for plates
within the edge stress range of 75% to 100% of
the plate yield stress. These are determined using
and effective plate slenderness, βe that replaces the
proof or yield stress definition in the plate slen-
derness, β, from Equation 13 by the current edge
stress. This approach is conceptually similar to the
Stowell buckling approach presented previously for
maximum strength obtained by the plate is deter-
ministic, and given when the edge stress reaches
the yield stress of the material.
This approach forms the basis of Gordo and
Guedes Soares (1993) extension to the method over
a wider range of edge stresses. Gordo and Guedes
Soares also extended this method to predict the
entire load-shortening curve of a column in com-
pression by assuming that the approach can be gen-
eralized to any strain level. This in turn assumed that
Faulkner et al.,’s equations for effective breadth and
tangent effective breadth are valid outside of the
range initially proposed, however the resulting load-
shortening curves have proven to be quite accurate.
The conceptual approach taken by Faulkner
et al., and the modifications by Gordo and Guedes
Soares for predicting the entire load-shortening
curve were used as the basis of a load-shortening
curve methodology in the present work. In this
approach, the load-shortening curve accounting
for both plate and column buckling failure modes
was computed as follows.
First, the edge strain at which column failure
occurs was computed using Equation 14. In this
equation, the area term was adjusted for the gen-
eral extrusion cross-section case where multiple
plating elements may buckle by replacing the area
term at the end of the expression by:
AEfff (ε ) ∑ σ Efff i (ε ) Ai
, AEfff = i (17)
ATotal ∑ σ M (ε ) Ai
i
2/18/2011 5:42:39 PM
 where σEff is the average compressive stress for each Table 1. Performance of simplified panel strength
plate element, 1..i in the cross section calculated by method.
the plate strength approach given previously, σM is
Ultimate strengh Ultimate strain
the fully-effective edge stress based on the mate-
rial stress-strain curve in compression, and A is Experiment Mean Cov Mean Cov
the cross-sectional area of each plate element in
the overall cross section. The dependence on the SSC-451 1.12 0.13 1.04 0.28
applied edge strain, ε, indicates that the area ratio NTNU “Hat” 1.05 0.04 0.90 0.14
changes with applied strain. The effective width A.R.E. 0.97 0.20 0.81 0.20
and tangent effective width for Equations 14–16
are calculated directly from the theoretical defi-
nitions of each quantity and the load-shortening
curve determined for each plate element as out-
lined previously.
This method is then extended to predict the
entire load-shortening curve, following the general
approach of Gordo and Guedes Soares. At strains
below the column failure strain, εCOL, the net resist-
ing force in the column is calculated by summing
the individual plate load-shortening curves for
each plate element in the panel cross-section.
At strains above the column failure strain, εCOL,
the net resisting force in the column is determined
by applying Equation 14 at the instantaneous
value of edge strain. This approach gives a rapid
method of determining the load-shortening curve Figure 4. Comparison of prediction and experimental
of aluminum panels and complex extrusions in load-shortening curve, NTNU panel P.
compression.

be σ (ε ) The results of the panel method are shown in

(ε ) = Efff i
b σ M (ε )
⎛ σ Efff i (ε ) ⎞ (18)
∂⎜ ⎟
⎝ σ M (ε ) ⎠
′ for the panel ultimate strength, and slightly con-
be
(ε ) =
b ∂ (ε )
2.4 Panels and extrusions verification
The proposed panel method was verified against
three different experimental programs. The first
was a recent large-scale test program of 5083,
5383 and 6082 alloy panels conducted by Profes-
sor Paik and published as Ship Structure Com-
mittee report SSC-451 (Paik et al., 2008). These
were conventionally-stiffened panels, with fusion
welded stiffeners. The second test program was a
group of 4 extruded panels in 6082 alloy, tested
at NTNU (Aalberg et al., 2001). These stiffen-
ers were “hat” shaped hollow extrusions joined
together by friction stir welding to form a 3 or
5-stiffener wide panel, two replicates were tested
at two different overall panel lengths. The final
group consisted of three tee-stiffened panels and
one flat bar panel constructed in the 5083 alloy
and tested in mid 1980s in the United King-
dom (Clarke and Swan 1985) by the Admiralty
Research Establishment (A.R.E.).
Table 1 below, using the same definition of bias
and covariance as for the plate equations above.
The initially proposed method is slightly optimistic
servative for the panel ultimate strain. An example
load-shortening curve is given in Figure 4.
In general, it appears that updating the effec-
tive width and tangent effective width of plates
post-failure by Equation 17 is overly conservative,
and was causing the reduction in ultimate strain
shown in Table 1 and load-shortening curves that
were too steep in the post ultimate strength region.
Equation 17 reduces the effective properties of the
panel much more quickly than Gordo and Guedes
Soares extension of the Faulkner method.
While the current method does not include stiff-
ener tripping, and should not be used on panels
where stiffener tripping could occur, with further
improvements to Equation 17, it does appear to be a
promising method of predicting the load-shortening
curve of complex extrusions that cannot be ideal-
ized as plate-stiffener combinations.
3 TENSILE RESPONSE
A key component of any progressive collapse
approach is the prediction of the response of the

114

MARSTRUCT.indb 114 2/18/2011 5:42:41 PM

 tension flange of the hull girder. For steel vessels,
this is normally assumed to follow the material
stress-strain curve in tension. However, an alumi-
num vessel will have regular HAZ transverse to the
longitudinal bending stress, from unit and block
joints as well as HAZ from fillet welds joining web
frames to the shell plate. Thus, the tensile stiffness
of an aluminum hull girder depends upon the base
material properties as well as the size and distribu-
tion of these HAZ. Previous work has examined
a range of simplified models and compared them Figure 5. Idealized HAZ model for tensile response.
to limited experimental tests with mixed results
(Collette 2007), as well as examining the risk of
fracture in the HAZ as a collapse mode that could
induce hull girder collapse (Collette 2005).
In this work, non-linear finite element analysis
of idealized HAZ in a butt weld, ignoring any
weld metal reinforcement was made as a way of
exploring the distribution of strain under uniform
tensile extension. From such analysis, the required
Figure 6. 2-D FEA model with centerline symmetry for
features of a simplified model were determined.
tensile response.
A picture of the idealization of the weld is shown
in Figure 5. A key to understanding how the HAZ
responds in tension is the concept of constraint.
When loaded in tension, weaker HAZ material
will deform plastically before the surrounding
base material. In isolation the plastic HAZ mate-
rial would shrink both through the thickness of
the plate and along the length of the weld, but is
prevented from doing so by being fused to the still-
elastic base metal. Thus, the base metal constrains
the HAZ and induces a hydrostatic stress state in
the HAZ material which can retard yielding of the
HAZ. The HAZ joint therefore appears stronger
and stiffer than the HAZ material properties deter-
mined from small weld or coupon tests without
adequate surrounding material.
The initial study concentrated on 2-D plane-
strain FEA models built with the ABAQUS Figure 7. HAZ width influence on through-thickness
finite-element code. The plane-strain assumption constraint.
represents complete constraint along the length of
the weld and addresses constraint in the through-
thickness direction only. In ABAQUS, eight-node wide 6082 alloy butt weld models was computed
CPE8R quadratic elements were used along with allowing the weld HAZ width to vary from 40 mm
the deformation plasticity model for material to 2 mm (5 times the plate thickness to 0.25 times
properties. All models used a symmetry boundary the plate thickness). The results from this study are
condition at the weld centerline with enforced dis- shown in Figure 7. The HAZ proof strength was
placement at the free edge of the weld. Square ele- kept as 50% of the base material proof strength
ments were used with a minimum of eight elements in this study. It appears that the through-thickness
through the thickness of the plate. The plate was constraint becomes significant only for HAZ
assumed to be 8 mm thick with an overall width of whose width is less than the plate thickness. For
125 mm. Within this 125 mm overall width, the rel- very narrow HAZ, the HAZ strength approaches
ative width of the HAZ, strength of the HAZ, and that of the base material as the constraint stresses
shape of the material stress-strain curve were var- become quite large. For most fusion welds, the
ied parametrically. A sketch of the 2-D FEA model HAZ width is assumed to be roughly three times
and boundary conditions is shown in Figure 6. the thickness; based on these result the through-
To evaluate the effect of through-thickness con- thickness constraint does not need to be included
straint forces, the load-extension curve for 125 mm in the simplified tensile response model.

115

MARSTRUCT.indb 115 2/18/2011 5:42:43 PM

 The effect of constraint along the direction
of the weld was studied next by removing the
plane-strain assumption of the 2-D model. In a
large stiffened panel, this constraint can likely be
approximated by a plane strain condition on the
HAZ alone, as differential deformations between
the HAZ and base material along the length of
the weld cannot develop. However, such constraint
forces are likely to be missing in small-scale tensile
test specimens typically used to develop material
properties for structural modeling. Thus, a method
of moving from the small-scale measured proper-
ties to the response of a larger structure is needed.
To study this effect, the 2-D FEA model shown
in Figure 6 was replaced by a series of 3-D mod-
els where the depth of the model out of the plane Figure 8. Load-extension curves for 6082 butt weld,
Figure 6 was progressively increased. For this HAZ width three times plate thickness, HAZ strength
model, 20-node quadratic elements were used with varies from 50% to 90% of base plate strength.
the same boundary conditions as Figure 6. Based
on this study, it appears that a simple correction
factor, c, accounting for the constraint stresses in 4 CONCLUSIONS
the transverse direction applied to both the uni-
axial proof stress and elastic modulus of the HAZ A series of simplified ultimate strength models
used in Equation 12 would be adequate to approxi- have been presented for the compressive and ten-
mate the behavior: sile response of aluminum welds, plates, and stiff-
ened panels. These models are designed to allow
1 Smith-type progressive collapse approaches to be
c= ≈ 1.13 (19) implemented for aluminum vessels. These models
1 − υ + υ2 take into account important differences between
aluminum and steel, such as alloy-dependent stress-
With this correction factor applied to the strain curve shape, weaker material near fusion
HAZ material properties, a very simple “series” welds, and the ability to extrude aluminum into
model (Collette 2007) can be proposed for the complex extrusions. In limited comparison of the
load-extension of the HAZ, where the strain, results to FEA studies and published experiments
ε, in the HAZ material and base material are found the methods generally perform encouragingly, and
separately by requiring equal stress (and hence net represent a rapid, aluminum-specific approach to
force) in each material, using Equation 12 with ultimate strength calculations.
proof stress and elastic modulus modified by the
c factor computed from Equation 18. These strains
are then combined in proportion to their lengths, ACKNOWLEDGEMENTS
L, to give the overall strain of the joint:
The author would like to acknowledge support of
ε LHAZ + ε BASE LBBASE Dr. Paul Hess at the U.S. Office of Naval Research,
ε = HAZ H (20)
Code 331, who has supported this work through
LHA
H Z + LB
HAZ BASE
S
several different projects and grants, and encour-
A comparison between this simplified model aged its development.
and the plane-strain 2-D FEA analysis for a 6082
HAZ joint with a HAZ width three times the plate
thickness and varying levels of proof strength in REFERENCES
the HAZ from 50% to 90% of the base strength is
shown in Figure 8. As can be seen from the figure, Aalberg, A., Langseth, M. & Larsen, P. 2001. Stiffened
the simplified model performs remarkably well aluminium panels subjected to axial compression.
Thin-Walled Structures 39:861–885.
until through-thickness deformations reduce the
Anderson, R. & Anderson, M. 1956. Correlation of crip-
cross-sectional area in the HAZ thus reducing the pling strength of plate structures with material prop-
strength predicted by the FEA model. This sim- erties. NACA Technical Note 3600. Washington, DC:
plified model can be used to determine the load- NACA.
extension curve of the hull girder in the presence Benson, S., Downes, J. & Dow, R. 2010. A semi-analytical
of transverse HAZ. method to predict the ultimate strength and

116

MARSTRUCT.indb 116 2/18/2011 5:42:46 PM

 collapse behavior of orthogonally stiffened panels.
In Proceedings 11th International Symposium on Prac-
tical Design of Ships and Other Floating Structures,
PRADS 2010, Rio de Janeiro: 1091–1101.
Clarke, J. & Swan, J. 1985. Interframe buckling of alu-
minium alloy stiffened plating. Admiralty Research
Establishment Dunfermline Report AMTE(S) R85104.
Dunfermline: ARE.
Collette, M. 2005. Strength and reliability of aluminium
stiffened panels. PhD Thesis, School of Marine Science
and Technology. Newcastle: University of Newcastle
upon Tyne.
Collette, M. 2007. Impact of fusion welds on the ultimate
strength of aluminum structures. In Proceedings 10th
International Symposium on Practical Design of Ships
and Other Floating Structures, PRADS 2007, Houston,
Texas: 944–952.
Conley, W., Becker, L. & Allnutt, R. 1963. Buckling and
ultimate strength of plating loaded in edge compres-
sion progress report 2—unstiffened panels. David
Taylor Model Base, Structural Mechanics Laboratory,
R&D Report 1682. West Bethesda, MD: DTMB.
Faulkner, D., Adamchak, J., Snyder, G. & Vetter, M.
1973. Synthesis of welded grillages to withstand com-
pression and normal loads. Computers & Structures 3:
221–246.
Gordo, J. & Guedes Soares, C. 1993. Approximate load
shortening curves for stiffened plate under uni-axial
compression. In Proceedings of the 5th International
Conference on the Structural Integrity of Offshore
Structures, Glasgow, UK: 189–211.
117
MARSTRUCT.indb 117
Guedes Soares, C. 1988. Design equation for the
compressive strength of unstiffened plate elements
with initial imperfections. J. Construct. Steel Research
9: 287–310.
Hopperstad, O., Langseth, M. & Tryland, T. 1999. Ulti-
mate strength of aluminum alloy outstands in com-
pression, Thin-Walled Structures 43: 279–295.
Mofflin, D. 1983. Plate buckling in steel and aluminum.
PhD Thesis, Trinity College. Cambridge: University of
Cambridge.
Paik, J., Thayamballi, A., Ryu, J., Jang, J., Seo, J.
Park, S., Soe, S., Renaud, C. & Kim, N. 2008.
Mechanical collapse testing on aluminum stiffened
panels for marine applications. Ship Structure Com-
mittee Report SSC-451. Washington, DC: Ship Struc-
ture Committee.
Smith, C.S. 1977. Influence of local compressive failure
on ultimate longitudinal strength of a ship’s hull.
In Proceedings International Symposium on Practical
Design in Shipbuilding, PRADS 77, Tokyo: 73–79.
Stowell, E.Z. 1948. A unified theory of plastic buckling
of columns and plates, NACA Technical Note 1556.
Washington, DC: NACA.
2/18/2011 5:42:48 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

A revisit on design and analysis of stiffened shell structures

for offshore applications

Purnendu K. Das, K.K. Subin & Paul C. Pretheesh

Department of Naval Architecture and Marine Engineering, Universities of Glasgow and Strathclyde,
Glasgow, UK

ABSTRACT: Advanced design methods and procedures are getting published every now and then from
all corners of the world. But the practicing codes could not always append all the refined or essential
recommendations timely. Stiffened cylinders are one among those structural components. Most of the
offshore floating platform components are made as stiffened cylinders and an improved model in the
design process can affect the total construction cost and schedule to a great extent. Reliability based
design approach is now seems to be advantageous over the deterministic type of structural designing
process as it addresses uncertainties in the design variables and leads to consistent level of safety. The Reli-
ability based approach still needs a robust strength model to predict the capacity with respect to random
design variables. Numerical analysis methods are suitable for this purpose but the time and effort involved
are quite high and hence a robust analytical approach is preferred for reliability analysis. DNV and API
are the most widely used design codes which offer strength models for stiffened cylindrical shell under
different loading conditions. This paper establishes a strength model for ring, stringer and orthogonally
stiffened cylindrical shells which is actually a modified version of a strength model proposed earlier. The
proposed model shows better agreement with the experimental results compared to the practicing DNV
and API design codes. The model uncertainty factor and the strength model can be utilised for the reli-
ability analysis of similar structures.

1 INTRODUCTION a tool to predict the structural capacity very accu-

Cylindrical shells are one of the major structural
components in offshore engineering world.
Researchers from the last century (Thimoshenko
and Gere (1961), Windenburg and Trilling (1934),
Von Mises (1929) etc.) rigorously investigated
the underlying mechanisms of this category of
structures and predicted the structural behaviour
under various loading conditions. Many of these
closed form relations in terms of the basic geo-
metrical and material design parameters predicts
the behaviour reasonably accurate. The revolution-
ary developments in the computing realm within
the last century increased the power of numerical
analysis to a great extent that it can predict results
so close to the reality. It has the capability to analyse
the structures with all its geometrical and material
complexities under static or dynamic loading situ-
ation with prevailing environmental conditions.
It can even perform coupled type of analysis to
combine various physical phenomenons such as
mechanical, thermal, electrical etc.
The modern design approaches consider struc-
tural reliability as one of the essential criteria to
be satisfied for structural integrity. The design
optimisation with reliability based approaches need
rately. Hence the strength analysis of structures
with a higher degree of accuracy is quite important
and crucial in the overall design process. Numeri-
cal analysis tools calibrated with reasonable model
uncertainty factor are absolutely suitable for this
purpose.
The structural reliability analysis needs to do
the capacity assessment of the structure numerous
times with variations in the design parameters
to evaluate the structural reliability. Although
the numerical methods can be used for reliability
analysis, the time and expense involved is quite
high. It further demands great effort and exper-
tise for acceptable results. Considering the above
facts, an analytical approach in terms of basic
structural design parameters to predict the struc-
tural capacity is more suitable for the reliability
analysis. Moreover, a component level reliability
assessment for a huge structure with number of
local structural parts at a preliminary design stage
cannot afford much time and expense. The neces-
sity of a good analytical strength model for ini-
tial design process is hence very important at this
instance.
There are various rule based design codes
available for the assessment of structural capac-

119

MARSTRUCT.indb 119 2/18/2011 5:42:48 PM

 ity of stiffened cylindrical structures under differ-
ent loading conditions. DNV-RP-C202 and API
BUL 2U are two of the major industry recom-
mended codes in practice. This paper proposes a
formulation for the strength assessment of ring
stiffened and ring-stringer stiffened cylinders. The
codes and the proposed formulation are compared
statistically with respect to mean and COV of a
large population of screened test data collected
over the years.

2 BUCKLING OF STIFFENED
CYLINDERS

The stiffened cylinders are one of the most

important structural components in buoyant
semi-submersible and Tension leg type offshore
platforms. The legs of these structures are gener-
ally and most likely made of stiffened cylinders
because of its inherent capability to resist high
axial loads and bending moments with lateral pres-
sure loads.
The stiffened cylinders are classified as ring stiff-
ened, stringer stiffened and ring-stringer stiffened
cylinders which is also known as orthogonally stiff-
ened cylinders. Ring stiffened cylinders are made
of fabricated cylinder with ring frames welded
externally or internally at wide spacing. Stringer
Figure 1. Modes of local and overall buckling in stiffened
stiffened cylinders will have equally spaced longi- cylinders.
tudinal stiffeners known as stringers welded inter-
nally or externally around the fabricated cylinder
throughout the length. Orthogonally stiffened cyl- Snap-through buckling occurs by a sudden reverse
inders will have both of these stiffeners. The stiff- of the curvature locally at certain combination of
eners can be of many types like flat bar, angle bar axial loads and the successive bending moments
and T bar etc. The structure is fabricated by butt and results in a total failure as there is no chance
welding process from cold or hot-formed plates so of moment redistribution. Other failure type is
that the structural continuity of the stiffeners and the classical type of bifurcation buckling. Various
the cylinder is established. The welding introduces local and overall buckling modes of stiffened cylin-
geometrical distortion and residual stresses in the ders are shown in Figure 1.
structure in addition to the pre-fabrication and
mechanical handling imperfections. The strength of
the structure is mainly dependent on the basic geo- 3 ANALYTICAL STRENGTH
metrical and material structural design parameters. MODELLING
At certain ideal conditions, the strength prediction
considering the basic structural parameters could In the simplest way, a good analytical strength
be reasonably accurate. But this approach never model should predict the strength of the structure
can represent any real life situation. It involves a accurately under the imposed loading and support
lot of known and unknown parameters which conditions. As mentioned earlier, because of the
potentially governs the structural behaviour. Some assumptions and approximations considered in
of them are the effect of geometrical imperfections, the analytical relations along with the unaccounted
residual stresses, type and direction of stiffeners factors, there always remain a certain percentage
(whether internal or external) etc. The contribu- of error in the structural strength prediction. So
tion of these parameters on the structural behav- a strength model can be rated based on the devia-
iour at different loading and support conditions tion from the experimental results. The best way
will be surprisingly different. to quantify this uncertainty is with the modelling
Basically the stiffened cylinder structure parameter i.e., ratio of experimental value and
can buckle and eventually fail in two ways. the theoretically predicted value. This modelling

120

MARSTRUCT.indb 120 2/18/2011 5:42:48 PM

 parameter is also known as the model uncertainty
⎧1; Zl ≥ 2.85
factor Xm. In structural reliability analysis, the ⎪
model uncertainty factor is incorporated in the C = ⎨1.425 + 0.17575; Zl < 2.85
failure surface equation as follows ⎪⎩ Zl

Z = XmR−S ⎧ ⎛ R ⎞
⎪0.75 − 0.142(Zl 1)
04
0 003Zl 1 − ;
⎝ 300t ⎟⎠
⎪
Xm is the model uncertainty factor associated ρn = ⎨ 1 ≤ Zl < 20
with strength R ⋅ S is the load and Z is the g(.) ⎪ R
function in the first order Second order reliabil- ⎪0.35 − 0.0003
0
t
; Zl ≥ 20
ity analysis. Z represents the safety margin in the ⎩

{
structural component.
1.3
Xm can be calculated for various theoretical B=
1 + 0.3λ n λ n ≥ 1
strength models comparing with the experimental
data. A good analytical strength model will have A quadratic interaction of λσy and σe can be
mean of Xm tends to unity and the coefficient of used to predict the inelastic collapse stress.
variation will be small. The strength models can be
σc = φσy
compared based on these values for different load-
ing cases as axial, radial and combined. where
1
φ=
4 STRENGTH MODELLING OF RING (1 + λ e4 )
STIFFENED SHELLS
σy
The design strength of the ring stiffened cylin- λe =
ders under axial, radial and combined loading are σe
computed based on the limit state approach. This
formulation focuses on the shell collapse between Now the Model uncertainty factor for the axial
ring stiffeners. The recommended formulation is load,
similar the one suggested earlier, Das et al. (2003) σ
with some modification on the knockdown factors Xm =
σc
so that the experimental results are getting closer
to the prediction. 4.2 Under hydrostatic pressure
For hydrostatic pressure, the proposed formula-
4.1 Under axial compression tion, Faulkner et al. (1983) is identical with the
The limit state approach estimate the elastic buck- approach in BS5500. The inelastic hydrostatic col-
ling strength of a ring stiffened cylinder subjected lapse pressure is estimated as,
to axial compression as,
⎧0.5 phm ; py ≥ phm
σe = BρnCσcr ⎪
phc = ⎨ ⎛ py ⎞
⎪ py ⎜⎝1 − 0.5 p ⎟⎠ ; py < phm
where ⎩ hm

Von Mises, (1929) propose the solution for

1 Et Et elastic hydrostatic buckling pressure of an unsup-
σ cr = = 0.605 ,
2 R
3(1 − ν ) R ported cylinder is as follows.

Et
Classical elastic axial buckling stress,
phm = R
Timoshenko & Gere, (1961) 2
1 ⎛ πR ⎞
n2 − 1 +
L2 2⎝ L ⎠
Zl = 1 − ν2 ⎧ ⎫
Rt
⎪ 2
⎪
⎪⎪ t2 ⎡ 2 ⎛ π R ⎞ ⎤ ⎪⎪
2
σy ×⎨
1
+ ⎢n − 1 + ⎥
λn = 2 2 2
⎪ ⎡ 2 ⎛ L ⎞ 2 ⎤ 12R 1 − υ ⎢⎣ ( ) ⎝ L ⎠ ⎥ ⎬⎪
ρnCσ cr ⎦
⎪ ⎢ n ⎝ ⎠ + 1⎥ ⎪
⎪⎩ ⎢⎣ πR ⎥⎦ ⎪⎭
C is a Length dependent coefficient,

121

MARSTRUCT.indb 121 2/18/2011 5:42:49 PM

 Windenburg and Trilling, (1934) proposed a p
Xm =
simplified expression as follows. phc
2
1 ⎛t⎞
E
( − )( + ⎝ R⎠ 4.3 Under radial pressure
)
phm phhm = Assuming a linear interaction between axial load
L and radial pressure load, the proposed model pre-
− 0.636
Rt dicts the elastic radial collapse pressure as follows.
2
⎛t⎞
0.919E
⎝ R⎠ phm
= prm =
L ⎛ R⎞
− 0.636
Rt ⎝t⎠
1 − 0.5 phm
σe
The above expression does not provide satis-
factory results for too small or too large values
of L / Rt . The above expression assumes pinned Now the inelastic collapse pressure under radial
boundary condition at the supported cylinder end. pressure load is computed similar to the hydro-
Even though more advanced analytical expressions static case as below.
are available, the above expression is widely used
because of the simplicity and the parameter Phm
⎧0.5 prm ; py ≥ prm
has low influence in the prediction of inelastic col- ⎪
lapse pressure. prc = ⎨ ⎛ py ⎞
Wilson (1966), proposed relatively simple linear ⎪ py ⎜⎝1 − 0.5 p ⎟⎠ ; py < prm
⎩ rm
equation for the circumferential yield stress of the
cylindrical shell.
The Model uncertainty factor for the radial
⎛t⎞ pressure load,
σy
⎝ R⎠
py =
1− γG Xm =
p
where, prc

⎛ υ⎞
J 1− 4.4 Under combined axial compression
⎝ 2⎠ and radial pressure
γ =
2 Nt
J + trht + Most of the design codes handle the combined load-
α ing based on an interaction approach. The general
2 interaction expression is in the following form.
⎛ R⎞
J Ar ⎜
⎝ Rcr ⎟⎠ m n
⎛σ⎞ ⎛ p ⎞
⎜⎝ σ ⎟⎠ + ⎜⎝ p ⎟⎠ = 1
1.285 c rc
α=
Rt
The above expression demonstrates the limiting
cosh ( L ) cos ( L ) criteria for the structure stability under combined
N= ,
sinh ( L ) sii ( L ) loading. Hence, there exist a number of load-
ing combinations which can cause the collapse
⎡ ⎛ αL ⎞ ⎛ αL ⎞ ⎛ αL ⎞ ⎛ αL ⎞ ⎤ of the structure. Eventually, the above expression
2 ⎢sinh cos + cosh siin
⎣ ⎝ 2 ⎠ ⎝ 2 ⎠ ⎝ 2 ⎠ ⎝ 2 ⎠ ⎥⎦ provides the model uncertainty factor for com-
G= bined loads at which a structural collapse occurs.
sinh ( ) + sin
si ( )
The best results with the available data is noticed
G and N are transcendental functions of with m = 1 and n = 2.
αL, i.e., which cannot be expressed in terms of Therefore,
algebraic operations or satisfy a polynomial
equation. 2
⎛σ⎞ ⎛ p ⎞
The Model uncertainty factor for the hydrostatic Xm = ⎜ ⎟ + ⎜
pressure load, ⎝ σ c ⎠ ⎝ prc ⎟⎠

122

MARSTRUCT.indb 122 2/18/2011 5:42:53 PM

 5 STRENGTH MODELLING OF RING ⎧ 4 2
2η λr λr
AND STRINGER STIFFENED SHELLS ⎪⎪1 − ; λ r ≥ 0.53
Kr = ⎨ ⎛ s⎞
− 2η ( ) (
2
r )
5.1 Under axial compression ⎪ ⎝ t⎠
⎪⎩1; λ r < 0.53
Das et al. (1992) proposed the design strength
of the ring and stringer stiffened cylinders under
vii. Effective width (minimum)
axial, radial and combined loading. It is basically
Shell effective width,
RCC formulation with a revised value for the bias
of knockdown factor.
⎧⎛ 1.05 0.28 ⎞
The steps to calculate the axial strength are as sem ⎪⎜ − 2 ⎟ K r ; λ r ≥ 0.53
follows. = ⎨⎝ λ r λr ⎠
s ⎪1; λ < 0.53
i. The elastic buckling stress for perfect shell under ⎩ r
curved panel formulation,
Shell reduced effective width,
⎧ ⎛t⎞ ⎛
2
3Z 2 ⎞
⎪0.904 E ⎜ ⎟ ⎜ 4 + 4s ⎟ ; Zs ≤ 11.4
⎪ ⎝ s⎠ ⎝ π ⎠ ′ ⎧⎛ 0.53 ⎞
=⎨ sem ⎪ K ; λ ≥ 0.53
= ⎨⎜⎝ λ r ⎟⎠ r r
cr
⎪0.605E ⎛ t ⎞ ; Z > 11.4
⎪⎩ ⎜⎝ ⎟⎠ s s ⎪1; λ < 0.53
R ⎩ r

s2 viii. MI of stringer and the reduced effective width

Zs = 1 − ν2 of the shell
Rt
′ 3
As dcs + .5t )2 semt
ii. The lower bound knockdown factor, I e′ Is + +
As 12
1+ ′
⎧ ⎛ R ⎞ semt
⎪1 − 0.919Zs
1 25
0 0024Zs ⎜1 − ;
⎝ 300t ⎟⎠ ix. Elastic stress for stringer stiffened cylinder is
⎪
⎪ Z ≤ 11.4 the sum of column elastic stress considering
ρn = ⎨ s
⎪0.27 +
1.5 27 ⎛ R ⎞ the effective shell width and product of criti-
+ + 0.008 Zs 1 − ;
⎪ Zs Zs2 ⎝ 300t ⎟⎠ cal stress for smeared un-stiffened shell and
⎪⎩ 11.4 s 70 shell knock down factor. The shell knockdown
factor is assumed to be 0.75

iii. Bias for knockdown factor, ⎛ ⎛ t ⎞⎞

⎛ π 2 EI e′ ⎞ ⎜ 0.605E ⎝ R ⎠ ⎟
B= {1.6; λ n > 1
1 + 0.6λ n λ n ≤ 1
σe = ⎜ 2 ⎟ + ρs ⎜
⎝ L ( As + semt ) ⎠ ⎜ ⎛1 + As ⎞ ⎟
⎜⎝ ⎝⎜ ⎟
st ⎠ ⎟⎠
⎟

σy x. Imperfect elastic buckling parameter

λn =
ρnσ cr
σe
ψ=
iv. Elastic buckling stress for imperfect shell, σy

σes = Bρnσcr xi. Ostenfeld-Bleich tangent modulus approach

to find inelastic stress.
v. Shell reduced slenderness parameter,
⎧⎛ ps ( ps ) ⎞
⎪ 1−
σ c = ⎨⎜⎝ ψ ⎟ σ y ψ ≥ ps
σy ⎠
λr = ⎪ψσ σ ; ψ < p
σ es ⎩ y s

σ pps
vi. The weld induced residual stress is incorporated ps =
using the width of the tension block η. For σy
continuous structural fillet welds, η = 4.5. For The structural proportional limit ps is 0.75
light fillets or for significant shake down situation, for stress relieved structures and 0.5 for all other
η = 3. For stress relieved structures, η = 0. cases.

123

MARSTRUCT.indb 123 2/18/2011 5:42:57 PM

 xii. Revised shell reduced slenderness parameter where
and revised effective shell width Effective pressure correction factor,

σc σ es ⎧ 0.85
λ re = = λr
σ es σy ⎪0.25 + 500 g g ≤ 500
Kp = ⎨
0.12
⎪0.98 + g 500 < g < 2500
⎧⎛ 1.05 0.28 ⎞ ⎩ 500
se ⎪⎜ − 2 ⎟ K r ; λ re
re ≥ 0.53
=⎨ λ λ re ⎠
s ⎪⎝ re g=
M x Mθ LtAs
, Mθ =
s
⎩1; λ re < 0.53 Is Rt
xiii. Average ultimate collapse stress
iv. Bay instability stress
⎛A set ⎞
σu σc ⎜ s (R t)
⎝A s st ⎟⎠ σθ =p Kθ L
t
The Model uncertainty factor for the axial where
load,

Xm =
σ
σu
Kθ L = { 1; M x ≥ 3.42
1 − ξε ; M x < 3.42

⎧1; x ≤ 1.26
5.2 Under radial pressure ⎪
ξ = ⎨1.58 − 0.46M x ;1.266 < x < 3.42
The proposed strength formulation for radial pres- ⎪⎩0; x ≥ 3.42
sure is according to API Bul 2U with some changes
in the effective pressure correction factor.
ε=
( − )
The steps to calculate the axial strength are as ⎛ Let ⎞
follows. ⎜⎝1 + ⎟
J ⎠
i. Local buckling pressure of un-stiffened shell
Nφ
⎧⎛ 1.27 ⎞ ⎛ t ⎞ 2 k2 =
Nθ
⎪⎜ 1.18 ⎟ E ⎜ ⎟ ; M x 1.5; H < 2.5
⎪⎝ H + 0.5 ⎠ ⎝ R ⎠
⎪ 0.92 ⎛ t ⎞ 2 N
⎛ R⎞ Nφ =
⎪ E ⎜ ⎟ ; 2.5 < H < 0.208 ⎜ ⎟ 2π R
⎪ A ⎝ R⎠ ⎝t⎠
peL =⎨ 3
⎪0.836C −11.061E ⎛ ⎞ ; 0.208 < C < 2.85
t Nθ = p(R + 0.5t)
⎪ p ⎜⎝ ⎟⎠ p
R
⎪ 3 ⎛ R ⎞
2
⎪0.275E ⎛ t ⎞ ;C > 2.85 J Ar ⎜
⎪⎩ ⎝ R⎠ p ⎝ Rcr ⎟⎠

where Le 1 56 Rtt twr ≤ L

L Ht
Mx = , Cp = The Model uncertainty factor for the radial
Rt R pressure load,
H = Mx − 1.17 + 1.068k1 (k1 = 0 for radial pres- p
sure and 0.5 for hydrostatic pressure) Xm =
pθcB
ii. Plastic collapse pressure of stiffener shell
combination
5.3 Under combined axial compression
16 and radial pressure
pcs = 2 As dcs σ y
sL
L The proposed interaction equation for the com-
iii. Bay instability pressure bined axial and radial loads is similar to the
API Bul 2U with a different definition for the
pcB = (peL + pcs)Kp factor Cc.

124

MARSTRUCT.indb 124 2/18/2011 5:43:02 PM

 2 2 R/t – (15–529),
⎛ Rx ⎞ ⎛ Rθ ⎞ s/t – (29–132),
⎜⎝ φ ⎟⎠ + Cc Rx Rθ + ⎜⎝ φ ⎟⎠ = 1
x θ Zs – (4–34),
Zl – (1.5–1550),
In which,
L/R – (0.07–5.82)

2 ( − x )2 ( − )2
Cc = −1 7 STRENGTH ANALYSIS AND RESULTS
φxφθ
where The data collected are carefully arranged and tab-
ulated with all the necessary inputs for the code
σ based design. The data is then pushed through the
Rx = analytical relations of DNV, API and the Recom-
σy mended Models for stiffened cylinders. The strength
predicted by each of the models is then compared
p (R + t) with the experimental results to evaluate the model
Rθ =
tσ y uncertainty factor for each set of data. The mean
and COV of the model uncertainty factor is then
σu computed for each codes for ring stiffened and ring-
φx = stringer stiffened cases based on three loading con-
σy
ditions, axial, radial and combined. The predicted
and experimental strength are then represented in
σ θcB
φθ = a graphical form which is normalised in terms of
σy yield strength. For combined loading cases, the
model uncertainty is plotted against L/R ratio as it
Since the above expression for combined loading is not straight forward to represent the strength.
characterises the limiting structural stability crite-
ria, it represents the model uncertainty factor of 7.1 Ring stiffened cylinders
the structure for the combined loading condition.
The model uncertainty factor for combined The ring stiffened cylinders are basically checked
loading is, against the local shell buckling which is the domi-
nant failure mode in this type of structures. Other
2 2 modes of failure and its interactions also have been
⎛R ⎞ ⎛R ⎞ taken into account.
X m = ⎜ x ⎟ + Cc Rx Rθ + ⎜ θ ⎟
⎝ φx ⎠ ⎝ φθ ⎠
7.1.1 Under axial compression
Table 1 shows the statistical results of the ring stiff-
6 EXPERIMENTAL DATA ened cylinders under axial compression for a popu-
lation of 40 for DNV, API and the recommended
The experimental test results are collected from a strength model. Figure 2 to Figure 4 shows the
wide literature survey over the last century. It is comparison of predicted and experimental data
observed that majority of the experimental works for the different approaches. The strength predic-
on stiffened cylinders are being undertaken during tion of the recommended model is more accurate
1960’s to 1980’s and there is not much experimen- compared to the other two approaches in terms of
tal works available recently as the researches are its statistical measures. Figure 4 shows the spread
comfortable with the numerical results with the of the results about its mean line which is having
increased capabilities and accuracy. In this paper, a better bias to the unity with low COV and it is
only the valid experimental data collected from evident visually also.
the documentation mentioned in the references
are considered. The data collected from various Table 1. Statistical results of ring stiffened cylinder
technical documents mentioned in the reference under axial loading.
are subjected to critical examination to avoid any
unreliable data. The data collected can be classified Recommended
based on various factors like geometrical proper- DNV API model
ties, material properties, Method of production, Mean 1.28 1.15 1.05
Test conditions etc. The proposed formulation COV 17.94% 11.84% 10.01%
considered here is applicable to structures with Population 40
parameters of the following range.

125

MARSTRUCT.indb 125 2/18/2011 5:43:08 PM

 7.1.2 Under radial pressure
1.00 Table 2 shows the statistical results of the ring stiff-
ened cylinders under Radial pressure for a popula-
0.80 tion of 65 for DNV, API and the recommended
strength model. Figure 5 to Figure 7 shows the
φ - Test

0.60
Table 2. Statistical results of ring stiffened cylinder
0.40
under radial loading.
Mean - 1.28
COV - 17.94% Recommended
0.20 DNV API model

Mean 0.98 1.35 1.01

0.00 COV 19.43% 19.09% 17.51%
0.00 0.20 0.40 0.60 0.80 1.00
φ - Predicted
Population 65

Figure 2. DNV prediction and test results of ring

stiffened cylinders under axial compression.
1.50

1.25
1.00

1.00
0.80
φ - Test

0.75
φ - Test

0.60

0.50
0.40
Mean - 1.15
COV - 11.84% 0.25
Mean - 0.98
0.20 COV - 19.43%
0.00
0.00 0.50 1.00 1.50
0.00
0.00 0.20 0.40 0.60 0.80 1.00 φ - Predicted
φ - Predicted
Figure 5. DNV prediction and test results of ring stiffened
cylinders under radial pressure.
Figure 3. API prediction and test results of ring stiff-
ened cylinders under axial compression.

1.50

1.00
1.25

0.80
1.00
φ - Test
φ - Test

0.60 0.75

0.40 0.50
Mean - 1.05
COV - 10.01%
0.20 0.25
Mean - 1.35
COV - 19.09%
0.00 0.00
0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.50 1.00 1.50
φ - Predicted φ - Predicted

Figure 4. Prediction of recommended model and test Figure 6. API prediction and test results of ring stiffened
results of ring stiffened cylinders under axial compression. cylinders under radial pressure.

126

MARSTRUCT.indb 126 2/18/2011 5:43:10 PM

 1.50 Table 3. Statistical results of ring stiffened cylinder
under combined loading.
1.25
Recommended
DNV API model
1.00
Mean 1.45 1.10 1.16
φ - Test

COV 20.79% 21.86% 17.02%

0.75
Population 25

0.50

2.50
0.25
Mean - 1.01
COV - 17.51%
2.00
0.00
0.00 0.50 1.00 1.50
φ - Predicted 1.50

Xm
Figure 7. Prediction of recommended model and test
results of ring stiffened cylinders under radial pressure. 1.00

0.50
comparison of predicted and experimental data
Mean - 1.45
for the different approaches. The average and COV - 20.79%
spread of the population is much better for the 0.00
recommended model. The recommended model 0.00 1.00 2.00 3.00 4.00 5.00
shows better central tendency compared to the L/R
other two approaches.
Figure 8. DNV prediction and test results of ring
7.1.3 Under combined axial compression stiffened cylinders under combined loading.
and radial pressure
Table 3 shows the statistical results of the ring
stiffened cylinders under combined axial com- 1.80

pression and Radial pressure for a population of 1.60

25 for DNV, API and the recommended strength
model. Figure 8 to Figure 10 shows the compari- 1.40

son of predicted and experimental data for the 1.20

different approaches. Here the mean of the mod-
elling parameter for the recommended model 1.00
Xm

is slightly high compared to API model, but the 0.80

COV is pretty low compared to that approach.
It confirms that the recommended approach is 0.60
more stable in the prediction of strength than the 0.40
other models.
0.20 Mean - 1.10
COV - 21.86%
7.2 Ring and stringer stiffened cylinders 0.00
0.00 1.00 2.00 3.00 4.00 5.00
The stringer stiffened cylinders and orthogonally L/R
stiffened cylinders are considered in one category
as the analysis is considering the effect of ring stiff- Figure 9. API prediction and test results of ring
eners with the ring stiffener spacing and its geo- stiffened cylinders under combined loading.
metrical and material parameters.

7.2.1 Under axial compression
Table 4 shows the statistical results of the ring and
stringer stiffened cylinders under Axial compres-
sion for a population of 32 for DNV, API and
the recommended strength model. Figure 11 to
Figure 13 shows the comparison of predicted and
experimental data for the different approaches.
The recommended model predicts the strength
almost similar to that of the API model and which
is better when compared to DNV model.

127

MARSTRUCT.indb 127 2/18/2011 5:43:12 PM

 1.80 1.40

1.60
1.20
1.40
1.00
1.20

1.00 0.80

φ - Test
Xm

0.80
0.60
0.60

0.40 0.40

0.20 Mean - 1.16

COV - 17.02% 0.20 Mean - 1.06
0.00 COV- 14.92%
0.00 1.00 2.00 3.00 4.00 5.00 0.00
L/R 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40
φ - Predicted
Figure 10. Prediction of recommended model and
test results of ring stiffened cylinders under combined
Figure 12. API prediction and test results of ring and
loading.
stringer stiffened cylinders under axial compression.

Table 4. Statistical results of ring and stringer

stiffened cylinder under axial loading. 1.40

Recommended 1.20
DNV API model
1.00
Mean 1.00 1.06 1.00
COV 23.18% 14.92% 14.99%
0.80
φ - Test

Population 32
0.60

1.40
0.40

1.20
0.20 Mean - 1.00
COV - 14.99%
1.00
0.00
0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40
0.80
φ - Test

φ - Predicted
0.60
Figure 13. Prediction of recommended model and test
0.40
results of ring and stringer stiffened cylinders under axial
compression.
0.20 Mean - 1.00
COV - 23.18%
Table 5. statistical results of ring and stringer
0.00 stiffened cylinder under radial pressure.
0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40
φ - Predicted Recommended
DNV API model
Figure 11. DNV prediction and test results of ring and
stringer stiffened cylinders under axial compression. Mean 1.33 1.12 1.06
COV 47.38% 21.54% 18.38%
Population 9
7.2.2 Under radial pressure
Table 5 shows the statistical results of the ring
and stringer stiffened cylinders under Radial
pressure for a population of 9 for DNV, API and experimental data for the different approaches. The
the recommended strength model. Figure 14 to recommended model has low bias and low COV
Figure 16 shows the comparison of predicted and compared to the other two models.

128

MARSTRUCT.indb 128 2/18/2011 5:43:13 PM

 7.2.3 Under combined axial compression and
1.00 radial pressure
Table 6 shows the statistical results of the ring and
stringer stiffened cylinders under combined Axial
0.75
compression and Radial pressure for a popula-
tion of 25 for DNV, API and the recommended
φ - Test

strength model. Figure 17 to Figure 19 shows the

0.50

Mean - 1.33
Table 6. Statistical results of ring and stringer
0.25 COV - 47.38%
stiffened cylinder under combined loading.

Recommended
0.00 DNV API model
0.00 0.25 0.50 0.75 1.00
φ - Predicted Mean 1.84 1.33 1.26
COV 43.82% 22.19% 20.12%
Figure 14. DNV prediction and test results of ring and Population 25
stringer stiffened cylinders under radial pressure.

4.00

1.00 3.50 Mean - 1.84

COV - 43.82%
3.00
0.75
2.50
φ - Test

Xm

2.00
0.50

1.50
Mean - 1.12
0.25 COV - 21.54%
1.00

0.50
0.00
0.00
0.00 0.25 0.50 0.75 1.00
0.00 0.20 0.40 0.60 0.80 1.00 1.20
φ - Predicted
L/R

Figure 15. API prediction and test results of ring and

stringer stiffened cylinders under radial pressure. Figure 17. DNV prediction and test results of ring and
stringer stiffened cylinders under combined loading.

1.80
1.00
1.60

1.40
0.75
1.20
φ - Test

1.00
Xm

0.50
0.80
Mean - 1.06
COV - 18.38% 0.60 Mean - 1.33
0.25
COV - 22.19%
0.40

0.20
0.00
0.00 0.25 0.50 0.75 1.00 0.00
φ - Predicted 0.00 0.20 0.40 0.60 0.80 1.00 1.20
L/R
Figure 16. Prediction of recommended model and test
results of ring and stringer stiffened cylinders under Figure 18. API prediction and test results of ring and
radial pressure. stringer stiffened cylinders under combined loading.

129

MARSTRUCT.indb 129 2/18/2011 5:43:16 PM

 1.80 API Bulletin 2U. (May 1997), Bulletin on Stability
Design of Cylindrical Shells, 1st ed., (ANSI/API/Bull
1.60
2U-1992).
1.40 British Standard Institution, (1976), Specification for
Unfired Fusion Welded Pressure Vessels, BS 5500,
1.20 B.S.I., Section 3.
Carl, T.F., Ross, J.R. & Sadler. (2000), Inelastic shell
1.00
instability of thin-walled circular cylinders under
Xm

0.80 external hydrostatic pressure, Ocean Engineering 27,

0.60 Mean - 1.26
COV - 20.12%
0.40
0.20
0.00
0.00 0.20 0.40 0.60 0.80 1.00 1.20
L/R
Figure 19. Prediction of recommended model and test
results of ring and stringer stiffened cylinders under com-
bined loading.
comparison of predicted and experimental data for
the different approaches. The recommended model
is showing low bias and COV compared to DNV
and API models.
8 CONCLUSIONS
The analyses with the experimental results illustrate
the fact that the recommended model predicts the
structural capacity more accurately in most of
the individual cases compared to API and DNV
codes.
The statistical parameters of the analysis show
that the recommended model is more stable in
predicting the strength of the stiffened cylinders
compared to the DNV and API codes.
The experimental data available for the radial
pressure load cases for ring-stringer stiffened cyl-
inders are very low and it is required to do further
investigation to acquire more data.
The design equations and the model uncertainty
factors derived in this paper is suitable for reliabil-
ity analysis and evaluating the partial safety factors
for similar structures.
REFERENCES
ABS (February 1984), Model Code for Structural Design
of Tension Leg Platforms, Conoco/ABS TLP Rule
Case Committee (RCC).
ABS (March 2005), Commentary on the guide for buck-
ling and ultimate strength assessment for offshore
structures.
765–774.
Das, P.K., Faulkner, D. & Zimmer, R.A. (June 1992),
‘Selection of Robust Strength Models for Efficient
Design of Ring and Stringer Stiffened Cylinders
Under Combined Loads’, Proc. of OMAE-92,
Calgary, p. 12
Das, P.K., Frieze, P.A. & Faulkner, D. (1984), Structural
Reliability Modelling of Stiffened Components of
Floating Structures, Structural Safety, 2, 3–16.
Das, P.K., Thavalingam, A. & Bai, Y. (2003), Buckling
and ultimate strength criteria of stiffened shells under
combined loading for reliability analysis, Thin-Walled
Structures 41, 69–88.
Das, P.K., Thavalingam, A., Hauch, S. & Bai, Y. (June
2001), ‘A New Look at the Model Uncertainty of
Stiffened Cylinders for Reliability Analysis’, Proceed-
ings of the 20th Intl Conference on Offshore Mechan-
ics and Arctic Engineering (OMAE 2001), Rio de
Janeiro, Brazil.
Das, P.K., Zanic, V. & Faulkner, D. (May 1993),
‘Reliability Based Design Procedure of Stiffened Cyl-
inders Using Multiple Criteria Optimisation Tech-
niques’, Offshore Technology Conference (OTC) ‘93,
p. 17.
Derek Graham, (2007), Predicting the collapse of exter-
nally pressurised ring-stiffened cylinders using finite
element analysis, Marine Structures 20, 202–217.
Det Norske Veritas. (October 2002), Recommended
Practice RP-C202. Buckling Strength of Shells.
Faulkner, D., Chen, Y.N. & De Oliveira, J.G. (June 1983),
‘Limit state design criteria for stiffened cylinders of
offshore structures’. American Society of Mechanical
Engineers of the National Congress of Pressure Ves-
sels and Piping Technology, Portland, Oregon.
Faulkner, D., Guedes Soares, C. & Warwick, D.M.
(1988), Modelling requirements for structural design
and assessment. Integrity of Offshore Structures-3,
IOS-87, Elsevier Applied Science.
Frieze, P.A., Das, P.K. & Faulkner, D. (1983), Partial
safety factors for stringer stiffened cylinders under
extreme compressive loads, PRADS 83, The 2nd
International Symposium on Practical Design in Ship
Building, Tokyo and Seoul, 475–482.
Kendrick, S.B. (March 1955), Analysis of results of
static pressure tests of Chatham submarine mod-
els. Naval Construction Research Est. (now ARE),
Dumfermline, Ref No. R218.
Krasovskya,V.L. & Kostyrko, V.V. (2007), Experimen-
tal studying of buckling of stringer cylindrical shells
under axial compression, Thin-Walled Structures 45,
877–882.
Miller, C.D. (March 1977), Buckling of axially
compressed cylinders. J Struct Div Trans ASCE,
Vol. 103, No, ST3, 1977:695–721.

130

MARSTRUCT.indb 130 2/18/2011 5:43:18 PM

 Miller, C.D. (1982), Summary of buckling tests
on fabricated steel cylindrical shells in the U.S.A. In:
Harding, Dowling and Agelidis, Editors, Buckling of
Shells in Offshore Structures, pp. 429–472.
Morandi, A.C., Faulkner, D. & Das, P.K. (1996), ‘Frame
Tripping in ring stiffened externally Pressurised
Cylinders’, Journal of Marine Structures, Vol. 9,
No. 6, pp. 585–608, p. 24.
Odland, J. (Feb. 1981), An experimental investigation
of the buckling strength of ring-stiffened cylindrical
shells under axial compression. Norwegian Maritime
Research, No. 4, pp. 22–39.
Paul, A. Frieze, (1994), The Experimental Response of
Flat-Bar Stiffeners in Cylinders under External Pres-
sure, Marine Structures, 7, 213–230.
Ralph, E. Ekstrom. (1963), Buckling of Cylindrical Shells
Under Combined Torsion and Hydrostatic Pressure,
Experimental Mechanics, Vol. 3, 192–197.
Ross, C.T.F. & Johns, T. (1998), Plastic Axisymmetric
Collapse of Thin-Walled Circular Cylinders and
Cones Under Uniform External Pressure, Thin-Walled
Structures Vol. 30, Nos. 1–4, pp. 35–54.
Seleim, S.S. & Roorda, J. (1986), ‘Buckling Behaviour
of Ring-Stiffened Cylinders; Experimental Study’,
Thin-Walled Structures, 4, 203–222.
Sridharan, P. & Walker, A.C. (February, 1980),
Experimental investigation of the buckling behaviour
of stiffened cylindrical shells. UK Department of
Energy Report OT-R7835.
131
MARSTRUCT.indb 131
Thimoshenko, P. & Gere, J. (1961), “Theory of Elastic
Stability”, McGraw-Hill Book Company Inc.,
New York-Toronto-London.
Voce, S.J. (1969), Buckling under external hydrostatic
pressure of orthotropic cylindrical shells with evenly
spaced equal strength circular ring frames, Ocean
Engineering. Vol. 1, pp. 521–534. Pergamon Press.
Von Mises, R. (1929), Stodola Festschrift. Zurich,
p. 418.
Walker, A.C. & McCall, S. (1987), Strength of Damaged
Ring and Orthogonally Stiffened Shells-Part I: Plain
Ring Stiffened Shells, Thin-Walled Structures, 5,
425–453.
Walker, A.C. & McCall, S. (1988), Strength of Damaged
Ring and Orthogonally Stiffened Shells Part II: T-Ring
and Orthogonally Stiffened Shells,Thin-Walled Struc-
tures, 6, 19–50.
Wilson, L.B. (1966), The elastic deformation of a circu-
lar cylindrical shell supported by equally spaced ring
frames under uniform external pressure. Trans RINA,
Vol. 108.
Windenburg, D.F. & Trilling, C. (1934), Collapse of insta-
bility of thin cylindrical shells under external pressure.
Trans ASME Vol. 56, p. 819.
2/18/2011 5:43:19 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Finite elements modeling of delaminations in composite

laminates

M. Gaiotti & C.M. Rizzo

DINAEL, Faculty of Engineering, University of Genova, Italy

K. Branner & P. Berring

Risø National Laboratory for Sustainable Energy, Technical University of Denmark, Denmark

ABSTRACT: The application of composite materials in many structures poses to engineers the problem
to create reliable and relatively simple methods, able to estimate the strength of multilayer composite struc-
tures. Multilayer composites, like other laminated materials, suffer from layer separation, i.e., delamina-
tions, which may affect the stiffness and stability of structural components. Especially deep delaminations
in the mid surface of laminates are expected to reduce the effective flexural stiffness and lead to collapse,
often due to buckling behaviour. This paper deals with the numerical modelling of the buckling strength
of composite laminates containing delaminations. Namely, non-linear buckling and post-buckling analy-
ses are carried out to predict the critical buckling load of elementary composite laminates affected by
rectangular delaminations of different sizes and locations, which are modelled by finite elements using
different techniques. Results obtained with different finite element models are compared and discussed.

1 INTRODUCTION buckling, where the sub laminates on both sides

1.1 Delaminations in multilayer laminates
In a multilayer composite, the delamination can be
defined as an area with lack of bonding between
two adjacent layers. This can originates either in
the manufacturing process or from damage during
production, transport or service. Typical damage
that gives origin to delamination is impact, but also
stresses concentrations around structural disconti-
nuities and three dimensional stress conditions on
free edges may cause the initiation and growth of
the delaminations. Delaminations may also be gen-
erated in the composite laminating process due to
lack of impregnability of the fibers, or to thermal
and chemical shrinkage of composite components
during the matrix polymerization (Bolotin 1996).
Delamination is usually the most critical type
of damage that composite and sandwich structure
experience under compressive loads (Abrate 1991,
Pavier & Clarke 1995).
When a delaminated panel is subjected to a
compressive in plane load then, depending on
delamination size and position, different behavior
can be observed; as expected, after a critical load
limit is reached, the panel starts buckling but the
mode in which the panel buckles has an important
influence onto the panel failure load. Typically
two different conditions are expected: global mode
of the delamination moves to the same side, and
local mode buckling where the sub laminates move
towards opposite direction. As observed by Peck &
Springer (1991) and by Pavier & Clarke (1995),
when local buckling occurs, it introduces bending
in the plies on the other side of the delamination so
that they are subjected to both bending and com-
pressive stress resulting in a reduced failure load.
Much numerical and experimental work has
already been carried out to define the ultimate
strength of delaminated panel; in particular
Sørensen et al. (2009) and Sørensen et al. (2010)
from Risø DTU have conducted a sensitivity anal-
ysis using a 3D solid element model to evaluate
the critical non linear buckling load of flat panels
made of unidirectional layers.
1.2 Aim of the present paper
This paper focuses on two different modeling
approaches of delaminated multilayer composite
laminates: the former one, applied in Risø DTU,
adopts 3D 20-node orthotropic solid elements
with 3 Degrees Of Freedom (DOF) per node and
uses two or three elements through the thickness,
depending on the through thickness position of the
delamination, and the latter one, developed by the
University of Genoa, adopts shell elements with

133

MARSTRUCT.indb 133 2/18/2011 5:43:19 PM

 6 DOF per node and simulates the delamination
by off-setting two shells in the delaminated area
and by connecting them to the non delaminated
region (modeled by one shell only) through rigid
links, as better described in the following.
Results are in both cases presented as a map of
the deduction factor on the critical load that the
delamination introduces in the panel with respect b a A
to a non delaminated plate, depending on delami-
nation size and depth.

B
2 DIFFERENT MODELLING
APPROACHES

2.1 Material properties and specimen geometry

Mechanical properties representative of unidi- T
rectional fiber glass laminates produced in the
laboratory is used in this work, see Hansen et al.
t
(2009): Figure 1. Definition of panel and delamination
geometry.
E11 = 46.5 GPa G12 = 4.1 GPa ν12 = 0.25
E22 = 13.4 GPa G23 = 2.6 GPa ν23 = 0.25
The plate taken into account for the numeri- Also the delaminated area has the same aspect
cal analysis is rectangular with an aspect ratio ratio of the panel a/b = 1.36; where a, b are the edges
A/B = 1.36 where A is the length of the edge along of the rectangular delaminated area (Fig. 1).
the load direction, having the same orientation
of the fibers, i.e., direction 1. See Fig. 1 for size 2.2 3D solid elements model
definitions.
The selected aspect ratio is obtained considering The 3D solid finite-element models were modeled
the critical buckling load evaluated according to with 20-node orthotropic elements. Two or three
the analytical formulation proposed by Bisagni & elements were used through the thickness depend-
Vescovini (2008): ing on the through thickness position of the dela-
mination in the Abaqus finite element software
(SIMULIA, 2009).
N0 ( m,11) = F ⎡⎣ A, B, Dij ⎤⎦
A small out-of-plane displacement corresponding
⎡ ⎛ m⎞
2
1 to the first buckling mode shape was appropriately
= π 2 ⎢ D11 ⎜ ⎟ + 2 ( D12 2 D66 ) applied as an initial imperfection of the delaminated
⎢⎣ ⎝ A⎠ B2
sub-laminate. The amplitude of the initial imperfec-
⎛ 1 ⎞ ⎛ A⎞ ⎤
2
tions was approximately 0.5% of the panel thick-
+ D22 ⎜ 4 ⎟ ⎜ ⎟ ⎥ (1)
⎝ B ⎠ ⎝ m⎠ ⎥ ness for the flat panels and approximately 0.5% of
⎦ the panel thickness for the curved panels.
The elements were all joined in the interfaces,
where Dij are elements of the bending stiffness except for the delaminated area where quadratic
matrix obtained by the Classic Laminate Theory contact conditions were applied to prevent penetra-
(CLT). First buckling mode (n = 1) is considered tion. The average number of degrees of freedom
since it always gives the minimum value. (DOFs) was approximately 150.000. Nonlinear
In facts, by calculating the first derivative with geometric analyses were conducted with minimum
respect to m and equaling the right side of Eq. 1 100 increments to ensure that a well described and
to 0 gives: smooth graph could be made for in-plane force vs.
out-of-plane displacement. An explicit solution algo-
N0 ( m,1)
dN ⎛ D m D22 A2 ⎞ rithm was adopted to avoid convergence problems.
= π 2 ⎜ 2 112 2 ⎟ =0
dm ⎝ A B 4 m3 ⎠
1 1 (2) 2.3 6 DOF shell elements model
A⎛ D ⎞ 4 A ⎛ D ⎞4
m = ⎜ 22 ⎟ ; if
i m = 1 => = ⎜ 11 ⎟ The shell element model presented in the paper is
B ⎝ D11 ⎠ B ⎝ D22 ⎠ defined by a single shell surface for the intact path

134

MARSTRUCT.indb 134 2/18/2011 5:43:19 PM


Figure 2. Shell model with offset translation of the
delaminated layers.
and by two surfaces for the delaminated area, each
surface representing one side of the delamination,
as shown in Fig. 2.
The middle plane of the model is defined
at the half thickness of the physical plate, and on
this plane shell elements are meshed to describe
the intact part of the model. The two sides of the
delamination are modeled onto two surfaces, offset
above and below the middle plane.
The offset h1 and h2 are defined as:
1⎛ t⎞ 1
h1 T− ⎟ h2 (T − t ) (3)
2⎝ 2⎠ 4 corresponding in plane force reactions.
where T is the intact laminate thickness; t sub lami-
nates thicknesses in the delaminated area as shown
in Fig. 1.
The nodes along the edge of the delamination
are coupled to the nodes on the main surface by a
rigid link constraint, where the master node is the
node lying on the edge of the intact middle plane
and the nodes on both sub laminates edges are its
slave.
A 9-nodes shell elements mesh has been gener-
ated on the surfaces, for a total number of about
4800 elements; the element type is the MITC9 as
suggested by Buncalem & Bathe (1993), and Bathe
et al. (2000), to prevent element locking problems
for thin laminates.
Element locking is, as widely discussed in related
literature, the phenomenon of an element being
much too stiff compared with reality: in essence,
the phenomenon arises because the interpolation
functions used for an element are not able to rep-
resent zero (or very small) shearing or membrane
strains. If the element cannot represent zero shear-
ing strains, but the physical situation corresponds
to zero (or very small) shearing strains, then the
element becomes very stiff as its thickness over
length ratio decreases. (see e.g., ADINA, 2008).
In the considered problem the MITC9 use is
justified when modeling delaminations very close
135
MARSTRUCT.indb 135
Figure 3. Particular of the rigid link coupling between
intact and delaminated path.
to the surface, where the thickness of one of the
sub-laminates is very small if compared to other
dimensions.
To prevent penetrations between the surfaces
defining the sub laminates in the delaminated area,
a contact algorithm has been taken into account.
Similarly to the solid models, small out-of-plane
displacement field derived from a preliminary lin-
ear buckling analysis was appropriately applied to
trigger buckling in the non-linear analysis, as sug-
gested in ADINA (2008).
The panel is considered simply supported on
all its edges; a distributed compressive line load is
applied on the short edges.
The analysis is performed applying a fixed in-
plane displacement, in 120 identical load steps to
ensure a resulting smooth curve, then plotting the
3 A ROBUST METHOD TO DEFINE
THE BUCKLING LOAD
The robust method proposed by Sørensen et al.
(2009, 2010) has been adopted to define the buck-
ling critical load in both modeling approaches: this
method consists in the analysis of the in-plane dis-
placements vs. in-plane load curve obtained from a
non linear static finite element model having the ini-
tial imperfections obtained by a preliminary linear
eigenvalue analysis performed on the same model.
The first part of the curve (in the criterion up to
the 30% of critical load) is linear: the criterion con-
sists in the prolongation of the linear curve along
the in-plane displacements axis; the new curve
is then shifted down by a quantity equal to 2.5%
of the linearized critical load; eventually, the point
where the down-shifted linear curve meet the non-
linear curve defines the non linear buckling load
(Fig. 4).
Due to the weak non linearity observed in the
early loading history, in particular for cases where
local buckling occurs, instead of considering the
2.5% of the linearized buckling load of each model,
the offset calculated for the non delaminated case
has been taken into account and used for all the
other models considered.
2/18/2011 5:43:21 PM
 Figure 6. Typical local buckling behavior for a 3D solid
element finite element model.

Figure 4. Robust method to define buckling load in

non linear analysis.

Figure 5. Buckling load offset dependency for close to

surface delaminated models.

Other definitions of the criterion would result in

underestimating the critical load for those models
with delaminations very close to the surface of the Figure 7. Delamination modes shown with the corre-
panel, see Fig. 5. sponding central out-of-plane displacement for the two
delaminated sub-laminates.

4 CONSIDERATIONS FOR 3D SOLID Buckling Mode Map, panel aspect ratio A/B=1.36, UD, (flat)
1
MODELS FE − Global
0.9
Non dimensional delamination size, b/B

FE − Local
FE − Global/Local I
A panel with a delamination subjected to uni-axial 0.8
FE − Global/Local II
compression can buckle in local or global modes as 0.7 Global limit
Sub−mode limit
described in the introduction. A typical local buck- 0.6
ling mode is shown in Fig. 6 for solid models. 0.5
However, results from the 3D solid models also 0.4
show that combinations of the global and local
0.3
modes can appear as so-called combined modes
0.2
or sub-modes. The sub-modes are found to appear
for special combination of delamination size and 0.1

through thickness position. In these studies five 0

0 0.1 0.2 0.3 0.4 0.5
different mode types have been observed. Non dimensional through thickness position, t/T
These modes are illustrated in Fig. 7, where the
relation between the out-of-plane displacement at Figure 8. Buckling mode map for a 100% unidirectional
the centre of the two delaminated sub-laminates fiber composite laminate from solid model FE analyses.
are plotted.
The different modes can be shown in a buck- • Local buckling, occurring for large delamina-
ling mode map in a similar manner as reported tions close to the surface,
by Short et al. (2001) where only local and global • Global buckling, occurring for small and deep
modes were considered. delaminations,
The buckling map shown in Fig. 8 is divided • Sub-mode, occurring for large and deep
into the following 3 areas: delaminations.

136

MARSTRUCT.indb 136 2/18/2011 5:43:23 PM

 5 COMMENTS ON SHELL MODELS
RESULTS
5.1 Buckling modes for deep delaminations
In the first series of analyses, the delamination is
placed in the mid thickness of the laminate, i.e.,
t/T = 0.5.
Under these condition, relatively small delami-
nations evolve into a mode 1 buckling shape; as
soon as the delamination size grows, a progres-
sive transition from mode 1 to mode 2 is observed
when increasing the in-plane load; thereafter a pure
buckling mode 2 occurs when the delamination is
relatively larger. Fig. 9 shows an example.
This fact is rather unexpected since the model
has been designed, in agreement with the analyti-
cal formulation proposed by Bisagni & Vescovini
(2008), having an aspect ratio supposed to induce a
mode 1 buckling shape, i.e., a single half wave onto
the plate length. However, the above mentioned
analytical formulation applies to perfect rectan-
gular plates and does not account for the effects
introduced by the delamination.
The different buckling mode does not affect the
load displacements curve, which appear smooth
for any considered delamination size.
The critical buckling load dependency on the
delamination size is easily observed from Fig. 10,
where it is also evident, as expected, that the dela-
mination does not have any effect on the panel
stiffness until the panel starts buckling, thus intro-
ducing non linearities in the problem at this point.
Figure 9. Different mode shapes (mode 1 left, mode 2
right) obtained increasing the delamination size in shell
models.
Figure 10. Load displacements curves for panels pre-
senting deep delaminations, for different delamination
sizes.
137
MARSTRUCT.indb 137
5.2 Unstable behavior moving the delamination
towards the surface
Considering the panel having t/T = 0.4, an unex-
pected behavior was observed: when the dela-
mination size is b/B = 0.5, a drop in the in-plane
load is observed when increasing the in-plane
displacement.
Such behavior is attributed to the instability of
the model, see Fig. 11. The physical reason seems
to lie in the buckling mode suddenly shifting from
local to global, with an abrupt out of plane dis-
placement of the thickest laminate, after remain-
ing plane in the first part of the loading history.
This behaviour could explain some instant col-
lapses observed during experimental tests being
conducted in Risø DTU National Laboratories.
5.3 Local buckling mode for delaminations
close to the laminate surface
When the delamination is very close to the sur-
face of the laminate, the thinnest sub-laminates
buckles in a very early stage of the loading history,
introducing a rather weak non linearity in the sys-
tem. The obtained curves therefore differ from the
intact plate curve in practice since the beginning of
the calculation, and do not show a sudden steep-
ness change when buckling begins.
This fact makes a buckling criterion definition
quite hard since the used offset-based robust crite-
rion does not fit very well with a weak non linear
behavior of the curve without sudden changes in
the first derivative, as shown in Fig. 5. For this rea-
son the same offset distance of the intact model
has been taken into account for all the geometrical
conditions considered.
In Fig. 12 the typical behavior of panels buck-
ling in local mode is presented; it is also worth not-
ing that local modes start from a delamination size
b/B > 0.4: the mode is still global with a rapid change
in steepness if delamination is at 40% of the thick-
ness; the difference can be appreciated in the plot.
It is interesting to plot on the same chart
the curves obtained from the deep delaminated
Figure 11. Mode transition with instability observed
for panels having t/T = 0.4; b/B = 0.5.
2/18/2011 5:43:28 PM
 Table 1. Compressive strength function of delamina-
tion size and through-thickness position.
Delamination depth t/T
0.00 0.10 0.20 0.30 0.40 0.50
Delam. 0.00 1.00 1.00 1.00 1.00 1.00 1.00
Size 0.10 1.00 0.86 0.92 0.98 0.99 0.99
b/B 0.20 1.00 0.79 0.82 0.93 0.91 0.90*
0.30 1.00 0.72 0.70 0.92 0.83 0.79*
0.40 1.00 0.62 0.58 0.77 0.77 0.68*
0.50 1.00 0.42 0.44 0.51 0.69 0.55*
* Buckling mode 2.
Figure 12. Load displacements curves for panels local ¬ Local/Global buckling limit.
modes.

Figure 14. Reduced compressive strength map function

Figure 13. Comparison of load-displacements curves of delamination size and position.
for deep and surface delaminated models.
interpolations (Fig. 14), similarly to what reported
by Sørensen et al. (2010).
model (t/T = 0.5) and the ones from the surface
When looking at the map, it is interesting to note
delaminated model (t/T = 0.9) as shown in Fig. 13.
that there is an increase of the residual strength
Deep delaminated models buckle at higher
when the delamination position through the thick-
loads, but in the post buckling phase they show a
ness assumes the value t/T = 0.4. At first this fact
lower derivative with respect to the in-plane dis-
could suggest a safer situation, but, accounting for
placement of the in-plane load (i.e. ∂ x ∂∂U U x ) in
the previous considerations, this is also the condi-
comparison to the surface delaminated models.
tion of instable buckling, where the in-plane load
This fact implies that the curves relevant to the
suddenly drops after buckling when increasing the
surface delaminated models overlap the ones cor-
in-plane displacement.
responding to the deep delaminated models at high
loads.
At this stage of the work it is deemed impossible
6 MODEL COMPARISON
to predict if in reality panels with delaminations
closer to the surface could show a better resistance
The residual strength map, obtained from both
to post-buckling collapse or if the collapse hap-
the models considered, is plotted in the same
pens suddenly once the critical load is reached.
‘mirrored’ chart in Fig. 15. The center of the chart
corresponds to a mid-thickness delamination: on
5.4 Compressive strength the left side is presented the 3D solid model while
on the right side the shell model.
Applying the previously described robust method
On the x axis it is possible the residual strength
to determine the critical load, the buckling strength
for a delamination getting closer to the surface of
normalized with respect to the critical load for the
the panel is defined proceeding left and right from
intact panel is presented as function of the delami-
centre, while on the y axis the normalized delami-
nation size and its through-thickness position.
nation size is plotted.
Table 1 shows the limit between global modes,
From the results presented in Fig. 15 the follow-
typical for deep delaminations, and local modes,
ing considerations arise:
involving deeper delaminations when extending
the delamination size. • The shell model is more sensitive to small delam-
From Table 1 it is possible to plot the linearized inations: the residual strength, function of b/B,
residual strength map obtained through numerical starts reducing earlier than for the 3D model,

138

MARSTRUCT.indb 138 2/18/2011 5:43:29 PM


Figure 15. Reduced compressive strength map for both
the models presented in the paper.
• Both models show a minimum for t/T ≈ 0.1 and a
maximum for t/T ≈ 0.4, whose amplitude is any-
way much greater if shell model is considered,
• Deep delaminations have greater influence on
shell models,
• Wide and shallow delaminations effects
(t/T < 0.25; b/B > 0.5) are predicted almost iden-
tically in both models.
For the time being, it is not possible to establish
which numerical model better describes the physi-
cal problem, since very little comparison have been
made with experimental results.
As far as the computation costs is concerned,
a non linear analysis performed in 100 time steps
as above described takes up to 45 minutes when
running a shell element finite element model on a
4 processors and 16 GB RAM memory worksta-
tion, being less than one half of the 2 hours needed
to run the same calculation of 3D 20 nodes solid
element model on a workstation having the same
capabilities.
The computational effort for the 3D solid model
is basically due to the need of using at least 3 layers
of elements through the panel thickness, in order
to catch bending effects, whose influence is basic
in the problem.
7 CONCLUSIONS
In the present paper various sizes and depths of the
delamination were assessed by two sets of finite ele-
ments calculations: at first a more intuitive mod-
elling strategy was employed using 20 nodes solid
elements to model each layer of the laminate and
thus requiring rather large computation efforts; then
a different modelling approach adopting multilayer
shell elements to model the laminate was applied,
requiring more limited computational efforts.
In both cases plots summarizing the influence
of the geometrical parameters of the panel and of
the delamination onto the collapse buckling behav-
ior were produced.
139
MARSTRUCT.indb 139
Both the applied models were able to produce
a residual strength chart depending on the geo-
metrical parameters of the delaminations taken
into account.
The residual strength prediction is the same
considering shallow and wide delaminations, while
the shell model looks more sensitive to small or
deep delaminations. The trend of the curves is
similar for both approaches and show maximum
(t/T ≈ 0.4) and minimum (t/T ≈ 0.1) values of resid-
ual strength function of delamination depth.
These results seem to indicate that shallow dela-
minations result in more dangerous conditions, but
this is true only if no post-buckling life of the panel
is considered, as clearly shown in section 5.3.
The results provided by the shell model analysis
appear rather remarkable, especially considering
the lower computational efforts, and suggest to
proceed to a step forward in the work, involving
deeper experimental comparison.
REFERENCES
Abrate, S. 1991. Impact on laminated composite materi-
als. Applied mechanical review 44:155–190.
ADINA 2008. Theory and modeling guide, v. 8.5.3.,
Watertown (MA), ADINA R&D Inc.
Bolotin, V.V. 1996. Delaminations in composites struc-
tures: its origin, buckling, growth and stability. Com-
posites: Part B 27B:129–145.
Bisagni, C. & Vescovini, R. 2008. Analytical formulation
for local buckling and post-buckling analysis of stiffened
laminated panels. Thin-Walled Structures 47:318–334.
Hansen, A.L., Lund, E., Pinho, S.T. & Branner, K. 2009.
A Hierarchical FE approach for simulation of geo-
metrical and material induced instability of composite
structures. Composites 2009, 2nd ECCOMAS thematic
conference on the mechanical response of composites,
Imperial college London.
Pavier, M.J. & Clarke, M.P. 1995. Experimental Tech-
niques for the Investigation of the Effects of Impact
Damage on Carbon-Fibre Composites. Composites
Science & Technology 55:157–169.
SIMULIA (2009). ABAQUS/Standard Users Manual,
Rising Sun Mills, 166 Valley Street, Providence,
RI, USA.
Sørensen, B.F., Branner, K., Lund, E., Wedel-Heinen, J. &
Garm, J.H. 2009. Improved Design of large wind tur-
bine blade of fibre composites (phase 3). Summary
Report, Risø-R-1699(EN), Risø National Laboratory
for Sustainable Energy, Denmark.
Sørensen, B.F., Toftegaard, H., Goutanos, S., Branner, K.,
Berring, P., Lund, E., Wedel-Heinen, J. & Garm, J.H.
2010. Improved Design of large wind Turbine Blade
of fibre composites (phase 4). Summary Report, Risø-
R-1734(EN), Risø National Laboratory for Sustainable
Energy, Denmark.
Short, G.J., Guild, F.J. & Pavier, M.J. 2001. The effect of
delamination geometry on the compressive failure of
composite laminates. Composites Science and Technol-
ogy 61:2075–2086.
2/18/2011 5:43:33 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Shakedown of welding-induced residual stress and effect

on stiffened plate strength and behaviour

L.G. Gannon, N.G. Pegg & M.J. Smith

Defence Research and Development Canada — Atlantic

Y. Liu
Dalhousie University, Nova Scotia, Canada

ABSTRACT: Numerical simulation is used to study the influence of welding-induced residual stress
in welded, tee-stiffened plates focusing on the effect of shakedown. Residual stresses are simulated using
3D thermo-elasto-plastic finite element analysis. The influence of strain hardening and number of load
cycles on residual stress shakedown is then investigated. Load versus end-shortening curves are used to
characterize the strength and behaviour of stiffened plates under axial compression both before and after
shakedown. Results show that the reduction in residual stress due to shakedown occurs entirely during the
first load cycle provided that the magnitude of that load is not subsequently exceeded. Both the tensile
and compressive welding residual stresses are reduced by as much as 40% when the applied load causes an
average stress equal to 50% of the yield stress. This level of shakedown increased the ultimate strength of
tee-stiffened plates by as much as 6%.

1 INTRODUCTION asymmetric and the structure accumulates plastic

The primary load for which ship hull girders are
designed is wave-induced longitudinal bending.
The bending stresses are resisted by longitudi-
nally stiffened plates that that also contain residual
stresses caused by welding during fabrication of the
structure. Several studies (Faulkner 1975; Guedes
Soares 1988; Gordo & Guedes Soares 1993) have
shown that residual stresses can have a detrimental
effect on the ultimate strength of stiffened plates
and consequently on the strength of a hull.
Welding-induced residual stresses may be relieved
to some degree by stretching of stiffened plates
under cyclic loads during service. This process is
commonly referred to as shakedown. Abdel-Karim
(2005) identifies three types of shakedown that
occur when the magnitude of the cyclic load lies
between the first yield and plastic collapse loads.
Elastic shakedown occurs when a finite amount of
plastic deformation occurs during the first few load
cycles, after which any further deformation is purely
elastic. In plastic shakedown, the structure experi-
ences equal and alternating plastic strains during
each load cycle and continues to experience shake-
down in the form of non-cumulative cyclic plastic
straining which eventually leads to failure by low-
cycle fatigue. If the alternating plastic strains are
strain with each load cycle, incremental plastic col-
lapse of the structure will occur. This incremental
accumulation of plastic strain during cyclic load-
ing is called ratcheting.
Shakedown in ship structures has been investi-
gated both experimentally and numerically; how-
ever most research has focused on the effects of
shakedown on fatigue behavior in welded struc-
tures. Latrou et al. (2005) studied the behavior of
welded joints accounting for residual stress and
shakedown using numerical models. A rectangu-
lar plate was modelled assuming plane stress and
residual stresses were simulated by application of a
non-uniform displacement on one side of the plate.
They found that after a low number of load cycles,
the behavior of the joint became elastic. Similar
results were reported by Liang et al. (2007). They
also studied the effect of residual stress relaxation
on the fatigue behavior of welded joints and found
that under constant amplitude cyclic loads, residual
stress relaxation was limited to the first load cycle.
In an experimental study of shakedown in
butt-welded aluminum plates subjected to 3-point
bending, Paik et al. (2005) found that measured
longitudinal residual stresses were reduced by 36%
and 21% in tension and compression respectively.
The load in that experiment was applied for three

141

MARSTRUCT.indb 141 2/18/2011 5:43:34 PM

 cycles, producing an extreme fiber stresses equal to
88% of the yield stress.
In the analysis of stiffened plates, simplified
residual stress distributions are sometimes assumed
based on measurements available in literature.
In design, fabrication related imperfections are
typically accounted for using an empirical method
such as the Johnson-Ostenfeld correction for plas-
ticity in beam columns. In some cases however,
residual stresses are neglected under the assump-
tion that they are relieved by shakedown.
A welding-induced residual stress distribution
commonly used in stiffened plate analysis assumes
that residual stress is constant along the length of
the stiffened plate. Based on experimental data
and assuming that the residual stresses acting over
the cross-section of the plate in a welded stiffened
plate are in equilibrium, Faulkner (1975) proposed
the idealized longitudinal residual stress distribu-
tion shown in Figure 1, where b is the plate width,
σy is the yield stress, t is the plate thickness and
η is a parameter describing the width of the ten-
sile stress block at the weld in the middle of the
plate. Faulkner (1975) suggested initial values for
η in the range of 4.5–6 and values ranging from
3–4.5 to allow for shakedown. From equilibrium
requirements, the compressive residual stress σc is
given by:
2 η tσ y
σc =
b − 2η
The primary objective of this study is to investi-
gate the influence of residual stress shakedown on
the strength and behaviour of longitudinally stiff-
ened plates typical of ship hull girders. The finite
element modelling technique used for the analyses
allows the complex three-dimensional distribution
of welding-induced residual stress and distortion
to be accounted for in assessing shakedown and
ultimate strength of stiffened plates.
Welding-induced residual stress and distor-
tion in the stiffened plates were determined using
sequential 3D, nonlinear thermal and mechanical
Figure 1. Idealized longitudinal residual stress distribu-
tion in a welded plate.
MARSTRUCT.indb 142
finite element analyses. Assuming that the stiffened
plates are not subjected to any type of stress relief
such as annealing, a cyclic axial load was then
applied, causing shakedown of the residual stress.
An ultimate strength analysis was then conducted
and load-shortening curves that characterize the
behaviour of the stiffened plates under axial load
were obtained. The finite element method was also
used to determine the effect that strain hardening
of the material and variable amplitudes of load-
ing have on the shakedown of welding-induced
residual stresses.
2 FINITE ELEMENT MODELLING
2.1 Welding simulation
A finite element simulation was used to predict
welding-induced residual stress and distortion in
tee-stiffened plates. The simulation consisted of
sequentially coupled nonlinear thermal and struc-
tural analyses. The model made use of two element
types; 8-node, linearly interpolated hexahedrons to
mesh the solid volume, and 2-node nonlinear springs
to model contact between the stiffener base and
the plate. The material used for the stiffened plates
was AH-36 shipbuilding steel with a nominal yield
stress of 360 MPa and elastic modulus of 210 GPA.
An elastic, perfectly plastic material model was used
(1) with temperature dependent properties adopted
from Michaleris and DeBiccari (1997).
Twelve elements were required through the
thickness of the plate in the vicinity of the weld
in order to accurately characterize the severe ther-
mal gradient in that region. The mesh density was
decreased in regions further away from the weld
where the thermal gradient was small so that the
analysis could be run in a reasonable amount of
time with the computational resources available.
Figure 2 shows a finite element mesh typical of
those used for the tee-stiffened plates in this study.
A nonlinear thermal analysis was used to predict
the transient temperature field produced by the
moving heat source. A circular heat source with a
Gaussian energy distribution representing the heat
Figure 2. Finite element mesh of tee-stiffened plate.
142
2/18/2011 5:43:35 PM
 from a welding torch, was moved along the weld numerical modelling was adopted from Deng et al.
path in 10 mm increments. After the continuous (2008). Figure 4 shows that the vertical deflection
fillet weld on one side was finished, the model was of the plate predicted by the welding simulation
allowed to cool for 30 minutes before the weld on method is in good agreement with experimen-
the other side of the stiffener web was started. The tal measurements provided by Deng et al. (2006).
thermal analysis considered temperature depend- Figure 5 shows the longitudinal residual stress dis-
ent material properties including thermal conduc- tribution at the mid-length, mid-plane of the plate.
tivity, specific heat and density. Heat loss due to Although residual stress measurements from the
convection and radiation was accounted for using experiment were not available, the residual stress
a film coefficient given by Goldak et al. (1984) to
account for heat loss by both mechanisms. Latent
heats of melting and fusion were also accounted for.
User-defined subroutines were created using the
ANSYS® parametric design language to model the
moving heat source and to control the activation
of weld elements as the heat source progressed.
For the second stage of the analysis, the tran-
sient temperature field from the thermal analysis
was used as a series of load steps in a structural
analysis. Each load step consisted of an incremental
progression of the heat source along the weld path.
As the heat source advanced along the weld path,
the ANSYS® element birth and death feature was Figure 3. Test specimen used by Deng et al. (2006).
used to activate the weld elements behind the heat Dimensions in mm.
source once their temperature fell below the solidi-
fication temperature, taken as 1450°C. This ena-
1
bled the model to simulate the dynamic coupling Position (mm)
of the stiffener to the plate as the weld progressed. 0
Vertical deflection of plate (mm)

0 100 200 300 400 500

The application of sequential thermal loads, and -1
the element activation scheme were controlled via -2
user-defined subroutines. Plate centerline
-3
Boundary conditions in the structural analysis (stiffener location)
were representative of the level of restraint in stiff- -4

ened panels in ship hull girders where a panel with -5

multiple evenly spaced stiffeners can be divided into -6
several individually stiffened plates. Longitudinal
-7
edges were constrained to remain straight, but free Experimental (Deng et al., 2006)
FEM
to move in the plane of the plating. This produced -8
a level of restraint similar to that provided by adja-
cent panels in ship hulls (Dow et al., 1981). Simple Figure 4. Comparison of vertical deflection of plate
supports were applied at the end cross-sections by with measured values.
constraining displacements along all three coordi-
nate axes at the centroid of one end and the verti- 350
cal and transverse displacements at the centroid of 300
Longitudinal residual stress (MPa)

the opposite end. An elastic-plastic material model 250

was used with von Mises failure criteria and asso-
200
ciated flow rule. Nonlinearities due to large strain
150
and displacement were considered.
The accuracy of the simulation was verified 100
using the methods described above to simulate 50
a welding experiment carried out by Deng et al. 0
(2006). In the experiment, a steel flat-bar stiffener -50
0 100 200 300 400 500

was connected to a steel plate by sequential, 6 mm

-100
fillet welds. Figure 3 shows the weld sequence and
-150
direction along with the test specimen dimensions. Position (mm)
The material used in the experiment was SM400A
shipbuilding steel with a yield stress of 300 MPa. Figure 5. Longitudinal residual stress at mid-span from
Temperature dependent material data necessary for welding simulation.

143

MARSTRUCT.indb 143 2/18/2011 5:43:38 PM

 distribution determined by finite element analysis Table 1. Stiffened plate dimensions.
is consistent with measured values available in lit-
erature (Michaleris & DeBiccari, 1997; Nagaraja Model b t d w tw tf β λ
Rao & Tall, 1961; Kenno et al., 2010). For a more
T1 350 12 180 100 10 12 1.21 0.34
comprehensive description of the welding simula-
T2 550 12 180 100 10 12 1.90 0.37
tion method, the reader is referred to Gannon et al.
T3 750 12 180 100 10 12 2.59 0.40
(2010). T4 950 12 180 100 10 12 3.28 0.43
T5 550 12 100 100 10 12 1.90 0.64
2.2 Shakedown analysis T6 550 12 140 100 10 12 1.90 0.47
T7 550 12 220 100 10 12 1.90 0.31
At the end cross-sections of the stiffened plate,
every node in the section was connected to a single * All dimensions in mm
node at the cross-section centroid using stiff beam
elements. These beam elements were deactivated
during the welding simulation using the ANSYS®
element birth and death feature. For the subsequent
shakedown analysis, the beam elements were re-
activated so that they forced the end cross-section
nodes to remain planar during the shakedown and
ultimate strength analyses. This constraint repre-
sents the support that would be provided by trans-
verse frames added after the stiffeners are welded
to the plate. For the shakedown analysis, an axial
load was applied to the centroidal node of one end
cross-section while the other remained pin sup-
ported at its centroid. Figure 6. Stiffened plate dimensions.

3 PARAMETRIC STUDY Table 2. Maximum distortions in FE models.

Plate (mm) Stiffener (mm)

3.1 Geometry
The study considered tee-stiffened plates with Model Vertical Vertical Horizontal
geometries chosen to cover a range of values of
T1 2.5 0.86 0.27
two non-dimensional parameters commonly used
T2 3.3 0.29 0.22
to characterize stiffened plate geometry. These are
T3 4.5 0.24 0.14
the plate slenderness β, and column slenderness λ,
T4 5.9 0.24 0.22
given by: T5 3.0 −0.30 0.38
T6 3.2 0.24 0.30
b σy T7 3.5 0.42 0.12
β= E (2)
t
⎛ a ⎞ σy
λ =⎜ ⎟ (3) in ship structures, where the plating between
⎝ πr ⎠ E stiffeners deflects in a single half-wave towards the
stiffener side of the plate. The welding simulation
where b is the plate width, t is the plate thickness, also produced column-type vertical, and lateral
σy is the yield stress, E is the elastic modulus, a is distortion of the stiffeners. While maximum out-
the plate length and r is the radius of gyration. of-plane distortions of the plate fell between the
A summary of the stiffened plate geometries is slight and average levels defined by Smith et al.
given in Table 1 with dimensions defined as shown (1992), vertical distortions of the stiffeners along
in Figure 6. All stiffened plates were 2000 mm long the length of the stiffened plates were slight in all
with 7 mm, continuous fillet welds on each side of models except T1. A summary of maximum dis-
the stiffener base, connecting it to the plate. Welds tortions in the stiffened plates is given in Table 2
were deposited in the sequence shown in Figure 3. where the direction of distortions is consistent
with the orientation shown in Figure 2. It should
be noted that distortions from hot-rolling were not
3.2 Geometric imperfection
considered, nor were distortions due to any other
Welding-induced distortions in the stiffened plates fabrication steps following welding of the stiffener
produced the familiar “hungry horse” shape seen to the plate.

144

MARSTRUCT.indb 144 2/18/2011 5:43:40 PM

 3.3 Shakedown 400
After welding

Longitudinal residual stress (MPa)

350
Shakedown of residual stress in the tee-stiffened 1 cycle at 25% of yield stress
300
plates was facilitated by application of a single
cycle of axial load at the centroid of the simply 250 1 cycle at 50% of yield stress
supported stiffened plates. For each model, two 200
load amplitudes were considered; one producing 150
an average stress in the stiffened plate equal to 100
0.25 σy, and the other producing an average stress
of 0.5 σy. The load cycles began with a linearly 50

increasing compressive load. This was followed by 0

-300 -200 -100 0 100 200 300
a linear transition to the same load in tension, after -50
which the load was reduced linearly back to zero. -100
Table 3 contains a summary of normalized Distance from centerline (mm)
maximum tensile σ t , and compressive σ c , residual
stresses in the plates of the finite element mod- Figure 7. Longitudinal residual stress in plate of
els before and after shakedown. To exemplify model T2.
the change in the longitudinal residual stress dis-
tribution due to shakedown, Figure 7, Figure 8 200
After welding
and Figure 9 show the residual stress distribu- Distance from top of plate (mm) 180
tions in the plate, web and flange respectively, of 160
1 cycle at 25% of yield stress
model T2. The results indicate that residual stresses 140 1 cycle at 50% of yield stress
are reduced significantly by shakedown. Where the
120
applied load produced a stress equal to 0.25 σy, ten-
100
sile and compressive residual stresses were reduced
by approximately 20% and 15%, respectively. When 80

the applied load produced a stress equal to 0.5 σy, 60

welding-induced residual stresses were reduced by 40

around 43% in tension and 40% in compression. 20
It is evident from the results in Table 3 that the 0
geometry of the stiffened plates had little influence -100 0 100 200 300 400

on the effects of shakedown. It is also noted that Longitudinal residual stress (MPa)
due to the low magnitude of compressive residual
stress and the amplitude of the applied load, no Figure 8. Longitudinal residual stress in web of
plastic straining occurred during the compressive model T2.
portion of the load cycle. Shakedown in this case
occurred only during the tensile part of the load. 0
The residual stress distribution in the plate of -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60
model T2 before shakedown is shown in Figure 7.
Longitudinal residual stress (MPa)

The plate has a tensile stress zone approximately -1

66 mm wide, so that the tensile stress block
parameter η (Figure 1) is equal to 2.75. After
-2

Table 3. Normalized maximum longitudinal resid-

ual stresses. -3
After welding
Initial 0.25 σy 0.5 σy 1 cycle at 25% of yield stress
-4 1 cycle at 50% of yield stress
Model σ t σc σt σc σt σc Distance from centerline (mm)

T1 1.00 0.20 0.79 0.17 0.57 0.13 Figure 9. Longitudinal residual stress in flange of
T2 1.01 0.14 0.79 0.12 0.57 0.09 model T2.
T3 1.02 0.10 0.79 0.08 0.57 0.06
T4 1.02 0.08 0.79 0.07 0.57 0.05
T5 1.01 0.13 0.79 0.11 0.57 0.08 welding, the width of the equivalent tensile stress
T6 1.02 0.14 0.79 0.11 0.56 0.08 block decreased so that η = 2.17 and η = 1.57 for
T7 1.01 0.13 0.79 0.11 0.56 0.09 shakedown stresses of 0.25 σy and 0.5 σy, respec-
tively. This is consistent with Faulkner’s (1975)
* Normalized with respect to yield stress of 360 MPa statement that the range of η decreases from 4.5–6

145

MARSTRUCT.indb 145 2/18/2011 5:43:41 PM

 to 3–4.5 after shakedown, however in this case displacement was used in order to allow the
η values were lower than Faulkner's suggestion. post-ultimate portion of the load-shortening curve
Distortion in stiffened plates was also affected by to be calculated.
shakedown. Figure 10 and Figure 11 show changes Load-shortening curves were also calculated for
in distortion in model T2 following one load cycle each stiffened plate considering welding-induced
at 0.5 σy. These changes in distortions, along distortions, but no residual stresses. This was done
with the reduced state of residual stress shown in to evaluate the ultimate strength when it is assumed
Figures 7 through 9 indicate that there are signifi- that residual stresses have been completely relieved
cant changes in fabrication-related imperfections by shakedown. The ultimate strength analyses
once a hull girder has been in subjected to longitu- considered large strains and displacements and
dinal bending moments during service. used the same material properties as the weld-
ing simulation and shakedown analysis described
previously.
3.4 Ultimate strength
Table 4 provides a summary of normalized ulti-
The strength and behaviour of the stiffened plates mate stress values for the stiffened plates calculated
is characterized by plots of average axial strain ver- by finite element analysis. Ultimate stress values are
sus average axial stress, known as load-shortening given before shakedown (σ u,0 ), after one load cycle
curves. An axial displacement was applied at the at 0.25 σy (σ u,25 ), after one load cycle at 0.5 σy (σ u,50 )
centroidal node of one end cross-section while the and with no residual stress (σ 0RS ). An example of
opposite end remained pin-supported at its cen- load-shortening curves before and after shake-
troid, producing a compressive load on the stiff- down is given in Figure 12 for model T5, and the
ened plate that was uniformly distributed over the deformed shape of the model in the post-ultimate
cross-section through stiff beam elements connect- stage is shown in Figure 13. This overall column
ing the nodes of each end cross-section. An applied type of buckling failure is typical of the stiffened
plates considered in this study where the effective
width of the plate is reduced until the cross-section
3.5
can no longer sustain the applied load.
Vertical deflection of plate (mm)

2.5
Table 4. Normalized ultimate strengths.
2

Model σ u,0 σ u,25 σ u,50 σ 0RS

1.5

1 T1 1.00 0.99 0.99 1.00

After welding T2 0.87 0.89 0.92 1.00
0.5
1 cycle at 50% of yield stress
T3 0.65 0.67 0.68 0.73
0
-300 -200 -100 0 100 200 300
T4 0.58 0.59 0.61 0.63
Distance from centerline (mm)
T5 0.86 0.89 0.92 1.00
T6 0.86 0.89 0.92 1.00
Figure 10. Vertical distortion of plate at mid-span. T7 0.86 0.88 0.91 1.00

0.3
1
No shakedown (1)
0.9 (3)
(2)
Normalized axial stress, σ/σy
Vertical distortion of plate at

0.25 (1) 1 cycle at 25% of yield stress (3)

0.8 1 cycle at 50% of yield stress (2)
0.2 No Residual Stress
mid-plane (mm)

0.7

After welding 0.6

0.15
1 cycle at 50% of yield stress 0.5
0.1 0.4
0.3
0.05
0.2
0 0.1
0 500 1000 1500 2000
0
-0.05
Distance along axis of stiffener (mm) 0 1 2 3 4
Normalized axial strain, ε/εy
Figure 11. Vertical distortion of plate along axis of
stiffener. Figure 12. Load-shortening curves for model T5.

146

MARSTRUCT.indb 146 2/18/2011 5:43:44 PM


Figure 13. Deformed shape of Model T5 after ultimate
load.
Shakedown of residual stress increased the ulti-
mate strength of the stiffened plates by a maximum
of 3.4% for an applied stress of 0.25 σy and 6.5%
for an applied stress of 0.5 σy (models T5 and T6).
Results of the ultimate strength analyses indicate
that the percentage change in ultimate strength due
to shakedown does not vary significantly with the
geometry of the stiffened plate. Figure 12 shows
that although shakedown increased the ultimate
load of model T5, it had little influence on the
shape of the load-shortening curve. The shape of
load-shortening curves calculated for the other
stiffened plates were also similar before and after
shakedown.
used, which is typical of mild steels used in ship
structures. The difference in shakedown behav-
iour between the two analyses with different plas-
tic material properties is shown by comparing the
strain histories at two nodes located at the mid-
plane and at the mid-length of the model. Node 1
is located at the mid-width of the plate in the ten-
sile stress zone, and node 2 is located near the edge
of the plate where the residual stress is compres-
sive. The node locations are shown in Figure 14.
Figure 15 shows the strain history at these points
over 3 load cycles represented by 8 steps with a
stress amplitude of 0.25 σy. Figure 16 shows the
same for the case where the material is perfectly
plastic. A summary of residual stresses at node 1
and node 2 at the points indicated in Figure 15 and
Figure 16, is provided in Table 5.
Figure 14. Location of nodes for results comparison.
400
1 3,5, 7
300
2 8 Node 1
4,6 Node 2
Stress (MPa)

200
A comparison of ultimate strengths after shake-
down at a stress of 0.5 σy against values calculated 100
without considering residual stress reveals that the 3, 5, 7

remaining residual stresses may still decrease the 0

-500 -300 -100 100 300
ultimate strength by as much as 10%. This shows
-100
that a potentially unsafe design may result if it 2
1 8
4, 6
is assumed that residual stresses are completely
-200
relieved by shakedown in order to simplify an ulti- Strain (με)
mate strength analysis.
Figure 15. Stress-strain history during shakedown with
kinematic hardening.
4 EFFECT OF HARDENING
AND NUMBER OF LOAD CYCLES 400
1
3,5,7
The shakedown analyses in Section 3 assumed an 300
elastic perfectly plastic material so that it would be 2 8
Node 1
consistent with the material behaviour used in the 4,6 200
Stress (MPa)

Node 2
welding simulation. In order to understand how
those results might differ from those of a shake- 100
3,5,7
down analysis where kinematic strain hardening
0
is considered, shakedown analyses of a stiffened -400 -300 -200 -100 0 100 200 300 400
plate with no strain hardening and with a kin- -100
1 8

ematic hardening material were run and results are 2

4,6
compared below. -200
Model T5 (Table 1) was chosen for the analysis Strain (με)
and the initial residual stress was similar to that
show in Figures 7 through 9. A notional hard- Figure 16. Stress-strain history during shakedown with
ening modulus of 5 GPa (Andersen 2000) was no hardening.

147

MARSTRUCT.indb 147 2/18/2011 5:43:49 PM

 The results in Table 5. indicate that there is little and kinematic hardening models, all plastic strain-
difference (0.7%) in the tensile residual stress after ing and thus, residual stress shakedown occurred
shakedown between the two models considering entirely during the first load cycle.
perfectly plastic and kinematic hardening materi-
als. The change in compressive residual stress due
to shakedown was virtually the same for both mod- 5 VARIABLE AMPLITUDE LOAD CYCLES
els. Furthermore, examination of Figure 16 and
Figure 17 reveals that for both the perfectly plastic In order to determine whether residual stress relief
by elastic shakedown is controlled only by the
maximum applied load and not by the load his-
Table 5. Stress history comparison considering tory, model T5 was subjected to 3 variable ampli-
perfect plasticity and kinematic hardening. tude load cycles. The complete load history is
illustrated in Figure 17 and the strain histories at
Kinematic hardening Perfectly plastic
nodes 1 and 2 (Figure 14) are plotted in Figure 18.
Step Node 1 Node 2 Node 1 Node 2 The maximum applied stress was 0.25 σy and the
material model used for the analysis assumed per-
1 350 −50 350 −50 fectly plastic behaviour.
2 269 −132 269 −132 After application of the variable amplitude
3 365 38 363 38 cyclic load, the tensile stress at node 1 and the
4 204 −126 202 −126 compressive stress at node 2 were reduced to
5 365 38 363 38 283 MPa and −44 MPa, respectively. Referring to
6 204 −126 202 −126 Table 5, where shakedown occurred entirely dur-
7 365 38 363 38 ing the first cycle at a stress of 0.25 σy, it is evident
8 285 −44 283 −44 that the maximum applied stress governs the final
magnitude of residual stress and that the sequence
*All values in MPa of loads is not relevant.

5
0.25

0.2
6 CONCLUSIONS
3 7
0.15
Applied stress, σ/σy

Welding-induced three-dimensional residual stress

0.1
and distortion fields in tee-stiffened plates were
0.05
1
simulated using finite element analysis. Cyclic axial
0
8 loads were applied to the stiffened plates and the
-0.05 resulting reduction in residual stress due to elastic
-0.1 shakedown was studied. Following the shakedown
-0.15
2 6 analysis, a compressive axial load was applied to
-0.2 the stiffened plates and their strength and behavior
-0.25 was characterized by load-shortening curves. The
4
effects of strain hardening and variable amplitude
load cycles on shakedown were also investigated.
Figure 17. Variable amplitude load history.
The following summarizes key conclusions drawn
400
from this study.
1 3 5 Welding-induced residual stresses in stiffened
2 300 7 plates typical of ship hull girders may be signifi-
8 cantly reduced when subjected to axial loads while
in service. For applied axial stresses of 0.25 σy and
4 6
200
Stress (MPa)

Node 1
Node 2 0.5 σy, longitudinal residual stresses were decreased
100 by around 20% and 40%, respectively. Residual
3 7
5 stresses are relieved entirely during the first load
0
-400 -300 -200 -100 0 100 200 300 400
cycle of a constant amplitude cyclic load. When
-100 1 8 the load amplitude is varied, the magnitude of the
4 2 6 highest load is the primary factor controlling the
-200 amount of residual stress shakedown achieved.
Strain (με) After partial stress relief by shakedown under
average axial stresses of 0.25 σy and 0.5 σy, the
Figure 18. Stress-strain history for variable amplitude ultimate strengths of tee-stiffened plates increased
loading. by 1.5–3.5% and 4.5–7%, respectively. When

148

MARSTRUCT.indb 148 2/18/2011 5:43:51 PM

 welding-induced residual stresses were neglected
and only distortions included, the ultimate stren-
gths of the stiffened plates increased by 8–16%.
This suggests that although residual stress relief
by shakedown in stiffened panels is beneficial, it
should not be assumed in a hull girder analysis
that residual stresses are entirely removed due to
shakedown while in service, as this may lead to
overly optimistic estimates of hull girder ultimate
strength.
A comparison of strain hardening models dem-
onstrated that for hardening moduli typical of
shipbuilding steels, kinematic hardening has little
influence on shakedown of welding-induced resid-
ual stress.
REFERENCES
Abdel-Karim, M. 2005. Shakedown of complex struc-
tures according to various hardening rules. Inter-
national Journal of Pressure Vessels and Piping 82:
427–548.
Andersen, L.F. 2000. Residual stresses and deformations
in steel structures. PhD. thesis, Technical University of
Denmark, Department of Naval Architecture and Off-
shore Engineering.
Deng, D., Liang, W. & Murkawa, H. 2006. Determina-
tion of welding deformation in fillet-welded joint by
means of numerical simulation and comparison with
experimental measurements. Journal of Materials
Processing Technology 183: 219–225.
Deng, D. & Murakawa, H. 2008. Prediction of welding
distortion and residual stress in a thin plate butt-
welded joint. Computational Materials Science 43:
353–365.
Dow, R., Hugill, R., Clark, J. & Smith, C. 1981. Evalua-
tion of ultimate ship hull strength. Extreme loads and
response symposium. SNAME, Arlington VA, October
1981.
149
MARSTRUCT.indb 149
Faulkner, D. 1975. A review of effective plating for use in
the analysis of stiffened plating in bending and com-
pression. Journal of Ship Research 19(1): 1–17.
Gannon, L.G., Liu, Y., Pegg, N.G. & Smith, M. 2010.
Influence of welding sequence on residual stress and
distortion in flat-bar stiffened plates. Marine Struc-
tures 23: 1–20.
Goldak, J., Chakravarti, A. & Bibby, M. 1984. A new
finite element model for welding heat sources. Metal-
lurgical Transactions B 15B: 229–305.
Gordo, J.M. & Guedes Soares, C. 1993. Approximate load
shortening curves for stiffened plates under uniaxial
compression. In D. Faulkner et al. (eds), Integrity of
Offshore Structures—5, 189–211.
Guedes Soares, C. 1988. Design equation for the com-
pressive strength of unstiffened plate elements with
initial imperfections. Journal of Constructional Steel
Research 9: 287–310.
Kenno, S.Y., Das S., Kennedy, J.B., Rogge, R.B. &
Gharghouri, M. 2010. Residual stress distributions in
ship hull specimens. Marine Structures 23: 263–273.
Latrou, N., Thevenet, D. & Cognard, J.Y. 2005. A fatigue
crack initiation approach for naval welded joints.
Oceans 2005—Europe, Vol. 2; Proc. IEEE, Brest,
20–23 June 2005.
Michaleris, P. & DeBiccari, A. 1997. Prediction of weld-
ing distortion. Welding Journal 76(4): 172–180.
Nagaraja, R. & Tall, L. 1961. Residual stresses in welded
plates. The Welding Journal 40: 468 s–480 s.
Paik, J.K., Hughes, O.F. & Renaud, C. 2005. Ultimate
limit state design technology for aluminum multi-hull
ship structures. Transactions, SNAME 113: 1–37.
Smith, C.S., Anderson, N., Chapman, J.C., Davidson,
P.C. & Dowling, P.J. 1992. Strength of stiffened plat-
ing under combined compression and lateral pressure.
Transactions, RINA 134: 131–147.
2/18/2011 5:43:52 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Geometrical effects on strength and deformability of corroded

steel plates

Muhammad Rabiul Islam

Graduate School of Engineering, Yokohama National University, Yokohama, Japan

Yoichi Sumi
Faculty of Engineering, Yokohama National University, Yokohama, Japan

ABSTRACT: Strength and deformability of steel plates for marine use are studied from the viewpoint
of geometry of corrosion pits and the size effect of corroded plates. The actual shape of corrosion pit
depends on the surrounding environment, which may result in a variety of pitting shapes such as coni-
cal and ellipsoidal shapes. In the present study, the effect of the two shapes has been investigated by
the non-linear, large deformation and three-dimensional finite element analyses for simulated corrosion
surfaces generated by a probabilistic model of a corrosion process. The strength of corroded plates with
semi-ellipsoidal pits is found to be estimated by the empirical formula obtained from that with conical
pits, where the estimation is based on the minimum cross sectional area of the plate. The deformability
and energy absorption of the corroded plates could be estimated by the surface roughness represented by
the difference of averaged plate thickness and that at the section of minimum cross sectional area. Having
investigated the size effect of corroded plates probabilistically, it has been quantitatively shown that the
strength and deformability reduce with increasing plate length, while they may increase with increasing
plate width. The size effect is more pronounced for deformability with the change of width.

1 INTRODUCTION proposed by Yamamoto and Ikegami (1998) for

Corrosion wastage is a prominent cause of age
related deterioration of steel structures. Metal
degrades locally in pit forms reducing strength and
deformability, which are main salient features for
integrity of steel structures. Since the effect of cor-
rosion is due to the geometric change of structures
where chemistry does not come into play (Oka et al.,
1990), studies related to pitting corrosion should
take into account of the actual shape of pitting,
which is considerably affected by the surrounding
environment. Nakai et al. (2004a, b) observed cir-
cular cone-shaped pits in the hold frames of bulk
carriers, and ellipsoidal-shaped pits in the bottom
shell plates of a tanker. They investigated actual
pitting corrosion observed on hold frames of
bulk carriers in different studies (2004a, b, 2005 &
2006). Paik et al. (2003 & 2004) studied the ulti-
mate strength behavior of corroded plates, whereas
Sumi (2008) estimated tensile strength and deform-
ability by using replica specimen.
The present study focuses on the effects due to
pit geometry and plate size, where various corro-
sion conditions with conical and ellipsoidal pits
have been simulated using probabilistic model
conical pits.
Strength and deformability reduction under
quasi-static uni-axial tensile load is estimated by a
series of non-linear implicit finite element analy-
sis. Commercial code LS DYNA 971 is used with
the material type of piecewise linear plasticity.
In non-linear FE analyses it is required to define
materials behavior under large strain. Material true
stress-true strain relationship is investigated with
the help of tensile test by using vision-sensor tech-
nology. Inhomogeneous strain field is calculated
by the effective strain defined by Scheider et al.
(2004), and tri-axial state of stress is considered
using both Bridgman (1964) and Ostsemin (1992)
correction factors. Results of FE analyses are also
verified by experiments in the case of ellipsoidal
pits. The material used is a conventional structural
steel SM490A. Empirical formulae for strength
reduction (Paik et al., 2003) and deformability
reduction (Ahmmad and Sumi, 2010) have been
examined for ellipsoidal pits.
Size effects of strength, deformability and energy
absorption of corroded plates are also investigated
by the numerical simulation with the use of the
central limit theorem.

151

MARSTRUCT.indb 151 2/18/2011 5:43:52 PM

 Table 1. Material properties.

Yield Mass Young’s Tensile

strength density modulus strength Failure
3
Material (MPa) (kg/m ) (GPa) (MPa) Elongation strain

SM490A 365.0 7853.6 206.9 509.4 28.73 0.97

(1998), we shall assume the phenomena of

corrosion by the three sequential processes.
Life of paint coating (T0) follows lognormal dis-
tribution with parameters μ0 (mean of ln(T0)) and
σ0 (standard deviation of ln(T0)):

1 ⎧⎪ ( t − )2 ⎫⎪
fT0 (t ) = exp ⎨ − 0
⎬. (1)
2πσ 0t 2σ 02 ⎪⎭
⎩

Transition time between active and progressive

pitting (Tr) follows exponential distribution with
Figure 1. True stress—true strain relation. parameter α (inverse of mean transition time):

2 MATERIAL PROPERTY IN LARGE gTr (t ) = α p ( − t ). (2)

STRAIN
Geometric configuration of a specimen changes
considerably under large deformation due to neck-
ing, which makes the measuring process of mate-
rial response during the large plastic deformation
a great challenge. Inhomogeneous strain field may
result in a tri-axial stress state. In the case of large
and non-uniform deformation, the true form of
quantities of stress and strain should be consid-
ered. As far as the deformation is uniform the true
form can be calculated from engineering quanti-
ties, but after the initiation of diffuse necking,
it requires the precise measurement of the instan-
taneous deformation.
In the present paper, the vision-sensor technol-
ogy (Ahmmad and Sumi, 2010) is used in obtaining
the relationship between the true stress and the true
strain beyond the onset of localized necking, which
can be accomplished by introducing Bridgman
(1964) and/or Ostsemin (1992) correction factors.
The mechanical properties of the material used in
the present work are listed in Table 1. In the current
study local necking is observed at effective strain,
0.41. The average percentage of stress correction
at this value is 0.8% (1% by Bridgman and 0.6%
by Ostsemin) so that the effect is not so significant
(see Figure 1).
In the third process, depth of pit (z) progresses
with time (τ) elapsed after the generation of pro-
gressive points as:
z(τ) = a.(τ)b; a, b coefficients, (3)
where the coefficient b varies from 1 to 1/3 depend-
ing on materials, environmental conditions and etc.
The coefficient a follows lognormal distribution
with parameters μa(mean of ln(a)) and σa(standard
deviation of ln(a)):
1 ⎧⎪ ( x − )2 ⎫⎪ .
ha ( x ) = exp ⎨ − a
⎬ (4)
2πσ a x 2σ a2
⎩⎪ ⎭⎪
They estimated the parameters governing their
probabilistic model from a survey carried by
ClassNK on 50 bulk carriers. These values are esti-
mated for four different locations of those ships, in
which values corresponding to bulkhead plates in
cargo holds are used in this study. Using ‘Monte
Carlo Simulation’ 5 data-sets for each process are
generated.
The shape functions for conical and ellipsoi-
dal pits are defined in the following form; conical
shape:

3 SIMULATION OF PITTED SURFACE

W0 ( 0 , r0 ,δ 0 )

According to the description of a probabilistic cor-

rosion model proposed Yamamoto and Ikegami
⎛
{
= δ 0 max 0 r0
⎝
(x − x0 )2 + ( y − y0 )2 }
⎞ (5)
⎟⎠

152

MARSTRUCT.indb 152 2/18/2011 5:43:52 PM

 ellipsoidal shape: implicit finite element code of LS-DYNA using
an elasto-plastic material where an arbitrary stress
W02 ( 0 , r0 ,δ 0 )
verses strain curve can be defined. This material
= δ 02 ( {
max 00, r02 {(x − x ) + ( y − y ) }}), (6)
0
2
0
2 model is based on the J2 flow theory with isotropic
hardening (Hallquist, 1998). A constant velocity,
V(t) = 3 mm/min, is applied in the loading direc-
where X0 and X are the position vector of a pit tion. Fracture is introduced by allowing elimina-
center and that of an evaluation point, z0 and r0 are tion of elements when strain to failure is achieved.
the depth and the radius of a corrosion pit at X0. FE results are validated by experiments using the
The parameter δ0 ( = z0/r0) follows normal distribu- test specimens with ellipsoidal pits.
tion with parameters μδ (mean of δ) and σδ (stand-
ard deviation of δ): 4.1 Mesh size effect

1 ⎧⎪ ( x − )2 ⎫⎪ Mesh size always has significant effects on finite ele-

fδ ( x ) = exp ⎨ − δ
⎬.
2πσ δ ⎪⎩ 2σ δ2 ⎪
⎭
Each sample of the corroded surfaces is defined
by using equally spaced data points. Pit cusps are
located by using uniform random variants. CAD-
software ‘RHINOCEROS’ is used to model the
corrosion surface by the NURBS surfaces. Cor-
roded surfaces with different DOP (Degree of pit-
ting intensity) are generated by varying the time (τ)
of equation (3).
4 FINITE ELEMENT ANALYSES
The test specimen is illustrated in Figure 2, whose
finite element model is generated by ‘MSC Patran’
using 8-node hexahedron elements (Figure 3).
Analyses have been carried out by the nonlinear
Figure 2. Test specimen for tensile test.
(7) ment results. Strain to failure is generally increases
with finer finite element meshes. So, it was aimed
to find out an appropriate element size along with
strain to failure for which experimental total elon-
gation of specimen can be achieved. In the case
of current model, it has been found that the effect
of element sizes except for the loading direction
are not so significant (Ahmmad and Sumi, 2010).
Therefore, the element size is kept constant (1 mm)
in width direction in the present analysis. Since the
element size in the thickness direction may vary
for specimens with corroded surfaces, the effects
of element sizes in the thickness along with length
(loading) directions are investigated.
As is shown in Figure 4, the total elongation
may change with the element size in the loading
direction. In the present analysis, the failure crite-
rion in terms of the effective strain in an element
is determined in such a way that the total elonga-
tion in FE analysis reaches that of the experiment,
28.73% (Table 1) in a flat specimen. It is seen that
the total elongation reaches the experimental value
at effective strain 0.932 in the case of element
size, 1 mm, so that this value with 1 mm mesh is
used for the failure effective strain in the following
analyses. In Figure 5 effect of loading and thick-
ness directional element size is shown, where the
strain to failure is normalized by the value 0.932.

Figure 3. Finite element modeling: (a) Boundary con-

ditions. (b) The minimum value of hz is 1 mm under deep- Figure 4. Effect of the mesh size in the loading direction
est pit cusp. for stress-strain relationship.

153

MARSTRUCT.indb 153 2/18/2011 5:43:55 PM

 Figure 6. Cross sectional average corrosion diminution
Figure 5. Effect of the mesh size in the loading and of 5 ellipsoidal pitted specimen.
thickness directions with regard to the failure strain.

From this figure it is found that the element size in

the thickness direction does not exhibit significant
effect. In the case of corroded specimens, a mini-
mum element size 1 mm is maintained by defining
a middle surface (Figure 3b), while the element size
is kept 1 mm in the other two directions.

4.2 Comparison between FE and experimental

results
Five specimens of different corrosion conditions Figure 7. Locations of failure in simulations (top) and
(DOP 20%, DOP 50%, DOP 78%, DOP 93% and experiments (bottom).
DOP 99%) with ellipsoidal pits are processed in the
present study. Figure 6 shows the averaged cross
sectional area along the specimen length, in which
the locations of failure are also indicated. It may
be seen that they show very strong correlation with
the location of minimum thickness of the cross
section. From FE analysis it is observed that stress
concentration occurs at each pit cusp, where a
shear band begins to form. The final failure occurs
at the position of the minimum thickness in a shear
band. Locations of failure in FE analyses exactly
coincide with those of experiments (Figure 7)
except for DOP 50%, in which there exist the two
possible sections. (see the insets of Figure 6). Figure 8. Experimental results for nominal stress—
In comparing the experimental and FE results nominal strain (ellipsoidal pits).
(Figures 8 and 9), they generally exhibit fairly good
agreement with each other, where the nominal
stresses are based on the intact cross section of the
flat specimen. Since the continuous measurement
by extensometer is not available for DOP 78% and
DOP 93%, the total elongations are measured from
fracture specimens in these cases. The experimental
results and FE analysis are listed in Table 2.

4.3 Effect of pit shape on strength

and deformability
A number of non-linear implicit finite element
analysis have been carried out for different corro- Figure 9. FE analysis of experimental specimens
sion conditions considering conical and ellipsoidal (ellipsoidal pits).

154

MARSTRUCT.indb 154 2/18/2011 5:43:59 PM

 Table 2. Strength and elongation of plates with
ellipsoidal pits.

DOP (%) 0 20 50 78 93 99

Tensile Exp. 509.4 498.1 484.0 452.3 432.8 373.4

strength
(MPa) FE. 507.8 495.8 480.8 455.4 429.3 373.0
Elongation Exp. 28.7 24.0 24.0 18.5 19.0 18.0
(%) FE. 28.7 21.5 23.5 18.5 17.6 18.0

Figure 10. FE results for nominal stress—nominal

pits using 5 sample data sets of probabilistic strain relationship (conical pits).
corrosion model. From Figures 10 and 11 it is clear
that total elongation of pitted specimen decreases
considerably with increasing DOP whereas tensile
strength decreases moderately. Considering the
case where DOP = 99%, deformability decreases
51% and strength decreases 18% in case of coni-
cal shape whereas those amounts are 56% and 33%
respectively in the case of ellipsoidal pit. Higher
material loss due to elliptical shape is responsible
for this phenomenon.
Paik et al. (2003) derived an empirical formula
for predicting the ultimate compressive strength
and shear strength based on damage (Dm):
Figure 11. FE results for nominal stress—nominal
Ru = (1 − Dm)0.73, (8) strain relationship (ellipsoidal pits).

where Ru is the ultimate tensile strength of pitted

plates normalized by that of an intact plate. The
damage is defined by

A0 AP
Dm = , (9)
A0

where A0 is the intact sectional area and AP is the

smallest cross sectional area due to surface pits.
The present study shows very good agree-
ment with the formula for both types of pit shape
(Figures 12 and 13). Experimental results are also
plotted in Figure 13. It is found that strength Figure 12. Strength reduction at different damages for
reduces approximately 20% in 10 years due to coni- specimens with conical pits.
cal pits in a bulk carrier.
Ahmmad & Sumi (2010) derived two empiri-
cal formulae for predicting deformability reduc-
tion due to conical pit based on surface roughness
given by

Rd 1 8 14RS + 26.4RS2 f 0 Rs ≤ 0.15, (10)

and

Rd 1 0 2RP − 5.3RP2 f 0 RP ≤ 0.35. (11)

In equation (10) and (11) Rd is the ratio of total Figure 13. Strength reduction at different damages for
elongation of pitted plate to that of an intact plate specimens with ellipsoidal pits.

155

MARSTRUCT.indb 155 2/18/2011 5:44:03 PM


Figure 14. Change of deformability reduction with
surface roughness due to ellipsoidal pits.
in minimum section by relating the width and
length of plate to their representative quantities.
In this study the size effect coefficient so obtained
will be used to predict the size effect of strength,
deformability and energy absorption reduction
due to pitting corrosion.
According to Yamamoto (2008), a plate sample
is discretized by N number of cross sections where
each cross section consists of M number of points,
which means that N and M are the numbers rep-
resenting plate length and width, respectively. The
average sectional corrosion diminution ( ) can
be defined as:
M
1
ZM =
M
∑ z (x ym ), (14)
m =1

where z(x, ym) is the corrosion diminution of mth

sampling point in a transverse section located at
position x. In the present analysis, the sampling
points are equally spaced in the x- and y-direction
with the distance 1 mm. The value, Z M , is assumed
to be a random variable, so that the central limit
theorem leads it to follow a normal distribution
regardless of the original distribution,, with a
standard deviation of mean as σ M σ M . ( )
Therefore, we can obtain
Figure 15. Change of deformability reduction with
σ2
maximum surface roughness due to ellipsoidal pits. M= = Representative Width, (15)
σM
2

and the variables representing surface roughness where σ is the standard deviation of corrosion
are given by diminution of the whole domain. The distribution
function for the maximum Z M among N sections
zavg Pmax zavgg can be given by:
RS Dm − ; RP = , (12), (13)
T T
{ }
N −1
respectively, where zavg is the average corrosion gN max ( z ) = N FZ M ( z ) fZ M ( z ) , (16)
diminution, T is the thickness of the intact plate
and Pmax is the depth of the deepest pit. where fZ M ( z ) is the probability density function
As illustrated in Figure 14, current study shows of the normal distribution of Z M . The mode
that equation (10) is also applicable for ellipsoidal value of the above distribution can be expressed by
pits with the specified range, whereas equation (11) using the inverse of standard normal distribution
gives good estimation in a very limited range of function Φ:
application, say surface roughness (RP) up to 0.1
for ellipsoidal pits (Figure 15). With increasing RP ⎛ 1⎞
value, the scatter of Rd increases, so that equation z MN mod = FZ−M1 1 − ⎟
⎝ N⎠
(10) is recommended for use in the estimation of
σ ⎛ 1⎞
deformability. Deformability is reduced more than =μ+ Φ −1 ⎜1 − ⎟ , (17)
60% due to ellipsoidal pit for a corroded surface M ⎝ N⎠
with DOP = 99%.
where μ is the mean corrosion diminution of
the whole domain. Equation (17) shows that the
5 SIZE EFFECT maximum average sectional corrosion diminu-
tion decreases and converges to μ with increasing
Yamamoto (2008) studied size effect of pitted plate representative width, M, while it monotonically
in the evaluation of average thickness diminution increases with increasing representative length, N.

156

MARSTRUCT.indb 156 2/18/2011 5:44:08 PM

 Rewriting equation (17), the representative length
can be expressed as:
1
N= . (18)
(
⎛ M z MN
1− Φ⎜
− )⎞⎟
⎜⎝ σ ⎟⎠
5.1 Effect of width and length on maximum
sectional average corrosion diminution
The relationships expressed by equations (15) and
(18) are examined by varying both actual width
and actual length from 200 mm to 1,000 mm by
100 mm interval for six sample specimens with
conical pits with seven corrosion conditions. The
averaged relations are shown in Figures 16 and 17,
in which they show linear correlation.
The maximum average sectional corrosion dimi-
nution (equation (17)) can be evaluated by adding
size effect coefficient, SC, multiplied by standard
deviation of corrosion diminution of whole area to
the average corrosion diminution of whole area.
Figure 16. Linear approximation for relation between
actual and representative width.
Figure 17. Linear approximation for relation between
actual and representative length.
157
MARSTRUCT.indb 157
Figure 18. Size effect on maximum sectional average
corrosion diminution.
1 ⎛ 1⎞
SC = Φ −1 1 − ⎟ . (19)
M ⎝ N⎠
The size effect coefficient is illustrated in
Figure 18, extending the length and width up to
3000 mm based on the results of Figures 16 and
17 in an averaged manner. This may cover the
standard size of stiffened panel in marine struc-
tures. The size effect of width is more pronounced
than that of length, and it decreases gradually
with increasing the width, while it increases with
increasing the length.
5.2 Size effect on strength, deformability
and energy absorption reduction
Ahmmad & Sumi (2010) proposed an empirical
formula for estimating energy absorption reduc-
tion due to pitting, which is given by
Re = Ru ⋅ Rd (20)
where Ru and Rd represent strength reduction
(equation (8)) and deformability reduction (equa-
tion (10)), respectively. Estimation of Ru solely
depends on damage of surface, whereas Rd can
be estimated from damage and average corrosion
diminution. In order to investigate the size effect
on Ru, Rd and Re, damage can be redefined as:
z MN mod
Dm = (21)
T
The mean and standard deviation appearing in
equation (17) are statistically independent regard-
less of location. As an illustrative example, here, we
shall investigate the size effect of strength, deform-
ability and energy absorption reduction of a speci-
men with plate thickness, T = 16 mm, μ = 3.09 mm
and σ = 1.33 mm, and the results are shown in
Figures 19–21. As can be seen from Figure 19,
2/18/2011 5:44:14 PM

Figure 19. Size effect on strength reduction factor.
Figure 20. Size effect on deformability reduction
factor.
Figure 21. Size effect on energy absorption reduction
factor.
the size effect on strength is not so significant for
plates of wider than 500 mm, while deformability
reduction and energy absorption reduction are
influenced in a wider range.
The results also show that strength and deform-
ability reduce with increasing plate length, while
they increase with increasing plate width. This
phenomenon may be interpreted in the following
way; the parallel load-bearing paths strengthen
the structural redundancy with increasing a plate
width. On the contrary, a least durable section may
appear by increasing the specimen length.
158
MARSTRUCT.indb 158
6 CONCLUSIONS
Two types of pit geometry and the size effect of
corroded plates have been studied numerically
and experimentally to examine the strength and
deformability of corroded plates. The following
findings are obtained:
1. The empirical formulae for strength reduction
and deformability reduction derived for conical
pits are also applicable to ellipsoidal pits.
2. Strength and deformability reduction is higher
for ellipsoidal pits than those for conical pits
because of the geometrical effect.
3. By combining the size effect coefficient of the
maximum average sectional corrosion dimi-
nution with the strength and deformation
reduction factors, the size effect of strength
reduction, deformability reduction and energy
absorption reduction due to pitting corrosion
can be estimated.
4. Strength and deformability increase with
increasing the plate width, while they reduce
with increasing length.
5. The size effect of strength reduction is not so
significant for a plate wider than 500 mm,
while deformability is affected in a much wider
range.
ACKNOWLEDGEMENTS
The authors express their appreciation to
Mr. S. Michiyama and Mr. H. Arakaki for their
supports during the present work. This work has
been supported by Grant-in-Aid for Scientific
Research (No. A(2) 22246109) from the Ministry
of Education, Culture, Sports, Science and Tech-
nology to Yokohama National University. The
authors are grateful for the support.
REFERENCES
Ahmmad Md. M. & Sumi, Y. (2010). Strength and
deformability of corroded steel plates under quasi-
static tensile load. J Mar Sci Technol vol. 15(1),
pp. 1–15.
Bridgman, P.W. (1964). Studies in Large Plastic Flow and
Fracture. Cambridge, Massachusetts.
Hallquist, J.O. (1998). LS-DYNA Theoretical Manual.
Livermore Software Technology Corporation.
Nakai, T., Matsushita, H. & Yamamoto, N. (2004). Effect
of pitting corrosion on local strength of hold frames
of bulk carriers (2nd report)—Lateral-distortional
buckling and local face buckling. Marine Structures
vol. 17, pp. 612–641.
Nakai, T., Matsushita, H. & Yamamoto, N. (2005).
Pitting corrosion and its influence on local strength
of hull structural members. Proceedings of the ASME
2/18/2011 5:44:19 PM
 24th International Conference on Offshore Mechanics
and Arctic Engineering OMAE2005-67025.
Nakai, T., Matsushita, H. & Yamamoto, N. (2006). Effect
of pitting corrosion on the ultimate strength of steel
plates subjected to in-plane compression and bending.
J Mar Sci Technol vol. 11(1), pp. 52–64.
Nakai, T., Matsushita, H., Yamamoto, N. & Arai, H.
(2004). Effect of pitting corrosion on local strength
of hold frames of bulk carriers (1st report). Marine
Structures vol. 17, pp. 403–432.
Oka, M., Kitada, H. & Watanabe, T. (1990). Experimen-
tal study on statistical strength of corrosive mild steel.
Journal of the Japan Society of Naval Architects and
Ocean Engineers vol. 167, pp. 229–235 (in Japanese).
Ostsemin, A.A. (1992). Stress in the least cross section
of round and plane specimen. Strength of Materials
vol. 24(4), pp. 298–301.
Paik, J.K., Lee, J.M. & Ko, M.J. (2003). Ultimate
strength of plate elements with pit corrosion wast-
age. J Engineering for Maritime Environment vol. 217,
pp. 185–200.
159
MARSTRUCT.indb 159
Paik, J.K., Lee, J.M. & Ko, M.J. (2004). Ultimate shear
strength of plate elements with pit corrosion wastage.
Thin-Wall Structures vol. 42, pp. 1161–1176.
Schider, I., Brocks, W. & Cornec, A. (2004). Procedure
for the determination of true stress-strain curves from
tensile tests with rectangular cross-sections. Journal
of Engineering Materials and Technology vol. 126,
pp. 70–76.
Sumi, Y. (2008). Strength and deformability of corroded
steel plates estimated by replicated specimens. Journal
of Ship Production vol. 24(3), pp. 161–167.
Yamamoto, N. (2008). Probabilistic model of pitting
corrosion and the simulation of pitted corroded
condition. Proceedings of the ASME 27th Interna-
tional Conference on Offshore Mechanics and Arctic
Engineering OMAE2008-57623.
Yamamoto, N. & Ikegami, K. (1998). A study on the
degradation of coating and corrosion of ship’s hull
based on the probabilistic approach. Journal of Off-
shore Mechanics and Arctic Engineering vol. 120,
pp. 121–128.
2/18/2011 5:44:22 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Methods to cope with up heaval buckling of high temperature

offshore pipelines in Mexico’s Bay of Campeche

J. Ochoa Z., J.E. Iturriaga F. & S. Meléndez P.

PEMEX Exploración y Producción, México

ABSTRACT: In Mexico, offshore pipelines are installed inside trenches. This has the effect of restraining
their movements to some extent, due to the weight of the backfill and the friction of the soil around the
pipe. When high temperature hydrocarbons start flowing through the pipelines, the friction around them
prevents the elongation and an axial force builds up. This force reaches a point where the restraining forces
are not enough to hold the pipe and buckling occurs. For buried pipelines the buckling is upwards, thus
the name Up Heaval Buckling (UHB). The mitigation measure for this problem is to add weight in the
form of rock or concrete mattresses at the places where UHB might occur. In Mexico a group of factors
have resulted in very expensive solutions, thus, PEMEX’s engineering department is working on finding
the most adequate methods to cope with UHB.

1 INTRODUCTION The type of buckling described above is known

When steel pipelines are heated they tend to increase
their dimensions, and particularly their length, due
to thermal expansion. When a pipeline is installed,
either offshore or inland, the contact between the
soil and the pipe restrains the movement to some
extent, due to the friction produced by this con-
tact. When high temperature hydrocarbons start
flowing through the pipe, the temperature of the
pipe starts to increase, but the restraining force
produced by the friction prevents the elongation,
and an axial force starts building-up along the pipe.
As time passes, the pipeline temperature starts
approaching the temperature of the hydrocarbons.
If the temperature of the hydrocarbons is above a
certain limit, the axial force reaches a point where
the restraining forces are not enough to hold the
pipeline in place and elongation occurs.
Normally, the restraining forces that keep the
pipeline in place are not constant along the pipe,
so, the increasing axial force produced by heating
the pipe will be higher than the restraining forces at
certain locations first along the pipeline. These local
elongations between still-fixed sections implicate
that the pipeline will buckle where the restraining
forces are lower, thus exceeded first by the increas-
ing axial force. Having a sideways or upwards
curvature along the axis of the pipeline makes
the location of the curvature prone to buckling.
As the curvature increases in a specific location,
so does the risk of buckling. When the curvature
of the pipeline’s axis is downwards (concave),
the development of the buckle is prevented by the
contact between the soil and the pipe.
as global buckling. Local buckling is when, in a
localized section, the plate that forms the pipeline
buckles. The scope of this work focuses on global
buckling.
Although high temperature is the factor that
affects the most the elongation of the pipeline and
therefore its eventual buckling, high internal pres-
sure has a minor contribution. Generally, the pipe-
lines that are prone to buckling are those known
as high temperature, high pressure or HT/HP
pipelines.
“Global buckling is a load response and not a
failure mode as such. Global buckling may, how-
ever, imply an ultimate failure mode such as:
− local buckling
− fracture
− fatigue”. (DNV, 2007)
Global buckling not being a failure mode, in
some cases it is not just allowed but even needed
and designed to happen at predefined locations to
keep the stresses and response of the pipeline under
the allowable limits. Letting the pipeline buckle at
pre-defined locations can “… achieve significant
CAPEX reduction over the very few viable alter-
natives, such as trench and rockdump”. (Hooper
et al., 2004).
DNV (2007) groups the possible pipeline buck-
ling scenarios in 3 categories:
− Exposed pipelines on even seabed
− Exposed pipelines on un-even seabed
− Buried/covered pipelines.

161

MARSTRUCT.indb 161 2/18/2011 5:44:22 PM

 For exposed pipelines, global buckling is
permitted, and the analysis and design philosophy
is based on allowing the pipe to buckle at specified
locations. The design for these scenarios includes
horizontal curvatures meant to induce buckling,
which liberates the expansion forces. This method
is known as Snake-lay Design, and, as Hooper et al.
(2004) explain it, the philosophy involved is to “…
work with the response rather than against it”.
Sometimes, in order to reduce the friction
between the soil and the pipeline in the locations
where buckling has to occur, thus facilitating the
process, the pipeline is laid over steel pipes installed
perpendicular to the axis of the pipeline, in the
locations where the curvature is specified. These
steel pipes over which the pipeline is installed are
known as “Sleepers”.
For buried or covered pipelines, the analysis
and design philosophy is to prevent buckling com-
pletely; this is achieved by maintaining the restrain-
ing forces always higher than the driving forces.
When a pipeline is installed inside a trench, the
buckling tends to be upwards, in a vertical plane,
as this direction presents less resistance to the pipe
movement, thus the name Up Heaval Buckling
(UHB).
In Mexico, the code for designing offshore pipe-
lines (PEMEX, 2009) specifies that they have to
always be installed inside trenches, with a minimum
depth of lowering (DOL, vertical distance between
the top of the pipeline and the undisturbed seabed
elevation) of 1 m. This indicates that the HT/HP
pipelines prone to buckle would present UHB,
and they have to be designed as to prevent it from
happening.
Mexico has been producing hydrocarbons in off-
shore facilities for almost 40 years. PEMEX is the
Mexican National Oil Company, and by law, is the
only entity allowed to produce oil and gas within
the Mexican borders. Despite PEMEX’s long
experience in offshore activities, the first HT/HP
pipeline prone to UHB in the Mexican part of the
Gulf of Mexico came around 2005 when some
oilfields from the Southwest Marine Region pre-
sented very hot oil.
Being a new problem for PEMEX, technical
support was hired from international engineer-
ing firms. A group of factors have resulted in very
expensive solutions, thus, PEMEX’s engineering
department is working on finding the most ade-
quate methods for Mexico to cope with UHB.
2 BASIC DEFINITIONS FOR UHB
ANALYSIS
As in most of the structural analysis and design
situations, UHB design is about making sure that
162
MARSTRUCT.indb 162
the restraining forces that prevent buckling from
happening are safely above the driving forces that
induce it.
DNV (2007) suggests that the design process
should be performed in two stages:
− Pre-installed design phase
− Post-installed design phase.
The post-installed design phase, which is the
final design against UHB is performed with Finite
Element Modeling (FEM) and non-linear analy-
ses. Although there are good commercial soft-
ware packages capable of performing the analysis,
PEMEX is writing and testing its own software
in order to adapt it to its particular needs. A few
years ago, the post-installed design phase against
UHB was done using the theoretical prop shape
configuration, and the expressions derived from
this assumption.
For the pre-installed design phase, the expres-
sions derived from the theoretical prop shape
configuration are still in use. A widely accepted
approach is that of Palmer et al. (1990) which
sets the bases for latter works. Expression 1 below
(DNV, 2007) is one example, and it is going to be
used as a support to what is being discussed and
analyzed in this work:
E⋅I
SF ⋅ | eff
f | (Rmax + wp + ⋅ wo ) 4 ⋅ δ ⋅ wo
(1)
where the left-hand side of the inequality is the
driving force in Newtons that induces UHB, and
the right-hand side is the restraining force that pre-
vents it.
SF is the safety factor and has to be greater than
1 in order to prevent UHB; Seff is called effective
axial force and is the driving force that induces
UHB, it is measured in Newtons; Rmax is the addi-
tional required downward force to prevent UHB, in
Newtons per metre; wp and wo are the submerged
weights during operation and during installation
respectively, measured in Newtons per metre; E
is the Young’s modulus of the steel in Pa; I is the
moment of inertia of the transverse section of the
steel pipe and δ is height of the theoretical imper-
fection (prop).
The effective axial force Seff is obtained with
Equation 2 below (DNV, 2007):
Seff = H − Δpi ⋅ Ai (1 − 2 ⋅ v) − As ⋅ E ⋅ α ⋅ ΔT (2)
where H is the residual tension from the instal-
lation of the pipe in N, which is normally taken
as null to be conservative; Δpi is the difference in
internal pressure compared to as laid in Pa; Ai is
the internal transverse area in m2; v is the steel’s
2/18/2011 5:44:22 PM
 poisson ratio, As is the steel transverse area in m2;
α is the steel’s expansion coefficient in K−1 and ΔT
is the difference in the steel’s temperature com-
pared to as laid in K.

2.1 Prop type imperfection Figure 2. Flexural moment diagram and deformed
shape showing the convex zone and wave-length.
Imagine that while laying a pipeline on a flat sea-
floor, a segment of it rests on an imperfection on
the trench bottom that is in contact with the pipe-
line just at one point, as depicted in Figure 1.
As seen in Figure 1, the soil is modelled as
Winckler springs, and the imperfection is modelled
as a simple support at a certain node, that has an
upwards ground displacement of δ m. Note that
the theoretical configuration is symmetrical with
respect to the simple support. For constructing
the model shown in Figure 1, springs in tension
are taken out, to have a better representation of
Figure 3. Transverse section of a trenched pipeline
the pipe/soil interaction. There is a direct relation- mitigated with rock dump.
ship between the magnitude of the curvature and
the propensity to UHB, the higher the curvature,
the lesser is the restraining force preventing UHB. Manipulating Expression 1 and taking the limit,
Besides the height of the theoretical imperfection Equation 3 below is obtained:
δ, other important parameters affecting the prop
shape and the curvature are the submerged weight 4 δ wo
wo, the stiffness E ⋅ I of the pipeline, and the stiff- R max | Sefffff | − w p 11 w o (3)
ness of the soil. It should be clear now that in real- E I
ity, when laying a pipeline inside a trench on the
when Rmax is null or negative means that there is
bottom of the sea, the occurrence of a theoretical
no requirement for additional weight to prevent
imperfection configuration is very rare.
UHB, the own weight and stiffness of the pipe are
enough to keep the pipeline stable. If Rmax is posi-
2.2 Mitigation measures against UHB tive, means that in order to prevent UHB, a down-
ward force of Rmax magnitude has to be applied on
Two important definitions are derived from the
top of the pipeline in the convex zone. The most
flexural moment diagram of the prop imperfection
common way to add weight on top of the pipeline
configuration. The so-called wave-length, which is
to prevent UHB is by dumping gravel inside the
two times the horizontal distance between the max-
trench, on top of the pipeline. Figure 3 depicts a
imum and the minimum elevations of the pipeline,
transverse section of a trenched pipeline mitigated
which are the points with horizontal tangents; and
with rock dump.
the convex zone which can be determined as the
When the length of the pipeline to be mitigated
horizontal distance between the inflection points
against UHB is relatively short, other solutions
of the longitudinal axis of the pipeline. At those
to achieve additional weight are used, as concrete
inflection points the bending moment is null.
mattresses or geotextile bags filled with sand or
In Figure 2 below, the flexural moment diagram of
other material.
the theoretical imperfection configuration is pre-
Analyzing Equations 2 and 3, it can be seen that
sented with the prop shape. The convex zone and
the parameters inducing UHB are:
wave-length are indicated.
− High internal operating pressure Δpi
− steel transverse area As
− high internal operating temperature ΔT
− height of the imperfection δ.
The parameters preventing UHB are:
− Submerged weight of the pipeline wo
− submerged weight of the fluids being trans-
Figure 1. Theoretical configuration for an imperfection ported, included in wp
(prop) of δ m of height. − bending stiffness E ⋅ I.

163

MARSTRUCT.indb 163 2/18/2011 5:44:23 PM

 Besides just adding extra weight Rmax over the
pipeline to prevent UHB, affecting the parameters
inducing it has been thought of frequently. The
operating conditions Δpi and ΔT are difficult to
reduce. In some cases, not in Mexico, ΔT has been
manipulated by using cooling spools. The idea
of heating the pipeline while installing it, during
pipe laying thus reducing ΔT has been discussed.
Increasing the Specified Minimum Yield Strength
(SMYS), hence being able to reduce the steel trans-
verse area As is something common. Reducing the
height δ of the imperfection is directly related to
the methods and quality of the trenching.
3 METHODS TO COPE WITH UHB
IN MEXICO
Mexico has been producing hydrocarbons from
offshore fields for almost 40 years, but the expe-
rience with HT/HP pipelines prone to buckling is
limited to some years, and not more than five pipe-
lines. Although the number of pipelines designed
against UHB is very limited, they were long, and
the mitigating measures to prevent UHB proved to
be very expensive. PEMEX’s engineering depart-
ment was assigned to reduce the amount of miti-
gating measures and to investigate ways to reduce
the cost of implementing them.
3.1 Changing the design code?
The Mexican design code for offshore pipelines
(PEMEX, 2009) stipulates that offshore pipelines
have to be trenched. This responds to the fact that
the code was developed for shallow water pipelines,
where hydrodynamic stability on the sea bottom
of the pipeline is hard to achieve for hurricane
conditions. As explained before, when the pipe-
line is inside a trench, UHB has to be forcefully
prevented.
PEMEX’s engineering department is exploring
the possibility of changing the code to allow for
exposed pipelines, while still maintaining hydro-
dynamic stability. As mentioned before, exposed
pipelines are allowed to buckle at specified loca-
tions, and there is the possibility of not using miti-
gating measures against buckling at all.
The Mexican code (PEMEX, 2009) specifies
that the pipeline has to be trenched with a DOL of
1 m, following, parallel, the profile of the seabed.
When the seabed is un-even, there might be crests,
and the code is asking the designer to impose the
same crest in the pipeline profile as there is in the
seabed. The possibility and advantages of specify-
ing in the code a more rectilinear pipeline profile,
with a minimum DOL of 1 m is being studied and
its feasibility established.
164
MARSTRUCT.indb 164
The Mexican Petroleum Institute (IMP) is
proposing PEMEX to include in the future HT/HP
to insert expansion loops (“omegas”) each some
km that would be above the seafloor, and would
allow the pipeline to expand, liberating the internal
force created by the increasing temperature.
3.2 Modifying laying and trenching practices
In Mexico, offshore pipelines are first laid on the
seafloor, and then, with water-jetting, trenched
to the specified DOL. This trenching method has
served well its original purpose, which was to pro-
tect the pipeline against hydrodynamic instability,
and therefore little has changed in the technology
involved.
However, water-jetting the pipeline to trench it
produces a profile that has a significant amount
of imperfections and in previous paragraphs it has
been discussed the implications and consequences
of the imperfection heights for UHB mitigation.
It has become clear to PEMEX that in order to
reduce as much as possible the amount of imper-
fections and their height, and hence the amount of
mitigating measures against UHB, the laying and
trenching practices have to be improved, starting
to use the latest technology available. Improv-
ing the flatness of the trenching in Mexico’s Bay
of Campeche will present challenges, as the soil
is extremely soft in most areas, intermingled with
paleochannels and dead coral zones.
3.3 Improving the smoothing of the raw
survey data
Once the pipeline is trenched in place, and after
the hydro-test has been performed, a survey of
the profile of the pipeline has to be performed.
This information is paramount to perform the
post-installation UHB analysis. No matter how
accurate the survey is, there is always some “noise”
in the measurements, so, in order to perform the
UHB analysis, the raw data has to be “smoothed”
by fitting a polynomial to it. Fitting one polyno-
mial to several kilometres of pipeline would not
give a correct answer, so, the polynomial fitting has
to be performed by steps in much shorter lengths
of pipeline. International engineering firms have
the length over which the polynomial fitting is
to be performed predefined, irrespectively of the
exact characteristics of the pipeline and seabed.
PEMEX’s engineering department is exploring
the advantages of fitting the polynomials ever the
exact length of the wave-length (see Fig. 2).
The polynomial’s degree for the fitting is some-
thing that is being studied. International engi-
neering firms, and some literature specify a cubic
polynomial for the fitting. The opinion of PEMEX’s
2/18/2011 5:44:25 PM
 engineering department is that according to beam
theory, the regression polynomial’s degree has to
be at least 4.
The regression polynomial represents the eleva-
tion of the pipeline profile, in other words, the
deformed position of the neutral axis. Assuming
that the un-deformed neutral axis of the pipeline
is collinear with the X axis, and that the Y axis
represents the vertical direction, a general regres-
sion polynomial of third degree for smoothing
data for UHB analysis would be as presented in
Equation 4 below:
y(x) = a ⋅ x3 + b ⋅ x2 + c ⋅ x + d (4)
from beam theory, it is known that the curvature of
the pipeline axis is related to the flexural moment
by Equation 5 below:
d 2 y (x)
M (x) = E I ⋅ (5)
dx 2
also from beam theory, it is known that the deriva-
tive of the flexural moment along the axis of the
pipeline is the shear force, and that the derivative
of the shear force along the axis is the distributed
load, as presented in Equations 6 and 7 below:
dM ( x )
V (x) = (6)
dx
dV ( x )
q (x) = (7)
dx
assuming that the pattern of the distributed load
on the pipeline is constant, the shear force would
be linear, the flexural moment would be quadratic,
as would the curvature of the pipeline, the slope
of the pipe line would be cubic, and the polyno-
mial representing the deformed position of the
pipeline’s neutral axis would be of fourth degree,
as opposed to the third degree polynomial used by
the international engineering firms that assisted
PEMEX. The load pattern that affects an offshore
trenched pipeline must be at least quadratic, which
leads to believe that the “real” polynomial repre-
senting the pipeline longitudinal axis should be of
sixth degree.
It is important to mention that reducing the
length over which the regression is made, and
increasing the degree of the fitting polynomial,
makes it more compliant with the raw data, which
might introduce risks that are being studied.
3.4 Improving survey quality
In the past, for pipelines that were not prone to
UHB, the post installation survey was the means
165
MARSTRUCT.indb 165
to have As-laid information for the Geographic
Information System and for future interventions.
The equipment used for the survey was not part
of state-of-the-art technology, but again, for
the intended purposes it served well. The gen-
eral position of the pipeline was known, which
was what was being needed. For UHB analysis,
the precision needed for the pipeline profile is
completely different. Lack of precision in a sur-
vey for UHB analysis has as a consequence that
continuous rock dump has to be applied along
the axis of the pipeline, which implies massive
amounts of gravel and money. The uncertain-
ties on the exact location of the pipeline profile
impede the designer to know where the profile
is convex, hence eventually needing mitigating
measures. If the designer cannot identify the
locations where mitigating measures would be
needed, he will specify them for the whole sus-
ceptible length of pipeline.
For the common operating conditions in
Mexico, if the standard deviation of the differ-
ences between the smoothed profile and the raw
data is around 2.5 cm, there would not be the need
for continuous mitigating measures.
3.5 Design of other mitigation measures
It has been mentioned that the mitigation measures
for UHB, once the pipeline is already trenched, are
rock dumping (see Fig. 3) and installing concrete
mattresses or geotextile bags filled with sand or
concrete.
If the total pipeline length to be mitigated is
relatively long, the solution is rock dumping, which
implies mobilization and demobilization of a spe-
cial vessel, generally from the North Sea. The cost
of these is above five million dollars without taking
into account the cost of the rock, which is bought
in Mexican ports. For a pipeline that presents a
shorter length to be mitigated, the adopted solu-
tions are installing, one by one, either concrete
mattresses or geotextile bags.
The problem with rock dumping, besides the
cost of bringing the equipment from the North
Sea, is that a huge amount of rock is not directly
working on top of the pipeline, it fills the sides of
the trench or simply rests on the sea floor. The poor
accuracy for locating the rock, added to the lack of
cohesion make that the amount of rock wasted is
considerable.
Using concrete mattresses or geotextile bags
present two problems, first, the geometry of both
do not match properly the top half of the pipeline,
and a big portion of them is not resting on top of
the pipe but on the seabed. Second, they have to
be installed one by one, with a diver directing and
participating in the maneuver.
2/18/2011 5:44:25 PM
 The idea was to design a solution that could
address and minimize the disadvantages presented
by the existing mitigating solutions. The chosen
name is “alforjas” which means saddlebags in
Spanish. Figure 4 shows a transverse section of a
pipeline with the saddlebags on top.
A pair of saddlebags is two high density con-
crete pieces connected to each other by chain or
steel rope. In figure 5 a 3D image of a single sad-
dlebag is presented.
The saddlebags can be mass-produced and stored.
The geometry is designed to take into account:
− Different pipeline outside diameters. The articu-
lation on top of the pair of saddlebags allows
them to accommodate a range of outside
diameters. Figure 5. 3D image of a single saddlebag.
− Sedimentation area. The upper part of the pair
of saddlebags is designed to create a region of
low fluid velocity, hence allowing sedimentation
of suspended particles on the seawater, which
increases the downward force and the stability.
In extreme cases, this region would be used as
a base to other mitigating measures as rock or
concrete mattresses.
− Different trench materials. The design allows
changing the outer wall inclination to accom-
modate trenches in soils with different angle
of rest.
Besides, the geometrical design implies a low
position of the centroid, giving stability to the sad-
dlebags, once installed on top of the pipeline.
The concrete’s density can be increased, incre-
menting the weight of each saddlebag, or reducing Figure 6. Isometric image of an ensemble of saddlebags.
their sizes.
Some very interesting advantages of the sad-
dlebags are being found, but the most promising
is the reduction of time of installation. It has been
mentioned that a pair of saddlebags is connected
on top by chain or steel rope. Longitudinally, a
series of pairs of saddlebags are also connected by
chain or steel wire. In previous paragraphs it was
commented that the additional downward force
needed to prevent UHB from happening has to Figure 7. Top view of an ensemble of saddlebags.
be applied in the convex zone of the imperfection.

Pairs of saddlebags are connected longitudinally

Figure 4. Transverse section showing the mitigation
measure designed in PEMEX’s engineering department.
to obtain a length equal to the convex zone. The
distance between pairs of saddlebags is such as to
have the downward force per unit length needed.
This ensemble of pairs of saddlebags that has the
length of the convex zone can be installed on top
of the pipeline from a vessel, in a very similar way
as the pipeline is laid. In Figures 6 and 7 a group of
saddlebags, connected between them is depicted.
For installation purposes, the first end of the
saddlebags ensemble would have to be connected
to an anchor or dead-weight at the sea bottom.

166

MARSTRUCT.indb 166 2/18/2011 5:44:26 PM

 The installation vessel would have to start sailing Table 2. Rock cover and rock weight to prevent
towards the imperfection, lowering the ensemble. UHB for different imperfection heights.
When the first end, composed by a pair of sad-
dlebags reaches the seabed a diver or ROV would Imperfection Rock weight
height (m) Rock cover (m) (kN/m)
have to direct it to land directly on top of the pipe,
as the vessel sails and lowers the rest of the sad- 0.1 0.5 45
dlebags ensemble, the ROV or diver would have to 0.2 0.55 50
direct the vessel in order to place all the saddlebags 0.3 0.75 66
pairs on top of the pipeline. 0.4 0.9 78
At the current time, PEMEX is evaluating tech- 0.5 1 93
nically and economically the merits of pursuing
the saddlebags idea.

5 CONCLUSIONS
4 RESULTS
The analysis and design of HT/HP pipelines
For the technical merit of the saddlebags or other
prone to buckling is a new challenge for the oth-
solutions, an analysis has been made of the tradi-
erwise offshore experienced PEMEX’s engineering
tional rock-dumping.
department.
PEMEX designed and installed a 30″ pipeline
Allowing the design of exposed offshore pipe-
that is 21 km long, and will be transporting oil
lines could be a means of addressing HT/HP pipe-
at around 120°C. During the design phase of the
lines, but it requires changing the Mexican design
project (PEMEX 2010), the downward force Rmax
code for offshore pipelines (PEMEX, 2009), and
required to keep the pipeline stable, thus preventing
solving hydrodynamic stability in hurricane condi-
UHB was determined. In table 1 in the next page,
tions in different ways than trenching it.
the values of Rmax for different heights of the theo-
Laying, trenching, surveying and smoothing
retical imperfection δ, and 120°C are presented:
practices have a very high impact on the amount
The values presented in the second column of
of mitigating measures needed for a specific pipe-
table 1 are then converted to either rock quanti-
line prone to UHB.
ties or number of concrete mattresses. Using as
Commonly used mitigating measures against
example rock-dump, for each value of downward
UHB in other parts of the world might not be the
force, a rock cover height is determined. From the
most adequate for Mexico’s Bay of Campeche. The
rock cover height the amount of rock is computed
lack of operators, other than PEMEX, in Mexico
(PEMEX 2010). In table 2 below the values of rock
makes mobilization and demobilization costs
cover height and amount of rock are presented:
extremely high. The use of alternative solutions, as
Comparing the downward force required to
the saddlebags presented here, could be adopted by
prevent UHB from happening (second column of
countries where the traditional solutions are not as
table 1) against the weight of rock needed to provide
convenient as they are for the countries that origi-
such downward force (third column of table 2), it is
nally developed them.
found that the amount rock not directly acting on
top of the pipeline, thus not “working” in the UHB
solution ranges from 60% to 85% of the total. REFERENCES
The saddlebags solution, by design, applies all the
mobilized weight on top of the pipeline, thus mak- DNV 2007. Global buckling of submarine pipelines.
ing it a more efficient solution in this aspect also. Structural design due to high temperature/high pressure
DNV-RP-F110. Norway: Det Norske Veritas.
Table 1. Downward forces Hooper, J., Maschner, E. & Farrant, T. 2004. HT/HP
required to prevent UHB for Pipe-in-Pipe Snaked Lay Technology—Industry Chal-
different imperfection heights lenges. OTC 16379. Houston: OTC.
and 120°C. Palmer, A.C., Ellinas, C.P., Richards, D.M. & Guijt, J.
1990. Design of Submarine Pipelines Against Upheaval
Imperfection Buckling. OTC 6335. Houston: OTC.
height (M) Rmax (kN/m) PEMEX 2009. Diseño de líneas submarinas en el Golfo de
México. Mexico: PEMEX.
0.1 4.17 PEMEX 2010. 30″ Pipeline Between—and—Predictive
0.2 11.21 Upheaval Buckling Report. London: KW LTD.
0.3 15.64
0.4 19.07
0.5 21.71

167

MARSTRUCT.indb 167 2/18/2011 5:44:31 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Benchmark study on use of ALPS/ULSAP method to determine

plate and stiffened panel ultimate strength

J.K. Paik, S.J. Kim, D.H. Kim & D.C. Kim

The Lloyd’s Register Educational Trust Research Centre of Excellence, Pusan National University,
Busan, Korea

P.A. Frieze, M. Abbattista & M. Vallascas

PAFA Consulting Engineers, Hampton, UK

O.F. Hughes
Virginia Tech, Blacksburg, USA

ABSTRACT: The objective of this paper is to check the accuracy of the ALPS/ULSAP (Analysis of
Large Plated Structures/Ultimate Limit State Assessment Program) method’s use to determine the ulti-
mate strength of plates and stiffened panels. The details of the ALPS/ULSAP method and theory are
presented in both Ultimate Limit State Design of Steel-plated Structures, co-authored by J.K. Paik and
A.K. Thayamballi, and Ship Structural Analysis and Design, co-authored by O.F. Hughes and J.K. Paik.
In this benchmark study, the accuracy of the plate and stiffened panel ultimate strength obtained with
the ALPS/ULSAP method is ascertained through comparison with that obtained using nonlinear finite
element methods and the DNV/PULS method.

1 INTRODUCTION 2 THEORY OF THE ALPS/ULSAP

It is now well recognized that ultimate strength
is a much better basis for the design and strength
assessment of ship structures than allowable work-
ing stress (Paik & Thayamballi 2003, 2007; ISO
2007; Hughes & Paik 2010), which also holds true
for the condition assessment of aged structures
(Paik & Melchers 2008).
Plates and stiffened panels are the basic struc-
tural components that govern the overall failure of
ships and offshore structures. Their accurate and
efficient calculation is thus a very important task
in the design and safety assessment of ships and
offshore structures.
The ultimate strength algorithms for plates and
stiffened panels developed by Paik and his col-
leagues (Paik & Thayamballi 2003; Hughes & Paik
2010) have been implemented in ALPS/ULSAP
software (2010). The present paper reports the
results of a benchmark study comparing the
ultimate strength of plates and stiffened panels
obtained using the ALPS/ULSAP method with
that obtained with nonlinear finite element meth-
ods and the DNV/PULS method (2009).
METHOD
The stiffened plate structure shown in Fig. 1 is con-
sidered. This structure is subject to the combined
in-plane and lateral pressure load shown in Fig. 2.
2.1 Ultimate strength of plates
The membrane stress-based method (plastic edge-
oriented plate hinge approach) is applied (Paik &
Thayamballi 2003, 2007; Hughes & Paik 2010).
The membrane stress inside a deflected or buck-
led plate is non-uniform. Figure 3 depicts a typical
example of the axial membrane stress distribution
inside a plate that is subject to uniaxial compres-
sive loading before and after buckling occurs. For
simplicity, the case of a single bulge in the middle
of the plate is shown.
The membrane stress distribution in the load-
ing (x) direction becomes non-uniform as the plate
starts to deflect (e.g., due to buckling). That in
the y direction also becomes non-uniform if the
unloaded plate edges remain straight, although no
membrane stresses will develop in this direction

169

MARSTRUCT.indb 169 2/18/2011 5:44:31 PM

 (a) Before buckling.

Figure 1(a). Stiffened plate structure—nomenclature

of the plate panel.
(b) After buckling, the unloaded edges move freely in plane.
b b b

tp tp tp

N. A. N. A. N. A.
hw hw hw
tw tw tw
tf tf

bf bf

Figure 1(b). Stiffened plate structure—nomenclature (c) After buckling, the unloaded edges remain straight.
of the stiffeners.
Figure 3. Membrane stress distribution inside a plate
subject to uniaxial compressive loads in the case of one
bulge in the middle of the plate.

middle of the plate, where a membrane tension

field is formed by the plate deflection because the
plate edges remain straight.
A similar nonlinear membrane stress distribu-
tion may appear inside a deflected plate that is
subject to combined axial compression and lateral
pressure loads. Edge shear loading may render the
membrane stress distribution pattern more com-
plex than that under biaxial and lateral pressure
load conditions, but as long as the edge shear is
a secondary load component, the basic membrane
stress distribution pattern inside the plate is likely
to be similar to that in Fig. 4(c).
With an increase in plate deflection, the mem-
brane stress is redistributed as in Fig. 4(c), but,
although the stress in the mid-width of the plate
remains lower, that in the upper and/or lower faces
in the mid-width will initially yield through bend-
Figure 2. Stiffened panel subject to a combined in-plane ing action.
and lateral pressure load.
As long as it is possible to redistribute the
stress to the straight plate boundaries through
under longitudinal compression only if these edges membrane action, however, the plate will not col-
move freely in plane. It should be noted that the lapse. Collapse will occur when the most stressed
unloaded edges of a plate that is part of a stiffened boundary locations yield because the plate can no
panel are likely to remain straight. longer keep the boundaries straight, thus result-
The maximum compressive membrane stresses ing in a rapid increase in lateral deflection, which
develop around the plate corners, and the mini- corresponds to the ultimate limit state or ultimate
mum (tensile) membrane stresses occur in the strength (Paik & Thayamballi 2003).

170

MARSTRUCT.indb 170 2/18/2011 5:44:31 PM


(a) Plasticity at the corners.
(b) Plasticity at the longitudinal mid-edges.
(c) Plasticity at the transverse mid-edges.
Figure 4. Three possible locations for the initial plastic
yield at plate edges subject to combined loads (•: Expected
plasticity location; T: Tension; C: Compression) (Paik &
Thayamballi 2003).
Because of the nature of the combined mem-
brane axial stresses in the x and y directions, there
are three possible locations for the initial yield at
the edges, namely, the plate corners, the longitu-
dinal mid-edges, and the transverse mid-edges, as
shown in Fig. 4. The stress at the two mid-edge
locations, i.e., that at each longitudinal or trans-
verse mid-edge, is expected to be the same as long
as the longitudinal or transverse axial stresses are
uniformly applied, i.e., without in-plane bending.
Depending on the predominant half-wave mode
in the length direction, the location of possible
plasticity may vary at the long edges because the
location of the minimum membrane stresses may
differ, whereas it is always at the mid-edges in
the short direction. In this regard, the membrane
stress-based method can also be called the plate
edge-oriented plastic hinge approach.
The occurrence of plasticity can be assessed
using the von Mises yield criterion. The three
following ultimate strength criteria for the most
171
MARSTRUCT.indb 171
probable yield locations can be found once the
maximum or minimum membrane stresses are
defined, as shown schematically in Fig. 4.
a. Plasticity at plate corners:
σ eq1 σ x2 max − σ x maxσ y max + σ 2y max + 3τ 2 = σY . (1)
b. Plasticity at longitudinal mid-edges:
σ eq 2 σ x2 max − σ x maxσ y min + σ 2y min + 3τ 2 = σY . (2)
c. Plasticity at transverse edges:
σ eq3 σ x2 min − σ x minσ y max + σ 2y max + 3τ 2 = σY . (3)
As the applied loads increase, the plate will col-
lapse if any one of the three foregoing equivalent
stresses, namely, σeq1, σeq2, or σeq3, reaches material
yield stress σY. The minimum value among all of
the applied load components that satisfy the three
equations must then be the real ultimate strength
of the plate.
The maximum and minimum membrane stresses
in equations (1) to (3) can be formulated as func-
tions of the various parameters of influence (Paik &
Thayamballi 2003; Hughes & Paik 2010).
2.2 Ultimate strength of stiffened panels
The possible collapse modes for a stiffened panel
subject to a combined in-plane and lateral pressure
load, such as that shown in Figs. 1 and 2, can be
categorized into the following six types (Paik &
Thayamballi 2003, 2007; Hughes & Paik 2010).
• Collapse mode I: Overall collapse of the plating
and stiffeners as a unit; see Fig. 5(a).
• Collapse mode II: Plate-induced collapse without
distinct failure of the stiffeners; see Fig. 5(b).
• Collapse mode III: Stiffener-induced collapse by
beam-column-type collapse; see Fig. 5(c).
• Collapse mode IV: Stiffener-induced collapse by
local buckling of the stiffener web; see Fig. 5(d).
• Collapse mode V: Stiffener-induced collapse by
flexural-torsional buckling or tripping of the
stiffeners; see Fig. 5(e).
• Collapse mode VI: Gross yielding.
This classification of collapse modes is applica-
ble to any load combinations, including uniaxial
compressive loads and combined in-plane loads
with or without lateral pressure loads.
Collapse mode I represents overall collapse after
overall buckling. In this mode, the stiffeners and
the plating buckle as a unit, and overall buckling
often occurs under an elastic regime. This collapse
2/18/2011 5:44:35 PM
 Figure 5(a). Collapse mode I: Overall collapse of the
plating and stiffeners as a unit (shaded areas represent
yielded regions).
Figure 5(b). Collapse mode II: Plate-induced collapse
mode without distinct failure of the stiffeners (shaded
areas represent yielded regions).
Figure 5(c). Collapse mode III: Stiffener-induced col-
lapse mode by beam-column-type failure (shaded areas
represent yielded regions).
Figure 5(d). Collapse mode IV: Stiffener-induced
collapse mode by local buckling of the stiffener web
(after the buckling collapse of the plating between the
stiffeners).
172
MARSTRUCT.indb 172
Figure 5(e). Collapse mode V: Stiffener-induced collapse
mode by flexural-torsional buckling of the stiffeners
(after the buckling collapse of the plating between the
stiffeners).
mode typically occurs when the stiffeners are rela-
tively weak relative to the plating.
Collapse mode II occurs when the panel is sub-
jected predominantly to biaxial compressive loads,
thereby causing it to collapse due to yielding along
the plate-stiffener intersection at the panel edges,
with no distinct stiffener failure. In contrast to col-
lapse modes III, IV, and V, this mode assumes that
the stiffeners do not fail first.
When the stiffener dimensions are neither weak
nor strong, the stiffened panel is likely to behave
as a plate-stiffener combination that is representa-
tive of the entire panel, thus reaching its ultimate
strength via collapse mode III, beam-column-type
collapse.
When the height to thickness ratio of the stiff-
ener web is large, local buckling is likely to take
place in the web. Collapse mode IV occurs when
the stiffener web buckles in conjunction with the
inception of failure in the plating between the
stiffeners.
When the stiffener flange is of a type that is
unable to remain straight, the stiffeners twist side-
ways, a phenomenon known as flexural-torsional
buckling or tripping. Collapse mode V constitutes
the pattern of failure in which the panel collapses
due to the lateral-torsional buckling or tripping of
the stiffeners.
The stiffened panel reaches its ultimate strength
in collapse mode VI when the panel is stocky or
subjected predominantly to axial tensile loading,
such that neither local nor overall buckling occurs
until the panel cross-section yields either entirely
or to a large extent.
Although these collapse modes are illustrated
separately here, some of them may interact and
occur simultaneously. For the sake of simplicity,
however, a stiffened panel is considered to reach
its ultimate strength via the first, and predomi-
nant, of the six collapse modes to occur. Hence,
2/18/2011 5:44:37 PM
 the ultimate strengths of the panel are calculated Trans. frames
separately for each of these collapse patterns, with
b/2

Long. stiffeners
the smallest value among those computed taken as b/2
its real ultimate strength.
b
b/2
b/2
3 BENCHMARK STUDY
a/2 a/2 a a/2 a/2
3.1 Study methods
The three following methods were employed for Figure 6(a). Nonlinear finite element method
comparison purposes in this benchmark study. modelling—extent of the analysis for the plate.

• Nonlinear finite element method

• ALPS/ULSAP method (2010)
• DNV/PULS method (2009)
For the nonlinear finite element method analy-
sis, which was considered to be the most refined
approach, ANSYS (2010), Abaqus (2010), and
MSC/MARC (2010) were employed.

3.2 Ultimate strength of plates

The geometric and material properties of the plates
considered in this study are as follows.
Figure 6(b). Nonlinear finite element method
• Plate length, a = 2550 mm modelling—mesh size.
• Plate breadth, b = 850 mm
• Plate thickness, tp = 11, 16, 22, 33 mm
• Yield stress, σYp = 313.6 MPa 1.0
a××bb==2550×
a × 850 (mm)
2550 850(mm)
• Elastic modulus, E = 205800 MPa tp= 11
11mm
mm
• Poisson’s ratio, v = 0.3 0.8

The plates are subject to biaxial compressive

0.6
loads, and all edges are assumed to be simply sup-
σyu/ σY

FEM
ported. No welding residual stresses are consid- ALPS/ULSAP
DNV/PULS
ered, although the plates have initial deflection wopl, 0.4
which corresponds to the plate buckling mode, as
follows. 0.2

• wopl = 0.1β 2tp for ANSYS and ALPS/ULSAP;

• wopl = b/200 for DNV/PULS, 0.0

( )
0 0.2 0.4 0.6 0.8 1.0

where β p σYp E . σxu/ σY

The magnitude of the ANSYS and ALPS/
ULSAP method analyses differs from that of the Figure 7. Ultimate strength interaction relationship
DNV/PULS because the latter implicitly consid- between biaxial compressive loads for the plate with
tp = 11 mm.
ers the initial imperfections, whereas both of the
former deal with them as parameters of influence,
and the present benchmark study considers an
with the ANSYS analyses, whereas the DNV/PULS
average level of plate initial deflection.
tends to overestimate the plate ultimate strength.
Figure 6 presents the nonlinear finite element
method modeling for the plate in terms of the
analysis extent and mesh size. 3.3 Ultimate strength of stiffened panels
Figures 7 to 10 present the ultimate strength
The geometric and material properties of the stiff-
interaction relationships between biaxial com-
ened panel (denoted by panel type C) considered in
pressive loads for plates with tp of 11 mm, 16 mm,
this study are as follows.
22 mm, and 33 mm, respectively.
The comparisons show that the ALPS/ULSAP • Panel length, a = 4750 mm
method computations are in very good agreement • Panel breadth, B = 8550 mm

173

MARSTRUCT.indb 173 2/18/2011 5:44:39 PM

 1.0 • Number of stiffeners = 8
a ×bb==2550×
a× 2550 ×850(mm)
850 (mm) • Plate breadth, b = 950 mm
tp= 16
16mm
mm
0.8 • Plate thickness, tp = 11, 12.5, 15, 18.5, 25
37 mm
• Yield stress of plate, σYp = 313.6 MPa
0.6 • Yield stress of stiffeners, σYs = 313.6 MPa
σyu/ σY

FEM
ALPS/ULSAP • Elastic modulus, E = 205800 MPa
0.4
DNV/PULS
• Poisson’s ratio, v = 0.3
No welding residual stresses are considered. The
0.2
following condition is applied for the plate initial
deflection.

0.0
• wopl = 0.1β 2tp for FEA and ALPS/ULSAP;
0 0.2 0.4 0.6 0.8 1.0
• wopl = b/200 for DNV/PULS,
σxu/ σY where wopl = the plate initial deflection ampli-
tude corresponding to the buckling mode,
Figure 8. Ultimate strength interaction relationship
between biaxial compressive loads for the plate with
β p (
σYp E . )
tp = 16 mm. For the initial distortions of the stiffeners,
the following condition is applied for all of the
1.0
methods.
a ×bb==2550×
a× 2550 ×850(mm)
850 (mm)
tp= 22
22mm
mm • Column-type initial distortion, woc = 0.0015a
0.8
• Sideways initial distortion, wos = 0.0015a
Three types of stiffeners, namely, flat-bar, angle-
0.6
bar, and T-bar stiffeners, are considered. The four
σyu/ σY

stiffener sizes shown in Table 1 are considered for

0.4 each of the stiffener types. The size of transverse
frames and longitudinal girders is not addressed
herein, but it is considered to be large enough so that
0.2 FEM
ALPS/ULSAP
neither lateral deformation nor failure occurs before
DNV/PULS the stiffened panel reaches the ultimate strength.
0.0 Figure 11 represents the ANSYS nonlinear finite
0 0.2 0.4 0.6 0.8 1.0 element method modeling in terms of the analysis
σxu/ σY extent and mesh size. A ½ + 1 + ½ span model in
the longitudinal (x) direction and ½ + 1 + ½ bay
Figure 9. Ultimate strength interaction relationship model in the transverse (y) direction are applied.
between biaxial compressive loads for the plate with A finer mesh is applied as the stiffener web height
tp = 22 mm. increases, based on the results of a convergence
1.2
study. Table 2 indicates the boundary conditions
a × b = 2550 × 850 (mm) applied for the ANSYS nonlinear finite element
tp = 33 mm method analysis of the stiffened panel.
1.0
The present benchmark study was undertaken
0.8
in association with the activities of ISSC (Inter-
national Ship and Offshore Structures Congress)
σyu/ σY

0.6

Table 1. Dimensions of the stiffeners considered

0.4
(panel C).
FEM
0.2 ALPS/ULSAP Flat-bar Angle-bar
DNV/PULS
(mm) (mm) T-bar (mm)
0.0 Size hw × tw hw × bf × tw/tf hw × bf × tw/tf
0 0.2 0.4 0.6 0.8 1.0 1.2
σxu/ σY Size 1 150 × 17 138 × 90 × 9/12 138 × 90 × 9/12
Size 2 250 × 25 235 × 90 × 10/15 235 × 90 × 10/15
Figure 10. Ultimate strength interaction relationship Size 3 350 × 35 383 × 100 × 12/17 383 × 100 × 12/17
between biaxial compressive loads for the plate with Size 4 550 × 35 580 × 150 × 15/20 580 × 150 × 15/20
tp = 33 mm.

174

MARSTRUCT.indb 174 2/18/2011 5:44:42 PM

 D y
D'
Tran
s. fra of the analysis. Six elements in the plate breadth
mes
C D''
between the stiffeners are assigned, and three ele-
er
s D''' ments in the stiffener web height.
rd
gi One element is used for the angle-type stiffener
gi.
n
Lo
C' flange, and two for the T-type, in all of the non-
B
linear finite element method analyses. No elements
z
are assigned for the longitudinal girders or trans-
A
A'
B' verse frames, although lateral deflections along
A''
them are not allowed.
A''' x

3.3.1 Stiffened panels under longitudinal

Size 1 Size 2 Size 3 Size 4
compression
Flat type
Figures 12 to 14 show the normalized panel ulti-
(hw× tw) mate strength under longitudinal compression as a
150× 17(mm)
13800 elements
250× 25(mm)
13800 elements
350× 35(mm)
25000 elements
550× 35(mm)
41000 elements function of the slenderness coefficient β of the plate
between the longitudinal stiffeners by comparison
Angle type
(hw× bf× tw /tf)
with the nonlinear finite element analysis (FEA),
138× 90× 9/12(mm)
15400 elements
235× 90× 10/15(mm) 383× 100× 12/17(mm)
15400 elements 26600 elements
580× 150× 15/20(mm)
42600 elements
ALPS/ULSAP, and DNV/PULS for the flat-bar,
angle-bar, and T-bar stiffeners, respectively.
Tee type The figures show, and in line with expectations,
(hw× bf× tw /tf)
138× 90× 9/12(mm) 235× 90× 10/15(mm) 383× 100× 12/17(mm) 580× 150× 15/20(mm) that the normalized panel ultimate strength char-
15400 elements 15400 elements 26600 elements 42600 elements
acteristics are significantly dependent on the panel
geometry, among other factors. It is interesting to
Figure 11. ANSYS nonlinear finite element method note that the maximum value of the normalized
modeling in terms of the analysis extent and mesh size.
panel ultimate strength appears at a certain plate
slenderness coefficient with size 2 stiffeners.
Table 2. Boundary conditions applied for the ANSYS Figure 15 shows the variation in panel ultimate
nonlinear finite element method analysis. strength as a function of the column slenderness
Boundary Description ratio for a plate-stiffener combination with flat-,
angle- and T-bar stiffeners. This figure shows that
A-A′″ and D-D′″ Symmetric condition with the panel ultimate strength clearly decreases as the
Rx = Rz = 0 and uniform column slenderness ratio increases in the range of
displacement in the y direction moderate and small stiffeners representing rela-
(Uy = uniform), coupled the tively large column slenderness ratio. For stiffened
plate part panels with small column slenderness ratios, how-
A-D and A′″-D′″ Symmetric condition with ever, no clear relationship between the panel ulti-
Ry = Rz = 0 and uniform mate strength and the column slenderness ratio
displacement in the x direction
(Ux = uniform), coupled with
does exist.
longitudinal stiffeners
A′-D′, A″-D″, Uz = 0 1.2
Panel C: hw × tw = 150 × 17 (mm) (F)
B-B′ and C-C′
FEA (ANSYS)
FEA (ABAQUS)
1.0
FEA (MSC/MARC)
Design Formula (ALPS/ULSAP)
Technical Committee TTT.1 Ultimate strength, Design Formula (DNV/PULS)
0.8
and some different FE codes with different FE
σxu/ σYeq

modeling techniques were considered. For the

MSC/MARC nonlinear finite element method 0.6

analysis, a ½ + 1 + ½ span model in the longitudi-

nal (x) direction and 1 bay model in the transverse 0.4
(y) direction are taken as the extent of the analy-
III
sis. Ten plate-shell elements in the plate breadth 0.2 Mode III
III III
III III

between the stiffeners are assigned, and six plate-

shell elements are used in the stiffener web height 0.0
direction regardless of the stiffener web height. 0 1 2 3 4
For the Abaqus nonlinear finite element (b / t p ) σYp / E
method analysis, a ½ + 1 + 1 + ½ span model in the
longitudinal (x) direction and 1 bay model in the Figure 12(a). Ultimate strength of the panels under
transverse (y) direction are taken as the extent longitudinal compression for flat-bar stiffeners (size 1).

175

MARSTRUCT.indb 175 2/18/2011 5:44:45 PM

 1.2 1.2
Panel C: hw × tw = 250 × 25 (mm) (F) Panel C: hw × bf × tw/tf = 138 × 90 × 9/12 (mm) (A)
FEA (ANSYS)
FEA (ANSYS)
FEA (ABAQUS)
1.0 FEA (MSC/MARC) 1.0
FEA (MSC/MARC)
Design Formula (ALPS/ULSAP)
Design Formula (ALPS/ULSAP)
Design Formula (DNV/PULS)
Design Formula (DNV/PULS)
0.8 0.8

σxu/ σYeq
σxu/ σYeq

III IV IV
0.6 III 0.6
III
Mode III
III III
0.4 0.4
III
III
Mode III III
0.2 0.2

0.0 0.0
0 1 2 3 4 0 1 2 3 4
(b / t p ) σYp / E (b / t p ) σYp / E

Figure 12(b). Ultimate strength of the panels under Figure 13(a). Ultimate strength of the panels under lon-
longitudinal compression for flat-bar stiffeners (size 2). gitudinal compression for angle-bar stiffeners (size 1).

1.2
1.2 Panel C: hw× bf× tw/tf = 235 × 90 × 10/15 (mm) (A)
Panel C: hw × tw = 350 × 35 (mm) (F)
FEA (ANSYS)
1.0 FEA (MSC/MARC)
1.0 Design Formula (ALPS/ULSAP)
Design Formula (DNV/PULS)
0.8
0.8
σxu/ σYeq

III IV
σxu/ σYeq

Mode III III

IV V V
II
II
0.6
0.6 III V
Mode III
0.4
0.4

FEA (ANSYS) 0.2

0.2 FEA (MSC/MARC)
Design Formula (ALPS/ULSAP)
Design Formula (DNV/PULS) 0.0
0.0 0 1 2 3 4
0 1 2 3 4
(b / t p ) σYp / E
(b / t p ) σYp / E

Figure 13(b). Ultimate strength of the panels under lon-

Figure 12(c). Ultimate strength of the panels under gitudinal compression for angle-bar stiffeners (size 2).
longitudinal compression for flat-bar stiffeners (size 3).

1.2 1.2
Panel C: hw × tw = 550 × 35 (mm) (F) Panel C: hw × bf × tw/tf = 383 × 100 × 12/17 (mm) (A)

1.0 1.0

Mode III IV
0.8 0.8 III
σxu/ σYeq

σxu/ σYeq

IV Mode III
V
IV IV V
II V V
0.6 0.6

0.4 0.4
FEA (ANSYS)
FEA (ABAQUS) FEA (ANSYS)
0.2 0.2 FEA (MSC/MARC)
FEA (MSC/MARC)
Design Formula (ALPS/ULSAP) Design Formula (ALPS/ULSAP)
Design Formula (DNV/PULS) Design Formula (DNV/PULS)
0.0 0.0
0 1 2 3 4 0 1 2 3 4
(b / t p ) σYp / E (b / t p ) σYp / E

Figure 12(d). Ultimate strength of the panels under Figure 13(c). Ultimate strength of the panels under lon-
longitudinal compression for flat-bar stiffeners (size 4). gitudinal compression for angle-bar stiffeners (size 3).

176

MARSTRUCT.indb 176 2/18/2011 5:44:47 PM

 1.2 1.2
Panel C: hw × bf × tw/tf = 580 × 150 × 15/20 (mm)(A) Panel C: hw × bf × tw/tf = 383 × 100 × 12/17 (mm) (T)

1.0 1.0

Mode III V
0.8 0.8
V V

σxu/ σYeq
σxu/ σYeq

Mode III
II V
II V
II V V
0.6 0.6

0.4 0.4
FEA (ANSYS)
FEA (ABAQUS) FEA (ANSYS)
0.2 0.2 FEA (MSC/MARC)
FEA (MSC/MARC)
Design Formula (ALPS/ULSAP) Design Formula (ALPS/ULSAP)
Design Formula (DNV/PULS) Design Formula (DNV/PULS)
0.0 0.0
0 1 2 3 4 0 1 2 3 4
( b / t p ) σYp / E ( b / t p ) σYp / E

Figure 13(d). Ultimate strength of the panels under lon- Figure 14(c). Ultimate strength of the panels under
gitudinal compression for angle-bar stiffeners (size 4). longitudinal compression for T-bar stiffeners (size 3).

1.2 1.2
Panel C: hw× bf× tw/tf = 138 × 90 × 9/12 (mm) (T) Panel C: hw × bf × tw/tf = 580 × 150 × 15/20 (mm) (T)
FEA (ANSYS)
1.0 FEA (ABAQUS) 1.0
FEA (MSC/MARC)
Design Formula (ALPS/ULSAP)
Mode V V
Design Formula (DNV/PULS) 0.8
0.8 V
σxu/ σYeq
σxu/ σYeq

II
II
II
0.6 0.6

III 0.4
0.4
III
III FEA (ANSYS)
III
Mode III III FEA (ABAQUS)
0.2 0.2
FEA (MSC/MARC)
Design Formula (ALPS/ULSAP)
Design Formula (DNV/PULS)
0.0 0.0
0 1 2 3 4 0 1 2 3 4

(b / t p ) σYp / E ( b / tp ) σYp / E

Figure 14(a). Ultimate strength of the panels under Figure 14(d). Ultimate strength of the panels under
longitudinal compression for T-bar stiffeners (size 1). longitudinal compression for T-bar stiffeners (size 4).

1.2 Flat bar under longitudinal compressive loads

1.2 FEA (ANSYS)
Panel C: hw× bf× tw/tf = 235 × 90 × 10/15 (mm) (T)
Design formula (ALPS/ULSAP)
FEA (ANSYS) 1.0
1.0 FEA (MSC/MARC)
Design Formula (ALPS/ULSAP)
Design Formula (DNV/PULS) 0.8
σxu/ σYeq

0.8
σxu/ σYeq

III
V
0.6
V V
0.6
III
Mode III 0.4
0.4
0.2
0.2
0.0
0.0 0.0 0.5 1.0 1.5 2.0 2.5
0 1 2 3 4 (a / π r ) σYeq / E
( b / t p ) σYp / E
Figure 15(a). Variation in normalized panel ultimate
Figure 14(b). Ultimate strength of the panels under strength under longitudinal compression as a function of
longitudinal compression for T-bar stiffeners (size 2). the column slenderness ratio with flat-bar stiffeners.

177

MARSTRUCT.indb 177 2/18/2011 5:44:54 PM

 1.2 Angle bar under longitudinal compressive loads
1.2
Panel C: hw × tw = 150 × 17 (mm) (F)
FEA (ANSYS) FEA (ANSYS)
Design formula (ALPS/ULSAP) Design Formula (ALPS/ULSAP)
1.0 1.0

0.8 0.8
σxu/ σYeq

Mode I

σyu/ σYeq
0.6 0.6

III
0.4 0.4 III
III
III III
0.2 0.2

0.0 0.0
0.0 0.5 1.0 1.5 2.0 2.5 0 1 2 3 4
(a / π r ) σYeq / E ( b / t p ) σYp / E

Figure 15(b). Variation in normalized panel ultimate Figure 16(a). Ultimate strength of the panels under
strength under longitudinal compression as a function of transverse compression for flat-bar stiffeners (size 1).
the column slenderness ratio with angle-bar stiffeners.

1.2
1.2 Panel C: h w× tw = 250× 25 (mm) (F)
Tee bar under longitudinal compressive loads
FEA (ANSYS)
FEA (ANSYS) Design Formula (ALPS/ULSAP)
Design formula (ALPS/ULSAP) 1.0
1.0

0.8
0.8
σyu/ σYeq

Mode III
σxu/ σYeq

0.6
0.6
III
0.4
0.4 III
III
IV IV
0.2 0.2

0.0 0.0
0.0 0.5 1.0 1.5 2.0 2.5 0 1 2 3 4

(a / π r ) σYeq / E ( b / t p ) σYp / E

Figure 15(c). Variation in normalized panel ultimate Figure 16(b). Ultimate strength of the panels under
strength under longitudinal compression as a function of transverse compression for flat-bar stiffeners (size 2).
the column slenderness ratio with T-bar stiffeners.
1.2
Panel C: h w× tw = 350× 35 (mm) (F)
Through the wide range of panel dimensions, the FEA (ANSYS)
Design Formula (ALPS/ULSAP)
1.0
comparisons show that the ALPS/ULSAP method
solutions are in very good agreement with those
of the nonlinear FEA, though slightly on the con- 0.8
σyu/ σYeq

Mode III
servative side, whereas the ALPS/ULSAP method
tends to underestimate the ultimate strength of the 0.6
panels with large flat-bar stiffeners, as shown in III
Figs. 12(c) and 12(d). 0.4 IV
IV
IV IV
3.3.2 Stiffened panels under transverse 0.2
compression
Figures 16 to 18 show the normalized panel ultimate 0.0
strength under transverse compression as a function 0 1 2 3 4
of the slenderness coefficient β of the plate between (b / t p ) σYp / E
the longitudinal stiffeners by comparison with
nonlinear FEA and ALPS/ULSAP for flat-bar, Figure 16(c). Ultimate strength of the panels under
angle-bar, and T-bar stiffeners, respectively. transverse compression for flat-bar stiffeners (size 3).

178

MARSTRUCT.indb 178 2/18/2011 5:45:02 PM

 1.2 1.2
Panel C: hw× tw = 550× 35 (mm) (F) Panel C: hw × bf × tw/tf = 383 × 100 × 12/17 (mm) (A)
FEA (ANSYS) FEA (ANSYS)
Design Formula (ALPS/ULSAP) Design Formula (ALPS/ULSAP)
1.0 1.0

0.8 0.8

σyu/ σYeq
σyu/ σYeq

Mode III Mode III

0.6 0.6

IV V
0.4 IV
0.4
V
IV IV
IV IV IV IV
0.2 0.2

0.0 0.0
0 1 2 3 4 0 1 2 3 4
(b / t p ) σYp / E ( b / t p ) σYp / E

Figure 16(d). Ultimate strength of the panels under Figure 17(c). Ultimate strength of the panels under
transverse compression for flat-bar stiffeners (size 4). transverse compression for angle-bar stiffeners (size 3).

1.2 1.2
Panel C: hw × bf × tw/tf = 138 × 90 × 9/12 (mm) (A) Panel C: hw × bf × tw/tf = 580 × 150 × 12/17 (mm) (A)
FEA (ANSYS) FEA (ANSYS)
Design Formula (ALPS/ULSAP) Design Formula (ALPS/ULSAP)
1.0 1.0

0.8 0.8
Mode V
σyu/ σYeq
σyu/ σYeq

Mode III

0.6 0.6

IV
III
0.4 0.4 IV
III
IV
III IV IV
III III
0.2 0.2

0.0 0.0
0 1 2 3 4 0 1 2 3 4

(b / t p ) σYp / E ( b / t p ) σYp / E

Figure 17(a). Ultimate strength of the panels under Figure 17(d). Ultimate strength of the panels under
transverse compression for angle-bar stiffeners (size 1). transverse compression for angle-bar stiffeners (size 4).

1.2 1.2
Panel C: hw× bf× tw/tf = 235× 90× 10/15 (mm) (A) Panel C: hw× bf× tw/tf = 138× 90× 9/12 (mm) (T)
FEA (ANSYS) FEA (ANSYS)
Design Formula (ALPS/ULSAP) Design Formula (ALPS/ULSAP)
1.0 1.0

0.8 0.8
σyu/ σYeq
σyu/ σYeq

Mode III Mode III

0.6 0.6

III III
0.4 III
0.4
III
V III
V V III III
0.2 0.2

0.0 0.0
0 1 2 3 4 0 1 2 3 4
( b / t p ) σYp / E ( b / t p ) σYp / E

Figure 17(b). Ultimate strength of the panels under Figure 18(a). Ultimate strength of the panels under
transverse compression for angle-bar stiffeners (size 2). transverse compression for T-bar stiffeners (size 1).

179

MARSTRUCT.indb 179 2/18/2011 5:45:07 PM

 1.2
Panel C: hw× bf× tw/tf = 235× 90× 10/15 (mm) (T)
Figure 19 shows the variation in normalized
FEA (ANSYS) panel ultimate strength under transverse com-
Design Formula (ALPS/ULSAP)
1.0 pression as a function of the column slenderness
ratio of the longitudinal stiffeners with attached
0.8 plating. The panel ultimate transverse compres-
σyu/ σYeq

Mode III
sive strength tends to remain unchanged for the
0.6 column slenderness ratio of the longitudinal stiff-
III
eners as long as the plate is identical. It is interest-
0.4 ing to note that a clear relationship between the
III
V
V
panel ultimate transverse compressive strength
V
0.2 and the column slenderness ratio of the longitu-
dinal stiffeners exists, representing that the panel
0.0
ultimate transverse compressive strength increases
0 1 2 3 4 as the column slenderness ratio of the longitudinal
( b / t p ) σYp / E stiffeners with attached plating increases when the

Figure 18(b). Ultimate strength of the panels under

transverse compression for T-bar stiffeners (size 2). 1.2 Flat bar under transverse compressive loads
FEA (ANSYS)
Design formula (ALPS/ULSAP)
1.2 1.0
Panel C: hw × bf × tw/tf = 383 × 100 × 12/17 (mm) (T)
FEA (ANSYS) Size 4 Size 3 Size 2
Design Formula (ALPS/ULSAP) 0.8
1.0 Size 1
σxu/ σYeq

0.6
0.8
σyu/ σYeq

Mode III

0.4
0.6

V
0.2
0.4
V
IV
IV IV 0.0
0.2 0.0 0.5 1.0 1.5 2.0 2.5

( a / π r ) σYeq / E
0.0
0 1 2 3 4
(b / t p ) σYp / E Figure 19(a). Variation in normalized panel ultimate
strength under transverse compression as a function of
the column slenderness ratio with flat-bar stiffeners.
Figure 18(c). Ultimate strength of the panels under
transverse compression for T-bar stiffeners (size 3).

1.2 Angle bar under transverse compressive loads

1.2 FEA (ANSYS)
Panel C: hw × bf × tw/tf = 580 × 150 × 15/20 (mm) (T)
Design formula (ALPS/ULSAP)
FEA (ANSYS) 1.0
Design Formula (ALPS/ULSAP)
1.0 Size 4 Size 3
Size 2 Size 1
0.8
σxu/ σYeq

0.8
Mode V
σyu/ σYeq

0.6
0.6

IV 0.4
0.4 IV
IV
IV IV
0.2
0.2
0.0
0.0 0.0 0.5 1.0 1.5 2.0 2.5
0 1 2 3 4 (a / π r ) σYeq / E
( b / t p ) σYp / E
Figure 19(b). Variation in normalized panel ultimate
Figure 18(d). Ultimate strength of the panels under strength under transverse compression as a function of
transverse compression for T-bar stiffeners (size 4). the column slenderness ratio with angle-bar stiffeners.

180

MARSTRUCT.indb 180 2/18/2011 5:45:13 PM

 1.2 Tee bar under transverse compressive loads
1.0
Panel C: tp=18.5 mm
FEA (ANSYS)
hw× tw = 150× 17 (mm) (F)
Design formula (ALPS/ULSAP)
1.0 FEA (ANSYS)
0.8 Design Formula (ALPS/ULSAP)
Size 4 Size 3
Size 2 Size 1
0.8
σxu/ σYeq

σyu/ σYeq
0.6
0.6

0.4
0.4 Mode III

0.2 0.2

0.0 0.0
0.0 0.5 1.0 1.5 2.0 2.5
0.0 0.2 0.4 0.6 0.8 1.0
( a / π r ) σYeq / E σxu/ σYeq

Figure 19(c). Variation in normalized panel ultimate Figure 20(a). Ultimate strength interaction relationship
strength under transverse compression as a function of between biaxial compressive loads for tp = 18.5 mm and
the column slenderness ratio with T-bar stiffeners. flat-bar stiffeners (size 1).

1.0
Panel C: tp= 18.5 mm
plate thickness is changed but with identical longi- hw× tw = 250× 25 (mm) (F)
tudinal stiffeners. 0.8
FEA (ANSYS)
Design Formula (ALPS/ULSAP)
The comparisons show that the ALPS/ULSAP
method solutions for the panel ultimate strength
σyu/ σYeq

under transverse compression are in very good 0.6

agreement with those of the nonlinear FEA except
for the panels with a very thick plate and relatively 0.4
Mode III
weak stiffeners. See Figs. 16(a) and 16(b) in which
the ALPS/ULSAP method overestimates the
panel ultimate strength due to the overestimation 0.2
of the load-carrying capacity of the longitudinal
stiffeners. 0.0
0.0 0.2 0.4 0.6 0.8 1.0
3.3.3 Stiffened panels under biaxial compression σxu/ σYeq
Figures 20 to 22 show the panel ultimate strength
interaction relationships between biaxial compres- Figure 20(b). Ultimate strength interaction relation-
sive loads by comparison with nonlinear FEA and ship between biaxial compressive loads for tp = 18.5 mm
ALPS/ULSAP for flat-bar, angle-bar, and T-bar and flat-bar stiffeners (size 2).
stiffeners, respectively.
1.0
Figure 23 shows the panel ultimate strength Panel C: tp= 18.5 mm
interaction relationship between biaxial compres- hw× tw = 350× 35 (mm) (F)
FEA (ANSYS)
sive loads by comparison with nonlinear FEA and 0.8 Design Formula (ALPS/ULSAP)
ALPS/ULSAP for size 4 T-bar stiffeners with dif-
ferent plate thicknesses.
σyu/ σYeq

0.6
The comparisons show that the ALPS/ULSAP
method computations are in very good agreement
0.4
with those of the nonlinear FEA for the wide Mode III
Mode II
range of plate and stiffener dimensions with differ-
ent types of stiffeners. 0.2
Mode IV

3.3.4 Stiffened panels under combined 0.0

longitudinal compression and lateral 0.0 0.2 0.4 0.6 0.8 1.0
pressure loads σxu/ σYeq
The panel ultimate strength under combined lon-
gitudinal compression and lateral pressure loads Figure 20(c). Ultimate strength interaction relationship
is now analyzed, as shown in Fig. 24. Two kinds between biaxial compressive loads for tp = 18.5 mm and
of lateral pressure loading, namely, the plate-sided flat-bar stiffeners (size 3).

181

MARSTRUCT.indb 181 2/18/2011 5:45:18 PM

 1.0 1.0
Panel C: tp= 18.5 mm Panel C: tp= 18.5 mm
hw × bf × tw/tf = 383 × 100 × 12/17 (mm) (A)
hw× tw = 550 ×35 (mm) (F)
FEA (ANSYS) FEA (ANSYS)
0.8 Design Formula (ALPS/ULSAP) 0.8 Design Formula (ALPS/ULSAP)
σyu/ σYeq

σyu/ σYeq
0.6 0.6

0.4 Mode IV
0.4 Mode III
Mode II Mode II

Mode III
0.2 0.2
Mode IV Mode V

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
σxu/ σYeq σxu/ σYeq

Figure 20(d). Ultimate strength interaction relationship Figure 21(c). Ultimate strength interaction relationship
between biaxial compressive loads for tp = 18.5 mm and between biaxial compressive loads for tp = 18.5 mm and
flat-bar stiffeners (size 4). angle-bar stiffeners (size 3).

1.0
1.0 Panel C: tp= 18.5 mm
Panel C: tp= 18.5 mm
hw× bf× tw/tf = 580× 150× 15/20 (mm) (A)
hw× bf× tw/tf = 138× 90× 9/12 (mm) (A) FEA (ANSYS)
FEA (ANSYS) 0.8 Design Formula (ALPS/ULSAP)
0.8 Design Formula (ALPS/ULSAP)
σyu/ σYeq

0.6
σyu/ σYeq

0.6

Mode III Mode III

0.4 0.4 Mode IV
Mode II

0.2 0.2
Mode IV
Mode V

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
σxu/ σYeq σxu/ σYeq
Figure 21(a). Ultimate strength interaction relationship Figure 21(d). Ultimate strength interaction relationship
between biaxial compressive loads for tp = 18.5 mm and between biaxial compressive loads for tp = 18.5 mm and
angle-bar stiffeners (size 1). angle-bar stiffeners (size 4).

1.0 1.0
Panel C: tp= 18.5 mm Panel C: tp= 18.5 mm
hw× bf× tw/tf = 235× 90× 10/15 (mm) (A) hw× bf× tw/tf = 138× 90× 9/12 (mm) (T)
FEA (ANSYS) FEA (ANSYS)
0.8 Design Formula (ALPS/ULSAP) 0.8
FEA (MSC/MARC)
Design Formula (ALPS/ULSAP)
σyu/ σYeq

σyu/ σYeq

0.6 0.6

0.4 Mode III

0.4
Mode III

0.2 0.2

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
σxu/ σYeq σxu/ σYeq

Figure 21(b). Ultimate strength interaction relationship Figure 22(a). Ultimate strength interaction relationship
between biaxial compressive loads for tp = 18.5 mm and between biaxial compressive loads for tp = 18.5 mm and
angle-bar stiffeners (size 2). T-bar stiffeners (size 1).

182

MARSTRUCT.indb 182 2/18/2011 5:45:22 PM

 1.0 1.0
Panel C: tp= 18.5 mm Panel C: tp = 11mm
hw × bf× tw/tf = 235 ×90 ×10/15 (mm) (T) hw × bf × tw/tf = 580 × 150 × 15/20 (mm) (T)
FEA (ANSYS) FEA (ANSYS)
0.8 FEA (MSC/MARC) 0.8 FEA (MSC/MARC)
Design Formula (ALPS/ULSAP) Design Formula (ALPS/ULSAP)
σyu/ σYeq

σyu/ σYeq
0.6 0.6

0.4 Mode III 0.4

Mode IV

0.2 0.2
Mode II

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
σxu/ σYeq σxu/ σYeq

Figure 22(b). Ultimate strength interaction relationship Figure 23(a). Ultimate strength interaction relationship
between biaxial compressive loads for tp = 18.5 mm and between biaxial compressive loads for tp = 11 mm and
T-bar stiffeners (size 2). T-bar stiffeners (size 4).

1.0 1.0
Panel C: tp = 18.5 mm Panel C: tp= 12.5 mm
hw × bf × tw/tf = 383 × 100 × 12/17 (mm) (T)
hw× bf× tw/tf = 580× 150× 15/20 (mm) (T)
FEA (ANSYS) FEA (ANSYS)
0.8 FEA (MSC/MARC) 0.8 FEA (MSC/MARC)
Design Formula (ALPS/ULSAP) Design Formula (ALPS/ULSAP)
σyu/ σYeq

σyu/ σYeq

0.6 0.6

0.4 Mode III

0.4
Mode II Mode IV

0.2 0.2
Mode V Mode II

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
σxu/ σYeq σxu/ σYeq

Figure 22(c). Ultimate strength interaction relationship Figure 23(b). Ultimate strength interaction relationship
between biaxial compressive loads for tp = 18.5 mm and between biaxial compressive loads for tp = 12.5 mm and
T-bar stiffeners (size 3). T-bar stiffeners (size 4).

1.0 1.0
Panel C: tp= 18.5 mm Panel C: tp= 15 mm
hw× bf× tw/tf = 580× 150× 15/20 (mm) (T) hw× bf× tw/tf = 580× 150× 15/20 (mm) (T)
FEA (ANSYS) FEA (ANSYS)
0.8 FEA (MSC/MARC) 0.8 FEA (MSC/MARC)
Design Formula (ALPS/ULSAP) Design Formula (ALPS/ULSAP)
σyu/ σYeq
σyu/ σYeq

0.6 0.6

0.4 Mode IV
0.4
Mode IV Mode II
Mode II

0.2 0.2
Mode V

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
σxu/ σYeq σxu/ σYeq

Figure 22(d). Ultimate strength interaction relationship Figure 23(c). Ultimate strength interaction relationship
between biaxial compressive loads for tp = 18.5 mm and between biaxial compressive loads for tp = 15 mm and
T-bar stiffeners (size 4). T-bar stiffeners (size 4).

183

MARSTRUCT.indb 183 2/18/2011 5:45:27 PM

 1.0
Panel C: tp= 18.5 mm
hw× bf× tw/tf = 580× 150× 15/20 (mm) (T)
FEA (ANSYS)
0.8 FEA (MSC/MARC)
Design Formula (ALPS/ULSAP)
σyu/ σYeq

0.6

0.4 Mode IV
Mode II

0.2
Mode V

Figure 24. A stiffened panel under combined longitudi-

0.0 nal compression and lateral pressure loads.
0.0 0.2 0.4 0.6 0.8 1.0
σxu/ σYeq

Figure 23(d). Ultimate strength interaction relation-

ship between biaxial compressive loads for tp = 18.5 mm
and T-bar stiffeners (size 4).

1.0
Panel C: tp= 25 mm
hw× bf× tw/tf = 580× 150× 15/20 (mm) (T)
FEA (ANSYS)
0.8 FEA (MSC/MARC)
Design Formula (ALPS/ULSAP)
σyu/ σYeq

0.6
Mode III
Figure 25(a). Shape of the initial deflection before lateral
pressure loading (with an amplification factor of 20).
0.4 Mode II

0.2
Mode V

0.0
0.0 0.2 0.4 0.6 0.8 1.0
σxu/ σYeq

Figure 23(e). Ultimate strength interaction relationship

between biaxial compressive loads for tp = 25 mm and
T-bar stiffeners (size 4).

1.2 Figure 25(b). Shape of the initial deflection after lateral

Panel C: tp= 37 mm
pressure loading in the plate-sided pressure loading (with
hw× bf× tw/tf = 580× 150× 15/20 (mm) (T)
1.0 FEA (ANSYS)
an amplification factor of 20).
FEA (MSC/MARC)
Design Formula (ALPS/ULSAP)
0.8
σyu/ σYeq

Mode V
0.6

0.4

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2
σxu/ σYeq

Figure 23(f). Ultimate strength interaction relationship Figure 25(c). Shape of the initial deflection after lateral
between biaxial compressive loads for tp = 37 mm and pressure loading in the stiffener-sided pressure loading
T-bar stiffeners (size 4). (with an amplification factor of 20).

184

MARSTRUCT.indb 184 2/18/2011 5:45:31 PM

 pressure and the stiffener-sided pressure, are 1.0
Panel C: h w× bf× tw/tf = 383× 100× 12/17 (mm) (T)
considered. Design Formula (ALPS/ULSAP)
In the FEA employed in this study, the lateral FEA (ANSYS) with Plate-sided Pressure
FEA (ANSYS) with Stiffener-sided Pressure
pressure loads are applied first until the target 0.8
magnitude is achieved, and then the longitudinal Mode V Mode III
compressive loads are increased. This type of load-

σxu/ σYeq
ing order changes the shape of the initial deflec- 0.6
tions before longitudinal compression is applied,
as shown in Fig. 25. Plate-sided Pressure
Figure 26 shows the deformed shapes of the 0.4
panel with T-bar stiffeners (size 3) and tp = 15 mm
at the ultimate limit state under combined longi-
tudinal compression and lateral pressure loads, 0.2
where p = 0.25 MPa. Stiffener-sided Pressure

Figures 27 and 28 show the panel ultimate tp=18.5 mm

0.0
strength interaction relationships between lon-
0 0.1 0.2 0.3 0.4 0.5
p (MPa)

Figure 28. Ultimate strength interaction relationship

between longitudinal compression and lateral pressure
loads for the panel with T-bar stiffeners (size 3) and
tp = 18.5 mm.

gitudinal compression and lateral pressure loads

Figure 26. Deformed shape of the panel under com-
bined longitudinal compression and lateral pressure
loads for T-bar stiffeners (size 3) and P = 0.25 MPa at the
ultimate limit states (with an amplification factor of 10).
1.0
Panel C: hw× bf× tw/tf = 383× 100× 12/17 (mm) (T)
Design Formula (ALPS/ULSAP)
FEA (ANSYS) with Plate-sided Pressure
FEA (ANSYS) with Stiffener-sided Pressure
0.8 Mode V
Mode II Mode V
with T-bar stiffeners (size 3), by comparison with
nonlinear FEA and ALPS/ULSAP, for tp = 15 mm
and tp = 18.5 mm, respectively. The nonlinear FEA
results for both the plate- and stiffener-sided pres-
sure loading cases are compared.
It is observed that the panel collapse modes dif-
fer depending on both the loading ratio and the
pressure loading direction as well as the panel
dimensions. The comparisons show that the ALPS/
ULSAP method solutions are in very good agree-
ment with the nonlinear finite element method
computations.
4 CONCLUDING REMARKS

The objective of the benchmark study reported in

σxu/ σYeq

0.6
this paper was to check the accuracy of the ALPS/
ULSAP method’s use to determine plate and stiff-
0.4 ened panel ultimate strength, compared with non-
Plate-sided Pressure
linear FEA and the DNV/PULS method.
Mode III
The dimensions and material properties of
0.2 a real ship panel were selected as the standard
Stiffener-sided Pressure panel for testing purposes, and a wider range of
0.0
tp=15 mm plating and stiffener dimensions was considered
0 0.1 0.2 0.3 0.4 by varying the panel’s properties. Three types
p (MPa) of stiffeners, namely, flat-bar, angle-bar, and T-bar
stiffeners, were considered, and different loading
Figure 27. Ultimate strength interaction relationship conditions, including longitudinal compression,
between longitudinal compression and lateral pressure transverse compression, biaxial compression, and
loads for the panel with T-bar stiffeners (size 3) and combined longitudinal compression and lateral
tp = 15 mm. pressure loads, were applied.

185

MARSTRUCT.indb 185 2/18/2011 5:45:39 PM

 The ALPS/ULSAP method solutions were
found to be in very good agreement with the
nonlinear finite element method computations
through a wide range of panel dimensions and
different loading conditions. Because the ALPS/
ULSAP method is based on design formulations,
the computational time required is extremely short
compared to the nonlinear finite element method.
This will be of great advantage in the structural
design and safety assessment of ship structures
comprising a large number of plate panels.
ACKNOWLEDGEMENTS
The present study was undertaken at The Lloyd’s
Register Educational Trust (The LRET) Research
Centre of Excellence at Pusan National Univer-
sity, Korea. This benchmark study was performed
in association with the activities of International
Ship and Offshore Structures Congress (ISSC)
Technical Committee III.1 Ultimate Strength. The
authors are pleased to acknowledge the financial
support of both The LRET and The National
Research Foundation of Korea.
REFERENCES
Abaqus. 2010. SIMULIA, Providence, RI, USA
(www.simulia.com).
ALPS/ULSAP. 2010. A computer program for ultimate
limit state assessment of plate panels. Advanced Tech-
nology Center, DRS C3 Systems, Inc., MD, USA
(www.proteusengineering.com, www.maestromarine.
com).
186
MARSTRUCT.indb 186
ANSYS. 2010. Version 12.0, ANSYS Inc., Canonsburg,
PA, USA (www.ansys.com).
DNV/PULS. 2009. CSA—Direct analysis of ship struc-
tures. Classification Notes, No. 34.1, Det Norske Ver-
itas, April.
Hughes, O.F. and Paik, J.K. 2010. Ship structural analy-
sis and design, The Society of Naval Architects and
Marine Engineers, New Jersey, USA.
ISO. 2007. International standard ISO 18072-1, Ships and
marine technology—Ship structures, Part 1: General
requirements for their limit state assessment, Interna-
tional Organization for Standardization, Geneva.
MSC/MARC. 2010. MSC Software Corporation, Santa
Ana, CA, USA (www.mscsoftware.com).
Paik, J.K. and Melchers, R.E. 2008. Condition assessment
of aged structures, CRC Press, New York, USA.
Paik, J.K. and Thayamballi, A.K. 2003. Ultimate limit
state design of steel-plated structures. John Wiley &
Sons, Chichester, UK.
Paik, J.K. and Thayamballi, A.K. 2007. Ship-shaped off-
shore installations: Design, building, and operation.
Cambridge University Press, Cambridge, UK.
2/18/2011 5:45:40 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Modified Paik-Mansour formula for ultimate strength

calculations of ship hulls

J.K. Paik, D.K. Kim, D.H. Park & H.B. Kim

The Lloyd’s Register Educational Trust Research Centre of Excellence, Pusan National University,
Busan, Korea

A.E. Mansour
University of California, Berkeley, USA

J.B. Caldwell
Emeritus Professor, Newcastle University, Newcastle upon Tyne, UK

ABSTRACT: The objective of this paper is to develop a modified Paik-Mansour formula for the
ultimate strength calculations of ship hulls subject to vertical bending moments. The method is based on
a credible bending stress distribution over the hull cross-section presumed at the ultimate limit state. The
accuracy of this method is demonstrated through comparison with computations obtained using more
refined methods, such as nonlinear finite element method, intelligent super-size finite element method,
and idealized structural unit method. Statistical analysis of the hull girder ultimate strength based on
comparisons among the various computations is carried out in terms of their mean values and coefficients
of variation. The original Paik-Mansour method is found to be inapplicable to the case of a pure vertical
bending moment depending on the ship’s hull type and/or vertical bending direction, but the modified
Paik-Mansour method is more general and is able to resolve this issue.

1 INTRODUCTION loads applied. The most typical consequence of

It is now well recognized that ultimate strength
is a much better basis for the design and strength
assessment of ship structures than allowable work-
ing stress (Paik & Thayamballi 2003, 2007; ISO
2007; Mansour & Liu 2008; Hughes & Paik 2010).
The same is also true for the condition assessment
of aged structures (Paik & Melchers 2008).
As applied hull girder loads increase, the most
highly stressed structural components of the ship’s
hull buckle in compression or yield in tension.
A ship can withstand further hull girder loading
even after the buckling or yielding of a few struc-
tural components. However, local failures result
in a decrease in the structural effectiveness of the
hull, and hence the overall hull structure eventually
reaches the ultimate limit state as the hull’s redun-
dancy becomes exhausted due to the progressive
structural failures that occur when hull girder
loads are applied.
The collapse of a ship’s hull constitutes the most
catastrophic failure event because it almost always
entails the complete loss of the ship. Such collapse
can occur when the hull’s maximum load-carrying
capacity (or ultimate hull girder strength) is insuf-
ficient to sustain the corresponding hull girder
hull girder collapse is the breaking of the hull
into two parts due to the action of extreme verti-
cal bending moments that exceed the ultimate hull
girder strength.
The prevention of hull collapse is thus the most
important task in the design and safety assessment
of ship structures, and an accurate and efficient
method of computing ultimate hull girder strength
is a prerequisite for robust ship structural design.
In this paper, the original Paik-Mansour for-
mula method (Paik & Mansour 1995) is modified
to allow more accurate calculation of ultimate hull
girder strength. The accuracy of the method devel-
oped herein is demonstrated through comparison
with more refined method computations.
2 PRESUMED STRESS DISTRIBUTION-
BASED METHOD
2.1 Caldwell’s original formula method
Caldwell (1965) was the pioneer of the presumed
stress distribution-based method of calculating
the ultimate vertical bending moments of a ship’s
hull. As shown in Figure 1, he presumed a bending

187

MARSTRUCT.indb 187 2/18/2011 5:45:40 PM

 Figure 1. Caldwell’s (1965) presumption of the bending
stress distribution at the ultimate limit state under a verti-
cal bending moment for a simplified ship hull cross-section
subject to sagging or hogging (N.A. = neutral axis).
stress distribution over the hull cross-section at
the ultimate limit state under vertical bending
moments, in which all of the materials in com-
pression have reached their ultimate strength with
buckling and all of the materials in tension have
yielded. Caldwell then calculated the ultimate bend-
ing moments by integrating the presumed bending
stresses over the hull cross-section.
Such presumed stress distribution does not,
however, accurately represent the ultimate limit
states of modern ship structures, thereby result-
ing in overestimated calculations of ultimate hull
girder strength.
2.2 Original Paik-Mansour formula method
Experimental studies of large-scale ship hull mod-
els (e.g., Dow 1991) and numerical studies of full-
scale ships (e.g., Rutherford and Caldwell 1990;
Paik et al., 1996) have demonstrated that the over-
all collapse of a ship’s hull under a vertical bending
moment is governed by the collapse of the com-
pressed flange, although some degree of reserve
strength remains after the compressed flange has
collapsed.
This is the case because, after the buckling of
the compressed flange, the neutral axis (N.A.) of
the hull cross-section moves toward the tensioned
flange, and a further increase in the applied bend-
ing moment is sustained until this flange yields. At
later stages of this process, the vertical structures
around the compressed and tensioned flanges (e.g.,
the longitudinal bulkheads or side shell structures)
may also fail.
In the vicinity of the final N.A., however, the
vertical structures usually remain in a linear elastic
state until the overall collapse of the hull girder.
Depending on the geometrical and material prop-
erties of the hull’s cross-section, these parts may
of course fail, which corresponds with Caldwell’s
(1965) presumption.
Figure 2 shows an example of typical bend-
ing stresses across the hull cross-section of a
188
MARSTRUCT.indb 188
Figure 2. Example of typical bending stress distribution
across the cross-section of a ship’s hull at the ultimate limit
state under a hogging bending moment (+: tension: −:
compression), obtained through numerical investigations
(Paik et al., 1996).
single-hulled oil tanker at the ultimate limit state
under a vertical hogging bending moment, as
obtained through numerical investigations (Paik
et al., 1996). It is evident from this figure that
the compressed flange (the bottom panel) col-
lapses, and the tensioned flange (the deck panel)
yields, until the ultimate strength has been reached,
whereas the vertical structures in the vicinity of
the N.A. remain intact (linear elastic). Hence, the
approach based on Caldwell’s presumed bend-
ing stress distribution can result in the strength
of a ship’s hull against collapse being greatly
overestimated.
Paik and Mansour (1995) subsequently suggested
the bending stress distribution over the hull cross-
section at the ultimate limit state that is shown in
Figure 3. In the sagging condition, regions 1 and 2
are under tension, and regions 3 and 4 are under
compression. Region 1 represents the outer bottom
panels, which have yielded to reach yield stress σ x ,
Y
and region 4 the upper deck panels and upper part
of the vertical structures, which have buckled and
collapsed to reach ultimate stress σUx . Regions
2 and 3, however, remain in a linear elastic or
unfailed state, reaching an elastic stress of σ xE .
In the hogging condition, regions 1 and 2 are
under compression, and regions 3 and 4 are under
tension. Region 1, which represents the outer
bottom panels and the lower part of the vertical
structures, has buckled and collapsed to reach ulti-
mate stress σUx , and region 4, which represents the
upper deck panels, has yielded to reach yield stress
σYx . Regions 2 and 3 remain in the linear elastic
regime, reaching elastic stress σ Ex .
The height of region 4 (the upper part of the ver-
tical structures) in the sagging condition, or that of
region 1 (the lower part of the vertical structures)
2/18/2011 5:45:41 PM

z σ xY
σ Ux
+ Tens.
– Comp.
D-gus
σ xE
σ E
+ –
x
+ –
σ xE + Tens.
– Comp.
gus
σ xY
(a) Sagging
Figure 3. Paik and Mansour’s (1995) original pre-
sumption of the bending stress distribution across the
cross-section of a ship’s hull at the ultimate limit state
under sagging or hogging conditions (+: tension; −: com-
pression) (the superscripts U, Y, E denote the ultimate
strength, yielding, and elastic region, respectively).
in the hogging condition, after buckling and
collapse is assigned on the basis of the geometrical
and material properties of the ship’s hull structure.
Under a vertical bending moment, the summation
of axial forces over the entire cross-section of the
hull becomes zero, as follows.
∫σxdA = 0,
where ∫dA = integration across the entire cross-
section of the hull.
The height of region 4 in the sagging condition
or that of region 1 in the hogging condition can be
defined by solving Equation (1). The distance gu
from the ship’s baseline (reference position) to the
horizontal N.A. of the cross-section of the ship’s
hull at the ultimate limit state can then be obtained
as follows.
n
∑ σ xxi ai zi
i =1
gu = n
, (2)
∑ σ xxi ai
i =1
where zi = the distance from the ship’s baseline
(reference position) to the horizontal N.A. of the
ith structural component; σxi = the longitudinal
stress of the ith structural component following
the presumed stress distribution; ai = the cross-
sectional area of the ith structural component; and
n = the total number of structural components. gu
is denoted by gus in the sagging condition and by guh
in the hogging condition.
The ultimate vertical bending moment Mu is
then calculated as the first moment of the bending
stresses around the N.A. position, as follows.
MARSTRUCT.indb 189
z n
Mu ∑ xi ai ( zi gu ), (3)
i =1
D-guh where n = the total number of structural compo-
nents, and gu is as defined in Equation (2). Mu is
denoted by Mus (negative value) for the sagging
σ xE condition and by Muh (positive value) for the hog-
ging condition.
guh
2.3 Modified Paik-Mansour formula method
σ Ux
The original Paik-Mansour formula method does
(b) Hogging
not allow the expansion of the yielded area to the
vertical members under tensile loads, although it
presumes that the tension flange, i.e., the deck pan-
els in the hogging condition and the outer bottom
panels in the sagging condition, has yielded at the
ultimate limit state of the hull girders subject to
vertical bending moments.
However, depending on the geometric and/or
material properties of the hull cross-sections, the
vertical members close to the tension flange may
also have yielded until the hull girders reach the
ultimate limit state. Therefore, the bending stress
distribution at the ultimate limit state presumed in
the original Paik-Mansour method is now modi-
fied to that shown in Figure 4, where hY is the
height of the yielded area under axial tension and
(1) hC is the height of the collapsed area under axial
compression.
In contrast to the original Paik-Mansour for-
mula method, which involves only one unknown,
i.e., hC, in the bending stress distribution over the
hull cross-section, the modified method involves
two unknowns, i.e., hY and hC. Equation (1) is
insufficient to determine two unknowns, and thus
the following iteration process is required to deter-
mine hY and hC.
Figure 4. Modification of the Paik-Mansour pre-
sumption of the bending stress distribution across the
cross-section of a ship’s hull at the ultimate limit state
under sagging or hogging conditions (+: tension; −: com-
pression) (the superscripts U, Y, E denote the ultimate
strength, yielding, and elastic region, respectively).
189
2/18/2011 5:45:43 PM
 i. Develop the structural model with nodal Figure 6 shows the hull cross-section models
points for the hull cross-section using the developed for various types of ship structures. In
plate-stiffener combination elements and/or the present study, all types of ship hulls are mod-
plate elements. eled as an assembly of plate-stiffener combination
ii. Calculate the ultimate axial compressive models except for tanker hull structures, while
stresses of the individual elements. tanker hull structures under a sagging moment are
iii. Divide the ship’s depth into a number of seg- modeled as an assembly of plate-stiffener separa-
ments (parts). tion models, and those under a hogging moment
iv. Keeping hY (the height of the yielded hull are modeled as an assembly of plate-stiffener com-
part) at a constant value starting from hY = 0, bination models. The details of the ships, including
increase hC (the height of the collapsed hull the geometric and material properties of their hull
part) starting from hC = 0. cross-sections, are presented in Paik (2011).
v. Assign the linear elastic stresses of the individ-
ual elements in regions 2 and 3 linearly between
3.2 Ultimate strength of structural elements
the average values of the ultimate stresses in
the collapsed hull part (region 4 under sag- The ultimate strength of individual structural
ging or region 1 under hogging) and the yield elements under axial tension is considered to be
stresses in the yielded hull part (region 1 under
sagging or region 4 under hogging).
vi. Calculate the total axial forces (positive sign)
in tension and the total axial forces (negative
sign) in compression across the entire hull
cross-section.
vii. Repeat steps (iv) to (vi) varying hC together with
hY until the difference between the numerical
values of these axial forces should be accept-
ably small.

3 STRUCTURAL MODELING OF HULL

CROSS-SECTIONS AND INDIVIDUAL
ELEMENT ULTIMATE STRENGTH

3.1 Structural modeling

The hull cross-sections of a ship may be modeled
Figure 6(a). Structural modeling for Dow’s 1/3-scaled
as an assembly of plate-stiffener combination mod- frigate test hull using plate-stiffener combination
els (beam-column elements with attached plating) models.
and/or plate-stiffener separation models (plate
elements + beam-column elements without attached
plating), as shown in Figure 5.

Figure 5(a). Plate-stiffener combination models (beam-

column elements with attached plating) in a stiffened
plate structure.

Figure 5(b). Plate-stiffener separation models (plate

elements + beam-column elements without attached plat- Figure 6(b). Structural modeling for a container ship
ing) in a stiffened plate structure. hull using plate-stiffener combination models.

190

MARSTRUCT.indb 190 2/18/2011 5:45:44 PM

 Figure 6(c). Structural modeling for a bulk carrier hull
using both plate-stiffener combination models and plate- Figure 6(f). Structural modeling for a single-hull VLCC
stiffener separation models. hull using plate-stiffener combination models.

Figure 6(d). Structural modeling for a Suezmax-class

double-hull tanker ship hull using plate-stiffener combi-
nation models. Figure 6(g). Structural modeling for a single-hull VLCC
hull using plate-stiffener separation models.

Figure 6(e). Structural modeling for a Suezmax-class

double-hull tanker ship hull using plate-stiffener separa- Figure 6(h). Structural modeling for a double-hull
tion models. VLCC hull using plate-stiffener combination models.

191

MARSTRUCT.indb 191 2/18/2011 5:45:45 PM

 I = the moment of inertia with regard to the
corresponding nodal point for the plate-stiffener
combination element; and
t = plate thickness.
The foregoing equation implicitly takes into
account the effect of initial imperfections in terms
of plate initial deflection, column type initial
distortions of the stiffeners, sideways initial dis-
tortions of the stiffeners, and welding residual
stresses.

3.2.2 Rectangular plate elements

Figure 6(i). Structural modeling for a double-hull
VLCC hull using plate-stiffener separation models.
The ultimate compressive strength of plate ele-
ments is determined by the following equation
(Paik et al., 2004).
For plate elements with a/b ≥ 1, the ultimate axial
compressive strength with an average level of initial
imperfections should be determined as follows.

equivalent to the material yield stress, whereas the ⎧ 032β 4 β 2 + 1.00 f β ≤ 1.5
ultimate strength of those under axial compression σ xu ⎪ −0.03
= ⎨1.274 / β β ≤ 3.0 (5)
is assumed to be as follows. σYp ⎪
⎩1.248 / β 0.283 β > 3.0
2

3.2.1 Plate-stiffener combination elements

The ultimate compressive strength of the plate- where,
stiffener combination elements shown in Figure 6
are determined by the following Paik-Thayamballi b σYp
β= ;
empirical formula (Paik & Thayamballi 2003). t E

⎛ σYeq ⎞ (4) σYp = yield stress of plating;

σ xu in σ xu1, 2 ⎟ , b = spacing of the stiffeners or the plate breadth
⎝ λ ⎠ between the stiffeners;
where E = elastic modulus; and
t = plate thickness.
σ xu1 1
= ; For plate elements with a/b < 1, the ultimate axial
σYeq 0.995 + 0.936 λ
2 2 2 2
β + 0.188λ β − 0.067 λ 4 compressive strength with an average level of initial
imperfections should be determined as follows.
a σYeq
λ= ; σ xu a σ *xu 0.475 ⎛ a ⎞
πr E = + ⎜1 − ⎟ , (6)
σYp b σYp α2 ⎝ b⎠
b σYp
β= ; where
t E
⎧
σ*xu ⎪ −0.032α + 0.002α + 1.0 f α ≤ 1.5
4 2
I = ⎨1. /α f 1.5 < α ≤ 3.0 ;
r= ;
As σYp ⎪
⎩1 .24
48 / α 2
0 .283 f α > 3.0
σYeq = the equivalent yield stress of the plate-stiffener
combination element; a σYp
α= .
σYp = the yield stress of the plating; t E
a = the length of the plate-stiffener combination
element;
As = the cross-sectional area of the plate-stiffener 4 APPLIED EXAMPLES
combination element;
b = the spacing of the stiffeners or the plate breadth The ultimate hull girder strengths of the six types of
between the stiffeners; ship indicated in Section 3.1 are now calculated using
E = elastic modulus; the proposed modified Paik-Mansour method, with

192

MARSTRUCT.indb 192 2/18/2011 5:45:47 PM

 the results compared with the strengths obtained with 12

more refined methods. As previously mentioned,

Sagging bending moment (MNm)

10
tanker hull structures under a sagging moment are
modeled as an assembly of plate-stiffener separa- 8

tion models (plate elements + beam-column ele-

6
ments without attached plating, see Figure 5(b)),
while those under a hogging moment and other 4
Test result
ANSYS
types of ship hulls are modeled as an assembly of ALPS/HULL
plate-stiffener combination models. 2 CSR
Modified Paik-Mansour formula method
The ANSYS (2010) nonlinear finite element
0
method, ALPS/HULL (2010) intelligent super- 0.0 0.5 1.0 1.5 2.0 2.5 3.0
size finite element method, and CSR (Common Curvature (1/km)
Structural Rules) (IACS 2006a, 2006b) idealized
structural unit method are employed for this Figure 7(b). Ultimate strength behavior of the Dow’s
comparison. The test results for the Dow’s (1991) frigate test hull under a sagging moment.
frigate test hull model in the sagging condition are
also compared. Table 1. Heights of the collapsed and yielded
An average level of initial imperfections was parts of the Dow’s frigate test hull.
applied for the numerical computations, although
welding residual stresses were not considered. Hogging (mm) Sagging (mm)
Details of this ultimate strength comparison are
presented in Paik (2011). Method hC hY hC hY

Original P-M 210.0 0.0 760.2 0.0

4.1 Dow’s frigate test hull Modified P-M 210.0 0.0 760.2 0.0
Figure 7 shows the results of the ultimate strength
behavior comparison for the Dow’s frigate test hull
subject to a vertical bending moment. 9

Table 1 presents the heights of the collapsed and

Hogging bending moment (GNm)

yielded parts, as per Figure 4, with hY = 0 indicat- 7

ing that only the tension flange, i.e., the deck panel 6
in the hogging condition or the outer bottom panel 5
in the sagging condition, has yielded until the 4
ultimate strength is reached. It can be seen that the 3
original and modified Paik-Mansour methods ANSYS
ALPS/HULL
2
produce identical results for this case. CSR
1 Modified Paik-Mansour formula method

0
4.2 Container ship hull 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
Curvature (1/km)
Figure 8 shows the results of the ultimate strength
behavior comparison for the container ship hull Figure 8(a). Ultimate strength behavior of the con-
tainer ship hull under a hogging moment.
12

9
Hogging bending moment (MNm)

10
Sagging bending moment (GNm)

7
8
6

6 5

4
4
ANSYS 3
ALPS/HULL ANSYS
CSR 2 ALPS/HULL
2 CSR
Modified Paik-Mansour formula method
1 Modified Paik-Mansour formula method
0 0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
Curvature (1/km) Curvature (1/km)

Figure 7(a). Ultimate strength behavior of the Dow’s Figure 8(b). Ultimate strength behavior of the con-
frigate test hull under a hogging moment. tainer ship hull under a sagging moment.

193

MARSTRUCT.indb 193 2/18/2011 5:45:51 PM

 Table 2. Heights of the collapsed and yielded 20

parts of the container ship hull.

Hogging bending moment (GNm)

16
Hogging (mm) Sagging (mm)
12
Method hC hY hC hY
8
Original P-M 698.8 0.0 10330.8 0.0 ANSYS
Modified P-M 698.8 0.0 10330.8 0.0 4
ALPS/HULL
CSR
Modified Paik-Mansour formula method

0
subject to a vertical bending moment. Table 2 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
presents the heights of the collapsed and yielded Curvature (1/km)

parts, as determined by the original and modified

Figure 9(a). Ultimate strength behavior of the bulk car-
Paik-Mansour methods, which produce identical
rier hull under a hogging moment.
results for this case.
16
4.3 Bulk carrier hull

Sagging bending moment (GNm)

Figure 9 shows the results of the ultimate strength 12
behavior comparison for the bulk carrier hull sub-
ject to a vertical bending moment. Table 3 presents
8
the heights of the collapsed and yielded parts, as
determined by the original and modified Paik-
ANSYS
Mansour methods. 4 ALPS/HULL
It can be seen that the pure hogging bending CSR
Modified Paik-Mansour formula method
moment condition cannot be achieved for this case 0
with the original Paik-Mansour method, as it does 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
not permit the expansion of the yielded part except Curvature (1/km)

for the tension flange. However, the modified Paik-

Mansour method is able to achieve this condition
because the tension flange is allowed to expand,
with hY = 13.7 mm.
4.4 Suezmax-class double-hull tanker hull
Figure 9(b). Ultimate strength behavior of the bulk
carrier hull under a sagging moment.
Table 3. Heights of the collapsed and yielded
parts of the bulk carrier hull.

Figure 10 shows the results of the ultimate Hogging (mm) Sagging (mm)
strength behavior comparison for the Suezmax-
class double-hull tanker hull subject to a vertical Method hC hY hC hY
bending moment.
Original P-M – – 17935.0 0.0
Table 4 presents the heights of the collapsed and Modified P-M 1654.1 13.7 17935.0 0.0
yielded parts of the Suezmax double-hull tanker
hull. It can be seen that, in the hogging condition,
the height of the yielded part is larger than that of 16
the collapsed part because the double bottom struc-
Hogging bending moment (GNm)

14
tures are much heavier than the deck structures
in this type of ship. The modified Paik-Mansour 12

method is able to handle this aspect, whereas the 10

original such method is not suitable. 8

6
ANSYS
4.5 Single-hull tanker hull 4 ALPS/HULL
CSR
Figure 11 shows the results of the ultimate strength 2 Modified Paik-Mansour formula method

behavior comparison for the single-hull tanker 0

hull subject to a vertical bending moment. Table 5 0.00 0.05 0.10 0.15
Curvature (1/km)
0.20 0.25 0.30

presents the heights of the collapsed and yielded

parts determined by the original and modified Figure 10(a). Ultimate strength behavior of the
Paik-Mansour methods, which in this case pro- Suezmax-class double-hull tanker hull under a hogging
duce identical results. moment.

194

MARSTRUCT.indb 194 2/18/2011 5:45:53 PM

 14 Table 5. Heights of the collapsed and yielded
parts of the single-hull tanker hull.
Sagging bending moment (GNm)

12

10 Hogging (mm) Sagging (mm)

8
Method hC hY hC hY
6

ANSYS
Original P-M 7035.2 0.0 15225.5 0.0
4
ALPS/HULL Modified P-M 7035.2 0.0 15225.5 0.0
CSR
2 Modified Paik-Mansour formula method

0
0.00 0.05 0.10 0.15 0.20 0.25 0.30
Curvature (1/km) 30

Hogging bending moment (GNm)

Figure 10(b). Ultimate strength behavior of the 25
Suezmax-class double-hull tanker hull under a sagging
moment. 20

Table 4. Heights of the collapsed and yielded 15

parts of the Suezmax-class double-hull tanker
hull. 10
ANSYS
ALPS/HULL
Hogging (mm) Sagging (mm) 5 CSR
Modified Paik-Mansour formula method

Method hC hY hC hY 0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
Original P-M – – 16078.5 0.0 Curvature (1/km)
Modified P-M 12.1 2210.6 16078.5 0.0
Figure 12(a). Ultimate strength behavior of the double-
hull tanker hull under a hogging moment.
21
Hogging bending moment (GNm)

18

15 30
Sagging bending moment (GNm)

12
25
9

ANSYS 20
6
ALPS/HULL
CSR
3 Modified Paik-Mansour formula method 15
0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 10
Curvature (1/km) ANSYS
ALPS/HULL
5 CSR
Figure 11(a). Ultimate strength behavior of the single- Modified Paik-Mansour formula method
hull tanker hull under a hogging moment.
0

20 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40

Curvature (1/km)
Sagging bending moment (GNm)

16
Figure 12(b). Ultimate strength behavior of the double-
hull tanker hull under a sagging moment.
12

ANSYS
4.6 Double-hull tanker hull
ALPS/HULL
4 CSR Figure 12 shows the results of the ultimate strength
Modified Paik-Mansour formula method
behavior comparison for the double-hull tanker
0
hull subject to a vertical bending moment.
0.00 0.05 0.10 0.15 0.20 0.25 0.30
Curvature (1/km) Table 6 presents the heights of the collapsed and
yielded parts of the double-hull tanker hull, which
Figure 11(b). Ultimate strength behavior of the single- has similar characteristics to the Suezmax-class
hull tanker hull under a sagging moment. double-hull tanker hull (see Table 4).

195

MARSTRUCT.indb 195 2/18/2011 5:45:54 PM

 5 STATISTICAL ANALYSIS
The mean values and coefficients of variation
for the modified Paik-Mansour design formula
and the numerical methods are identified for all
six types of ship hull. In addition, the deviation
between the numerical methods is also provided as
a reference.
5.1 Formula method versus ANSYS nonlinear
finite element method
Table 7 presents the mean values and coefficient
of variations for the formula method versus the
ANSYS nonlinear finite element method for all six
types of ship hull. Figure 13 shows the trend of the
deviation in ultimate hull strength between these
two methods.
HULL intelligent super-size finite element method
for all six types of ship hull. Figure 14 shows the
trend of the deviation in ultimate hull strength
between these two methods.
5.3 Formula method versus CSR idealized
structural unit method
Table 9 presents the mean values and coefficients
of variation for the formula method versus the
CSR idealized structural unit method for all six
types of ship hull. Figure 15 shows the trend of the
deviation in ultimate hull strength between these
two methods.
30
25

(Mu)Formula (GNm)
5.2 Formula method versus ALPS/HULL 20
intelligent super-size finite element method
Table 8 presents the mean values and coefficients of 15

variation for the formula method versus the ALPS/

10
Hog / Sag
: Dow’s frigate test hull
Table 6. Heights of the collapsed and yielded :
:
Container ship
Bulk carrier
parts of the double-hull tanker hull. 0.0125
: D/H Suezmax
: S/H VLCC
: D/H VLCC
Hogging (mm) Sagging (mm) 0
0 0.0125 10 15 20 25 30
Method hC hY hC hY (Mu)ANSYS (GNm)

Original P-M – – 20240.7 0.0 Figure 13. Trend of the deviation in ultimate hull
Modified P-M 15.9 3816.0 20240.7 0.0 strength between the formula method and the ANSYS
nonlinear finite element method.

Table 7. Comparison between developed formula method and ANSYS nonlinear finite element method.

Hogging Sagging

Formula ANSYS Formula ANSYS

Mp Muh Muh Formula/ Mus Mus Formula/

Ship (GNm) (GNm) Muh/Mp (GNm) Muh/Mp ANSYS (GNm) Mus/Mp (GNm) Mus/Mp ANSYS

Dow’s test 0.013 0.010 0.772 0.011 0.840 0.920 0.009 0.697 0.011 0.793 0.879
hull
Container 9.220 6.400 0.694 6.969 0.756 0.918 7.077 0.768 6.951 0.754 1.018
ship
Bulk carrier 20.394 16.576 0.813 17.500 0.858 0.947 14.798 0.726 15.800 0.775 0.937
D/H 17.677 13.965 0.790 14.066 0.796 0.993 12.213 0.691 11.151 0.631 1.095
Suezmax
S/H VLCC 22.578 18.701 0.828 17.355 0.769 1.078 17.825 0.789 16.179 0.717 1.102
D/H VLCC 32.667 25.667 0.786 27.335 0.837 0.939 22.390 0.685 22.495 0.689 0.995
Mean 0.966 1.004
S-D 0.061 0.088
COV 0.063 0.087

Note: MP = fully plastic bending capacity, Muh = ultimate hogging moment, Mus = ultimate sagging moment,
S-D = standard deviation, COV = coefficient of variation.

196

MARSTRUCT.indb 196 2/18/2011 5:45:56 PM

 Table 8. Comparison between developed formula method and ALPS/HULL intelligent super-size finite element
method.

Hogging Sagging

Formula ALPS Formula ALPS

Mp Muh Muh Formula/ Mus Mus Formula/

Ship (GNm) (GNm) Muh/Mp (GNm) Muh/Mp ALPS (GNm) Mus/Mp (GNm) Mus/Mp ALPS

Dow’s test 0.013 0.010 0.772 0.011 0.799 0.966 0.009 0.697 0.010 0.743 0.939
hull
Container 9.220 6.400 0.694 6.916 0.750 0.925 7.077 0.768 6.635 0.720 1.067
ship
Bulk carrier 20.394 16.576 0.813 16.602 0.814 0.998 14.798 0.726 15.380 0.754 0.962
D/H 17.677 13.965 0.790 13.308 0.753 1.049 12.213 0.691 11.097 0.628 1.101
Suezmax
S/H VLCC 22.578 18.701 0.828 17.335 0.768 1.079 17.825 0.789 17.263 0.765 1.033
D/H VLCC 32.667 25.667 0.786 25.600 0.784 1.003 22.390 0.685 22.000 0.673 1.018
Mean 1.003 1.020
S-D 0.055 0.061
COV 0.055 0.060

Note: Mp = fully plastic bending capacity, Muh = ultimate hogging moment, Mus = ultimate sagging moment,
S-D = standard deviation, COV = coefficient of variation.

30 5.5 CSR idealized structural unit method versus

ANSYS nonlinear finite element method
25 Table 11 presents the mean values and coeffi-
cients of variation for the CSR idealized structural
unit method versus the ANSYS nonlinear finite
(Mu)Formula (GNm)

20
element method for all six types of ship hull.
Figure 17 shows the trend of the deviation in ultimate
15 hull strength between these two methods.

10
Hog / Sag
5.6 CSR idealized structural unit method versus
: Dow’s frigate test hull ALPS/HULL intelligent super-size finite
: Container ship
: Bulk carrier element method
0.0125 : D/H Suezmax
: S/H VLCC
: D/H VLCC
Table 12 presents the mean values and coeffi-
0
cients of variation for the CSR idealized structural
0 0.0125 10 15 20 25 30
unit method versus the ALPS/HULL intelligent
(Mu)ALPS/HULL (GNm) super-size finite element method for all six types of
ship hull. Figure 18 shows the trend of the devia-
Figure 14. Trend of the deviation in ultimate hull tion in ultimate hull strength between these two
strength between the formula method and the ALPS/ methods.
HULL intelligent super-size finite element method.

5.7 Summary of the comparison

5.4 ALPS/HULL intelligent super-size finite
Table 13 summarizes the results of the compari-
element method versus ANSYS nonlinear
son between the mean values and coefficients of
finite element method
variation obtained with the various methods. It is
Table 10 presents the mean values and coefficients evident that the formula method developed herein
of variation for the ALPS/HULL intelligent super- is in very good agreement with more refined meth-
size finite element method versus the ANSYS ods such as the ANSYS nonlinear finite element
nonlinear finite element method for all six types of method and ALPS/HULL intelligent super-size
ship hull. Figure 16 shows the trend of the devia- finite element method. It tends that the CSR ideal-
tion in ultimate hull strength between these two ized structural method overestimates the ultimate
methods. hull strength.

197

MARSTRUCT.indb 197 2/18/2011 5:45:57 PM

 Table 9. Comparison between developed formula method and CSR idealized structural unit method.

Hogging Sagging

Formula CSR Formula CSR

MP Mus Mus Formula/ Mus Mus Formula/

Ship (GNm) (GNm) Muh/Mp (GNm) Muh/Mp CSR (GNm) Muh/Mp (GNm) Muh/Mp CSR

Dow’s test 0.013 0.010 0.772 0.012 0.888 0.870 0.009 0.697 0.010 0.764 0.912
hull
Container 9.220 6.400 0.694 8.040 0.872 0.796 7.077 0.768 7.843 0.851 0.902
ship
Bulk carrier 20.394 16.576 0.813 17.941 0.880 0.924 14.798 0.726 14.475 0.710 1.022
D/H 17.677 13.965 0.790 15.714 0.889 0.889 12.213 0.691 12.420 0.703 0.983
Suezmax
S/H VLCC 22.578 18.701 0.828 19.889 0.881 0.940 17.825 0.789 17.868 0.791 0.998
D/H VLCC 32.667 25.667 0.786 28.352 0.868 0.905 22.390 0.685 24.798 0.759 0.903
Mean 0.887 0.953
S-D 0.051 0.054
COV 0.058 0.056

Note: Mp = fully plastic bending capacity, Muh = ultimate hogging moment, Mus = ultimate sagging moment,
S-D = standard deviation, COV = coefficient of variation.

30 flange, i.e., the deck panel in the hogging condi-

tion and the outer bottom panel in the sagging
25
condition. Depending on the geometrical properties
of the ship’s hull cross-section and/or the direction
of the vertical bending moment, the original Paik-
20 Mansour method is unable to accommodate the
(Mu)Formula (GNm)

pure vertical bending moment condition in which

15
10
0.0125
0 This benchmark study of the modified Paik-
0 0.0125 10
(Mu)CSR (GNm)
Figure 15. Trend of the deviation in ultimate hull
strength between the formula method and the CSR ideal-
ized structural unit method.
6 CONCLUDING REMARKS
The objective of this paper has been to develop a
modified Paik-Mansour formula method for the
ultimate strength calculations of ship hulls sub-
ject to vertical bending moments. The principles
of both the original and modified Paik-Mansour
formula methods have been presented.
The original Paik-Mansour method does not
allow the expansion of the yielded part in the ver-
tical members, but limits this part to the tension
the total axial forces over the hull cross-section
must be zero.
The modified Paik-Mansour method, in con-
Hog / Sag trast, does permit the expansion of the yielded
: Dow’s frigate test hull
: Container ship
part, thereby allowing the pure vertical bending
: Bulk carrier moment condition to be achieved regardless of the
: D/H Suezmax
: S/H VLCC
geometrical properties of the hull cross-sections or
: D/H VLCC the vertical bending loading direction.
15 20 25 30 Mansour method was undertaken with more
refined methods, such as the ANSYS nonlinear
finite element method, ALPS/HULL intelligent
super-size finite element method, and CSR ide-
alized structural unit method, by identifying the
mean values and coefficients of variation for the
modified method.
The comparisons showed that the modified
Paik-Mansour formula method is in very good
agreement with both the ANSYS nonlinear
finite element method and ALPS/HULL intel-
ligent super-size finite element method, and will
prove very useful as a simple formula for ultimate
strength predictions of ship hulls.
The modified Paik-Mansour method presented
in this paper is a logical step in the development of
a formula-based method for predicting the longitu-
dinal bending moment which will “break the back”

198

MARSTRUCT.indb 198 2/18/2011 5:45:58 PM

 Table 10. Comparison between ALPS/HULL intelligent super-size finite element method and ANSYS nonlinear
finite element method.

Hogging Sagging

ANSYS ALPS ANSYS ALPS

Mp Muh Muh ALPS/ Mus Mus ALPS/

Ship (GNm) (GNm) Muh/Mp (GNm) Muh/Mp ANSYS (GNm) Mus/Mp (GNm) Mus/Mp ANSYS

Dow’s test 0.013 0.011 0.840 0.011 0.799 0.952 0.011 0.793 0.010 0.743 0.936
hull
Container 9.220 6.969 0.756 6.916 0.750 0.992 6.951 0.754 6.635 0.720 0.955
ship
Bulk carrier 20.394 17.500 0.858 16.602 0.814 0.949 15.800 0.775 15.380 0.754 0.973
D/H 17.677 14.066 0.796 13.308 0.753 0.946 11.151 0.631 11.097 0.628 0.995
Suezmax
S/H VLCC 22.578 17.355 0.769 17.335 0.768 0.999 16.179 0.717 17.263 0.765 1.067
D/H VLCC 32.667 27.335 0.837 25.600 0.784 0.937 22.495 0.689 22.000 0.673 0.978
Mean 0.962 0.984
S-D 0.026 0.045
COV 0.027 0.046

Note: Mp = fully plastic bending capacity, Muh = ultimate hogging moment, Mus = ultimate sagging moment,
S-D = standard deviation, COV = coefficient of variation.

30
side structures and longitudinal bulkheads) to the
ultimate hull bending strength was recognised in
25 the original paper as being less straightforward;
and the assumption of a modified “strength
(Mu)ALPS/HULL (GNm)

20 reduction factor” in the compressed elements of

15
10
0.0125
0
0 0.0125 10
(Mu)ANSYS (GNm)
Figure 16. Trend of the deviation in ultimate hull
strength between the ALPS/HULL intelligent super-size
finite element method and the ANSYS nonlinear finite
element method.
of a ship’s hull. Since the 1965 paper introduced
the concept of ultimate longitudinal strength, there
has been a general acceptance that this limit state is
approached—but not necessarily reached—when
the compressed “flange” of the ship’s hull (i.e., the
horizontal material in the deck(s) or bottom struc-
ture) reaches its ultimate compressive resistance,
and the tension “flange” reaches the tensile yield
stress of the structural material.
The modelling of the contribution of the “web”
of the hull girder (i.e., the vertical elements including
the web, together with the occurrence of yielding
throughout the tension side of the final neutral
axis, received much discussion at the time. It was
accepted that the latter assumption would be likely
Hog / Sag to result in an overestimate of the ultimate bend-
: Dow’s frigate test hull
: Container ship ing strength of the hull—though probably only a
: Bulk carrier
: D/H Suezmax
small one in view of the dominance of the contri-
: S/H VLCC butions of the two “flanges” (decks and bottom
: D/H VLCC
structures).
15 20 25 30 30 years later, these assumptions were chal-
lenged in the Paik-Mansour model of the stress
distribution across the hull girder at the ultimate
bending moment, as seen in Figure 3 of this paper.
By postulating a linear elastic stress distribution
over a large part of the “web” of the hull girder—
and thus not calling on the full tensile strength of
the longitudinal material in this part of the struc-
ture, it seemed possible that this method would
underestimate—but again only slightly—the ulti-
mate bending strength of the hull. Most impor-
tantly, the original Paik-Mansour method is unable
to accommodate a pure bending moment depend-
ing on the geometrical properties of ship hulls and/
or bending moment direction.
In the present paper, a scenario lying between
the above two ‘extreme’ models is proposed, in
which some proportion—to be found by an itera-
tive procedure—of the “web” of the hull girder,

199

MARSTRUCT.indb 199 2/18/2011 5:45:58 PM

 Table 11. Comparison between CSR idealized structural unit method and ANSYS nonlinear finite element method.

Hogging Sagging

ANSYS CSR ANSYS CSR

Mp Muh Muh CSR/ Mus Mus CSR/

Ship (GNm) (GNm) Muh/Mp (GNm) Muh/Mp ANSYS (GNm) Mus/Mp (GNm) Mus/Mp ANSYS

Dow’s test 0.013 0.011 0.840 0.012 0.888 1.058 0.011 0.793 0.010 0.764 0.963
hull
Container 9.220 6.969 0.756 8.040 0.872 1.154 6.951 0.754 7.843 0.851 1.128
ship
Bulk carrier 20.394 17.500 0.858 17.941 0.880 1.025 15.800 0.775 14.475 0.710 0.916
D/H 17.677 14.066 0.796 15.714 0.889 1.117 11.151 0.631 12.420 0.703 1.114
Suezmax
S/H VLCC 22.578 17.355 0.769 19.889 0.881 1.146 16.179 0.717 17.868 0.791 1.104
D/H VLCC 32.667 27.335 0.837 28.352 0.868 1.037 22.495 0.689 24.798 0.759 1.102
Mean 1.090 1.055
S-D 0.056 0.091
COV 0.052 0.086

Note: Mp = fully plastic bending capacity, Muh = ultimate hogging moment, Mus = ultimate sagging moment,
S-D = standard deviation, COV = coefficient of variation.

30
30

25
25

20
(Mu)CSR (GNm)

20
(Mu)CSR (GNm)

15
15

10
Hog / Sag
10
Hog / Sag : Dow’s frigate test hull
: Dow’s frigate test hull : Container ship
: Container ship : Bulk carrier
: Bulk carrier 0.0125 : D/H Suezmax
0.0125 : D/H Suezmax : S/H VLCC
: S/H VLCC : D/H VLCC
: D/H VLCC
0
0 0 0.0125 10 15 20 25 30
0 0.0125 10 15 20 25 30 (Mu)ALPS/HULL (GNm)
(Mu)ANSYS (GNm)
Figure 18. Trend of the deviation in ultimate hull
Figure 17. Trend of the deviation in ultimate hull strength between the CSR idealized structural unit
strength between the CSR idealized structural unit method method and the ALPS/HULL intelligent super-size finite
and the ANSYS nonlinear finite element method. element method.

on the tension side of this web, is postulated to
have reached the material yield stress (see Figure 4)
when the ultimate bending strength is reached. It
is noted that the modified Paik-Mansour method
presented in the paper covers both the above two
‘extreme’ models. Furthermore, the modified
Paik-Mansour method is more general and is able
to accommodate a pure bending moment regard-
less of the geometrical properties of ship hulls and/
or bending moment direction.
In the Applied Examples summarised in Sec-
tion 4, it is evident that in certain cases this yielding
of the material in the “webs” of ship hull girders
might indeed extend for significant distances ver-
tically from the yielding “flange”. The amount
of this contribution to the ultimate longitudinal
strength of a ship from the tensioned material in
the cross-section will, of course, depend on the
proportion and disposition of “vertical” material
in the hull cross-section.

200

MARSTRUCT.indb 200 2/18/2011 5:45:59 PM

 Table 12. Comparison between CSR idealized structural unit method and ALPS/HULL intelligent super-size finite
element method.

Hogging Sagging

ALPS CSR ALPS CSR

Mp Muh Muh CSR/ Mus Mus CSR/

Ship (GNm) (GNm) Muh/Mp (GNm) Muh/Mp ALPS (GNm) Mus/Mp (GNm) Mus/Mp ALPS

Dow’s test 0.013 0.011 0.799 0.012 0.888 1.111 0.010 0.743 0.010 0.764 1.029
hull
Container 9.220 6.916 0.750 8.040 0.872 1.163 6.635 0.720 7.843 0.851 1.182
ship
Bulk carrier 20.394 16.602 0.814 17.941 0.880 1.081 15.380 0.754 14.475 0.710 0.941
D/H 17.677 13.308 0.753 15.714 0.889 1.181 11.097 0.628 12.420 0.703 1.119
Suezmax
S/H VLCC 22.578 17.335 0.768 19.889 0.881 1.147 17.263 0.765 17.868 0.791 1.035
D/H VLCC 32.667 25.600 0.784 28.352 0.868 1.108 22.000 0.673 24.798 0.759 1.127
Mean 1.132 1.072
S-D 0.038 0.087
COV 0.034 0.081

Note: Mp = fully plastic bending capacity, Muh = ultimate hogging moment, Mus = ultimate sagging moment, S-D = stand-
ard deviation, COV = coefficient of variation.

Table 13. Comparison of all methods considered: A summary.

Formula/ANSYS Formula/ ALPS Formula/ CSR ALPS/ ANSYS CSR/ ANSYS CSR/ ALPS

Ship Hog Sag Hog Sag Hog Sag Hog Sag Hog Sag Hog Sag

Dow’s test 0.920 0.879 0.966 0.939 0.870 0.912 0.952 0.936 1.058 0.963 1.111 1.029
hull
Container 0.918 1.018 0.925 1.067 0.796 0.902 0.992 0.955 1.154 1.128 1.163 1.182
ship
Bulk carrier 0.947 0.937 0.998 0.962 0.924 1.022 0.949 0.973 1.025 0.916 1.081 0.941
D/H 0.993 1.095 1.049 1.101 0.889 0.983 0.946 0.995 1.117 1.114 1.181 1.119
Suezmax
S/H VLCC 1.078 1.102 1.079 1.033 0.940 0.998 0.999 1.067 1.146 1.104 1.147 1.035
D/H VLCC 0.939 0.995 1.003 1.018 0.905 0.903 0.937 0.978 1.037 1.102 1.108 1.127
Mean 0.966 1.004 1.003 1.020 0.887 0.953 0.962 0.984 1.090 1.055 1.132 1.072
S-D 0.061 0.088 0.055 0.061 0.051 0.054 0.026 0.045 0.056 0.091 0.038 0.087
COV 0.063 0.087 0.055 0.060 0.058 0.056 0.027 0.046 0.052 0.086 0.034 0.081

Note: S-D = standard deviation, COV = coefficient of variation.

While useful methods have been developed in
the literature, it is however emphasized that the
scarcity of full-scale data from experiments or ship
failures makes it difficult to judge which method
might provide the best basis for ship design or
approval calculations. Back-breaking hull failure
involves so much buckled and fractured metal, that
it must be doubted whether the reality can ever be
modelled with precision. It might therefore be pru-
dent to take as the design “critical limit state” the
longitudinal bending moment at which the tension
“flange” of the hull structure first reaches the yield
stress after the compressed “flange” has reached its
limiting resistance. Any margin above this “critical
bending moment” provided by progressive yielding
of the remaining structure is likely to be small, and
can be seen as a small margin of safety reflecting
our uncertain knowledge of the anatomy of final
failure of ships’ hulls in bending.
ACKNOWLEDGEMENTS
The present study was undertaken at The Lloyd’s
Register Educational Trust (The LRET) Research
Centre of Excellence at Pusan National University,

201

MARSTRUCT.indb 201 2/18/2011 5:46:00 PM

 Korea in association with the activities of
International Ship and Offshore Structures
Congress (ISSC) Technical Committee III.1
Ultimate Strength. Thanks are due to Dr. C.H.
Huang, China Corporation Register of Ship-
ping, Taiwan, for his CSR idealized structural unit
method computations. The authors are pleased to
acknowledge the financial support of The LRET
and The National Research Foundation (NRF) of
Korea. This paper is a sequel to “Chapter 16: Ulti-
mate Strength of Ship Hulls” in Ship Structural
Analysis and Design, co-authored by O.F. Hughes
and J.K. Paik.
REFERENCES
ALPS/HULL. 2010. A computer program for progressive
collapse analysis of ship hulls. Advanced Technology
Center, DRS C3 Systems, Inc., MD, USA (www.pro-
teusengineering.com, www.maestromarine.com).
ANSYS 2010. Version 12.0, ANSYS Inc., Canonsburg,
PA, USA.
Caldwell, J.B. 1965. Ultimate longitudinal strength.
Trans. RINA, Vol. 107, pp. 411–430.
Dow, R.S. 1991. Testing and analysis of a 1/3-scale
welded steel frigate model. Proceedings of Interna-
tional Conference on Advances in Marine Structures,
Vol. 2, Scotland, May.
Hughes, O.F. & Paik, J.K. 2010. Ship structural analy-
sis and design, The Society of Naval Architects and
Marine Engineers, New Jersey, USA.
IACS 2008a. Common structural rules for double hull oil
tankers. International Association of Classification
Societies, London, UK, July.
IACS 2008b. Common structural rules for bulk carriers.
International Association of Classification Societies,
London, UK, July.
202
MARSTRUCT.indb 202
ISO 2007. International standard ISO 18072-1, Ships and
marine technology—Ship structures, Part 1: General
requirements for their limit state assessment, Interna-
tional Organization for Standardization, Geneva.
Mansour, A.E. & Liu, D. 2008. Strength of ships and
ocean structures, The Society of Naval Architects and
Marine Engineers, New Jersey, USA.
Paik, J.K. 2011. A benchmark study on the ultimate
strength of ship hulls. The Lloyd’s Register Educa-
tional Trust Research Centre of Excellence, Pusan
National University, Busan, Korea, January.
Paik, J.K. & Mansour, A.E. 1995. A simple formulation
for predicting the ultimate strength of ships. Journal
of Marine Science and Technology, Vol. 1, Issue 1,
pp. 52–62.
Paik, J.K. & Melchers, R.E. 2008. Condition assessment
of aged structures, CRC Press, New York, USA.
Paik, J.K. & Thayamballi, A.K. 2003. Ultimate limit
state design of steel-plated structures. John Wiley &
Sons, Chichester, UK.
Paik, J.K. & Thayamballi, A.K. 2007. Ship-shaped off-
shore installations: Design, building, and operation.
Cambridge University Press, Cambridge, UK.
Paik, J.K., Thayamballi, A.K. & Che, J.S. 1996. Ultimate
strength of ship hulls under combined vertical bend-
ing, horizontal bending and shearing forces. Trans.
SNAME, Vol. 104, pp. 31–59.
Paik, J.K., Thayamballi, A.K. & Lee, J.M. 2004. Effect
of initial deflection shape on the ultimate strength
behavior of welded steel plates under biaxial compres-
sive loads. Journal of Ship Research, Vol. 48, No. 1,
pp. 45–60.
Rutherford, S.E. & Caldwell, J.B. 1990. Ultimate longitu-
dinal strength of ships: A case study. Trans. SNAME,
Vol. 98, pp. 441–471.
2/18/2011 5:46:00 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Development of advanced designed formulation to estimate

the buckling/ultimate strength of curved plates

Joo-Shin Park, Min-Sung Chun & Yong-Suk Suh

Department of Structure Research, Marine Research Institutes, Samsung Heavy Industries Co., Ltd, Geoje,
Kyungnam, Korea

ABSTRACT: During the past decades a number of studies have been conducted in terms of evaluation
of the nonlinear buckling characteristics of curved plates. However any explicit expression or formula
for calculating buckling/ultimate strength of curved plates is not available yet. Therefore it is impera-
tive to develop practical formulas for estimating buckling/ultimate strength of curved plates. The aim
of this study is to analyze nonlinear buckling characteristics of curved plates. A series of FEM analyses
are performed on curved plates varying several parameters such as flank angle (curvature), plate thick-
ness, loading conditions, etc. According to the various conditions applied to the curved plates, buckling
strength and ultimate strength are calculated. And also the stress-strain curve is drawn for each set of
applied conditions from the numerical calculations. It is shown that buckling/ultimate strength formula
developed for a curved plate can give a reasonable estimate of strength for curved plate, when the newly
defined curvature correction parameter considering the increase of the buckling strength due to curvature
is applied. It is our hope that the obtained buckling/ultimate strength characteristics would be used as
practical design guide for estimating the nonlinear buckling strength of curved plates in the field of ships
and offshore structures.

1 GENERAL INSTRUCTIONS to guarantee the total safety of any type of ship

Cylindrically curved plates are extensively used in
ship structures. Curved plates such as deck plat-
ing with a camber, side shell plating at fore and aft
parts, bilge circle part are normally assembled. It
is believed that such curved plates can be modeled
fundamentally by a part of a cylinder subjected
to axial loading produced by hull girder bending.
From the estimations using cylinder models, it
is known that in general, curvature increases the
buckling strength of a curved plate subjected to
axial compression, and component of curvature
may cause an increase in the ultimate strength as
well as buckling strength [1]. However, in spite of
wide use of these structural members, the practi-
cal method for structural assessment on buckling/
ultimate strength of curved plate has not clearly
established yet. Some reference of finite element
analysis and class rules which deals with the buck-
ling strength of curved plate as a part of ship
structures exist, but the result of FEA and related
rules indicate different design values. In addition
to this contradiction, the absence of validation by
relevant experimental investigations invokes the
need for an improved design procedure of curved
plates. Therefore, development of well predictable
and integrated rules for curved plate are required
structure as an urgent task in the shipbuilding
industries. In the present study, to clarify and
examine the fundamental buckling behaviours of
cylindrically curved plate under axial compres-
sion, combined compression with lateral pressure,
a series of elasto-plastic large deflection analy-
ses are performed together with the comparisons
with the collapse behaviours including nonlinear
factors. On the basis of the calculated results, the
effects of curvature (R), initial deflection (w), slen-
derness ratio (β) and aspect ratio (a/b) on the char-
acteristics of the buckling and ultimate strength
behavior of cylindrically curved plates subjected
to axial compressive loading and combined axial
compression and lateral pressure are discussed.
Based on the results of a series of the nonlinear
finite element calculations for all edges simply sup-
porting plating, design formulae are derived in
empirical form in order to predict the buckling/
ultimate strength of curved plates. The predicted
results show a good accuracy comparing to the
results of finite element analysis.
1.1 Literature survey
At the beginning, a brief review is made on pre-
vious research works related to buckling and

203

MARSTRUCT.indb 203 2/18/2011 5:46:00 PM

 ultimate strength behaviour of cylindrically curved
plates and stiffened plates.
Maeno et al. (2003, 2004) performed a series of
elasto-plastic large deflection analysis to investigate
the buckling/plastic collapse behaviour of ship’s
bilge strakes, typical in Tanker and Bulk Carrier,
which are unstiffened curved thick plates sub-
jected to axial compression. Yumura et al. (2005)
investigated buckling/plastic collapse behaviour of
cylindrically curved plates under axial compres-
sion. Park, H.-J. et al. (2005) performed nonlinear
FE analyses on actual stiffened curved plates of a
container ship varying the curvature and spacing
of stiffeners. Kwen et al. (2003, 2004) performed
non-linear FE analyses for unstiffened curved
plates varying aspect ratio, slenderness ratio and
curvature under various loading conditions such
as longitudinal thrust, transverse thrust and
shear load.
Cho et al. (2007) performed both ultimate
strength tests on six curved stiffened plates and
nonlinear finite element analyses under axial com-
pression. An analytical approach was proposed by
Levy (1943) by using large deflection theory for
simulating the elastic behaviour of initially curved
sheet. Buermann et al. (2006) presented a fast semi-
analytical model for the post-buckling analysis of
stiffened cylindrical panels. Kundu et al. (2007)
investigated the geometrically nonlinear post-
buckling analysis of laminated composite doubly
curved shells using FEM. Applying the arc length
method, both snap-through and snap-back post-
buckling behaviour were well captured. Park, J.-S.
et al. (2005–2010) performed non-linear FE analy-
ses for unstiffened curved plates varying aspect
ratio, slenderness ratio and curvature under com-
bined loading conditions as well as development
of analytical method. However, further studies are
needed for facilitating more rational guidelines of
buckling/ultimate strength considering engineer-
ing concept of curved plates.
2 FE MODLE AND METHOD
The target curved plate structure is shown in Fig. 1.
The cylindrically curved plates have the dimen-
sions of a in length, b in width, t in thickness and θ
in flank angle, where the length b is kept constant
as 1,000 mm throughout the present study. There
is a relationship between the width b and the flank
angle θ as b = θR, where R is the radius of curva-
ture as shown in the Figure.
2.1 Initial imperfections
In advanced ship structural design, load-carrying
capacity calculations of plating should be considered
204
MARSTRUCT.indb 204
0.5a
0.5a a
θ
R
b
TRANS.frame
Figure 1. Cylindrically curved plate.
post-weld initial imperfections as parameters
of influence [4]. The initial buckling mode obtained
by eigen-value analysis by using FEM is used as
initial deflection mode.
2.2 Material properties
The curved plates analyzed using the one bay
model had material properties as shown in Table 1.
The analysis models incorporated an idealized
elastic-perfectly plastic stress and strain curve, and
the strain hardening rate is set as zero. Isotropic
hardening law is assumed employing von Mises’s
yield condition.
2.3 Calculation method
The conventional Newton-Raphson method fails
because of the singularity of the stiffness matrix
and a diverging solution. In general, the Arc-length
methods avoid this situation and it is suitable for
nonlinear static equilibrium solutions of unsta-
ble problems such as snap-through or snap-back
behaviour. In the present study considered mate-
rial and geometric nonlinear analysis should be
undertaken using commercial FEM code, MSC.
Nastran [5] which is a general purpose finite ele-
ment method package. The Arc-length method
involves the tracing of a complex path in the
load-displacement response into the buckling/post
buckling regimes. For problems with sharp turns
in the load-displacement curve or path depend-
ent materials, it is necessary to limit the arc-length
radius (arc-length load step size) using the ini-
tial arc-length radius. During the solution, the
Arc-length method will vary the arc-length radius
2/18/2011 5:46:00 PM
 Table 1. Material properties for curved
plate.

High tensile steel

Material (AH32)

Elastic modulus (E) 206 GPa

Poisson’ ratio (v) 0.3
Yield stress (σY) 315 MPa

at each arc-length sub step according to the degree

of nonlinearities that is involved.

3 NUMERICAL SIMULATION
AND RESULTS
(a) Longi.compression (b) Trans.compression
3.1 Eigen buckling mode
A series of eigen-value buckling analysis is carried Figure 2. Typical buckling modes obtained by elastic
out to evaluate buckling strength and to examine eigen-value analysis.
the significant buckling mode. The latter result is
used to produce initial deflection. Figure 2 shows
typical buckling modes varying flank angle and
slenderness ratio. When a flank angle is zero, that
is in case of a flat thinner plate, the plate buckles
into four half-waves in the longitudinal direction.
However, as the flank angle increases to 5 degrees,
the buckling mode changes so that the deflection
flattened out towards loading direction and deflec-
tion locally grow at the loading edges as shown in
Fig. 2(a)-flank angle 5 degrees, 10 mm. When a
flank angle is greater than these values, buckling
takes place in an irregular mode as indicated in
Fig. 2(a). For larger thickness of plate, five or four
Figure 3. Elastic buckling mode varying flank angle.
half-waves mode appears in the loading direction
as shown in Fig. 2(a). In general, buckling mode of
plate under transverse compressive loading takes
place one half-wave, however, plate with curvature the outward deflection develops from the begin-
appears in the circumferential direction at the local ning of compressive loading because of Poisson’s
parts as shown in Fig. 2(b)-flank angle 30 degrees, effect. It is noticed that the curved plates can have
20 mm. This is considered as a part of shell buck- a primary buckling mode that is different from the
ling of a cylinder wall in a so-called diamond buck- buckling mode of flat plate and that the buckling
ling mode [2, 3]. strength is generally larger than that of the flat
plate.
3.2 Fundamental buckling mechanism
3.3 Benchmark study
Figure 3 shows the corresponding elastic buck-
ling modes varying flank angles. The solid red line Figure 4 shows the results of critical buckling
represents compressive load, and solid blue line is strength for curved plate which represent compari-
tension. When the flank angle is less than or equal son between class rule and finite element analy-
to 2 degrees, the buckling takes place with three sis using commercial code ABAQUS. As shown
longitudinal half-waves as in the case of a rectan- in Fig. 4, a comparison of two results is made in
gular flat plate having the same aspect ratio. With which the critical buckling strength from both
further increase in the flank angles, the buckling FEM and class rule calculation is relatively unani-
mode changes in one longitudinal half wave with mous in case of thicker plate. But in case of thinner
additional swelled component of deflection near plate, the result of critical buckling strength shows
the transverse edges. In the case of curved plates, significant errors because curvature reduction

205

MARSTRUCT.indb 205 2/18/2011 5:46:01 PM

 1 1

ABS Flat plate

y Curved plate
BV 5 degrees
0.8 DNV 0.8 x Curved plate
45 degrees
GL
NK
RULE (σcr /σy)

0.6 RINA
0.6

σxu/σy
KR

A C
0.4 0.4

0.2
0 0
0 0.2 0.4 0.6
FEA(ABAQUS, σcr /σy)
Figure 4. Verification results of buckling strength for
curved plate under axial compression.
factor used in class’s equations currently is not in
the reflected buckling phenomenon.
B
0.2
A : Secondary Buckling
B : After 2nd Buckling
C : Ultimate Strength
0.8 1 0 1 2 3 4
ε xu /ε y
Figure 5. The relationships of between average stress
and average stain of plate with/without curvature under
longi.compression.

3.4 Progressive collapse behaviour

A series of elasto-plastic large deflection analy-
sis is performed to clarify the fundamentals in Non-symm.onehalf-wave mode
progressive collapse behaviour including occur-
rence of buckling/yielding of cylindrically curved
plates under axial compression. The thickness of
the plate varies from 7 mm to 35 mm. For each
case, calculated average stress-average strain rela-
tionships are summarised in Figure 5 including the
case of a flat plate. Non-symm.threehalf-waves mode
Firstly, progressive collapse behaviour of a thin
plate subjected to compressive load is explained
Fig. 5. When the flank angle is 5 degrees, secondary
buckling takes place at the A point as shown Fig. 5.
Then, load-carrying capacity rapidly decreases due
to the occurrence of secondary buckling accom-
panied by snap-back unloading phenomenon. At
that time, the buckling mode abruptly changes Non-symm.fivehalf-waves mode
from one half wave to three half-waves for loading
direction. The final collapse mode further changes
Figure 6. Change of buckling mode of the curved plate
from three to five half-waves against increase in the
with a flank angle of 5 degrees.
compressive loading as shown in Figure 6.
This is because of the occurrence of the sec-
ondary buckling after the primary buckling. It is compressive load. Especially, the ultimate strength
known that secondary buckling strength of a sim- of thin plate (t = 10 mm) is estimated lower than
ply supported plate is very high, but for the aspect that of flat plate due to the occurrence of second-
ratio at which buckling mode terminates; the sec- ary buckling [6]. In case of the ultimate strength of
ondary buckling strength is relatively low. Such the curved plate with small flank angle secondary
high-order buckling after the primary buckling buckling behavior should be carefully considered.
takes place because of the change in the in-plane With further increase in the curvature, buckling/
stress distribution due to large deflection. ultimate strength gradually increases and thus
This snap-back behaviour can be captured buckling/yielding starts to take place and the ulti-
using arc-length control with incremental force by mate strength is attained. Figure 7 shows average

206

MARSTRUCT.indb 206 2/18/2011 5:46:04 PM

 1 1
Flat plate
y Curved plate
5 degrees
0.8 x Curved plate 0.8
45degrees

0.6 0.6
σyu/σy

σXU/σY
0.4 0.4

Longitudinal
Compression
5 Deg.
0.2
0.2 10 Deg.
20 Deg.
30 Deg.
0
0
0 1 2 3 4 5 0 1 2 3 4
ε yu /εy Plate Slenderness Ratio, β

Figure 8. The relationships of between average stress

Figure 7. The relationships of between average stress
and slenderness ratio varying curvature under longi.
and average stain of plate with/without curvature under
compression.
trans.compression.
1

stress and average strain curve varying curvature Transverse

Compression
considering same thickness (t = 10 mm). When 0.8 5 Deg.
10 Deg.
transverse compression is applied on the curved 20 Deg.
30 Deg.
plate, buckling/yielding starts at the short-edges 0.6
and the ultimate strength is attained. In the post-
σYU/σY

ultimate strength, yielding is restricted only near

0.4
the end parts where buckling deflection is pro-
duced. This implies that the middle part of the
plate remains in the elastic range. Thus from the 0.2

results i.e. Fig. 7 the buckling/ultimate strength is

decreased with the increase in the curvature. How- 0
ever this characteristic is different for the loading 0 1 2 3 4
in the longitudinal direction. Plate Slenderness Ratio, β

Figure 9. The relationships of between average stress

3.5 Influence of slenderness ratio and slenderness ratio varying curvature under trans.
compression.
A series of FEM analysis are performed by chang-
ing the slenderness ratio of the curved plate from
1
1.18 to 4.14. This slenderness ratio is obtained by Model1
changing the thickness of the plate between 10 mm BC
10t

and 35 mm whereas keeping the breadth and the 0.8 12t

14t

length of the plate as 1,000 mm and 3,700 mm, b = 1000 mm, a/b = 3.7, ω0 = 0.05β2t
16t
20t

respectively. The flank angle is taken as 5, 10, 20 and Frank Angle = 10 deg.
24t
30t
0.6
30 degrees. The calculated ultimate strength is plot-
σYU/σY

35t

ted against slenderness ratio as shown in Figure 8.

It is seen that ultimate strength of curved plate 0.4
with flank angle of 5 degrees gives too conserva-
tive estimation. This is because of the occurrence
0.2
of the secondary buckling. Figure 9 shows rela-
tionship between average stress and slenderness
ratio varying curvature under trans.compression. 0
The differences of ultimate strength according to 0 0.2 0.4 0.6 0.8 1
change in the thickness is small, over 20 degrees of σXU/σY
flank angle.
Figure 10 shows the Ultimate strength inter- Figure 10. Ultimate strength interaction relationship of
action relationship of a curved plate under com- a curved plate under combined biaxial compressive loads
bined biaxial compressive loads as a function of as a function of the plate thickness.

207

MARSTRUCT.indb 207 2/18/2011 5:46:06 PM

 the plate thickness. The loading ratio (longitudinal 1

compressive loading by transverse compressive y

loading) is composed 9:1, 8:2, 7:3, 5:5 and 3:7. 0.8 x
With the increase in the plate thickness, the load-

Empirical Formula (σxcr /σ y )

carrying capacity also increased for various combi- Correlation : 0.979
nation of stress components. 0.6 Standard Deviation : 0.05

0.4
4 DEVELOPMENT OF FORMULATION

4.1 Buckling strength formulation 0.2

The buckling strength obtained by FEM analysis

is plotted against the slenderness ratio, and com- 0
0 0.2 0.4 0.6 0.8 1
pared with the empirical formula in Fig. 11. The FEA (σxcr /σy)
distribution of buckling strength by varying slen-
derness ratio almost remains same at the flank Figure 12. Correlation of the empirical formula with
angle over 10 degrees. However, in case of small buckling strength obtained by FEM of curved plate
flank angle the tendency is different. This is mainly under trans. compressive load.
affected due to the change the curvature. The scat-
ter of the buckling strength between FE-Analysis
and empirical formula is shown in Figures 11 from these results that the developed formula can
and 12. In the present formulation used classical predict the buckling strength that is in good agree-
Euler buckling format and re-adopted the correc- ment with the FEM results.
tion factor as namely, CB, CR and CJ by Eqs. (4.1)–
(4.8). The coefficient of buckling strength (CB) is • Buckling strength of curved plate under
assumed as function of curvature/slenderness ratio longitudinal /transverse compressive loads
of the curved plate and coefficient of reduction
2
(CR). This function of CB represents the effect of π 2E ⎛ t ⎞
curvature. The final coefficient (CJ) is assumed σE =k ⎜ ⎟ × CB (1)
12(1 − v 2 ) ⎝ b ⎠
LC
as correction factor to represent the buckling
β
mechanism. This formulation follows calculation
of plasticity correction by Johnson-Ostenfeld. In
CB ( 1 v2 × θ × ) σY / E
× CR (2)
general, Euler buckling stress (σE) is higher than −0.751
half the yield stress (σy), the critical buckling stress ⎛ R⎞
CR = 0 0323 ⎜ ⎟ × CJ (3)
is given by σJ−0 = [1−(σY / 4σE)] × σY, assuming that ⎝t⎠
the proportional limit is 0.5σY. It is generally found
⎧1.0 0 β < 2.0
CJ = ⎨ θ = 10 ~ 30 (4)
⎩ 1
1 .1 β ≥ 2.0
1
⎧ 1.00 β < 2.0
ψ
CJ = ⎨ θ = 0 10 (5)
x ⎩0.8888 β ≥ 2.0
0.8
Empirical Formula (σxcr /σ y )

Correlation : 0.991
Standard Deviation : 0.163
• Transverse compressive load
0.6
2
π 2E ⎛ t ⎞
σE =k ⎜ ⎟ × CB (6)
12(1 − v 2 ) ⎝ b ⎠
TC
0.4

β
0.2
CB ( 1 v2 × θ × ) σY / E
× CR (7)

−1.514
⎛ R⎞
CR = 0.003 ⎜ ⎟ (8)
0
0 0.2 0.4 0.6 0.8 1 ⎝t⎠
FEA (σxcr /σy)

4.2 Ultimate strength formulation

Figure 11. Correlation of the empirical formula with
buckling strength obtained by FEM of curved plate Based on the insights noted above, the ultimate
under longi. compressive load. strength of the cylindrically curved plate, for axial

208

MARSTRUCT.indb 208 2/18/2011 5:46:08 PM

 compression may be empirically derived by curve 1

fitting of the finite element results as follows; The y

new expression of the Frankland’s formula [7] 0.8 x
used including new factors. It was also adopted by

Empirical Formula (σxcr /σ y )

the US Navy. This expression has the same gen- Correlation : 0.999

eral form as the one due to Faulkner to Guedes 0.6 Standard Deviation : 0.08

Soares but the coefficients are different, leading

to a more conservative prediction. The coefficient 0.4
CF assumed as function of double slenderness ratio,
and correction factors (Ca, Cb and Cc) represents
effect of curvature with the change in the thickness 0.2

of plate. The accuracy of the present formula plot-

ted against slenderness ratio is checked by a com-
0
parison with finite element solutions for a range of 0 0.2 0.4 0.6 0.8 1

flank angle between 5 and 30 degrees, as shown in FEA( σxcr /σy)

Figures 13 and 14. It is noted that correlation ratio

and standard deviation of the error in the empiri- Figure 14. Correlation of the empirical formula with
cal formula against FEA is 0.99 and 0.133, of the ultimate strength obtained by FEM of curved plate
under trans. compressive load.
curved plate subjected to longitudinal compression
respectively. Also, the empirical formula correlated
with 0.99 and standard deviation is 0.08 against 2
the ultimate strength obtained by FEA of curved ⎛ R⎞ ⎛ R⎞
Cb = −4.138 ⎜ ⎟ + 1.934 ⎜ ⎟ − 1.023 (12)
plate under transverse compression. The newly ⎝t⎠ ⎝t⎠
developed ultimate strength formula for curved 2
plate can give a reasonable estimation comparing ⎛ R⎞ ⎛ R⎞
Cc = 1.001⎜ ⎟ − 0.181⎜ ⎟ + 1.382 (13)
FEM results. ⎝t⎠ ⎝t⎠
• Longitudinal compressive load
• Transverse compressive load
σU − LC ⎛ 2.25 1.25 ⎞
=⎜ − 2 ⎟ × CF (9) ⎛ R⎞ ⎛ R⎞
2
σY ⎝ β β ⎠ Ca = 2.596 ⎜ ⎟ − 1.712 ⎜ ⎟ + 0.415 (14)
⎝t⎠ ⎝t⎠
Ca Cb
CF = + + Cc (10) ⎛ R⎞ ⎛ R⎞
2
β2 β Cb = −2.095 ⎜ ⎟ + 0.929 ⎜ ⎟ − 0.136 (15)
⎝t⎠ ⎝t⎠
2
⎛ R⎞ ⎛ R⎞ 2
Ca = 3.434 ⎜ ⎟ − 1.989 ⎜ ⎟ + 0.646 (11) ⎛ R⎞ ⎛ R⎞
⎝t⎠ ⎝t⎠ Cc = 1.009 ⎜ ⎟ − 0.724 ⎜ ⎟ + 0.322 (16)
⎝t⎠ ⎝t⎠

• Ultimate strength of curved plate under com-

1 bined longitudinal compressive load and lateral
y
pressure
x
0.8
σU − LCLP ⎛ 2.25 1.25 ⎞
Empirical Formula (σxcr /σ y )

=⎜ − 2 ⎟ × CF
Correlation : 0.998
Standard Deviation : 0.133 CP (17)
0.6
σY ⎝ β β ⎠
Ca Cb
CF = + + Cc (18)
0.4 β2 β
∴ Range ( ) = 10 ~ 30
0.2
⎡ ⎛ b⎞
2
⎛ b⎞ ⎤
⎢ Ca = 3.434 ⎜ ⎟ − 1.989 ⎜ ⎟ + 0.646 ⎥
⎢ ⎝ R ⎠ ⎝ R ⎠ ⎥
0
0 0.2 0.4 0.6 0.8 1 ⎢ 2 ⎥ (19)
⎢C = −4.138 ⎛ ⎞
b ⎛ ⎞
b
FEA( σxcr /σy)
1 ⎜ ⎟ + 1.934934 ⎜ ⎟ − 1.023⎥
⎢ b
⎝ R⎠ ⎝ R⎠ ⎥
Figure 13. Correlation of the empirical formula with
⎢ 2
⎥
⎢ ⎛ b⎞ ⎛ b⎞ ⎥
ultimate strength obtained by FEM of curved plate ⎢ Cc = 1.001
001⎜ ⎟ − 0.181⎜ ⎟ + 1.382 ⎥
⎝ R⎠ ⎝ R⎠
under longi. compressive load. ⎢⎣ ⎥⎦

209

MARSTRUCT.indb 209 2/18/2011 5:46:12 PM

 ∴ Range ( ) ≤ 10 1
Model1
⎡Ca = −0.068⎤ TC+LP
FEM-Formula
⎢C = 0.181 ⎥ (20) Linear
⎢ b ⎥ 0.8
⎢⎣Cc = 0.883 ⎥⎦ b = 1000 mm, a/b = 3.7

Formula (σXU/σY)
ω0 = 0.05β2 t, P = 0.2 MPa

CP = −0.210β + 1.263 (21) 0.6

• Ultimate strength of curved plate under

combined-transverse compressive load and lat- 0.4
eral pressure
0.2
σU −TCLP ⎛ 2.25 1.25 ⎞
=⎜ − 2 ⎟ × CF CP (22) Correlation : 0.993
σY ⎝ β β ⎠ Standard Deviation : 0.105

0
Ca Cb
CF = + + Cc (23) 0 0.2 0.4 0.6 0.8 1
β2 β FEA( σXU/σY)
⎡ ⎛ b⎞
2
⎛ b⎞ ⎤ Figure 16. Correlation of the empirical formula with
⎢ Ca = 2.596 ⎜ ⎟ − 1.712 ⎜ ⎟ + 0.415 ⎥
⎢ ⎝ R ⎠ ⎝ R ⎠ ⎥
ultimate strength obtained by FEM of curved plate
under transverse compression and lateral pressure.
⎢ 2 ⎥
⎢C = −2.096 ⎛ b ⎞ ⎛ b ⎞
⎜ ⎟ 9 ⎜ ⎟ − 0.136 ⎥
+ 0.929 (24)
⎢ b
⎝ R⎠ ⎝ R⎠ ⎥
⎢ 2
⎥ formula plotted against slenderness ratio is checked
⎢ ⎛ b⎞ ⎛ b⎞ ⎥
⎢ Cc = 1.009
009 ⎜ ⎟ − 0.724 ⎜ ⎟ + 0.322 ⎥
⎝ R⎠ ⎝ R⎠
by a comparison with finite element solutions as
⎢⎣ ⎥⎦ shown in Figures 15 and 16. It is noted that cor-
relation ratio and standard deviation of the error
CP = −0.196β 2 + 0.893β + 0.314 (25) in the empirical formula against FEA is 0.99 and
0.27, of the curved plate subjected to combined
longitudinal compression and lateral pressure,
The coefficient CP assumed as function of slen- respectively.
derness ratio, and the other factors (Ca, Cb and Cc) The normalized ultimate strength by FEA is
represents effect of curvature with the change in plotted against the normalized ultimate strength
the thickness of plate. The accuracy of the present by empirical formula of curved plate under com-
bined transverse compression and lateral pressure
as shown in Figure 16. Good agreements are also
1 observed in both results.
Model1
LC+LP
FEM-Formula
Linear
0.8 5 CONCLUSION
b = 1000 mm, a/b = 3.7
Formula (σXU/σY)

ω0 = 0.05β2 t, P = 0.2 MPa The objective of the present paper is to clarify and
0.6 examine the fundamental buckling/plastic collapse
behaviour and ultimate strength of cylindrically
curved plate under a variety of loading conditions
0.4 (compression and combined compression and lat-
eral pressure).
On the basis of the calculated results, the effects
0.2
Correlation : 0.995
of curvature (R), slenderness ratio and loading
Standard Deviation : 0.271 effect on the buckling and ultimate strength have
0
been discussed. A simple formulation is developed
as an efficient method to predict the critical buck-
0 0.2 0.4 0.6 0.8 1
ling strength and ultimate strength. The following
FEA( σXU/σY) points can be concluded:
Figure 15. Correlation of the empirical formula with 1. Cylindrically curved plate under longitudinal
ultimate strength obtained by FEM of curved plate compression takes place the buckling mode
under longitudinal compression and lateral pressure. with several half waves for loading direction

210

MARSTRUCT.indb 210 2/18/2011 5:46:18 PM

 when secondary buckling occurs. When free
from secondary buckling, local region near
loading edges only collapses.
2. The buckling of the curved plate with rela-
tively large flank angle takes place in one-half
wave mode in loading direction with the local
swelled shape near the loading edges. After the
occurrence of primary buckling, the secondary
buckling takes place and deflection change to
the mode with the larger number of half-waves.
Also the in-plane rigidity is decreased.
3. The ultimate strength of curved plate is sig-
nificantly reduced when the secondary buck-
ing takes place. At that time, buckling mode
abruptly changed with redistribution of in-
plane stress.
4. The curved plate under trasversal compres-
sion underestimates the buckling and ultimate
strength with increase in the curvature. This is
mainly induced by collapse pattern.
5. A simple formula developed for a curved plate
can give a reasonable estimate of buckling/
ultimate strength of curved plate under a variety
of loading conditions (longitudinal/transverse
compressive load, combined biaxial compres-
sion and lateral pressure).
6. Good correlations are observed in the ultimate
strength including buckling strength by apply-
ing the proposed empirical formula.
ACKNOWLEDGMENTS
The research study reported in this paper was
undertaken with Samsung Heavy Industry and
seven kinds of classification of societies (Lloyd’s
Register, ABS, Class NK, Korean Register, Bureau
Veritas, Germanischer Lloyds and RINA). The
authors are glad to acknowledge their continuous
technical and other supports.
211
MARSTRUCT.indb 211
REFERENCES
[1] Joo-Shin, Park., Masahiko, Fujikubo., Iijima,
Kazuhiro & Tetsuya Yao, “Prediction of the second-
ary buckling strength and ultimate strength of cylin-
drically curved plate under axial compression”, The
International Journal Society of Offshore and Polar
Engineers (IJSOPE-ASME), 2009. 07.
[2] Timoshenko, S.P. & Gere, J.M. “Theory of elastic sta-
bility”, McGraw-Hill Book, New York. (1961).
[3] Timoshenko, S.P. & Woinosky-Krieger, S. “Theory
of plates and shells” McGraw-Hill Book, New York,
1959.
[4] Paik, J.-K. & Thayamballi, A.K. “Ultimate limit state
design of steel-plated structures”, John Wiley & Sons,
U.K., 2003.
[5] MSC. Nastran implicit nonlinear (SOL 600) User’s
Guide, “Solution methods and strategies in nonlinear
analysis”, (2005).
[6] Joo-Shin, Park., Iijima, Kazuhiro & Tetsuya Yao,
“Characteristics of Buckling and Ultimate Strength
and Collapse Behaviour of Cylindrically Curved Plates
subjected to Axial Compression”, Journal of Advanced
Material Research, 2008. 01. 30. pp. 1195–1200.
[7] Frankland, J.M. “The strength of ship plating under
edge compression,” US EMB Report 469, May, 1940.
2/18/2011 5:46:21 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Ultimate strength assessment of ageing steel plates subjected

to random non-uniform corrosion wastage

J.E. Silva, Y. Garbatov & C. Guedes Soares

Centre for Marine Technology and Engineering (CENTEC), Technical University of Lisbon,
Instituto Superior Técnico, Lisboa, Portugal

ABSTRACT: The aim of this study is to investigate the effect of non-linear randomly distributed
nonuniform corrosion on the ultimate strength of unstiffened rectangular plate subjected to axial
compressive loading. A series of 570 plate surface geometries are generated by Monte Carlo simulation
for different degree of corrosion, location and ages and nonlinear finite element analyses are carried out,
using a commercial finite element code. Based on a regression analysis, empirical formulae to predict
strength reduction because of corrosion have been derived demonstrating a good accuracy.

1 INTRODUCTION Jiang and Guedes Soares, (2008, 2009, 2010) and

Corrosion has always been one of the major
problems in marine industry. Many catastrophic
situations have been caused by corrosion damage,
even when all the design requirements are satisfied
(Nakai et al., 2004, 2006).
Some studies from the last decades considered
simplified models of general corrosion wastage,
linearly increasing with time (Hart et al., 1986;
Guedes Soares 1988a and Shi 1993). More recent
studies demonstrated the nonlinear time dependent
corrosion models are more appropriate (Guedes
Soares et al., 2009) and that corroded surfaces
could be modelled by random fields (Teixeira and
Guedes Soares, C. 2008)
Many models and studies have been carried out
to predict the behaviour of structural elements
affected by corrosion degradation in a determin-
istic way, focusing their attention on pitting corro-
sion as one of the most hazardous forms.
Paik et al. (2003, 2004) investigated the ultimate
strength of plate elements with pit corrosion
wastage under axial loads and in-plane shear loads.
They derived a closed-form solution to estimate
the ultimate strength of pitted plates by idealizing
corrosion pits as a cylindrical shape and by vary-
ing the degree of pits and intensity in a system-
atic way.
Duo et al. (2007) idealized corrosion pits as cylin-
drical cones and investigated the influence of local-
ized corrosion on the ultimate strength. Although
over 256 nonlinear finite element analyses were
conducted in a systematic way it was assumed that
corrosion was constrained to a rectangular area on
the plate.
Saad-Eldeen and Guedes Soares, (2009) focused
their attention on the influence of scattered pitted
plates on the collapse strength by using the math-
ematical model proposed by Daidola et al. (1997),
who developed a method to estimate the residual
thickness of pitted plates.
The study, presented here, instead of using any
idealization of special distribution of pitting cor-
rosion, treats it as a quasi-random distribution of
plate thickness by applying the approach of cor-
rosion wastage developed by Guedes Soares and
Garbatov (2009) and using Monte Carlo simula-
tion. Using non-linear finite element analysis, the
ultimate strength of a steel plate has been evalu-
ated. The scope of this work is to analyse the effect
of non-linear randomly distributed nonuniform
corrosion on the ultimate strength of unstiffened
rectangular plate subjected to axial compressive
loading.
2 CORROSION MODEL
Three fundamental approaches can be applied for
corrosion deterioration modelling. The conven-
tional approach is just to consider that corrosion
grows linearly with time but this is a crude model.
The second can be based on the results of exper-
iments in specific conditions which suggest laws
of growth of corrosion as a function of specific
parameters. The corrosion model can be devel-
oped by considering all those laws derived from
experiments in specific conditions. This approach
involves one difficulty in generalizing results from
laboratory tests to full-scale conditions. The other

213

MARSTRUCT.indb 213 2/18/2011 5:46:21 PM

 difficulty is related to the general lack of data on 2.0
the environmental conditions which affect corro- 1.8
sion in full-scale.
1.6

Corrosion depth, mm
The third approach, which is the one that is
adopted here, is to consider that a model should 1.4
provide the trend that is derived from for the domi- 1.2
nating mechanism and then it should be fit to the
field data. Guedes Soares and Garbatov, (1999) pro- 1.0
posed a model for the non-linear time-dependent 0.8
function of general corrosion wastage. This time- 0.6
dependent model separates corrosion degrada-
tion into three phases. In the first one there is no 0.4
corrosion because the protection of the metal sur- 0.2
face works properly. The second phase is initiated
0.0
when the corrosion protection is damaged and
0 4 8 12 16 20 24 28 32 36
corresponds really to the start of corrosion, which
decreases the thickness of the plate. The third t, years
phase corresponds to a stop in the corrosion proc-
Figure 1. Corrosion depth of deck plates of ballast
ess and the corrosion rate becomes zero.
tanks.
The model is based on the solution of a differen-
tial equation of the corrosion wastage:
1
⎧ ⎛ −t −τ c ⎞
⎪d ⎜ e τt
⎟ , t ≥ τc 0.9
d (t ) = ⎨ ∞ ⎜ (1)
⎟⎠
⎪ ⎝
0.8
Standard Deviation, mm

⎩0, t < τc 0.7 a = 0.384

b= 0.710
0.6
where d∞ is the long-term corrosion wastage,
d (t) is the corrosion wastage at time t, τc is the 0.5
time without corrosion which corresponds to the 0.4
start of failure of the corrosion protection coating
(when there is one), and τt is the transition time 0.3
duration. 0.2
The long-term wastage d∞ is defined as the
maximum value in the observed time interval 0.1
for ballast tanks and cargo tanks respectively. 0
The period without corrosion, or the time of ini- 0 4 8 12 16 20 24 28 32 36
tiation of corrosion τc, and the transition time, t, years
τt are defined based on performing a least squares
fit to the data using a quasi-Newton algorithm, Figure 2. Standard deviation of corrosion wastage of
which determines the direction to search used at deck plates of ballast tanks.
each iteration considering the mean value of cor-
rosion depth.
The parameters of the regressed line of corrosion corrosion wastage depth is fit by the Log-normal
depth as a function of time were determined under distribution.
the assumption that it is approximated by the The mean value and the variance of the
exponential function given in Eqn. (1) for ballast log-normal distribution for the corrosion wastage
tanks of tanker deck by Garbatov et al. (2007). of deck plates of ballast tanks are −0.544 and
The long-term corrosion wastage for deck plates of 0.919. The model just presented is used for deterio-
ballast tanks has been defined as d∞,ballast = 1.85 mm. rated plate surface modelling.
The time without corrosion is τc,ballast = 10.54 years
and the transition period τt,ballast = 11.14 years (see
Figure 1). 3 MONTE CARLO SIMULATION
The standard deviation as a function of time OF THE CORRODED PLATE SURFACE
has been fit to StDev(t) = a Ln(t)−b, which is
shown in Figure 2. Based on the analysis per- The corroded plate surface is modelled as random
formed by Garbatov and Guedes Soares, (2008) the plate thickness that results in the random vertical

214

MARSTRUCT.indb 214 2/18/2011 5:46:21 PM

 position of the coordinates of corroded surface for
equally spaced reference points positioned along
the x and y direction of the plate, as shown in
Figures 3 to 7. These reference points are defined
in a Monte Carlo simulation as being the nodes of
the finite element mesh on the p plate.
The plate thickness, Zijcorroded, at any reference
point with coordinates x, y for the corroded plate
surface, is defined by the random thickness of the
intact plate surface, Zijintj
act
affected by the random
vertical reduction resulting from the corrosion
depth, Zijcorrosion depth as:
Zcorroded Zint act − Zcorrosion depth
(2)
according to the previous procedure, using the
simulated random thickness at the reference nodes
to adjust the plate thickness at the nodes according
to Eqn. 2, as displayed in Figures 3 to 7. Corrosion
plate reduction is applied symmetrically on both
sides of plate. 570 successful simulations of cor-
roded plate surfaces are performed.
The mean value and the standard deviation of
the corrosion depth are considered as the ones
related to the deck plate of ballast tanks of tanker
ship. The mean value and the standard deviation of
the intact plate thickness are considered as 10 mm
and 1 mm respectively.

where Z are the matrixes of the corroded and intact

surface and corrosion depth.
This convention is used to derive the formulation
that describes the vertical position of the surface of
the non-linear corroded plate in the Monte Carlo
simulation resulting in randomly distributed
plate thicknesses for randomly defined reference
nodes at a specific year based on Eqn. (2) apply-
ing the corrosion degradation levels as defined by
Eqn. (1) and (2).
The vertical random coordinates (corrosion
depth) of the corroded and intact plate surfaces and
corrosion depths are modelled by a log-normal dis-
tribution. This probabilistic distribution is widely
known and can be found in the literature about
statistics. The intact plate surface coordinates and
corresponding corrosion depths are considered as
not correlated.
The modelling of the corroded plate sur- Figure 4. Modelled plate surface with an average thick-
face in the finite element analysis model is made ness of 9.6 mm at the 20th year.

Figure 3. Modelled plate surface with an average thick- Figure 5. Modelled plate surface with an average thick-
ness of 9.8 mm at the 15th year. ness of 9.2 mm at the 25th year.

215

MARSTRUCT.indb 215 2/18/2011 5:46:23 PM

 Table 1. Plate BC (C—Constrained; F-Free).

Ux Uy Uz rotx

y=0 F C C C
y=L F F C C
x=0 F F C F
x=b F F C F
x = b/2 and y = 0 C C C C

L is the plate length in y direction, b is the width

Figure 6. Modelled plate surface with an average thick-
ness of 8.9 mm at the 30th year.
along the x-axis and h0 is the intact thickness.
The plate is supported in such a way that there
is no displacement along z-axis in all the edges,
the rotation along an axis parallel to x-axis in the
edges y = 0 and y = L is constrained, there is no
displacement along y axis in the edge y = 0 and
the point (x = b/2; y = 0) is clamped to ensure the
symmetry. Table 1 summarizes the boundary con-
ditions, where “C” means constrained and “F” is
free displacement or rotation respectively.

4.2 Initial imperfections

The initial imperfections, which simulate the
presence of manufacture and welding defects, are
considered in this study as proposed by Smith
et al. (1987):
⎛x ⎞ ⎛ y ⎞
z w0 sin ⎜ π⎟ sin
i π (3)
⎝ b ⎠ ⎜⎝ L ⎟⎠

w0 h0β20 (4)

⎛ b⎞ σ
y
β0 = ⎜ ⎟ (5)
⎜⎝ h ⎟⎠ E
0

where, x, z, y are plate’s coordinates system,

Figure 7. Modelled plate surface with an average thick-
ness of 8.4 mm at the 40th year.
4 NONLINEAR FINITE ELEMENT
ANALYSIS
To evaluate the ultimate strength of a rectangular
steel plate subjected to compressive load, a non-
linear finite element analysis is performed using
ANSYS, (2009) commercial software. A large
deflection analysis with the arc length method has
been employed.
4.1 Plate geometry
The plate has a rectangular shape and is defined
with z being the perpendicular axis to plate’s plane.
w0 is the maximum amplitude of imperfection,
β0 denotes the intact plate slenderness as proposed
by Faulkner (1975) and E and σy are Young modu-
lus and yield stress respectively.
It is beyond the scope of this study to evalu-
ate the influence of these parameters on ultimate
strength and the only parameter that is varying
is the plate thickness as a function of corrosion
deterioration.
4.3 Finite element modelling
The finite element geometry is modelled using a
mesh with around 2520 rectangular elements and
2627 nodes accounting for some previously analyses
related to sensitivity analysis on ultimate strength,
as for example by Rigo et al. (2003). The defined

216

MARSTRUCT.indb 216 2/18/2011 5:46:26 PM

 Figure 9. Intact plate stress-strain response.

0.7

Figure 8. FEM mesh with initial imperfection.

0.6
Average Stress Ratio, ASR

mesh size proves a good quality of results and it is

0.5
not too dense to avoid endless of calculations.
The plate is modelled by nonlinear shell elements
with four nodes, SHELL181. This element permits 0.4
to use nodal properties for introducing thickness
on every node. Figure 8 illustrates FEM mesh, ini- 0.3
tial imperfections and plate coordinates system.
0.2

5 ULTIMATE STRENGHT ASSESSMENT 0.1

The stress-strain curve for the intact plate used for
FE analysis is shown in Figure 9. The axial load is 0
0 0.5 1 1.5
applied on the edge y = L and the average stresses Strain/ Yield Strain
it calculated based on the reaction forces in the
edge y = 0. Figure 10. Deteriorated plate stress-strain ratio
samples.
k
∑ Ry i
where ε is plate strain, εy = 0.001714 is the yield strain
i =1
ASR = (6) of material, Uyp is the displacement in y direction at
A σ point p = (b/2, L, 0) and L is the plate length.
0 yield
l
To evaluate the ultimate strength, the maximum
where ASR is the average stress ratio, Ry,i is the value of the stress strain curve of each plate is ana-
reaction forces, in y direction, at the ith node, which lysed. The ultimate strength ratio of intact plate
has the coordinates: (xi, 0, 0), k is the number of is defined as σu,0/σy = 0.6972 as can be seen from
nodes at y = 0, A0 = h0b is the sectional area of plate Figure 5. The average stress strain ratio samples
at y = 0 and σyield is the yield stress point of the for the deteriorated plate are shown in Figure 6.
material.
6 RESULTS AND DISCUSSION
ε U (7)
= y p
The ultimate strength ratio, σu/σy as a function of
εy Lε
y time is shown in Figure 11, where the resulting

217

MARSTRUCT.indb 217 2/18/2011 5:46:28 PM

 Figure 11. Ultimate strength ratio σu/σy as a function
of time. Figure 12. Standard deviation of ultimate strength
ratio, σu/σy as a function of time.

values of the 570 nonlinear finite element com-

putations are collected. The results clearly reveal
that a nonlinear curve is the best fitted one to the
collected data. Following this tendency an expo-
nential equation is used to define to the ultimate
strength ratio as:

⎧ σ u ,0
⎪ , t ≤ τC
σ u (t ) ⎪ σ y (8)
=⎨ ⎧ ⎛ ⎞ t⎫
n
σy ⎪ σ u,0 exp ⎪⎨− t − τC ⎪⎬ , t ≥ τC
⎪ σy ⎜ ⎟
⎩ ⎪⎩ ⎝ τt ,U ⎠ ⎪⎭

where t is time in years, τC = 10.54 years is the

coating life as discussed in Section 2, σ u,0 σ y
is the ultimate strength ratio when t = τC, τt,U is
the transition time to be adjusted and it has time
unit and n is a non dimensional parameter, which
represent the time decay capacity of the ultimate
strength of the plate. Figure 13. Ultimate strength ratio, σu/σy as a function
However, the parameters τt,U and nt depend of of plate slenderness, β.
plate ultimate strength. For the studied plate, the
parameters that best fit Eqn. (9) are τt,U = 49.92 where at and bt are defined based on regression
years and n = 1.42. analysis resulting in 0.0156 and −0.0411 respec-
The R2 value has been evaluated to check the tively. The R2 value has been calculated as 0.871.
accuracy of the regression analysis showing a good The ultimate strength ratio has been also ana-
agreement between the calculated and predicted lysed as a function of plate slenderness (see Eqn. 5),
values of ultimate strength ratio, R2 = 0.9769. where the mean value and standard deviation have
The standard deviation as a function of time has been modelled as:
been defined as (see Figure 12):
⎡ σ u (β) ⎤ σ u,0 ⎧ ⎛ β τ ⎞ nβ ⎫
⎡ σ (t ) ⎤ ⎧0, t ≤ τC ⎪ C ,β ⎪
E⎢ ⎥= exp ⎨− ⎜ ⎟ ⎬ , β ≥ τC ,β
StDev ⎢ u ⎥ = ⎨ (9) σ σ τ
⎢⎣ σ y ⎥⎦ ⎩at Ln (t ) + bt , t ≥ τC ⎢⎣ y ⎥⎦ y ⎪⎩ ⎝ β,U ⎠ ⎪⎭
(10)

218

MARSTRUCT.indb 218 2/18/2011 5:46:30 PM

 a function of plate slenderness can be also easily
⎡ σ (β) ⎤
StDev ⎢ u ⎥ aβ Ln (β) bβ β ≥ τC ,β (11) plotted.
⎢⎣ σ y ⎥⎦
For the studied plate, the parameters that
7 CONCLUSIONS
best fit Eqn. (10) are τβ,U = 1.11, τC,β = 2.9 and
nβ = 1.09 respectively. For the standard deviation,
The present study analyzed the effect of non-linear
the parameters aβ and bβ are 0.0976 and −0.1030
randomly distributed corrosion on the ultimate
respectively. The R2 coefficient for the mean value
strength of unstiffened rectangular plate subjected
and standard deviation of ultimate strength ratio
to axial compressive loading. A series of 570 plate
as a function of plate slenderness are 0.9885 and
surface geometries where generated by Monte Carlo
0.8542 respectively.
simulation for different degree of corrosion,
It is considered that the ultimate strength ratio
location and age and nonlinear finite element
can be described as a log-normal distribution func-
analyses were carried out. The random surface
tion, truncated at σ u σ y , f ⎡⎣σU σ y ⎤⎦ with
TR modelling, used in this study, allowed adequate
a mean value and standard deviation varying as
representation of the real corroded plate surface
a function of time or plate slenderness, β as:
accounting for the random origin of the location
and the form of corrosion depths. The plate slen-
⎡σ ⎤ derness varied from 2.9 to 3.45 as a function of
g⎢ U ⎥
⎡σ σ σ u ,0 ⎤ ⎢⎣ σ y ⎥⎦ corrosion degradation. For 30 years service life,
f ⎢ U 0< U ≤ ⎥ = σ u ,0 without replacement of plate, the ultimate strength
⎢⎣ σ y σy σ y ⎦⎥ ,
has lost his capacity from 0.69 to 0.44 which results
TR σy
⎡σ ⎤ σ in 37%. Based on a regression analysis empirical
∫ f ⎢ U ⎥d U
⎢⎣ σ y ⎥⎦ σ y
(12)
formulae to predict strength reduction because of
where 0
corrosion have been derived demonstrating a good
accuracy.
⎧ ⎡ σU ⎤ σ
⎡ σU ⎤ ⎪⎪ f ⎢ ⎥, 0 < U ≤ 1
σ
⎢⎣ y ⎥⎦ σy (13)
g⎢ ⎥ = ⎨ ACKNOWLEDGEMENTS
⎢⎣ σ y ⎥⎦ ⎪0 ,
σU
>1
⎪ σy
⎩ This work has been made under the plurianual
The time dependent probability density function funding of the Portuguese Foundation of Science
of the estimated ultimate strength ratio is given and Technology (Fundação para Ciência e Tec-
in Figure 14. The probability density function as nologia) to the Centre for Marine Technology and
Engineering (CENTEC).
The work reported here is a contribution to
the activities of the MARSTRUCT VIRTUAL
INSTITUTE, (www.marstruct-vi.com) in particu-
lar its Technical Subcommittee 2.3 on Ultimate
Strength.

REFERENCES

ANSYS. 2009, User’s manual (version 12.0).

Daidola, J., Parente, J. & Orisamolu, I. 1997. Residual
strength assessment of pitted plate panels. Report
SSC-394, Ship Structure Committee.
Duo, O.K., Yongchang, P.U. & Incecik, A. 2007. Compu-
tation of ultimate strength of locally corroded unstiff-
ened plates under compression, Marine Structures,
Vol. 20, pp. 100–114.
Faulkner, D. 1975. A review of effective plating for use in
the analysis of stiffened plating in bending and com-
pression, J. Ship Research, Vol. 19, pp. 1–17.
Garbatov, Y. & Guedes Soares, C. 2008. Corrosion
Wastage Modeling of Deteriorated Ship Struc-
Figure 14. Probability density function of ultimate tures, International Shipbuilding Progress, Vol. 55,
strength ratio, σu/σy and a function of time. pp. 109–125.

219

MARSTRUCT.indb 219 2/18/2011 5:46:33 PM

 Garbatov, Y., Guedes Soares, C. & Wang, G. 2007.
Non-linear Time Dependent Corrosion Wastage of
Deck Plates of Ballast and Cargo Tanks of Tankers,
Journal of Offshore Mechanics and Arctic Engineering,
Vol. 129, No. 1, pp. 48–55.
Guedes Soares, C. 1988. Uncertainty modelling in plate
buckling, Structural Safety, Vol. 5, pp. 17–34.
Guedes Soares, C. & Garbatov, Y. 1999. Reliability of
Maintained Corrosion Protected Plate Subjected
to Non-Linear Corrosion and Compressive Loads,
Marine Structures, Vol. 12, No. 6, 425–446.
Guedes Soares, C., Garbatov, Y., Zayed, A. & Wang, G.
2009. Influence of Environmental Factors on Corro-
sion of Ship Structures in Marine Atmosphere, Corro-
sion Science, Vol. 51, pp. 2014–2026.
Hart, D., Rutherford, S. & Wickham, A. 1986. Structural
reliability analysis of stiffened panels, Trans Roy Inst
Nav Architects (RINA), Vol. 128, pp. 293–310.
Jiang, X. & Guedes Soares, C. 2009. Nonlinear FEM
analysis of ultimate compressive strength of pitted
mild steel square plate, Journal of Ship Mechanics,
Vol. 13, No. 3, pp. 398–405.
Jiang, X. & Guedes Soares, C. 2010. Ultimate Compres-
sive Capacity of Rectangular Plates with Partial Depth
Pits, Proceedings of the OMAE, paper 2010-21050.
Jiang, X. & Guedes Soares, C. 2008. Nonlinear FEM
analysis of pitted mild steel square plates subjected
to in-plane compression, Proceedings IMAM Con-
ference, A Ergin (Ed), Istanbul Technical University,
pp. 123–130.
Nakai, T., Matsushita, H. & Yamamoto, N. 2006. Effect
of pitting corrosion on the ultimate strength of steel
plates subjected to in-plane compression and bending,
J Mar Sci Technol, pp. 52–64.
220
MARSTRUCT.indb 220
Nakai, T., Matsushita, H. & Yamamoto, N. 2007. Visual
assessment of corroded condition of plates with pitting
corrosion taking into account residual strength—in
the case of webs of hold frames of bulk carriers, Pro-
ceedings of the OMAE, paper OMAE2007-29159.
Paik, J., Lee, J. & Ko, M. 2004. Ultimate shear strength
of plate elements with pit corrosion wastage, Thin-
Walled Structures, Vol. 42, pp. 1161–1176.
Paik, J., Lee, J. & Ko, M. 2003. Ultimate compressive
strength of plate element with pit corrosion wastage,
J. Engineering for the Maritime Environment, Vol. 217,
pp. 185–200.
Rigo, R., Sarghiuta, P., Estefen, S., Lehmann, E.,
Otelea, S., Pasqualino, I., Simonsen, Wan, Z. & Yao, T.
2003. Sensitivity analysis on ultimate strength of alu-
minium stiffened panels, Marine Structures, Vol. 16,
pp. 437–468.
Saad-Eldeen, S. & Guedes Soares, C. 2009. Effect of pit-
ting corrosion on the collapse strength of rectangular
plates under axial compression. Analysis and Design
of Marine Structures, C. Guedes Soares and P.K. Das
(Eds), Taylor and Francis, pp. 231–236.
Shi, W. 1993. In-service assessment of ship structures:
effects of general corrosion on ultimate strength. Trans
Roy Inst Nav Architects (RINA), Vol. 135, pp. 77–91.
Smith, C., Davidson, P., Chapman, J. & Dowling, P.
1988. Strength and stiffness of ships’ plating under in-
plane compression and tension, Trans Roy Inst Nav
Architecs (RINA), Vol. 130, pp. 277–296.
Teixeira, A.P. & Guedes Soares, C. 2008. Ultimate
Strength of Plates with Random Fields of Corrosion,
Structure and Infrastructure Engineering, Vol. 4, No. 5,
pp. 363–370.
2/18/2011 5:46:35 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Comparison of numerical results with experiments on ultimate

strength of short stiffened panels

Mingcai Xu & C. Guedes Soares

Centre for Marine Technology and Engineering (CENTEC), Instituto Superior Técnico,
Technical University of Lisbon, Lisbon, Portugal

ABSTRACT: Short stiffened panels are simulated and compared with test results under axial compression
until collapse to investigate the influences of the stiffener’s geometry. The stiffened panels with different
combinations of mechanical material properties and geometric configurations are considered. Four type
stiffeners are made of mild or high tensile steel for bar stiffeners and mild steel for ‘L’ and ‘U’ stiffeners.
To produce adequate boundary conditions at the loaded edges in the experiments, three bays stiffened
panel are used in the test and in the FEM analysis. The influence of the stiffener’s geometry on the ulti-
mate strength of the stiffened panels under compression is analyzed.

1 INTRODUCTION panels instead of one single-bay panels allows to

It is very important to estimate the load carrying
capacity of stiffened panels from the viewpoints
of safety and economy. Non-conventional materi-
als that allow having the same strength of the hull
with a lighter ship structure are used in marine
structures. For example, composites, aluminum
alloys and high strength steel are used in different
ships for that purpose. The strength to weight ratio
is an important index to design economical and
efficient ship. The adoption of very high strength
steels satisfies these requirements allowing the use
of thinner plates, with the corresponding weight
reduction which is very important for high speed
vessels. The application of very high tensile steel
is a good option, but it requires explicit considera-
tion of the failure mechanisms, primarily fatigue
and buckling (Janssen 2000). However, it leads to
the use of thinner plate need to concerns about the
elasto-plastic buckling strength.
In order to reproduce adequate working con-
ditions on a ship structure, the boundary condi-
tions on the loaded top edges and unloaded lateral
edges should be considered carefully. Several ulti-
mate strength tests have been conducted in the
past on simple stiffened panels under compres-
sion (Faulkner 1977); (Horne 1976); (Mathewson
and Vinner 1962); (Smith 1979). However, this
approach raises difficulties in reproducing ade-
quate boundary conditions at the loaded edges in
the experiments.
To circumvent this problem, the tests of Gordo
and Guedes Soares (2008) used specimen with
three bays longitudinally. The use of three-bay
have more realistic results by avoiding boundary
conditions problems for the central plates related
to eccentricity of load and to include the interfer-
ence between adjacent panels (Luís et al 2008a, b).
To prescribe appropriate boundary conditions
is a main challenge in modeling stiffened pan-
els by experiment and finite elements. Because
the boundary of stiffened panels is supported by
strong members such as longitudinal girders and
transverse frames, the simply supported bound-
ary condition is often adopted. But the degree of
rotational restraints at the panel boundary is not
equivalent to zero. It is important to model the
panel edge condition in a relevant way.
The objective of these FEM simulations is to
compare the different structural solutions for pan-
els under compression adopted in the experimented
programmer. Comparison between the perform-
ances of S690, mild steel and hybrid solutions are
made. The base geometry is the one used on the
box girders tests of Gordo and Guedes Soares
(2007). In this regard, the results can be compared
with those of similar stiffened plates belonging to
much larger structures.
2 DESCRIPTION OF THE SPECIMENS
Figure 1 and Table 1 show the geometry of the dif-
ferent panels adopted in the experiment and in the
present study. The A–B means different number
stiffener.
Four series of experiments and FEM analysis
were carried out using two different types of steel
as follows:

221

MARSTRUCT.indb 221 2/18/2011 5:46:35 PM

 − Hybrid L structure: S690, on plating and mild
steel on L stiffeners.
− Hybrid U structure: S690, on plating and mild
steel on U stiffeners.
The S690 steel was supplied by Dillinger
Hutterwerke in the form of sheet of 4 mm thick-
ness, and the mild steel was supplied by Lisnave
Shipyard. The stiffened plates were manufactured
at Lisnave Shipyard according to the standard
techniques of the shipyard (Gordo and Guedes
Soares 2008).

3 EXPERIMENTAL WORK

The experiment work that is used here as reference

for the numerical calculations have been conducted
by Gordo and Guedes Soares (2008) and a brief
account is given here for completeness.
A 300 ton hydraulic press was used to perform
the tests of the panels under uniaxial compres-
sion. Figure 2 shows the general arrangement of
the tests (left) and a detailed view of the support
for the framing systems which intends to repro-
duce simply supported boundary conditions. The
Figure 1. Geometry of FS, BS, LS and US for stiffened lateral edges of the panels are totally free to move
panels.
out-of-plane and to rotate. This means that large
Table 1. Geometry and material of stiffened panels.
panels (B series) should be less affected by the lack
of effectiveness at the lateral plating edges during
Plate buckling. In fact, the percentage of the total cross-
section area with reduced effectiveness due to
Dim (mm) σY unsupported lateral edges is lower on the wide pan-
els than in the narrow ones and thus, the expected
FS3-A3 300 × 900 × 4 690
ultimate load is higher for the wide panels.
FS3-B3 600 × 900 × 4 690
The transverse framing system is simply sup-
BS3-A3 300 × 900 × 4 690
ported in a U bar in each side, allowing longi-
BS3-B3 600 × 900 × 4 690
tudinal displacement and rotation but avoiding
LS3-A3 300 × 900 × 4 690
out-of-plane displacement from the initial plane
LS3-B3 600 × 900 × 4 690
of load. The loaded top edges have full con-
US3-A3 300 × 900 × 4 690
tact with the steel beds corresponding to nearly
US3-B3 600 × 900 × 4 690
clamped conditions, at least until collapse, due to
Stiffener
Dim (mm) σY
FS3-A3 I 20 × 4 690
FS3-B3 I 20 × 4 690
BS3-A3 I 30 × 8 343
BS3-B3 I 30 × 8 343
LS3-A3 L38 × 19 × 4 296
LS3-B3 L38 × 19 × 4 296
US3-A3 U (40 × 150 × 40) × 2 200
US3-B3 U (40 × 150 × 40) × 2 200

− Fully S690 structure: S690, on plating and bar

stiffeners. Figure 2. Setup of the test of stiffened plates (left).
− Hybrid bar structure: S690, on plating and mild Details of the lateral support of the frames in vertical
steel on bars. guides (right).

222

MARSTRUCT.indb 222 2/18/2011 5:46:36 PM

 the bi-dimensional geometry of the cross section
of the panels. The hydraulic flow was controlled
manually due to limitations of the control device
which means that the shortening rate was not con-
stant during the tests.
There are six stiffened panel in experiment,
including FS3-B3, BS3-A3, BS3-B3, LS3-A3,
LS3-B3 and US3-B3.
4 NONLINEAR FINITE ELEMENT
ANALYSIS
4.1 Finite element model
The geometric and material nonlinearities are
both taken into account, including elastic-plastic
large deflection. The material property assumed
use the characteristic values of yield strength and
Young’s Modulus, where appropriate, a bi-linear
isotropic elastic-plastic material model excluding
strain rate effects is to be used. A plastic tangent
modulus of 1000 MPa is acceptable for normal
and higher strength steel (ABS 2006). The follow-
ing are the material properties: Young’s modulus,
E = 200 GPa; Tangent modulus, ET = 10 GPa;
Poison’s ratio, v = 0.3.
The FE code used for simulation is ANSYS/
Mechanical. This is a widely used finite element
code for nonlinear structural analyses. The shell
Figure 3. The FE model of ‘U’ Series stiffened panel.
223
MARSTRUCT.indb 223
element mesh should be fine enough to properly
describe the model shape (also after deformation).
Therefore, a balance between required accuracy
and efforts is needed. It is considered that the
element size to thickness ratio (usually at least 5)
(ISSC 2009). Figure 3 only shows three bays stiff-
ened panels which were used in the experiment.
4.2 The boundary condition and loading
To prescribe appropriate boundary conditions
is a main challenge in modeling stiffened panel
by experiment and finite elements. Because the
boundary of stiffened panels is supported by
strong members such as longitudinal girders and
transverse frames, the simply supported boundary
condition is often adopted. But the degree of rota-
tional restraint at the panel boundary is not equiv-
alent to zero. It is important to model the panel
edge condition in a relevant way. Figure 2 shows
that the loaded top edges have full contact with
the steel beds corresponding to nearly clamped
conditions, at least until collapse, due to the
bi-dimensional geometry of the cross section of
the panels. The coordinate and model is show in
Figure 4. The following are the boundary condi-
tion of stiffened panel:
− A–A1 at the stiffener and plate: UX, UY, UZ, RX,
RY and RZ.
− B–B1 at the stiffener and plate: UY, UZ, RX, RY,
RZ and equal x-displacement by coupling UX
degree.
− C–C1 and D–D1 at t transverse frame and plate
intersection: UZ.
The pressure in the z direction is applied on the
edge of the plate and stiffener. The pressure value
of the plate is Pp, and then the pressure value on
stiffener is Ps, Ps = Pp × tw/tp, where tw is the thick-
ness of web and tp is the thickness of plate.
Figure 4. The coordinate and model of stiffened panel.
2/18/2011 5:46:37 PM
 4.3 Geometrical initial imperfections
The imperfections are caused during a complex
fabrication process and are subject to significant
uncertainty related to the magnitude and spatial
variation. These initial imperfections affect sig-
nificantly the ultimate strength of stiffened panel
and should be accounted for. The buckling model
component of the deflected shape has the most
significant weakening effect (Guedes Soares and
Soreide 1983).
Kmiecik (1971) considered the initial deflection
as the superimposition of the Fourier components
for the first time. The behavior of plates subjected
to buckling loads depends to a considerable degree
on the shape of their initial deflection (Kmiecik
1995). The importance of their research is that
the buckling mode component can be isolated.
Three types of initial out-of-straightness should be
accounted for plate and stiffener imperfections, as
following (Paik 2009):
− Hungry horse mode initial deflection of local
plate panels.
The transverse imperfect displacement field of
plates can be normally represented by a double
Fourier series as follows:
b ⎛ πx⎞ ⎛ π y⎞
wopl = sin ⎜ ⎟ siin ⎜ ⎟ (1)
200 ⎝ a ⎠ ⎝ b ⎠
− Column-type initial deflection of stiffeners
a ⎛ πx⎞
w0c = sin ⎜ ⎟ (2)
1000 ⎝ a ⎠
− Side-ways initial deflection of stiffeners due to
angular rotation about panel-stiffener intersec-
tion line
a ⎛ πx⎞
w0 s = sin ⎜ ⎟
1000 ⎝ a ⎠ (3)
where a is the plate length, b is the plate width.
To get the initial imperfection in FEM analysis,
the shapes of initial imperfections are divided into
plate initial deflection, column-type initial distor-
tions of stiffeners and sideways initial distortions
of stiffeners. Firstly, linear buckling analysis is
performed for the target stiffened panel and find
out the related buckling modes of plating and stiff-
ener. Then the geometry properties, for example
the thickness of plates and stiffener, are changed
to decouple those deformations of interest from
224
MARSTRUCT.indb 224
lower eigenmodes and get desired shapes for plate
out-of-plane and stiffener out-of-plane deforma-
tions. The three types of initial distortions are
superimposed altogether FEM model.
5 COMPARISION OF RESULTS BETWEEN
EXPERIMENT AND FEM ANALYSIS
The strength of the panel was obtained by sum-
ming the reaction force on each node (Ri), on
the opposite boundary were the load is applied
and divided by the sectional area of the stiffened
panel (At):
N
∑ Ri
R i =1
σ= = (4)
At At
The following will compare the results of FEM
analysis and the test for three bays model. That
cause the value of dL/L between test and simula-
tion is different. So the initial shortening at low
stresses should be moved in test data.
5.1 FS series results
The very long initial shortening at low stresses
shown in the tests is due to the rearrangement of
the test setup until every part of the panel, support
and hydraulic machine is in full contact and it was
partially removed from the graphics due to lack of
interest.
Figure 5 and Figure 6 show the average stress
and shortening between experiment and FEM
analysis for ‘FS’ series. Average stress-shortening
curve shows a linear behavior until the ultimate
compressive stress was achieved. The ultimate
strength is similar between test and simulation, but
the stiffness in FEM analysis is bigger than in test.
Furthermore, the average stress-shortening curve
at unload phase are still different between them.
The ultimate strength of ‘F’ series is the lowest
in the four stiffener series. The stiffeners experience
lateral buckling and then induce panel failure from
Figure 7. It can be seen that the plate and stiffener
as a unit collapses. This phenomenon explains that
the collapse was sudden resulting from the yielding
of the stiffeners. It needs to the stronger stiffen-
ers to avoid the whole collapse when the spacing
of the panel is increased. The value of the 1three
bays model with different boundary condition are
all similar. It means that the support at the frame
z direction and the symmetric boundary condi-
tion on the lateral edge effect the ultimate strength
slightly for ‘FS’ series.
2/18/2011 5:46:37 PM
 350 FS3-B3 TEST 500 BS3-A3
FEM TEST
300 FEM
400
250
Stress (Mpa)

Stress (Mpa)
200 300

150
200
100
100
50

0 0
0 1 2 3 4 5 6 0 5 10
3 3
dL/L(10 ) dL/L(10 )

Figure 5. Average stress-shortening curve of experi- Figure 8. Average stress-shortening curve of experi-
ment and FEM for FS3-B3. ment and FEM for BS3-A3.

350

300 FS3-A3 500 BS3-B3

TEST
250 FEM
400
Stress (Mpa)

200
Stress (Mpa)

150 300

100
200
50
100
0
0 2 4 6 8 10 12
3
dL/L(10 ) 0
0 5 10
3
Figure 6. Average stress-shortening curve of FEM for dL/L(10 )
FS3-A3.
Figure 9. Average stress-shortening curve of experi-
ment and FEM for BS3-B3.

series specimen. The stiffeners are thick bars made

Figure 7. Von Mises stress distributions at the ultimate
limit state for ‘FS’.
of steel of 343 MPa yield stress and the associ-
ated plate is made of 690 MPa nominal yield stress
steel. There was a very great discharge of load after
the ultimate load was achieved during the devel-
opment of large out of plane deformations of the
panel between frames. Once the panels collapse,
from Figure 8 and Figure 9, the values of ultimate
strength drop down very quickly in the test and
FEM analysis. The ultimate strength of BS series
specimen is bigger than the FS series. The collapse
of the BS3 panel is due to stiffener induced fail-
ure and generates large transverse loads on the
transverse frame supports, shows in Figure 10 and
Figure 11.

5.2 BS series results 5.3 LS series results

Figure 8 and Figure 9 show the average stress and Figure 12 and Figure 13 show the experimental
shortening of test and FEM analysis for the ‘BS’ average stress and shortening curve and the FEM

225

MARSTRUCT.indb 225 2/18/2011 5:46:39 PM

 LS3-B3
500 TEST
FEM

400

Stress (Mpa)
300

200

100

0
0 5 10
3
Figure 10. Von Mises stress distributions at the ultimate dL/L(10 )
limit state for BS3-A3.
Figure 13. Average stress-shortening curve of experi-
ment and FEM for LS3-B3.

Figure 11. Von Mises stress distributions at the ultimate

limit state for BS3-B3.

Figure 14. Von Mises stress distributions at the ultimate

LS3-A3
TEST limit state for LS3-A3.
500
FEM

400
test. The main reason may be the residual stresses
Stress (Mpa)

300
in stiffened panel, and this aspect is not consid-
ered in FEM analysis. The ultimate strength of LS
200 series specimen is bigger than the BS series.
The buckling of plate occurs in the middle span
100 of the panels, and then induces stiffener failure
in the ‘LS’ series specimen from Figure 14 and
0 Figure 15. The series of panels reinforced with L
0 5 10 stiffeners allowed obtaining results as expected
dL/L(10 )
3
that the ‘L’ series have good capability to avoid
buckling.
Figure 12. Average stress-shortening curve of experi-
ment and FEM for LS3-A3.
5.4 US series results
Figure 16 and Figure 17 show the average stress
analysis for LS series specimen. The slope of the and shortening curve of the test and the FEM
average stress-shortening curve is different between analysis for the ‘US’ series specimen.
FEM analysis and the test. The stiffness of the The slopes of the average stress-shortening
FEM analysis is bigger than the stiffness of the curve are also different between FEM analysis and

226

MARSTRUCT.indb 226 2/18/2011 5:46:40 PM

 Figure 15. Von Mises stress distributions at the ultimate
Figure 18. Von Mises stress distributions at the ultimate
limit state for LS3-B3.
limit state for US3-B3.
US3-B3
TEST
400
FEM
in Figure 18, and then the overall collapse is very
300 sudden in US series specimen.
Stress (Mpa)

200 6 CONCLUSIONS

100 The ultimate strength of three bays stiffened panels

are compared with the experimental results. Several
modes of collapse were observed in each panel in
0 the A and B series numerical results which are the
0 1 2 3 4 5 6 7 8
same as the experiments. Column-induced collapse
3
dL/L(10 ) modes occur on FS and BS series. In the design,
the stiffener should be stronger to avoid column-
Figure 16. Average stress-shortening curve of experi- induced collapse and the aspect ratio should be
ment and FEM for US3-B3. considered. However, stiffener flange plate col-
lapses first on US models. The collapse is related
400
with not only the aspect ratio but also the type
US3-A3 of stiffener. The best results in terms of ultimate
strength were obtained for LS panels. ‘L’ series
300
stiffeners have good capability to avoid buckling
and they are worth of further research.
Stress (Mpa)

200
The ultimate strength of ‘U’ series stiffeners
drops down very quickly in both test and FEM
analysis. The ultimate strength stress distribution
100 pictures of the ‘U’ series show that buckling occur
in the flange occur very early. ‘U’ series stiffeners
have high ultimate strength value, but it’s necessary
0 to further research to determine their suitable
0 2 4 6 8 10 dimension.
3
dL/L(10 )

Figure 17. Average stress-shortening curve of FEM for ACKNOWLEDGEMENTS

US3-A3.
This work contributes to the activities of
the test. The ‘US’ series has U stiffeners of 2 mm MARSTRUCT VIRTUAL INSTITUTE, (www.
thickness. Thus, the slenderness of the flange plat- marstruct-vi.com) in particular its Technical Sub-
ing is higher than the slenderness of the associated committeee 2.3 on Ultimate Strength and 3.3
plating. The flange plate occurs buckling, showed Experiments on Ultimate Strength.

227

MARSTRUCT.indb 227 2/18/2011 5:46:43 PM

 The first author has been financed by the
Portuguese Foundation for Science and Technol-
ogy (Fundação para a Ciência e Tecnologia), under
contract SFRH / BD / 65120/ 2009.
REFERENCES
ABS 2006. Rules for building and classing, steel vessels.
Faulkner, D. 1977. Compression tests on welded eccen-
trically stiffened plate panels. In: Dowling P.J., et al.,
editors. Steel Plated Structures. London: Crosby
Lockwood Staples; pp. 130–9.
Gordo, J.M. & Guedes Soares, C. 2007. Experimental
evaluation of the behavior of a mild steel box girder
under bending moment. In: Guedes Soares, C., Das,
P.K., (Eds). Advancements in Marine Structures.
Taylor and Francis; pp. 377–383.
Gordo, J.M. & Guedes Soares, C. 2008. Compressive
tests on short continuous panels, Marine Structures,
21, 113–137.
Guedes Soares, C. & Soreide, T.H. 1983. Behaviour and
Design of Stiffened Plates under Predominantly Com-
pressive Loads, International Ship building Progress,
Vol. 300-January 1983 – No. 341.
Horne, M.R. & Narayanan, R. 1976. Ultimate capacity
of stiffened plates used in girders. Proc Inst Civil Eng
1976; 61:253–280.
ISSC 2009, 17th International ship and offshore structures
congress 16–21 August 2009, committee III.1 ultimate
strength.
228
MARSTRUCT.indb 228
Janssen, G.T.M. 2000. Fatigue based design rules for the
application of high tensile steel in ships. In: Proceed-
ings of the 7th international marine design conference,
Korea, 2000. pp. 317–328.
Paik, J.K. 2009. Nonlinear finite element method models
for ultimate strength analysis of steel stiffened-plate
structures under combined biaxial compression and
lateral pressure actions—Part II: Stiffened panels,
Thin-Walled Structures 47, 998–1007.
Kmiecik, M. 1971. Behaviour of axially loaded simply
supported long rectangular plates having initial defor-
mations, Report No. R84, Ship Research Institute,
Trondheim.
Kmiecik, M., Jastrzebski, T. & Kuzniar, J. 1995. Statis-
tics of Ship Plating Distortions, Marine Structures 8,
119–132.
Luís, R.M., Guedes Soares, C. & Nikolov, P.I. 2008a.
Collapse Strength of Longitudinal Plate Assemblies
with Dimple Imperfections. Ships and Offshore Struc-
tures. 3(4):359–370.
Luís, R.M., Witkowska, M. & Guedes Soares, C. 2008b.
Ultimate Strength of Transverse Plate Assemblies
Under Uniaxial Loads. Journal of Offshore Mechanics
and Arctic Engineering 130(2):021011-1-021011-7.
Mathewson, J. & Vinner, A. 1962. The strength and stiff-
ener of plating stiffened by flat bars. Part 1: axial com-
pressive loading tests. Report 392. UK: BSRA.
Smith, C. 1979. Compressive strength of welded steel
ship grillages. Trans RINA 117:325–59.
2/18/2011 5:46:45 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Numerical study of the effect of geometry and boundary

conditions on the collapse behaviour of short stiffened panels

Mingcai Xu & C. Guedes Soares

Centre for Marine Technology and Engineering (CENTEC), Instituto Superior Técnico,
Technical University of Lisbon, Lisbon, Portugal

ABSTRACT: A numerical study is conducted to analyze the influences of the stiffener’s geometry and
panel boundary conditions on the ultimate strength of stiffened panels under compression. The four types
of short stiffened panels analyzed are made of mild or high tensile steel and have bar, ‘L’ and ‘U’ stiffeners.
To understand the effect of finite element modeling on the ultimate strength of the stiffened panels, four
types of models with different geometry are investigated in the FE analysis. Moreover, different boundary
conditions for the same FE model are also investigated. From the results of the FE analysis, two of the
four models studied can produce adequate boundary conditions at the loaded edges. The stiffened panels
with different combinations of mechanical material properties and geometric configurations are consid-
ered. The initial geometric imperfection affects the collapse behaviour of stiffened panel and is analyzed
in FE simulation.

1 INTRODUCTION stiffened panels is supported by strong members

Stiffened panels are very popular structures in
marine structures and their load carrying capac-
ity are important from the viewpoint of safety and
economy. The strength to weight ratio is an impor-
tant index to design economical and efficient ship.
In order to reproduce adequate working condi-
tions on a ship structure, the boundary conditions
on the loaded top edges and unloaded lateral edges
should be considered carefully.
FE codes have been used to analyze the stress
distributions and deformation of very complicated
structures with the accuracy demanded in engineer-
ing applications under all kind of loading condi-
tions. They are also a suitable tool for assessing the
ultimate strength of ship structures. The advanced
buckling analysis method is to be based on nonlin-
ear analysis techniques or equivalent, which predict
the complex behaviour of stiffened and unstiffened
panels (IACS_CSR 2006). Namely, the extent of
the model used in the buckling assessment is to be
sufficient to account for the structure that is sur-
rounding the panel of interest, and to reduce the
uncertainties introduced through the boundary
conditions. In general, the model is to include more
than one stiffener span in the stiffener direction and
the portion between two primary support members
in the direction normal to the stiffeners. To pre-
scribe appropriate boundary conditions is a main
challenge in modeling stiffened panels by experi-
ment and finite elements. Because the boundary of
such as longitudinal girders and transverse frames,
the simply supported boundary condition is often
adopted. But the degree of rotational restraints at
the panel boundary is not equivalent to zero. It is
important to model the panel edge condition in a
relevant way.
To circumvent this problem, the tests of Gordo
and Guedes Soares (2008) used specimen with
three bays longitudinally. The use of three-bay
panels instead of one single-bay panels allows to
have more realistic results by avoiding boundary
conditions problems for the central plates related
to eccentricity of load and to include the interfer-
ence between adjacent panels, which was found to
be significant by Luís et al. (2008a, b).
In this study two kinds of models are investi-
gated which are 1 + 1 bays and 1/2 + 1 + 1/2 bays
in the longitudinal direction. The 1 + 1 bays model
consists of two full bays, while the 1/2 + 1 + 1/2
bays model consists of one full bay plus two half
bays.
A series of nonlinear finite element method
computations were carried out in ISSC (2009)
in two full bays (1 + 1 bays) model with various
parameters of influence to investigate the ultimate
strength characteristics of stiffened panels repre-
sentative of ship hulls.
Zhang & Khan (2009) and Fujikubo (2005) ana-
lyzed the ultimate strength of plates using non-linear
FE software by one full bay plus two half bays
(1/2 + 1 + 1/2 bays) model. Tanaka and Endo (1988)

229

MARSTRUCT.indb 229 2/18/2011 5:46:45 PM

 carried out a series of experimental and numerical
FE investigations on the ultimate compressive
strength characteristics of longitudinally stiffened
panels having three flat bar stiffeners with three
bays, which were intended to fail by local buckling
or tripping of longitudinal stiffeners.
To understand the influence of boundary condi-
tions and model geometry, the ultimate strengths
of stiffened panels under axial compression are cal-
culated for 120 specimens with different boundary
conditions and model geometry. These stiffened Figure 3. Geometry of 1 bay.
panel models include 3 bays, 1/2 + 1 + 1/2 bays,
1 + 1 bays and 1 bay. The plate is always very high Table 1. Geometry and material of stiffened panels.
strength steel (S690) but the stiffeners are made of
mild or high tensile steel for bar stiffeners and mild Plate
steel for ‘L’ and ‘U’ stiffeners. The base geometry
Sample Dim (mm) σY
is the one used on the box girders tests Gordo and
Guedes Soares (2007) (2008). FS3-I3 (300 × i) × (400 × 3) × 4 690
FS3-I21 (300 × i) × (200 + 400 + 200) × 4 690
FS3-I22 (300 × i) × (400 + 400) × 4 690
2 DESCRIPTION OF THE MODELS FS3-I1 (300 × i) × 400 × 4 690
FOR THE ANALYSIS BS3-I3 (300 × i) × (400 × 3) × 4 690
BS3-I21 (300 × i) × (200 + 400 + 200) × 4 690
Figures 1, 2 and 3 show the geometry of the dif- BS3-I22 (300 × i) × (400 + 400) × 4 690
ferent panels.
BS3-I1 (300 × i) × 400 × 4 690
Table 1 shows the geometry and material of
LS3-I3 (300 × i) × (400 × 3) × 4 690
stiffened panels. The A-E means different number
LS3-I21 (300 × i) × (200 + 400 + 200) × 4 690
of stiffener.
LS3-I22 (300 × i) × (400 + 400) × 4 690
LS3-I1 (300 × i) × 400 × 4 690
US3-I3 (300 × i) × (400 × 3) × 4 690
US3-I21 (300 × i) × (200 + 400 + 200) × 4 690
US3-I22 (300 × i) × (400 + 400) × 4 690
US3-I1 (300 × i) × 400 × 4 690

Stiffener

– Dim (mm) σY

FS3-I3 I 20 × 4 690
FS3-I21 I 20 × 4 690
FS3-I22 I 20 × 4 690
FS3-I1 I 20 × 4 690
Figure 1. Geometry of 1/2 + 1 + 1/2 and 3 bays. BS3-I3 I 30 × 8 343
BS3-I21 I 30 × 8 343
BS3-I22 I 30 × 8 343
BS3-I1 I 30 × 8 343
LS3-I3 L38 × 19 × 4 296
LS3-I21 L38 × 19 × 4 296
LS3-I22 L38 × 19 × 4 296
LS3-I1 L38 × 19 × 4 296
US3-I3 U (40 × 150 × 40) × 2 200
US3-I21 U (40 × 150 × 40) × 2 200
US3-I22 U (40 × 150 × 40) × 2 200
US3-I1 U (40 × 150 × 40) × 2 200

Note: The value of I and i: i = 1 when I = A, i = 2 when

I = B, i = 3 when I = C, i = 4 when I = D, i = 5 when I = E;
Figure 2. Geometry of 1 + 1 bays. Frame dimension—L50 × 20 × 6 (mm).

230

MARSTRUCT.indb 230 2/18/2011 5:46:45 PM

 Four series of FEM analysis were carried out to thickness ratio (usually at least 5) (ISSC 2009).
using two different types of steel as follows: Figure 4 only shows three bays stiffened panels.
− Fully S690 structure: S690, on plating and bar
stiffeners. 3.2 The boundary conditions and loading
− Hybrid bar structure: S690, on plating and mild
The stiffened panel models including 3 bays,
steel on bars.
1/2 + 1 + 1/2 bays, 1 + 1 bays and 1 bay, are simu-
− Hybrid L structure: S690, on plating and mild
lated with different boundary conditions, as shown
steel on L stiffeners.
in Table 2. To investigate the effect of model
− Hybrid U structure: S690, on plating and mild
geometry and boundary condition on the col-
steel on U stiffeners.
lapse behavior of stiffened panel, nine cases are
calculated in ANSYS.
3 NONLINEAR FINITE ELEMENT
ANALYSIS
1
ELEMENTS 1
ELEMENTS
Y FEB 4 2010 Y
21:39:54 FEB 4 2010
21:41:12
Z X Z X

3.1 Finite element model

To investigate the influence of different models and
boundary conditions, 3 bays, 1/2 + 1 + 1/2 bays,
1 + 1 bays and 1 bay stiffened panel are simulated
in FEM analysis. BS2-A BS2-B

The geometric and material nonlinearities are

(a) The FE model of ‘F’ and ‘B’Series
both taken into account, including elastic-plastic 1
ELEMENTS
1
ELEMENTS
FEB 4 2010

large deflection. The material properties assumed

FEB 4 2010 21:44:18
21:42:56
Y
Y
Z X
Z X

use the characteristic values of yield strength and

Young’s Modulus. Where appropriate, a bi-linear
isotropic elastic-plastic material model excluding
strain rate effects is to be used. A plastic tangent
modulus of 1000 MPa is acceptable for normal LS2-A LS2-B

and higher strength steel (ABS 2006). (b) The FE model of ‘L’ Series
The following are the material properties: 1
ELEMENTS
FEB 4 2010
1
ELEMENTS

Young’s modulus, E = 200 GPa; Tangent modu-

FEB 4 2010
21:44:40
Y Y 21:45:06

Z X Z X

lus, ET = 10 GPa; Poison’s ratio, v = 0.3. The FE

code used for simulation is ANSYS/Mechanical.
This is a widely used finite element code for non-
linear structural analyses. The shell element mesh
should be fine enough to properly describe the US2-A US2-A

model shape (also after deformation). Therefore, (c) The FE model of ‘U’ Series
a balance between required accuracy and efforts
is needed. It is considered that the element size Figure 4. The FE model of continuous stiffened panel.

Table 2. The boundary condition of FE model.

2 bays
3 bays 2 bays (1 + 1) 1bay

Boundary condition C1 C2 C3 C4 C5 C6 C7 C8 C9

A-A1: UX,UY,UZ,RX, RY and RZ √ √ √ × √ × √ × ×

A-A1: UX, UY, UZ, × × × √ × √ × √ √
B-B1: UY, UZ, RX, RY, RZ and equal Ux √ √ √ × √ × √ × ×
B-B1: UY, UZ, equal Ux × × × √ × √ × √ √
C, C1, D, and D1 on frame: UZ √ √ √ √ √ √ × × ×
C, C1 on frame: UZ × × × × × × √ √ ×
The intersection between frame and plate: UZ × √ √ √ √ √ √ √ ×
AB, A1B1 edge: UY, RZ and RX × × √ √ √ √ √ √ √

Note: Different model and boundary condition correspond to different location of stiffener.

231

MARSTRUCT.indb 231 2/18/2011 5:46:46 PM

 The pressure in the z direction is applied on the
edge of the plate and stiffener. The pressure value
of the plate is Pp, and then the pressure value on
stiffener is Ps, Ps = Pp × tw/tp, where tw is the thick-
ness of web and tp is the thickness of plate. Table 2
shows all kinds of boundary conditions with
different models. The coordinate and model is
show in Figure 5.
3.3 Geometrical initial imperfections
It has generally been found that initial imperfec-
tions tend to decrease the rigidity and ultimate
strength of plates. These initial imperfections affect
significantly the ultimate strength of stiffened
panel and should be accounted for. The imper-
fections are caused during a complex fabrication
process and are subject to significant uncertainty
related to the magnitude and spatial variation. The
most accurate method is to use real measured data.
But it’s not always available.
Kmiecik (1971) considered the initial deflection
as the superimposition of the Fourier components
for the first time. The behavior of plates subjected
to buckling loads depends to a considerable degree
on the shape of their initial deflection (Kmiecik
1995). So the equivalent initial imperfection is
used. In most of the initial theoretical studies ini-
tial deflection assumed to have the same shape as
the buckling mode.
The following three types of initial deflection
are accounted for (Paik 2009):
− Hungry horse mode initial deflection of local
plate panels
The adopted range of values for the plate out-
of-plane and stiffener lateral/flexural imperfection
magnitudes correspond to recommended values.
A statistical analysis of the initial distortions of the
ship plates shows that the majority of the plates,
(around 90%), particularity square plates, have
Figure 5. The FE model.
232
MARSTRUCT.indb 232
shapes of initial distortions with one half wave
in the longitudinal and transverse direction. The
transverse imperfect displacement field of plates
can be normally represented by a double Fourier
series as following:
b ⎛ πx⎞ ⎛ π y⎞
wopl = sin ⎜ ⎟ siin ⎜ ⎟ (1)
200 ⎝ a ⎠ ⎝ b ⎠
− Column-type initial deflection of stiffeners
a ⎛ πx⎞
w0c = sin ⎜ ⎟ (2)
1000 ⎝ a ⎠
− Side-ways initial deflection of stiffeners due to
angular rotation about panel-stiffener intersec-
tion line
a ⎛ πx⎞
w0 s = sin ⎜ ⎟ (3)
1000 ⎝ a ⎠
where a is the plate length, b is the plate width.
To get the initial imperfection in FEM analysis,
the shapes of initial imperfections are divided into
plate initial deflection, column-type initial distor-
tions of stiffeners and sideways initial distortions
of stiffeners. Firstly, linear buckling analysis is
performed for the target stiffened panel and find
out the related buckling modes of plating and stiff-
ener. Then the geometry properties, for example
the thickness of plates and stiffener, are changed
to decouple those deformations of interest from
lower eigenmodes and get desired shapes for plate
out-of-plane and stiffener out-of-plane deforma-
tions. The three types of initial distortions are
superimposed altogether FE model.
4 THE RESULTS OF THE FEM ANALYSIS
The total force at the boundary/compressed edge is
determined by the following:
N
R ∑ Ri (4)
i =1
where N is the number of node at boundary; Ri is
the reaction force on each node.
The stress at the boundary is that total reac-
tion force (R) is divided by the sectional area of
the stiffened panel (At), Eq. (6). These calculations
must be performed for each step of the non-linear
analysis.
R (5)
σ=
At
2/18/2011 5:46:47 PM
 Figure 6 shows the ultimate strength of the three
bays models with C1 boundary condition. From
the result of 3bay-C1 shown, increasing the width
of panel, namely number of stiffeners and length
of the frame, decreases the ultimate strength of the
stiffened panel. The displacements in the z direc-
tion of the frame are not equal to zero, which is a
result of the stiffness of the frame. The displace-
ment in the z direction of frame increases with the
length of frame increasing, as shown in Figures 7
and 8. So when designing the panel specimen, the
dimension of the frame should be considered care-
fully to make sure that the stiffness of the frame can
support the stiffener of panel. In other words, the
frame should be included in the ultimate strength
simulation of panel instead of boundary condition Figure 8. Von mises stress at the ultimate limit state
when the width of panel is very large. for C1.
The 3bay-C2 model is designed to understand
the effect of panel width on the ultimate strength
of the stiffened panel. To ignore the influence of at the intersection nodes of the plate and frame for
frame stiffness and to focus on the effectiveness at 3bay-C2 model. The ultimate strength increases
the lateral plating edges, the z degree is constrained slowly with increasing width of the panel in the
3bay-C2, shown in Figure 9, because the lateral
edges of the panels are totally free to move out-of-
FS3 plane and to rotate. This means that larger panels
500 BS3
LS3 should be less affected by the lack of effectiveness
450
US3 at the lateral plating edges during buckling. In fact,
the percentage of the total cross-section area with
400 reduced effectiveness due to unsupported lateral
Stress (MPa)

350 edges is lower on the wide panels than in the nar-

row ones and thus, the expected ultimate load is
300 higher for the wider panels. The ultimate strength
250 of 3bay-C2 is commonly bigger than the result of
3bay-C1. So it would be too optimistic if the inter-
200 section nodes between frame and plate is support
150 z direction displacement.
100 200 300 400 500 600 700 800 Figure 10 shows the ratio of ultimate strength
Width (mm)
between C2 and C1. It is observed that the mean
value is 1.14 and standard deviation is 0.2. The
Figure 6. Ultimate strength of three bays with C1. more the width of the panel increases, the big-
ger the ratio become between C2 and C1. When
the width of panel is 600 mm, the biggest ratio
is 1.42. When the width of the panel is increased
to 750 mm, the biggest ratio is 2.08. This means
that the frame is also important in simulation and
should be included in FEM analysis when the
width of panel or frame is not strong enough.
The symmetric boundary condition is applied
on the lateral edges of the panels in C3 and C4 with
fixed and simply supported boundary condition at
the end edge of panel. This can avoid the lateral
edges to move out-of-plane and to rotate. The
mean value of the ratio between 3bays C4 and C3
is 0.97. This shows that the boundary condition at
the end of panel is affecting slightly the ultimate
strength.
The magnitude of the ultimate strength is almost
Figure 7. Displacement at the ultimate limit state for C1. the same with increasing width of the panel with

233

MARSTRUCT.indb 233 2/18/2011 5:46:49 PM

 500 FS3 550 FS3
BS3 BS3
LS3 LS3
US3 US3
450 500
Stress (MPa)

Stress (MPa)
400 450

350 400

300 350

250 300
100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800
Width (mm) Width (mm)

Figure 9. Ultimate strength of three bays with C2. Figure 11. Ultimate strength for three bays with C3.

2.5 500 FS3

BS3
FS3
LS3
BS3 US3
LS3
2.0 US3 Stress (MPa) 450

400
C2/C1

1.5

350
1.0

300
100 200 300 400 500 600 700 800
0.5
100 200 300 400 500 600 700 800 Width (mm)
Width (mm)
Figure 12. Ultimate strength for three bays with C4.
Figure 10. C2 divided by C1 for three bays.
Table 3. The ultimate strength of FEM analysis
for three bays (MPa).
either C3 or C4, as shown in Figure 11, Figure 12
and Table 3. That illustrates that the width wouldn’t Case C1 C2 C3 C4
affect the ultimate strength of the stiffened panel in
this circumstance. So the ‘A’ series stiffened panel FS3-A 308 302 326 325
with C3 and C4 for three bays can get reasonable FS3-B 303 322 355 327
result. FS3-C 277 319 356 332
The following calculations will focus on the FS3-D 264 334 356 332
geometry of the model and the ‘A’ and ‘B’ series is FS3-E 220 340 356 332
further studied. Table 3 shows the ultimate strength BS3-A 399 404 450 441
of the stiffened panel for ‘A’ and ‘B’ series. The BS3-B 391 414 448 443
results are almost the same between the ‘A’ and ‘B’ BS3-C 348 412 448 442
series. Due to symmetry the boundary condition BS3-D 366 413 448 442
is applied on the AB and A1B1 edges shown in BS3-E 358 447 451 442
Table 3. The mean value of ratio between three LS3-A 460 441 495 487
LS3-B 486 481 492 488
and two bays with C5 /C3 and C6 /C4 is 1.13 and
LS3-C 479 486 492 490
1.16, namely the result of 1/2 + 1 + 1/2 bays big-
LS3-D 439 487 492 490
ger than three bays with same boundary condition.
LS3-E 422 489 486 490
The biggest ratio is 1.43 and 1.46 for clamp and
US3-A 356 364 416 415
simply supported boundary condition. Figure 13
US3-B 366 415 415 414
presents a comparison of the ultimate strength US3-C 325 412 414 416
for one/two bays panel with clamped and sim- US3-D 303 412 416 414
ply supported boundary conditions at the end US3-E 197 410 411 413
edges of the panel. The ultimate strength of the

234

MARSTRUCT.indb 234 2/18/2011 5:46:50 PM

 1bay-C9 is the biggest and the 3bay-BC1 is the 1
NODAL SOLUTION Y
smallest in all models. The mean value of the ratio STEP=1
SUB =11
Z
MX
X
APR 8 2010
23:59:36

of C4/C3, C6/C5 and C8/C7 is 0.97, 0.99 and 0.98 TIME=.411938

SEQV (AVG)
DMX =.005734
respectively. This indicates that the clamped or SMN =.290E+07
SMX =.690E+09

simply supported boundary conditions on longi-

tudinal edges have only a minor influence on panel
strength, but when the edges are clamped the panel MN

strength increases slightly. The ultimate strength of

the ‘LS’ series is the biggest in all the panels. From
Figure 13, the results of ultimate strength are very
different with different geometric model. It is very
important to investigate what kind of model is
reasonable. BS3-C3
.290E+07
.792E+08
.156E+09
.232E+09
.308E+09
.385E+09
.461E+09
.537E+09
.614E+09
.690E+09

The collapse modes of these stiffened panels

are also different between the different models Figure 15. Von Mises stress distributions of the stiff-
and boundary conditions as shown in Figure 14 ened panel at the ultimate limit state for 3bay-C2.
to Figure 22. The collapse is similar independ-
ently of whether or not the intersection nodes
between plate and frame are supported for UZ, 1
NODAL SOLUTION
APR 23 2010
shown in Figure 14 and Figure 15. It shows that STEP=1
SUB =10
TIME=.447899
Y
MX
22:04:13

the symmetry boundary condition at the lateral SEQV (AVG)

DMX =.005094
Z X

edges of the panels affects not only the ultimate SMN =.652E+07
SMX =.690E+09

MN

3Bay-C1
600 3Bay-C2
3Bay-C3
3Bay-C4
550 2Bay1-C5
2Bay1-C6
2Bay2-C7
500 2Bay2-C8
Stress (MPa)

1Bay-C9

450 .652E+07 .158E+09 .310E+09 .462E+09 .614E+09

.825E+08 .234E+09 .386E+09 .538E+09 .690E+09
BS3-B3
400

Figure 16. Von Mises stress distributions of the stiff-

350
ened panel at the ultimate limit state for 3bay-C3.
300
FS3-A BS3-A LS3-A US3-A 1
NODAL SOLUTION
Stiffener type STEP=1 APR 30 2010
Y 18:18:23
SUB =10
TIME=.442745
Z X
Figure 13. The ultimate strength of FEM analysis with SEQV (AVG)
DMX =.004615
SMN =.566E+07
different case. SMX =.690E+09

1
NODAL SOLUTION
Y
STEP=1 APR 5 2010
SUB =8 MX 17:43:55
TIME=.390886 Z X
SEQV (AVG) MN
DMX =.008786 MX
SMN =.943E+07
SMX =.690E+09

.566E+07 .158E+09 .310E+09 .462E+09 .614E+09

MN .817E+08 .234E+09 .386E+09 .538E+09 .690E+09
BS3-B3

Figure 17. Von Mises stress distributions of the stiffened

panel at the ultimate limit state for 3bay-C4.

.943E+07 .161E+09 .312E+09 .463E+09 .614E+09

BS3-B3
.851E+08 .236E+09 .388E+09 .539E+09 .690E+09
strengths but also the collapse mode indicated in
Figure 15 and Figure 16. The collapse modes of
Figure 14. Von Mises stress distributions of the stiffened three and 1/2 + 1 + 1/2 bays are similar, but they are
panel at the ultimate limit state for 3bay-C1. both different from the 1 + 1 bays.

235

MARSTRUCT.indb 235 2/18/2011 5:46:52 PM

 1 1
NODAL SOLUTION NODAL SOLUTION
STEP=1 APR 28 2010 STEP=1 APR 29 2010
SUB =11 16:57:16 SUB =9 16:41:52
TIME=.512624 Y Y
TIME=.46757
SEQV (AVG)
DMX =.004137
SEQV (AVG)
SMN =.285E+07
Z X DMX =.003913 Z X
SMX =.690E+09 MX SMN =.150E+07
SMX =.690E+09

MN

MN

MX
.285E+07 .156E+09 .308E+09 .461E+09 .614E+09
.792E+08 .232E+09 .385E+09 .537E+09 .690E+09
BS3-B2
.150E+07 .154E+09 .307E+09 .460E+09 .613E+09
.780E+08 .231E+09 .384E+09 .537E+09 .690E+09
BS3-B2

Figure 18. Von Mises stress distributions of the stiffened

panel at the ultimate limit state for 2bay1-C5. Figure 21. Von Mises stress distributions of the stiffened
panel at the ultimate limit state for 2bay2-C8.
1
NODAL SOLUTION
APR 27 2010
STEP=1
22:02:45
SUB =10 Y
TIME=.510623 1
SEQV (AVG) Z X NODAL SOLUTION
DMX =.004217 MX APR 30 2010
STEP=1
SMN =675114 Y 22:15:25
SUB =10
SMX =.690E+09
TIME=.519808 Z X
SEQV (AVG)
DMX =.004657
SMN =.107E+09
SMX =.690E+09

MN
MN

MX

675114 .154E+09 .307E+09 .460E+09 .613E+09

.773E+08 .230E+09 .384E+09 .537E+09 .690E+09
BS3-B2

Figure 19. Von Mises stress distributions of the stiffened .107E+09

.172E+09
.237E+09
.302E+09
.366E+09
.431E+09
.496E+09
.560E+09
.625E+09
.690E+09
BS3-B1
panel at the ultimate limit state for 2bay1-C6.

Figure 22. Von Mises stress distributions of the stiffened

1
NODAL SOLUTION panel at the ultimate limit state for 1bay-C9.
STEP=1 APR 29 2010
SUB =11 00:55:34
TIME=.469795 Y
SEQV (AVG)
DMX =.004732 Z X
SMN =.601E+07
SMX =.690E+09

MN The collapse mode of 1 bay is different from the

other models. The 1 bay cannot consider the inter-
ference between adjacent panels and is not recom-
MX
mended in ultimate strength analysis of stiffened
panels.

.601E+07 .158E+09
.820E+08 .234E+09
BS3-B2
Figure 20. Von Mises stress distributions of the stiffened
panel at the ultimate limit state for 2bay2-C7.
The unsymmetrical collapse modes occur for
1 + 1 bays in Figures 20 and 21. This phenomenon
will increase the uncertainty of the result. So 3 and
1/2 + 1 + 1/2 bays stiffened panels are more robust
than 1 + 1 bays in this circumstance.
5 CONCLUSIONS
.310E+09 .462E+09 .614E+09
.386E+09 .538E+09 .690E+09
Nine geometric configurations of stiffened panels
are investigated, including 3 bays, two (1/2 + 1 + 1/2)
bays, 1 + 1 bays and 3 bay with different boundary
conditions. From the results of the 120 calculations,
some conclusions can be obtained. The frame on
the boundaries of the plate is also important in
the simulation and should be included in the FEM
analysis when the width of panel or frame is not
strong enough, as resulted from the comparisons
between C1 and C2.

236

MARSTRUCT.indb 236 2/18/2011 5:46:54 PM

 The mean value of the ratio of C4/C3, C6/C5
and C8/C7 is 0.97, 0.99 and 0.98 respectively. This
indicates that the clamped or simply supported
boundary conditions on the longitudinal edges
have only a minor influence on panel strength,
but when the edges are clamped the panel strength
increases slightly. The loading rate of average
stress-shortening for 1bay-C9 is the biggest in
all cases and very different from the experiment.
The one bay can not account for the interference
between adjacent panels and is not recommended
in ultimate strength analysis of stiffened panel. The
unsymmetrical collapse modes occur in 1 + 1 bays.
This phenomenon will increase the uncertainty of
the result.
So, 3 and 1/2 + 1 + 1/2 bays stiffened panels are
more robust than 1 + 1 bays for this circumstance.
The result of 1/2 + 1 + 1/2 bays are higher than
3 bays with same boundary condition. Three-bay
panels allows to have more realistic results by
avoiding boundary conditions problems for the
central plates related to eccentricity of load and to
include the interference between adjacent panels.
ACKNOWLEDGEMENTS
This work contributes to the activities of
MARSTRUCT VIRTUAL INSTITUTE, (www.
marstruct-vi.com) in particular its Technical Sub-
committee 2.3 on Ultimate Strength.
The first author has been financed by the
Portuguese Foundation for Science and Technol-
ogy (Fundação para a Ciência e Tecnologia), under
contract SFRH / BD / 65120/ 2009.
REFERENCES
ABS, 2006. Rules for building and classing, steel vessels.
Fujikubo, M. 2005. Estimation of ultimate strength of
continuous stiffened panel under combined transverse
thrust and lateral pressure Part 2: Continuous stiff-
ened panel, Marine Structures 18, 411–427.
Gordo, J.M. & Guedes Soares, C. 2007. Experimental
evaluation of the behavior of a mild steel box girder
under bending moment. In: Guedes Soares, C. &
Das, P.K. (Eds). Advancements in Marine Structures.
Taylor and Francis; pp. 377–83.
237
MARSTRUCT.indb 237
Gordo, J.M. & Guedes Soares, C. 2008. Compressive tests
on short continuous panels, Marine Structures 21,
113–137.
IACS CSR 2006. Common structural rules for oil
tankers.
ISSC 2009. 17th International ship and offshore structures
congress 16–21 August 2009, committee III.1 ultimate
strength.
Kmiecik, M. 1971. Behaviour of axially loaded simply
supported long rectangular plates having initial defor-
mations, Report No. R84, Ship Research Institute,
Trondheim, 1971.
Kmiecik, M., Jastrzebski, T. & Kuzniar, J. 1995. Statis-
tics of Ship Plating Distortions, Marine Structures 8,
119–132.
Luís, R.M., Guedes Soares, C. & Nikolov, P.I. 2008a.
Collapse Strength of Longitudinal Plate Assemblies
with Dimple Imperfections. Ships and Offshore Struc-
tures. 3(4): 359–370.
Luís, R.M., Witkowska, M. & Guedes Soares, C. 2008b.
Ultimate Strength of Transverse Plate Assemblies
Under Uniaxial Loads. Journal of Offshore Mechanics
and Arctic Engineering 130(2): 021011-1-021011-7.
Paik, J.K. 2009. Nonlinear finite element method models
for ultimate strength analysis of steel stiffened-plate
structures under combined biaxial compression and
lateral pressure actions—Part II: Stiffened panels,
Thin-Walled Structures 47, 998–1007.
Smith, C. 1979. Compressive strength of welded steel
ship grillages. Trans RINA 117: 325–59.
Tanaka, Y. & Endo, H. 1988. Ultimate strength of stiff-
ened plates with their stiffeners locally buckled in
compression. J Soc Naval Archit Jpn 164: 456–67.
Zhang, K.I. 2009. Buckling and ultimate capability
of plates and stiffened panels in axial compression,
Marine Structures 22, 791–808.
2/18/2011 5:46:56 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Hydro-elastoplasticity approach to ship’s hull girder collapse

behavior in waves

Weijun Xu, Kazuhiro Iijima & Masahiko Fujikubo

Department of Naval Architecture and Ocean Engineering, Graduate School of Engineering,
Osaka University, Japan

ABSTRACT: A new approach to predicting the dynamic collapse behavior of a ship’s hull girder in
waves is proposed. The work is to evaluate the consequence of the collapse, which is important in evaluat-
ing the associated risk of the hull girder collapse. The progressive collapse analysis of the ship’s hull could
be followed by a hydro-elastoplastic response analysis. In this research, the whole ship hull is modeled as a
two-rigid-bodies system connected by a rotational spring, which represents the nonlinear relation between
the displacement and the moment. Nonlinear strip theory is used to solve the force equilibrium of the two
bodies. The approach is validated against tank tests, which show that the collapse increases rapidly after
the ultimate strength is reached, and the plastic deformation grows until unloading starts and the bending
rigidity recovers. It is concluded that the fundamental dynamic collapse behavior can be followed by the
numerical method.

1 INTRODUCTION components was taken into account. Yao (1991 &

Hull girder strength is the most fundamental to
secure the safety of ships. International Asso-
ciation of Classification Societies (IACS) has
strengthened the rules for hull girder safety these
years. Unified Requirement (UR) on hull girder
strength was firstly introduced in 1989. Very
recently, Common Structural Rules for new ship
construction standards (CSR) was developed and
taken into effect. In CSR, ultimate strength check
has been introduced as a new criteria based on ulti-
mate limit state design.
Many researchers have devoted themselves to the
study of the hull girder capacity. The first attempt
to evaluate the ultimate strength of ship structure
was made by Caldwell (1965). He applied Rigid
Plastic Mechanism Analysis to evaluate the ulti-
mate strength of a ship’s hull girder, and accounted
for the effect of buckling by reducing the yielding
stress of the material at the buckled part. However,
Caldwell’s method does not account for the post-
collapse strength of the structural members which
significantly influence the collapse strength. This
problem was further addressed by Smith (1977).
Smith proposed another simple but efficient
method to analyze progressive collapse behavior
of box girder structures under longitudinal bend-
ing. The progressive loss in stiffness of a cross
section due to buckling and yielding of structural
1992) proposed a simple and practical analytical
method to simulate progressive collapse behavior
of a ship’ hull subjected to longitudinal bending
based on Smith’s method, so as to estimate the load
carrying capacity of a ship’ hull as a whole includ-
ing post-ultimate strength behavior. In the method,
progressive loss in rigidity due to the occurrence of
local buckling and yielding were taken into account,
and the deflection mode for a stiffener was not only
considered as an Euler buckling, but also flexural-
torsional buckling of a stiffener. Recent develop-
ments may be found in Yao (2006).
In order to assess the safety of ship structure,
evaluation of extreme loads is another important
aspect. As far back as Smith (1966) has measured
the wave induced loads on destroyers, the non-
linear result has clearly shown that the vertical
bending moments in sagging are larger than that
in hogging. Yamamoto (1978) regarded the ship
hull as a rigid body, and calculated the motion and
longitudinal strength of a tanker in head sea tak-
ing account of the nonlinearities. Then, the author
regarded the ship hull as an elastic beam theoreti-
cally taking account of nonlinear hydrodynamic
pressure. Jensen & Pedersen (1978) proposed a
nonlinear strip theory formulated in frequency
domain based on a perturbation procedure. Later,
this approach was extended to account for the
hydro-elasticity by Jensen & Pedersen (1981).

239

MARSTRUCT.indb 239 2/18/2011 5:46:57 PM

 When it is assumed that the severity of a hull which represents a nonlinear relation between the
girder collapse is directly related to the consequence rotational displacement and the moment. Then,
of the collapse, clarifying how the hull girder col- a series of tank tests are conducted by using the
lapses in waves, i.e., to what extent it collapses in scaled model.
waves is important in evaluating the associated
risk of the hull girder collapse. Thus, focus should
be put more on the extent of the failure with the 2 HYDRO-ELASTOPLASTICITY THEORY
increase of interests in risk-based design.
Progressive collapse behavior of a ship’s hull 2.1 Theoretical model
girder is usually analyzed in a quasi-static man-
The whole ship hull is modeled as a two-rigid-bodies
ner over the cross section by either load control
system connected each other by a nonlinear
or displacement control. However, as Lehman
rotational spring between them. The rotational
(2006) pointed out, the input of a curvature does
spring represents the nonlinear relation between
not adequately represent the failure process. That
the displacement and the moment including
indicates the actual path of the progressive col-
reduction of load carrying capacity due to buck-
lapse behavior is not followed by imposing forced
ling and yielding, increase of the capacity due to
displacement/rotation, neither by imposing force.
material strain-hardening, and recovery of the
Instead, it can be followed only when the interac-
bending rigidity when unloaded. Such a nonlinear
tion between the collapsing structure and loads is
relation is exemplified in Fig. 1.
considered (Yao 2009).
The theoretical model of ship’s hull is illustrated
A research field in which the interaction
as in Fig. 2. F1w and F2w represent the external loads
between the fluid and structure in an elastic range
on the respective parts, M1w and M2w represent the
is considered is called as hydro-elasticity. In the
external moments measured around the connecting
same manner, the present problem may be called
points on the respective parts. f and M represent
as hydro-elastoplasticity. In order to predict the
the reaction force and moment due to the spring,
collapse behavior including the path after the
respectively. u1 and u2 are the vertical displacements
ultimate strength more accurately, it is necessary
of two parts measured at the connection, θ1 and θ2
to solve both simultaneously taking account of
show the rotational displacement of the respective
the interaction between fluid and structure. The
parts.
structural deformation including plastic defor-
mation may have a significant influence on the
equilibrium between the ship and the hydrostatic/
dynamic forces. The external loads which exceed M
the ultimate strength need to be compensated by
the inertia forces of the ship body. Such a method
can be called a hydro-elastoplastic response Mu
analysis.
Masaoka (2003) developed such an analysis
system for elastic and plastic dynamic behavior
of hull girders, and compared the dynamic col-
lapse behavior of a damaged ship in large regular Unloading path
waves with that of an intact ship. It is also shown
that the post-ultimate bending strength is impor-
tant in predicting the structural stability of the
collapsing hull girder. However, the results were 0 θ
not validated against experiments. In addition,
the analysis was performed only under regular Figure 1. Theoretical relation between bending moment
sea conditions. and rotational angle.
In this paper, the dynamic collapse behavior in a
large single wave is pursued numerically and experi-
mentally as the first step since the irregular extreme Aft part Fore part
f
seas can be regarded as a collection of large single M 2w M M
D θ1
waves. For numerical simulation, nonlinear strip θ2 M 1w
theory is used to solve the force equilibrium. For F2w
f
F1w
experimental validation, an experimental scaled L/2 L/2
model is designed and fabricated. The scaled model
consists of two rigid bodies connected by a device Figure 2. Free body diagram.

240

MARSTRUCT.indb 240 2/18/2011 5:46:57 PM

 The relation between the bending moment and
⎡ m11
1
+ m112
m121
m12 2 ⎤
⎧ Δu ⎫
the rotational angle can be expressed in Eq. (1). ⎢ ⎥⎪ 2⎪
⎢ m21 1
m221
0 ⎥ ⎨ Δθ1 ⎬
ΔM kR Δθ (1) ⎢ 2 2 ⎥ ⎪ Δθ  ⎪

⎢⎣ m21 0 m22 ⎥⎦ ⎩ 2 ⎭
where Δθ = θ1−θ2; kR is taken as a function of θ. ⎡C11 1
+ C111
2 1
C12 C1122 ⎤
⎧ Δu ⎫
⎢ ⎥⎪ 2⎪
The above expression is valid as long as it is single- + ⎢ C21 1 1
C222 0 ⎥ ⎨ Δθ1 ⎬
valued with respect to θ. ⎢ 2 2 ⎥ ⎪ Δθ ⎪
The equilibrium equations of motion can ⎣⎢ C21 0 C222 ⎦⎥ ⎩ 2 ⎭
be obtained by reference to their centers of ⎡ K11 + K11
1 2
K121 2
K12 ⎤ ⎧ Δu ⎫
gravity as, ⎢ ⎥⎪ 2⎪
+ ⎢ K 21 1
K 22 + kR
1
− kR ⎥ ⎨ Δθ1 ⎬
  ⎢ ⎥⎪ ⎪
[ 1 ]{ 1} + [C1 ]{ 1} + [ 1 ]{ 1} = { 1} (2)
⎢⎣ K 21
2
kR K 2 kR ⎥⎦ ⎩ Δθ 2 ⎭
 
[ 2 ]{ 2 } + [C2 ]{
{ 2} + [ 2 ]{ 2 } = { 2 } (3)
⎧F1w + F2w ⎫ ⎡ m11 + m11 m12 m12 ⎤ ⎧u2 ⎫
1 2 1 2

⎪ ⎪ ⎢ ⎥⎪ ⎪
u1 u2 (4) = ⎨ M1w ⎬ − ⎢ m21 1 1
m22 0 ⎥ ⎨θ1 ⎬
⎪ M ⎪ ⎢ ⎥⎪ ⎪
⎩ 2w ⎭
⎣⎢ m21
2
0 m222 ⎥⎦ ⎩θ2 ⎭
where,
⎡C11
1
+ C111
2 1
C12 2 ⎤
C112 ⎧u ⎫
⎢ ⎥⎪ 2⎪
C222 0 ⎥ ⎨θ1 ⎬
{} {}
u1 u2
− ⎢ C221 1 1

{U1} {U 2 } = ⎢ 2 2 ⎥ ⎪θ ⎪
1 2 ⎢⎣ C21 0 C222 ⎥⎦ ⎩ 2 ⎭
⎡ m1 m12 ⎤
1
⎡ m2 m2 ⎤ ⎡ K11 K111 K12
1 2 1
K112 ⎤ ⎧u2 ⎫
2
[M1 ] = ⎢ 111 ⎥ [M 2 ] = ⎢ 112 122 ⎥ ⎢
1
⎥⎪ ⎪
⎣ m21 m22 ⎦ ⎣ m21 m22 ⎦
1
− ⎢ K 21 1 1
K 222 0 ⎥ ⎨θ1 ⎬
⎡C111 C12 ⎤
1
⎡C112 C122 ⎤ ⎢ 2 ⎥ ⎪θ ⎪
[C1 ] = ⎢ 1 ⎥ [C2 ] = ⎢ 2 2 ⎥ ⎢⎣ K 21
2
0 K 222 ⎥⎦ ⎩ 2 ⎭ (5)
⎣C21 C 22 ⎦ ⎣C21 C22 ⎦
1

⎡K1 K112 ⎤
1
⎡ K112 K1122 ⎤
[K1 ] = ⎢ 111 ⎥ [K 2 ] = ⎢ 2 ⎥ Newmark β method is applied to perform the
⎣ K 21 K 22 ⎦ ⎣ K 21 K 22 ⎦
1 2
time-domain analysis in Eq. (5). The analysis
⎧F f ⎫ ⎧F f ⎫ results will be given later.
{F1 } = ⎨ 1w ⎬ {F2 } = ⎨ 2 w ⎬
⎩M1w M⎭ ⎩M 2 w M ⎭

{U} = displacement vector for the respective
bodies; [M] = Mass matrixes, which consist of
hull mass and added mass for the respective
bodies; [C] = damping matrixes, which includ-
ing structural damping and wave induced damp-
ing; [K] = restoring force matrixes; {F} = external
forces and moments. Subscript 1, 2 in equations
(2) and (3) correspond to fore part and aft part,
respectively. In the mass matrixes, damping
matrix and restoring force matrix, the former
1, 2 of the subscript corresponding to force and
moment, respectively. The later 1, 2 correspond-
ing to vertical displacement and rotational angle,
respectively.
For the evaluation of the external forces and
moments, a nonlinear strip theory proposed by
Fujino (1984) is employed.
From equations (1)–(4), a system of equa-
tions of motion may be written in incremental
form as,
3 EXPERIMENT
3.1 Model requirements and design idea
A box type ship is selected as a generic ship for
simplicity and clarity, and also for validation by the
experiments. The box-type ship are length 300 m,
breadth 40 m, depth 30 m and draft 10 m. Tank
tests using a scaled model need to be performed for
validation of the numerical method addressed in
the previous section.
The scaled model should follow the law of simil-
itude in terms of strength as well as geometry and
stiffness. Its scale ratio is assumed to be 1/100 as
conventional scaled models for tank tests. The main
particulars of model are length 3 m, breadth 0.4 m,
depth 0.3 m and draught 0.1 m. It is completely a
box-ship consisting of two rigid box bodies.
The requirements for the scaled model are, a) to
realize the strength model of the hull girder col-
lapse behavior in loading and unloading as shown

241

MARSTRUCT.indb 241 2/18/2011 5:46:58 PM


Figure 3. Strength model.
Figure 4. Final design of the scaled model.
in Fig. 3. b) to collapse in waves which can be
generated in the tank. c) to be free from severe
wave-induced vibrations and d) to be available for
repeated collapse tests. Concerning a), the reduc-
tion of strength in post-ultimate strength range,
shown as Fig. 3, is typical in the hull girder col-
lapse accompanied by buckling of stiffened panels,
and thus to be considered.
The requirement b) is interpreted as keeping
the law of similitude in ‘strength’ itself. The scaled
model must collapse under wave loads which are
scaled according to the Froude’s law. The model
strength must be low enough in this regard. The
design value of the wave bending moment Mw may
be roughly given for the design work in terms of
non-dimensional value as follows:
Mw
= 0.02 (6)
ρgL
L2 Bη
where ρ = density of water; g = gravity acceleration;
and η = wave amplitude. In the tank tests, the
maximum wave amplitude may be taken as
η = 100 mm.
The requirement c) necessitates keeping the law
of similitude in stiffness as well. The model must be
also rigid enough so that the natural frequency is
out of the wave frequency range which is normally
from 0.5 to 2 Hz in model scale. The two require-
ments (b) and (c) may contradict each other.
The design methodology is detailed in Wada
(2010). Various types of design alternatives have
242
MARSTRUCT.indb 242
been investigated. The final scaled model design
to meet the above requirements is shown in Fig. 4.
The two bodies are hinged together and a bottom-
fixed beam specimen is attached to the bottom of
the ship. The specimen bends and bears shear force
when hull girder deforms at the hinge to the bend-
ing moment loaded over the structure. The nonlin-
ear relation between the moments and rotational
angle may be represented by the collapse mecha-
nism of the specimen. The point where the yield-
ing occurs is at the upper end of the specimen, as
the bending moment is the largest there. A circu-
lar pillar specimen with diameter 10 mm cut out
from a block of steel is used. The material is SS400.
The collapse tests are conducted several times by
changing the sacrificial specimens.
3.2 Tank test
Dynamic collapse tests were conducted by using
a focused wave technique to realize large single
wave in the towing tank, Osaka University. This
towing tank has the plunger wave generator. The
tank is 100 m in length, 7.8 m in breadth and
4.30 m in depth. First, the calibrations of regular
and irregular waves were conducted to generate
the target focused wave. Then, the collapse test was
conducted in the focused waves.
The focused wave can be modeled by a sum of
regular elementary wave components taken from
wave energy spectrum for the prescribed probability
level (Jensen 2006). This technique starts from the
expression of irregular waves as in Eq. (7).
n
η( , t ) ∑ *
i i ω i ccos(
os( i ωi )
i =1
+ ui* Si Δω i i (k
(k i ω it )
n
= ∑ Si Δω i U i ( kxi it − εi ) (7)
i =1
where
U i 2 ui*2 + ui*2 ; ε i arctan((ui* ui* ); S(ω) = wave
spectrum; ui and are ui* uncorrelated, standard
normal distributed variables. Following Der
Kiureghian (2000), when we define a limit state
function g as,
g (u1 u1,uu2 u2 ,…, un , un ) ≡
(
η0 − η x0 ,tt0 u1, u1 u2 ,uu2 ,…, un , un ) (8)
A problem to seek for the probability of the
irregular wave exceeding the threshold η0 at t = t0
and x = x0 is formulated as limit state problem.
An approximate solution can be obtained by
First Order Reliability Theory (FORM). A set
of {u i ui *} can be given as the design point on
*
g = 0 minimizing the distance to the origin.
2/18/2011 5:47:00 PM
 The wave spectrum employed in the experiment measure the hull girder bending moments working
and numerical calculation is indicated in Table 1. on the cross section of the model excluding the
ISSC wave spectrum is employed. A sample wave contribution from the axial loads. Moreover, four
train employed in this analysis is shown as Fig. 5. acceleration transducers are mounted along the
At t = 80 s, the wave reaches its negative peak. model to measure the vertical and rotational accel-
Large bending moment is expected to work on the erations of the respective bodies. Two wave probes
hull girder, then. The wave shape at this instance is are also installed in the tank, one in front of the
supposed that the midship locates at trough of the wave maker and the other center of the tank.
single large wave and aft and fore of ship locate at As a preliminary check, the natural frequency
the crests. of the model was measured. When the ballasts
To measure the rotational angle between the were loaded, the natural frequency was found to
two bodies, a potentiometer is equipped at the be 6.88 Hz, or 43.2 rad/s. The natural frequency is
hinge part at midship (Fig. 6). Three load cells confirmed to be out of the wave frequency range
are equipped within a cross section (section 1) to from 2 rad/s to 15 rad/s in model scale. Then, the
still-water bending moment was 14.2 × 109 Nm
in real scale. The value of the still-water bending
Table 1. Irregular sea condition for the focused wave.
moment is subtracted from the results afterwards.
A time history of the wave elevation measured
Properties Symbol Values during the tank tests is shown in Fig. 7. Even
though the wave time history is not perfectly
Signification wave height Hs 12 m (0.12 m) symmetric with respect to the target time, the
Mean wave period Ts 13.9 s (1.39 s) negative peak appears as intended.
Wave direction ϕ 180° 180°

*Wave direction 180° indicates head sea. The value in

parenthesis represents the value in scaled model.

Figure 7. Time history of wave elevation measured.

Figure 5. A sample time history of focused wave.

Figure 8. Supposed relationship between bending

Figure 6. Measurement system. moment and rotational angle.

243

MARSTRUCT.indb 243 2/18/2011 5:47:03 PM

 4 COMPARISON C
D
4.1 Numerical calculation condition
In the numerical calculation, the relationship
between the bending moment and rotational angle
is supposed as in Fig. 8.
The rigidity of ship’s hull girder is kR = K for
B
elastic range. When rotational angle increases, A
yielding occurs in the specimen. The values of
the moment MU and the slope are 9.2 × 109 Nm
and 5.4 × 1012 Nm in real scale, respectively. As
the material of the specimen does not have clear
yielding point, the strength increases slightly even
more with the reduced slope S. When unloading Figure 10. Comparison between numerical and tank
starts, the bending rigidity recovers to kR = K, test for time histories of rotational angle.
again. These characteristics are measured from a
static four-point-bending test conducted prior to
the tank test.
C
B
4.2 Comparison with experiment
Numerical and experimental results are presented A
in this section. All results are given in the real ship
scale here considering the scale ratio is 1/100.
The time histories of the bending moment at the
midship obtained by the numerical analysis and the D
tank test are shown in Fig. 9. The time histories of
the rotational angle are compared in Fig. 10. We
can obtain the relationship between the moment
and the rotational angle shown as in Fig. 11.
From Fig. 9, we can observe that the bending Figure 11. Comparison between numerical and tank
moment reaches a flexion point slightly before test for moment-rotational angle relation.
t = 80 s. Then, it continues to rise slightly with the
advancement of the time, following the path BC
t = 82 s, the structure starts to be unloaded. Then,
with slight positive rigidity in Fig. 11. It is con-
at around t = 87 s the bending moment reaches
sidered that the wave-induced moment increase
maximum negative value. According to Fig. 10, the
is borne more by the moments induced by the
rotational angle changes within a short duration at
rotational accelerations of the bodies and less by
around t = 80 s in the order 0.012 rad. This indicates
the bending capacity because the bending capac-
that the hull girder collapse rapidly after reaching
ity can increase only slightly and the rest must be
ultimate strength. This can also be illustrated in
compensated by the rotational moment. At around
Fig. 11, which is showing the relationship between
the bending moment and rotational angle at the
midship section. After unloading starts at point C,
the rotational deformation does not change largely
due to the recovery of the rigidity, following the
path CD. The large oscillation after point C in
Fig. 10 maybe due to existence of external force
after the focused wave passed along the ship. In the
oscillation, the high frequency components can be
regarded due to whipping, because in the numeri-
cal calculation, the damping effect of spring was
neglected, which caused the hull appear whipping
phenomena in the single large wave.
In general, the numerical predictions compare
very well with the experimental results. However, it
Figure 9. Comparison between numerical and tank test needs to be noted that the wave amplitude in the cal-
for time histories of bending moment at the midship. culation is adjusted so that the maximum bending

244

MARSTRUCT.indb 244 2/18/2011 5:47:06 PM

 moments agrees with that of the measurement in
the tank test. This is rationalized by the fact that
the present nonlinear strip theory code tends to
overestimate the bending moment and the differ-
ence may also lead to overestimating the collapse
extent.
5 DISCUSSIONS
The extent of the collapse of the ship, or to what
extent the ship deforms in waves, is discussed based
on a series of calculations as parametric study.
5.1 The effect of the stability of the path
after ultimate strength
The stability of the path after the ultimate strength
may largely affect the extent of collapse. Three
cases with different rigidities after the ultimate
strength are investigated. First, in case (a), the
Figure 12. Relationships between bending moment and
rotational angle for the three cases.
Figure 13. Time histories of rotational angle for the
three cases.
245
MARSTRUCT.indb 245
rigidity after ultimate strength decreases but the
value is still positive. This path should be stable.
In the second case (b), the rigidity is equal to zero
after the ultimate strength. This path is neutral. In
the third case (c), the rigidity takes negative value
and the load carrying capacity decreases. This path
is unstable. The moment-rotational angle relation-
ships for these cases are shown in Fig. 12.
In Fig. 12, the collapse deformation is around
0.012 rad in case (a), 0.023 rad in case (b) and
0.063 rad in case (c). The time histories of rota-
tional angle are compared in Fig. 13, we can observe
that the deformations after the focused wave passes
along the ship correspond to cases (a), (b) and (c)
in Fig. 12. The extent of collapse becomes the larg-
est in case (c), which indicates an unstable path
with negative rigidity after the ultimate strength
leads to the larger collapse extent of the structure.
The structure is in a dynamical equilibrium state
with the contribution of hydrostatic restoring force
and inertia effects. This phenomenon indicates
Figure 14. Time histories of wave amplitude with
different amplitude.
Figure 15. Time histories of rotational angle with
different amplitude.
2/18/2011 5:47:08 PM
 that the extent of collapse is largely affected by the
stability of the path after the ultimate strength, in
particular, the rigidity after the ultimate strength.
Further research into parametric dependencies on
the severity of the collapse will be performed in
future.
5.2 The effect of the amplitude to the bending
moment
The amplitude of the bending moment may affect
the extent of collapse, too. The wave amplitude
of the focused wave is changed as shown in
Fig. 14. A curve for case (a) shown in Fig. 12 is
employed as the relationship between the bending
moment and the rotational angle. The correspond-
ing calculation results are compared and shown in
Fig. 15.
From Fig. 15, it is observed that the rotational
angle grows up more when the wave amplitude
becomes larger, because the increase of wave
amplitude will produce larger bending moment
acting on the hull girder, which thus induces larger
plastic deformation. Another point we need to pay
attention to is that although the wave amplitude
increases is small, the difference in the plastic
deformation is quite distinct. In Fig. 14, the wave
amplitude changes from 6.5 m to 8 m, then, the
plastic rotational angle changes by about 0.012 rad.
When we further increase the wave amplitude to
9 m, the plastic deformation becomes even the
double. This phenomenon indicates that the extent
of collapse is also largely affected by the wave
amplitude and the resulting bending moment.
6 CONCLUSIONS
In this study, a hydro-elastoplasticity approach
to predicting the dynamic collapse behavior of a
ship’s hull girder in waves is proposed. The focus
is especially placed on the post-ultimate strength
behavior, or to what extent the ship collapses in a
large wave. The approach is validated against the
tank test results. After the validation, it is con-
cluded that the fundamental dynamic collapse
behavior can be followed by the numerical method.
This approach can effectively predict the collapse
behavior of a ship’s hull girder in waves.
It is clarified that the hull girder collapses rap-
idly after the ultimate strength is reached. The
ship collapse is terminated when unloading starts
at the collapsed section and the bending rigidity
recovers. The extent of the collapse of a ship’s hull
girder depends both on its strength model and
load model. In particular, the reduction of the
246
MARSTRUCT.indb 246
load carrying capacity after the ultimate strength
largely affects the collapse behavior.
In the future work, the parametric dependencies
on the severity of the collapse must be clarified in
a more rational manner. The severity of the col-
lapse in irregular seas, in which successive collapse
may occur, must be clarified from the viewpoint
of deterministic and probabilistic aspects, too.
As regards to the tank tests, a specimen to repro-
duce the reduction of the load carrying capacity
after the ultimate strength must be developed and
implemented in the tank tests.
ACKNOWLEDGEMENTS
This research was partially supported by the
Ministry of Education, Science, Sports and Culture,
Grant-in-Aid for Young Scientist (B), (19760578),
2009.
REFERENCES
Caldwell, J.B. 1965. Ultimate longitudinal strength.
Trans. Royal Inst. Nav. Arch (107):411–430.
Der Kiureghian, A. 2000. The Geometry of Random
Vibrations by FORM and SORM, Probabilistic Engi-
neering Mechanics, 15(1):81–90.
Fujino, M., Yoon, B.S., Kawada, J. & Yoshino, I.
1984. A study on wave loads acting on a ship in
large amplitude waves. J. Soc. Nav. Arch. Japan Vol.
156:144–152.
Jensen, J.J. & Capuul, J. 2006. Extreme response
predictions for jack-up units in second order stochastic
wave by FORM, Probabilistic Engineering Mechanics
21:330–337.
Jensen, J.J. & Pedersen, P.T. 1978. Wave-induced Bending
Moments in Ships—a quadratic theory, Trans. Royal
Inst. Nav. Arch. 120:151–161.
Jensen, J.J. & Pedersem, P.T. 1981. Bending Moments
and Shear Forces in Ships Sailing in Irregular Waves,
Journal of Ship Resarch 25(4), 243–251.
Lehmann, E. 2006. Discussion on “Report of
Committee III.1: Ultimate Strength,” Proc. 16th ISSC
Vol. 3, Southampton, UK, 2006:121–131.
Masaoka, K. & Okada, H. 2003. A numerical approach
for ship hull girder collapse behavior in waves.
Proceedings of the thirteenth International Offshore
and Polar Engineering Conference, Honolulu, Hawaii,
USA, May 25–30, 2003:369–375.
Smith, C.S. 1966. Measurement of service stresses in
warships. Conference on Stresses in Service. London:
Inst. of Civil Engrs:1–8.
Smith, C.S. 1977. Influence of local compressive failure
on ultimate longitudinal strength of a ship’s hull. Proc.
PRADS, A-10, Tokyo, Japan:73–79.
Wada, R., Iijima, K., Kimura, K., Xu, W., & Fujikubo, M.
2010. Development of a design methodology for a
2/18/2011 5:47:13 PM
 scaled model for hydroplasticity of a hull girder in
waves, Proceedings of PAAMES 2010. (Submitted for
publication).
Yamamoto, Y. Fujino, M. & Fukasawa, T. 1978. Motion
and longitudinal strength of a ship in head sea and
the effect of non-linearities. J. Soc. Nav. Arch. Japan
Vol. 143:179–187.
Yao, T. et al. 2006. Report of Committee III.1: Ultimate
Strength. Proc. 16th ISSC, Vol. 1, Southampton, UK,
2006:356–437.
Yao, T. Fujikubo, M. Iijima, K. & Pei, Z. 2009. Total
system including capacity calculation applying ISUM/
FEM and loads calculation for progressive collapse
analysis of ship’s hull girder in longitudinal bending.
247
MARSTRUCT.indb 247
Proceedings of the 19th International Offshore and
Polar Engineering Conference, ISOPE2009, Osaka,
Japan. 2009:706–713.
Yao, T. & Nikolov, P.I. 1991. Progressive collapse analysis
of a ship’s hull under longitudinal bending. J. Soc.
Nav. Arch. Japan Vol. 170:449–461.
Yao, T. & Nikolov, P.I. 1992. Progressive collapse analysis
of a ship’s hull under longitudinal bending (2nd
Report). J. Soc. Nav. Arch. Japan Vol. 172:437–446.
2/18/2011 5:47:14 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

A study on the dynamic buckling strength of containership’s bow

structures subjected to bow flare impact force

S.H. Yang, H.L. Chien, C.M. Chou & K.C. Tseng

Department of Design, CSBC Corporation, Taiwan, Kaohsiung, Taiwan, R.O.C.

Y.J. Lee
Department of Engineering Science and Ocean Engineering, National Taiwan University,
Taipei, Taiwan, R.O.C.

ABSTRACT: The objective of this paper is to analyze the structural dynamic buckling strength of
containership bow structures subjected to impact force using finite element method. In this study, a con-
tainership bow structure designed and constructed by CSBC Corporation, Taiwan is modeled and solved
by geometric nonlinear and transient dynamic programs; and the wave impact pressures are obtained
through applying the semi-empirical formula established by Lloyd’s Register. Furthermore, one-time
thickness deformation method and buckling failure method are employed here to determine the initial
buckling load of bow structure. The results show that the impact force integration is the dominant factor
for structure buckling strength assessment. This conclusion offers the designer an effective and efficient
way to predict the structure dynamic buckling strength; moreover, it is can help the designer realize the
moment when the structure happens buckling under any different impact force case.

1 INTRODUCTION The prediction of wave impact loads for a run-

In recently decades of years, container carriers
have larger bow flares than before to stack more
containers on the deck of the bow. Because the
flare angle is so large, it is expected that the wave
impact loads acting on bow flares become very
large. According to the reviewing of ISSC 2009,
the slamming impact load is still an important rea-
son that causes the bow structure damage. There-
fore, the investigation of bow structure’s response
due to impact load is becoming much more impor-
tant and the knowledge for improving bow struc-
ture design is widely needed. It is attributed to the
benefits of the recent developments in numerical
analysis techniques and greatly advanced per-
formance of computers; a bow flare wave impact
phenomenon can be calculated more directly by
using numerical simulation method, consequently.
Based on the finite element theory, the commercial
software MSC Marc was used in this study to pro-
pose dynamic transient and geometric nonlinear
analysis. Additionally, a 1,700 TEU containership
designed and constructed by CSBC Corporation,
Taiwan, was taken as the reference ship in this
paper. Because it is so time-consuming to compute
time-domain problems, only bow structure of the
reference ship was modeled using MSC Patran,
which is a pre- and postprocessor of MSC Marc.
ning ship is also an important topic in this study.
Since it is too complex to predict wave impact force
directly, this kind of problem is usually simplified as
a two-dimensional solid dropped vertically onto a
free and initially calm water surface. Notwithstand-
ing these simplifications, the solution to this type of
problem remains complex, especially with regard to
transient fluid-structure interactions. Von Kármán
(1929) introduced significant work on this subject
as a pioneering study. He developed an analyti-
cal formula, which allowed the estimation of the
maximum pressure on seaplane floats during water
landing. Wagner (1932) modified the von Kármán
solution by taking into account the effect of water
splash on the body. Ochi and Motter (1969 & 1973)
used K-factor method, which is based on the use
of slamming coefficients. These coefficients related
the slamming pressure to the square of the impact
velocity, and they could be calculated, or obtained
experimentally prior to the analysis of ship motion.
More recently, Zhao and Faltinsen (1993) intro-
duced a complement to Wagner’s study. They used
linear approximation of the free-surface boundary
condition for the two-dimensional impact prob-
lem. The numerical method suggested in this paper
was valid for general section shapes. Mei et al.
(1999) proposed a purely analytical method of
resolution for the global two-dimensional impact

249

MARSTRUCT.indb 249 2/18/2011 5:47:14 PM

 problem of arbitrary bodies. They adopted the
same assumptions and the same linear formula-
tion of the problem as which were used by Zhao
and Faltinsen. In this paper, water impact pressure
was estimated in accordance with Lloyd’s Register
(LR) semi-empirical formula, which is also on the
basis of K-factor method.
The structural buckling strength has been inves-
tigated by several researchers for hundreds of
years. Levy (1942) gave the exact solution of von
Kármán equations to steel plates with built-in
edges under edge compression. In 1959, Yamaki
(1959) described the post-local buckling behavior
of clamped elastic steel plates with initial deflec-
tions. Furthermore, the ultimate strength of sim-
ply supported plates with geometric and material
nonlinearities was reported by Little (1981). Not
only static but also dynamic buckling behavior has
been discussed by many researchers. Bisplinghoff
(1957) used only three terms to represent the trans-
verse displacement of a buckled simply supported
plate. Yang and Han (1983) analyzed the same
question as Bisplinghoff, but using higher order
and more complicated triangular finite elements
to obtain more precise solutions. Extensive results
on the dynamic behavior of clamped plates under
in-plane compression were presented by White and
Teh (1980). In this study, the suggested methods,
including one-time thickness deformation method
and buckling failure method, were employed to
determine dynamic buckling load of bow struc-
tures under triangular impact forces; and the
impact force integration under impact force-time
curve was taken as a criterion of structure buckling
strength.
2 THEORY
Dynamic finite element analysis in time domain
and fully geometric nonlinear computation were
used to investigate the response of a container-
ship structure subjected to bow flare impact forces.
A brief description of the theory will be given in
this section.
2.1 Transient dynamic analysis
Dynamic analysis differs from static analysis in
two major aspects. First, loads applied are varying
with time. Second, structural responses, including
displacements, velocities, stresses and so on, also
react as a function of time. To solve a dynamic
time-domain problem by finite element method,
the governing equation should be written in dis-
crete form as follows:
Ma + Cv + Ku = F (1)
250
MARSTRUCT.indb 250
where M, C, and K are mass, damping and stiffness
matrices respectively; a, v, u and F are time varying
acceleration, velocity, displacement and force vec-
tors respectively.
2.2 Geometric nonlinearity
Structural nonlinearity can be considered as two
categories: material nonlinearity, which means the
material’s stress-strain relationships are nonlinear;
geometric nonlinearity, which is usually charac-
terized by large displacements and/or rotations.
The buckling behavior investigated here is a type
of large-displacement problem; for this reason, a
geometric nonlinear analysis is required. Geomet-
ric nonlinearity leads to two types of phenomena
including change in structural behavior and loss of
structural stability. The Lagrangian formulation
was chosen here to simulate buckling behavior.
The Lagrangian formulation is on the basis of
the virtual work equation for a continuum which
can be written as follows: (MSC. Software Corpo-
ration, 2008)
∫V d ε σdv = ∫ ρduT qdv
d + ∫ duT pda
T (2)
d
V a
where V is undeformed volume; σ is Piola-Kirchhoff
stress vector; dε is vertical Green’s strain vector due
to the virtual displacement du; ρ is mass density;
q is body force per unit mass; and p is surface trac-
tion acting on undeformed area a.
3 NUMERICAL STUDY
Finite Element (FE) method has become a stand-
ard tool of contemporary naval architects struc-
tural analysis. The commercial FE software MSC
Patran and MSC Marc were adopted here to carry
out the FE analysis.
3.1 Reference ship
The ship used in this study is a 1,700 TEU con-
tainer carrier. The principal particulars and the
general arrangement of this ship are displayed
in Table 1 and Figure 1. According to the survey
offered by CSBC Corporation, Taiwan, this ship
experienced extreme impacts on bow flare. Several
brackets happened external buckling and distor-
tion as shown in Figure 2. Figure 3 provides details
on failure area of brackets.
3.2 Finite element model
3D FE model was developed in MSC Patran. Only
bow structure was modeled as shown in Figure 4.
2/18/2011 5:47:14 PM
 Table 1. Main particular of 1,700 TEU container
carrier.

Particular Value

Length (L.O.A.) 175.100 m

Length (Design Draft L.B.P.) 164.900 m
Length (Scant. Draft L.B.P.) 164.900 m
Length (Scant.) 163.057 m Figure 4. Bow structure model.
Breadth (Mould) 27.9 m
Depth (Mould) 13.8 m
Draft (Design mould) 8.5 m
Draft (Scant. Mould) 9.5 m
CB (Scant. length & Draft mould) 0.68
VS (Scant. draft, MCR) abt. 20.5 kts

Figure 5. Finer mesh on 6 brackets.

Table 2. Material constants of the FE

model.

Figure 1. General arrangement of reference ship. Material constant Value

Density (ρ) 7850 kg/m3

Elastic modulus (E) 206 GPa
Poisson’s ratio (υ) 0.3
Yield stress 290 MPa

This model is about 33 m and consists of 2-D plate

and 1-D beam elements. To simulate local buckling
behavior more accurate, six brackets (Frame No.
95, 96, 98, 99, 100, 102) and surrounding structures
use finer mesh (about 20 × 20 mm2) with respect to
others with coarse mesh (about 700 × 700 mm2).
Figure 5 reveals information about the finer mesh
area. In this paper, material property is assumed as
perfectly plastic; besides, other primary constants
are listed in Table 2.
Figure 2. Buckling and distortion of brackets.

3.3 Loads and boundary conditions

As shown in Figure 6, only aft-side of the model
was fixed in 6 degree of freedom to ensure that
all forces can be balanced but with least bound-
ary effect on 6 brackets. Besides, a variety of loads
were taken into account in FE model, including
weight of hull, containers and deck machinery, still
water pressures, and wave impact loads. All forces
considered in the model are presented in Figure 7.

3.3.1 Bow flare wave impact pressure

To provide wave impact load information for the
FE model, an effective and correct way for estimat-
Figure 3. Brackets location under the upper deck. ing bow flare impact forces was needed. The bow

251

MARSTRUCT.indb 251 2/18/2011 5:47:14 PM

 where Kbf is hull form shape coefficient for wave
impact; Vbf is wave impact velocity; Krv is hull form
shape coefficient for impact due to forward speed;
Hrv is relative wave heading coefficient; and Vrv is
relative forward speed.
In general, the slamming force should distrib-
ute over the whole side shell. However, only par-
tial slamming forces are applied to highlight and
clarify the local buckling phenomenon. Figure 8
demonstrates the value and region of the slam-
ming force considered in this paper.

Figure 6. Fixed boundary condition of the FE model.

3.4 Formulation of the analysis case
Since the whole wave impact track is complex, the
wave impact forces was simplified and idealized to
a triangular dynamic impulse in this paper. A series
of cases were analyzed to compare the effects of
different dynamic parameters, including the peak
value of load, impact time, and impact load rate.
Figure 9 shows the relationships between these
parameters for a triangular impulse.
Figure 7. Distribution of forces acting on the FE To carry out the dynamic buckling assessment
model. properly, geometric nonlinear and implicit dynamic
computation codes are selected in this study.

4 SUMMARY

4.1 Buckling load criterion

Figure 10 shows the load-transverse deflection
curve of a buckling plate. It can be observed that
an apparent bifurcation point cannot be found eas-
ily to evaluate initial buckling load; consequently,
two novel and practical methods were suggested in
this study and are illustrated below.
Figure 8. The slamming force region and value.
4.1.1 One-time thickness deformation method
As shown in Figure 11, the initial buckling load
is defined as the load that causes the value of
structural lateral deflection to equal the value of
plate thickness. The main idea underlying the pro-
posed method is that one time thickness deflection

Figure 9. Triangular dynamic impulse.

flare wave impact pressure Pbf is defined in equation

(3): (Lloyd’s Register, 2007)

Pbf 0.5( K bfVbf2 K rv H rvVrv2 ) kN / m 2 (3) Figure 10. Load-transverse deflection curve.

252

MARSTRUCT.indb 252 2/18/2011 5:47:16 PM

 Figure 11. Description of one-time thickness deforma-
tion method (Plate thickness = 2 mm).

Figure 13. Bracket deformation distribution in

x-direction (wave impact time = 1 sec, wave impact
pressures = 1 time of LR forces).
Figure 12. Description of buckling failure method
(Material yield stress = 290 MPa).

can usually be seen as large deformation. This

mechanism response can therefore be regarded as
geometric nonlinear behavior.

4.1.2 Buckling failure method

As shown in Figure 12, the von Misses stress
instead of the structural lateral deflection was
used to determine initial buckling load. For many
structures, it is important that the material remains
within the linear elastic range to avoid permanent Figure 14. Von Misses stress distribution on bow (wave
deformations when the loads are removed. Accord- impact time = 1 sec, wave impact pressures = 1 time of
ingly, the buckling allowable stress may not exceed LR forces).
material yield stress anywhere in the structure. The
corresponding load to this stress is defined as ini-
tial buckling load.

4.2 Results and discussion

An example of the output from the FE analy-
sis is shown in Figure 13. It can be seen from the
figure that six brackets are not buckled at the same
time but one by one. Moreover, Figure 14 plots
Von Misses distribution on the bow structure;
and Figure 15 gives the information about lateral
deflection distribution on six brackets. It is worth Figure 15. Lateral deflection distribution on six brack-
to state outcomes observed from Figures 13–15 as ets (wave impact time = 1 sec, wave impact pressures = 1
follows: time of LR forces).

253

MARSTRUCT.indb 253 2/18/2011 5:47:17 PM

 – Six brackets are buckled under different impact Table 3. Impact force integration analysis on no. 96
forces and impact times. It can be found that bracket by one-time thickness deflection method.
three brackets in the middle buckled later than
other brackets. It may refer to different loca- Impact Initial Impact
Load rate time buckling load force integration
tions of these six brackets. Three brackets in the
middle can share the slamming force together (LR force/
whereas others have to undertake loads alone, sec) (Sec) (LR force) (LR force-sec)
therefore, it is reasonable to anticipate that the
middle three brackets have larger margin to 1.00 0.322 0.6440 0.1037
bear the slamming force. 1.11 0.308 0.6846 0.1055
– Citing bracket frame No. 102 as an example, 1.25 0.296 0.7392 0.1093
it is clear that the structure at upper deck side 1.67 0.273 0.9114 0.1246
appears to deform much more severely than 2.00 0.264 0.9454 0.1383
anywhere in the same bracket. Considering that 2.50 0.260 0.6976 0.1513
buckling phenomenon is a geometric behavior 3.33 0.259 0.2734 0.1444
and there is no any possible variable involved, 4.00 0.247 0.0214 0.1250
a reason about this result would probably be
attributed to the geometry of the bracket.
– In addition to six brackets, upper deck and side Table 4. Impact force integration analysis on no. 96
shell that are close to bow flare impact area have bracket by buckling failure method.
critical deformations due to bow flare impact
pressure. Although this paper concentrates on Initial
Impact buckling Impact force
the buckling of brackets, on the designer’s point Load rate time load integration
of view, the structure strength in these two areas
should be increased to prevent brackets from (LR force/
buckling as well. sec) (Sec) (LR force) (LR force-sec)
One of important objectives in this study is to 1.00 0.150 0.3000 0.0225
find the dominant criteria of structural dynamic 1.11 0.142 0.3150 0.0223
buckling strength. Four principal dynamic param- 1.25 0.134 0.3360 0.0226
eters were taken into account to evaluate structural 1.67 0.116 0.3864 0.0224
buckling strength, including load rate, impact time, 2.00 0.106 0.4242 0.0225
initial buckling load and impact force integration 2.50 0.097 0.4830 0.0233
under load-time curve. Table 3 and Table 4 show the 3.33 0.086 0.5712 0.0245
breakdown on buckling strength of frame No. 96 4.00 0.079 0.6342 0.0251
bracket by using one-time thickness deformation
method and buckling failure method, respectively.
According to these two tables, some discussion can
measurement. However, when structure stress is
be stated as follows.
over yield stress, it can be directly known that
– Compare with Table 3 and Table 4, we can see some permanent strain is occurred. For this rea-
immediately that critical buckling point decided son, buckling failure method can be considered
by displacement always happened later than by as a better way to decide initial dynamic buck-
stress. As mentioned in Section 3.2, material ling load.
property is assumed as perfectly plastic in this – Before this study, we have known that the impact
study. Therefore, if the critical buckling stress loads value and impact loads time can both
decided is higher than the yield stress, realis- affect structure buckling. However, it is hard for
tic material plastic property should be taken the designer to predict the specific time when
into account to get much more accurate criti- the structure buckling. According to Table 4, the
cal buckling point. Notwithstanding, one-time impact force integration causing the structure
thickness deformation method also provide an buckling remains nearly stable under each dif-
effective and efficient way to assess dynamic ferent impact condition. On that account,
buckling strength of any complicated structure. we may reach the conclusion that the impact
– Although two buckling criteria are both novel force integration under load-time curve is the
to use, buckling failure method may be the more dominant factor of dynamic buckling strength.
reasonable one. For different structures, struc- In general, the force integration is also a kind of
tural responses with one-time thickness defor- idea similar to energy. In other words, the results
mation can be critical or be nothing, since the showed in Table 4 reveal that a structure occurs
deformation of structure is not the normalized buckling within the same energy. Therefore, the

254

MARSTRUCT.indb 254 2/18/2011 5:47:19 PM

 calculation of impact force integration can be
seen as a practical and effective way for struc-
ture designers to decide structure strength.
Designers can consequently predict structure
dynamic buckling strength under any impact
force without the use of FE analysis.
5 CONCLUDING REMARKS
In this paper, dynamic transient and geometric
nonlinear FE methods were used to investigate the
response of bow structures to bow flare impact
forces. The wave impact pressures were estimated
through applying Lloyd’s Register half-empirical
formula. Two initial buckling load criteria, one-
time thickness deformation method and buckling
failure method, were established to determine ini-
tial buckling load. The primary objective from this
study is to realize whether the structure meets ulti-
mate buckling strength. According to the results,
we can conclude that impact force integration
under load-time curve is the dominant parameter
to decide dynamic buckling strength. The conclu-
sion offers structural designer an effective and effi-
cient way to predict structural buckling strength.
Although the designer still have to do FE analysis
to know the structure critical buckling strength for
the first time, this study conclusion can help the
designer predict the structure strength without the
use of FE analysis in any other different impact
load case. Therefore, the analysis time for dynamic
buckling problems can be greatly reduced but still
with reasonable strength prediction.
With regard to the future jobs, the following
related aspects are considered to be important for
this study.
– Finite element model: Only bow structure was
modeled on account of computation time;
therefore, the boundary condition effects of
the model could not be neglected. In addition,
selections of mesh and element types also had
great influence on simulation results. It is sug-
gested that the whole ship structure should be
modeled if any researcher wants to realize the
overall structure response due to wave impact
forces.
– Wave impact load: The magnitude of wave
impact pressures in this study were obtained in
accordance with half-empirical formula estab-
lished by Lloyd’s Register. However, it is also
important to know the information of impact
time since dynamic simulations were discussed
in this study. If a sea keeping experiment or a
CFD simulation is available, it is convinced
that a more accurate buckling behavior can be
255
MARSTRUCT.indb 255
predicted because of a more precise impact load
phenomenon.
– Buckling load criterion: Two initial buckling
load criteria were developed in this report,
including one-time thickness deformation
method and buckling failure method. By these
two methods, impact force integration could
therefore be regarded as the dominant factor for
determining structural buckling strength. How-
ever, since there is no effective way to validate
these two methods, more effort is needed to ver-
ify them or select a more reasonable criterion,
which would be used with more confidence in
buckling analysis.
REFERENCES
Bisplinghoff, R.L. & Pian, T.H.H. 1957. On the vibration
of thermally buckled bars and plates. In 9th Interna-
tional Congress for Applied Mechanics: 307–318.
July 2009. Rules and Regulations for the Classification of
Ships. Lloyd’s Register.
Levy, S. 1942. Buckling of rectangular plates with built-in
edges. Journal of Appplied Mechanics 9: 171–174.
Little, G.H. 1981. Collapse analysis of plates with strain
hardening. International Journal of Mechanical Sci-
ences 23: 561–576.
Mei, X., et al., 1999. On the water impact of general
two-dimensional sections. Applied Ocean Research
21: 1–15.
MSC. Software Corporation. 2008. Theory and User
Information.
Ochi, M.K. & Motter, L. 1969. Prediction of extreme
value of impact pressure associated with ship slam-
ming. Journal of Ship Research 13(2).
Ochi, M.K. & Motter, L. 1973. Prediction of slamming
characteristics and hull response for ship design.
Society Naval Architects and Marine Engineers 81:
144–176.
Von Kármán, T. 1929. The impact on seaplane floats
during landing. National Advisory Committee for Aer-
onautics. Washington, DC: USA.
Wagner, H. 1932. Ueber Stoss- und Gleitvorgaenge an
der Oberflaeche von Fluessigkeiten. Zeitschrift fuer
Angewandte Mathematik und Mechanik 12: 193–215.
White, R.G. & Teh, C.E. 1980. Dynamic response of iso-
tropic and anisotropic panels under simulated flight
load conditions. In First International Conference on
Recent Advances in Structural Dynamics.
Yamaki, N. 1959. Postbuckling behavior of rectangular
plates with small initial curvature loaded in edge
compression—2. In ASME Meeting: 8. 9–11 September
1959. New York: USA.
Yang, T.Y. & Han, A.D. 1983. Buckled plate vibrations
and large amplitude vibrations using high-order trian-
gular elements. AIAA journal 21: 758–766.
Zhao, R. & Faltinsen, O. 1993. Water entry of
2-dimensional bodies. Journal of Fluid Mechanics
246: 593–612.
2/18/2011 5:47:19 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Ultimate strength of aluminum Y-stiffened panels

M.R. Zareei
Chabahar Maritime University (CMU), Chabahar, Iran

ABSTRACT: In this paper, the results of an investigation into the post-buckling behavior of
high-strength aluminum alloy stiffened plates with Y-stiffener subjected to axial compression load using
non-linear finite element approach is presented. Both material and geometric non-linearities have been
taken into account. The principal variables studied are the geometries of stiffened plates. The influence
of Dimensions on the post-buckling behavior and ultimate strength of such stiffened plates has been
investigated in details.

1 INTRODUCTION stir welded aluminium-stiffened plate structures.

Stiffened plates are used as main supporting mem-
bers in many civil as well as marine structural
applications. They typically consist of a plate with
equally spaced stiffeners welded on one side, often
with intermediate transverse stiffeners or bulk-
heads. The most common stiffener cross-sections
are bulb, flat bar or T- and L-sections. Such struc-
tural arrangements are common for both steel and
aluminium structures. Aluminium panels have been
used in a variety of marine structures, with appli-
cations such as hull and decks in high-speed boats
and catamarans and superstructures for ships.
Recently, some new steel double hull structures
were invented to achieve better energy absorption
capacity. Y-, rectangular box section, trapezoidal
box section, X- and corrugated frames are some of
new innovation stiffened structures. The intention
of these new designs was to prevent early crack
occurrence during a collision or grounding. Some
examples of this type illustrates in Figure 1.
The ultimate strength of stiffened steel plate
panels has been the subject of many investigations,
both experimentally and numerically, with the
most significant contributions in the field of ship
structures and bridges. The literature on stiffened
aluminium panels is more limited. The ultimate
strength of stiffened aluminium AA6082-T6 plates
under the axial compression was investigated using
numerical and experimental methods (Aalber et al.
1999, 2001). Kristensen & Moan (1999) demon-
strated numerically the effect of HAZ and residual
stresses on the ultimate strength of rectangular alu-
minium plates (AA5083 and AA6082) under the
bi-axial loading of plates. Most recently, Paik et al.
(2009) studied buckling collapse testing of friction
Khedmati et al. (2009) studied the post-buckling
behavior of high-strength aluminium alloy stiff-
ened plates subjected to combined axial com-
pression load and different magnitudes of lateral
pressure using non-linear finite element approach.
Y-stiffener is discussed by Ludolphy (2001).
Badran et al. (2007) studied the stability of Y-stiffener
by calculating the elastic buckling coefficient of the
weakest part of Y-stiffener and calculated the criti-
cal buckling stress and buckling load of Y-stiffener
and using curve fitting of the analytically obtained
results, approximate expressions for calculation of
the elastic buckling coefficients of the T-part of
the Y-stiffener are obtained. Badran et al. (2008)
studied the stability of Y-stiffener and equivalent
T-stiffener under suitable two groups of boundary
conditions, and the obtained results showed that
the critical buckling loads for Y-stiffener is larger
than that for equivalent T-stiffener. A qualitative
assessment of 10 different steel sandwich alterna-
tives to identify potential novel crashworthy side
shell structures has been made by Klanac et al.
(2005) and the results are compared with the
Figure 1. Non-conventional rectangular box, hat, n,
and M stiffeners and its attached effective plating (Badran
et al. 2007).

257

MARSTRUCT.indb 257 2/18/2011 5:47:20 PM

 performance of a conventional double side. One
variant of a novel structure offered 40% higher
capacity to absorb collision energy. Badran et al.
(2009) use the multi-objective optimization with
real-coded genetic algorithm to design an optimum
Y-stiffener plate combination with consideration
of ultimate strength.
In most of aforementioned studies, only the
case of conventional stiffeners under in-plane
compression has been considered. In this paper,
the post-buckling behavior and ultimate strength
characteristics of stiffened aluminium plates under
axial compressive are investigated using non-linear
finite element method. Y-stiffener type is selected.
Stiffened panel dimensions are varied in a system-
atic manner in the analyses. In addition, average
initial weld-induced deflection and also softening
in heat-affected zone are considered throughout Figure 2. Cross-sectional geometry of aluminium stiff-
the investigation. ened plates.

2 FINITE ELEMENT ANALYSIS

Geometrically non-linear elastoplastic finite ele-

ment analysis is the only method capable of simu-
lating the succession of all buckling phenomena
that occur during the quasi-static compression of
a stiffened panel. To predict the ultimate strength
of thin-walled structures, plate-shell elements that
account for both membrane and bending stiffness
are employed. The mesh size must be fine enough
to capture long-wave-length buckling modes, such Figure 3. Extent of the continuous stiffened plate mod-
as torsional buckling, also accounting for continu- els for analysis.
ity through the transverse frames. Furthermore,
the mesh needs to be seeded with initial imperfec-
tions to make sure that buckling occurs and to
plastic collapse behavior of aluminium-stiffened
avoid unduly stiff behavior. Many methods exist to
plate with symmetrical stiffeners. Also for the
enhance the stability of non-linear buckling simu-
analysis on the stiffened plates with non-symmetrical
lations. Details need not be discussed in detail here,
stiffeners, a double span-triple bay (DS-TB) model
since FEA methods are extensively covered in exist-
(abgh in Figure 3) is applied.
ing literature. The buckling/plastic collapse behav-
The boundary conditions are given as follows:
ior and ultimate strength of aluminium-stiffened
panels are hereby assessed using ANSYS, in which • Periodically continuous conditions are imposed
the material and geometry nonlinearities are taken at the same x-coordinate along the longitudinal
into account. edges in triple bay model (i.e. along ab and gh).
• Symmetry conditions are imposed at the same
x-coordinate along the longitudinal edges in
2.1 Stiffened plates for analysis
double bay model (i.e. along ab and de).
The geometrical characteristics of the analyzed • Symmetry conditions are imposed at the same
aluminium stiffened panels are given in Table 1. y-coordinate along the transverse edges in dou-
Ten groups of models have been considered. In all ble span model (i.e. along adg and beh).
models Y-stiffener has been attached to the plat- • Although transverse frames are not modeled; the
ing, Figure 2. out-of-plane deformations of plate and stiffener
are restrained along the junction lines of them
and the transverse frame.
2.2 Extent of the models
• To consider the plate continuity, in-plane move-
A double span-double bay (DS-DB) model (abde ment of the plate edges in their perpendicular
in Figure 3) is applied for the analysis of buckling/ directions is assumed to be uniform.

258

MARSTRUCT.indb 258 2/18/2011 5:47:20 PM

 Table 1. Geometrical and ultimate strength of the aluminum stiffened plates with Y-stiffener all dimensions
in [mm].

Y-stiffener

Plate Hat section Tee section

Model ID L BT t1 B1 B2 H1 t2 H2 B3 t3 b 1 σ Ult σ Y

1 1500 500 3 110 70 80 3 80 70 3 10.123 0.505 0.523

2 4 125 80 90 3 90 70 3 7.592 0.467 0.552
3
5 150 100 100 4 100 80 4 6.074 0.411 0.608
4 6 175 115 110 5 110 90 5 5.062 0.369 0.671
5 6 200 135 120 5 120 110 5 5.062 0.332 0.694
6 7 225 150 135 6 135 120 6 4.339 0.294 0.741
7 8 250 170 150 7 150 130 7 3.796 0.265 0.858
8 2400 800 6 200 135 120 5 120 110 5 5.062 0.558 0.581
9 7 225 150 135 6 135 120 6 4.339 0.488 0.633
10 8 250 170 150 7 150 130 7 3.796 0.435 0.674

2.3 Material properties

The Young modulus of material is 70475 MPa and
its Poisson ratio is 0.3. The stress-strain relationship
of the original aluminium alloy (AA6082-T6) and
welded aluminium alloy (in HAZ region) is shown
in Figure 4-a. The breadth of HAZ is assumed to
be 50 mm in the plate and 25 mm in the stiffener
web, at the plate-stiffener junction, Figure 4-b.

2.4 Initial imperfections

In order to simulate the complex pattern of ini-
tial deflection stated in Khedmati et al. (2009), a
special procedure was employed. Simply speaking,
lateral pressure was applied first on the stiffened
plate model and a linear elastic finite element anal-
ysis was made. Such an analysis was repeated in
a trial and error sequence of calculations so that
maximum deflection of plate reaches to the value
given by Equation 1. After satisfying this condi-
tion, the information about the coordinates of Figure 4. (a) Stress-strain behavior of the material.
nodal points, element coordinates and boundary (b) extent of the HAZ.
conditions were exported in another finite element
mesh. The new model was used for the nonlinear
response analysis under the action of longitudinal
in plane compression. The procedure is shown in where b = BT; t = t1; sy = Yield stress and E = young
Figure 5. modulus.

W0 max = 0.05β 2t (1)

2.5 Types of elements
Among rich library of ANSYS elements, a four-
b σYp
β= (2) noded SHELL43 element is used in discretisation
t E of the stiffened plate models. In each case, about

259

MARSTRUCT.indb 259 2/18/2011 5:47:21 PM

 2500 to 3000 elements are used to model each local
plate panel (the panel surrounded by successive
longitudinal or transverse stiffeners). 15 to 18 and
5 to 6 and 5 to 6 elements are considered respec-
tively along Hat section, T-section stiffener web and
stiffener flange. Figure 6 shows typical examples
Figure 5. Procedure to generate initial deflection.
Figure 6. Typical examples of the meshed models.
Figure 7. Test set-up, schematic view of simply supported panel in test rig.
260
MARSTRUCT.indb 260
of the meshed models of stiffeners. In this figure,
dark elements are elements in HAZ.
2.6 Aalberg tests
A total of 21 stiffened aluminium panels were
tested by Alberg et al. (2001). Two sets of test
specimens were made by them; one with L shaped
stiffeners and the other with hat stiffeners, both
made by AA6082-T6 aluminum alloy. Stiffened
panels dimensions and boundary condition were
varied in each set. End plates of each specimen
were machined parallel to achieve uniform appli-
cation of load. A test rig was designed as shown
in Figure 7. The test specimen was mounted in
a vertical position. The axial compressive load-
ing was applied at the upper end of the specimen
through a rigid loading set, while the reaction force
was carried by the lower end support set. Before
testing, initial imperfections of the stiffened panels
were measured. During the tests, the axial com-
pressive load was applied by slowly imposing a
displacement subsequently the deformation of the
stiffened plate was measured. In addition of per-
forming tests, Aalberg also performed numerical
comparison by EURO code 9. More de tails are
given in the paper of Aalberg et al. (2001).
Among the test specimens of Aalberg et al.
(2001), two models P and O (closed section) were
chosen for validation purposes. For comparison
between numerical analysis with experimental
tests, above models simulate using ANSYS FEM
software by the author.
Stiffened panels dimensions, boundary condi-
tion along long side and initial imperfection are
2/18/2011 5:47:22 PM
 Figure 8. Cross-sectional geometry of extruded profiles.

Table 2. Test panels, specimen details and initial

imperfections (all dimensions in [mm]).
(a)

f02(MPa)

Panel a (mm) b (mm) plate stiffener

No support along longitudinal edge

O 2000 840 265 275

P 998 1400 261 275

(b)
Figure 10. Engineering stress-strain curves for panel
Cross section (mm) materials.
t1 t2 t3 s2 h

3.04 3.02 3.06 42 40.8

3.10 3.06 3.04 42 41.1

(c)

Imperfection (mm)

w0,1 w0,2 w0,3 w0,4 w0,5

1.2 1.8 1.8

0.1 0.2 0.1 0.2 0.1

Figure 9. Definition of imperfections.

shown in Figure 8 and Table 2. Meaning of initial

imperfection in Table 2 is shown in Figure 9.
Also material properties of these models were
shown in Figure 10. Figures 11(a) and 11(b) repre-
sent, respectively, the collapse modes of the O and
P models as obtained from numerical simulations.
Also, load-end shortening curves for these two Figure 11. Collapse modes of Aalberg tests that tested
models are shown in Figures 12 and 13. with ANSYS—(a) specimen O (b) specimen P.

261

MARSTRUCT.indb 261 2/18/2011 5:47:25 PM

 As can be seen, the corresponding load-end and then axial force of panel after ultimate load
shortening curves show that in both models decrease gradually.
ANSYS simulation results have a lower ultimate
strength than that obtained by tests.
The ultimate strength predicted by ANSYS dif- 3 FINITE ELEMENT RESULTS
fers, respectively, about 5.23% and 9.34% from that AND DISCUSSION
obtained experimentally for the P and O models.
Based on numerical analysis and Aalberg et al. 3.1 Results
(2001), the collapse modes with similar features
The results of average stress-average strain rela-
are obtained experimentally or numerically for
tionships for some models of stiffened plates
both models.
with Y-stiffeners are shown in Figure 14. Also the
Based on numerical analysis models P and O
numerical values are summarized in last column of
buckled in FB/LB and FB collapse modes respec-
Table 1. In Figure 15 the plots of collapse modes
tively (FB: flexural buckling; LB: local buckling).
and deflection modes at the final stage of calcula-
Similar collapse modes reported in Aalber et al.
tions, for models 2, 4 and 6 respectively, are shown.
article. In panel O according to overall buck-
Only some of results are shown in this paper.
ling sudden decrease in axial force is happened.
In what follows, the results are discussed.
In panel P, outset panel buckled in local zone

3.2 Discussion
From results of stiffened plates with Y-stiffener,
ultimate strength of panels increases with increas-
ing of bending rigidity. Also, we can see that, pan-
els with lower bending rigidity have the collapse
mode in a buckling mode. In these models with
increase in bending rigidity, the collapse mode
changes from buckling mode to quasi-clamped
mode. In these models, Stiffened plates with high
rigidity result in increasing post-ultimate strength.
The creations of the plastic hinges at the ends of
the stiffened plate are a common feature of the col-
lapse mode. Also the strength reserve after collapse
is significant. At the final stage of calculations, it
is observed that unloading (stress removal) takes
place in some part of the stiffened plate while in the
Figure 12. Load shortening curve for O model. remaining parts localized plastic deformations are
accumulated. We can see this phenomenon at mid-
dle of model 4 or quarter of model 6 in Figure 15.

Figure 13. Load shortening curve for P model. Figure 14. Average stress-average strain relationships.

262

MARSTRUCT.indb 262 2/18/2011 5:47:26 PM

 Figure 15. Plots of collapse modes at ultimate strength level (left column) and at the final stage of calculations (right
column) for models 2, 4 and 6.

4 CONCLUSIONS approach was investigated. Both material and

geometric non-linearity were taken into account.
Post-buckling behavior of high-strength aluminium It was found that with increase bending rigid-
alloy stiffened plates subjected to axial compres- ity of stiffened aluminum panels with Y-stiffener,
sion load using ANSYS non-linear finite element ultimate strength of panel will increase. Also in

263

MARSTRUCT.indb 263 2/18/2011 5:47:27 PM

 different bending rigidity, different collapse modes
were seen. Usually after ultimate strength of stiff-
ened panel unloading is occurs in some part of
stiffened panels.
REFERENCES
Aalberg, A., Langseth, M. & Larsen, P.K. 2001. Stiffened
aluminium panels subjected to axial compression.
Thin Walled Structure 39(10):861–885.
Aalberg, A., Langseth, M. & Malo, K.A. 1998. Ultimate
strength of stiffened aluminium plates. Department
of Structural Engineering, Norwegian University of
Science and Technology.
ANSYS user’s manual (Version7.1). Houston, Swanson
Analysis Systems Inc., 2003.
Badran, S.F., Nassef, A.O. & Metwalli, S. 2007. Stability
of Y-stiffeners in ship plating under uniaxial compres-
sive loads. Ships and Offshore Structures 2(1):87–94.
Badran, S.F., Nassef, A.O. & Metwalli, S. 2008. A com-
parison of buckling strength of Y-and T-stiffeners in
ship plating. Marine technology. 45(3):125–131.
Badran, S.F., Nassef, A.O. & Metwalli, S. 2009. Y-stiffened
panel multi-objective optimization using genetic
algorithm. Thin-Walled Structures. 47(11):1331–1342.
264
MARSTRUCT.indb 264
Khedmati, M.R., Zareei, M.R. & Rigo, P. 2009. Sensitivity
analysis on the elastic buckling and ultimate strength
of continuous stiffened aluminium plates under com-
bined in-plane compression and lateral pressure.
Thin-Walled Structures 47(11):1232–1245.
Klanac, A., Ehlers, S., Tabri, K., Rudan, S. &
Boekhuijsen, J. 2005. Qualitative design assessment of
crashworthy structures. Maritime transportation and
exploitation of ocean and coastal resources. London:
Taylor and Francis Group.
Kristensen, Q.H.H. & Moan, T. 1999. Ultimate strength
of aluminium plates under biaxial loading. In: Pro-
ceedings of the fifth international conference on fast sea
transportation, New York.
Ludolphy, Hans. 2001. The unsinkable ship-development
of the Y-shape support web. Royal Schelde Shipyard.
Netherlands Proceedings of the Second International
Conference on Collision and Grounding of Ships.
Copenhagen, Denmark.
Paik, J.K. et al. 2009. Mechanical collapse testing on
aluminum stiffened panels for marine applications.
Ship Structure Committee Report, SSC-4512008,
Washington D.C. (http://www.shipstructure.org).
2/18/2011 5:47:28 PM
 Fatigue strength

MARSTRUCT.indb 265 2/18/2011 5:47:28 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Fatigue of high-speed aluminium ships: A master curve

formulation

J.H. den Besten & R.H.M. Huijsmans

Delft University of Technology (Ship Hydromechanics and Structures Department), Delft, The Netherlands

ABSTRACT: Aim is to develop a design method concerning high-cycle fatigue of arc-welded joints
in high-speed aluminium ships. Accordingly, a Linear Elastic Fracture Mechanics (LEFM) based
non-similitude two-stage (micro- and macro) crack propagation model is proposed and elaborated into
a fatigue master curve formulation. First, an analytical, parametric expression of the micro-crack propa-
gation dominating Stress Concentration Factor (SCF) related weld notch stress distribution along the
(virtual) crack has been developed for welded basic joints. This expression is used to correct the macro-crack
propagation governing LEFM parameter, the Stress Intensity Factor (SIF). Subsequently, a crack prop-
agation model is developed and successfully verified using aluminium Centre Cracked Tension (CCT)
specimen experimental data. Finally, a fatigue master curve formulation is obtained and satisfactorily
validated using fatigue test data of aluminium welded basic joints.

1 INTRODUCTION moving crack micro-crack macro-crack final

dislocations nucleation propagation propagation fracture

crack initiation crack propagation

Fatigue is a governing damage mechanism in alu-
minium high-speed ship structures. In this respect,
welded joints are often the most critical parts.
The maritime innovation project “VOMAS”
has been initiated to develop a (high cycle) fatigue
design method for these type of structures. Part of
the scope is to develop a fatigue master curve for arc-
welded aluminium joints, inspired by the impressive
work of (Atzori et al. 1997, 2005), (Lazzarin et al.
1996) and (Dong et al. 2001, 2003a, 2003b). A sub-
stantial part of the work presented here is already
published in (den Besten & Huijsmans 2010). How-
ever, this part is considered to be essential for a
thorough understanding and hence included here
as well. The illustrative examples are different.
Fatigue, cyclic loading induced progressive
structural damage, is concerned with crack initia-
tion and propagation, as shown in (Figure 1).
Governing parameter in the crack initiation
period, a surface phenomenon, is the Stress Con-
centration Factor (SCF). Micro-crack propagation
is included in this period because of the low crack
propagation rate. When the crack penetrates into
the material, the material bulk property induced
macro-crack propagation resistance is taking
over control. Governing parameter in this crack
propagation period is the Stress Intensity Factor
(SIF) K. Both parameters, the SCF and SIF, are
geometry and loading determined.
For high-speed ship structures operating in the
high-cycle fatigue region, it is assumed that crack
SCF SIF
Figure 1. Fatigue fracture scheme (Schijve 2001).
propagation dominates, since it is inevitable that
flaws already exist because of arc-welding as join-
ing method. Consequently, the Linear Elastic Frac-
ture Mechanics (LEFM) principles are adopted.
Crack propagation consists of a micro-crack
and macro-crack part and both are considered to
be important. The micro-crack part in particular
because it may constitute a large part of the fatigue
life as a result of the low crack propagation rate.
However, the crack propagation rate, basis of
the fatigue life calculation equation, shows in a
number of cases non-similitude behaviour in the
micro-crack propagation region and even in the
lower part of the macro-crack propagation region
for base material as well as weld and Heat Affected
Zone (HAZ) material. It means that the crack
propagation rate is not fully determined by K. This
non-similitude behaviour has been modelled by
correcting the macro-crack propagation govern-
ing LEFM parameter, the SIF K, for micro-crack
propagation using the SCF related weld notch
stress distribution.
In this paper, weld notch stress formulations
are determined and used to define the weld
SIF’s for weld toe failure cases and to develop a

267

MARSTRUCT.indb 267 2/18/2011 5:47:28 PM

 non-similitude two-stage crack propagation model.
This model is verified using aluminium CCT speci-
men experimental data and elaborated to a fatigue
master curve formulation, which is validated using
some fatigue test data of aluminium standard
specimens.
2 WELD NOTCH STRESS DISTRIBUTIONS
Analytical, parametric weld notch stress formulations—
with its characteristic singularity—are developed
for basic welded joints, which take the involved
geometry parameters into account, e.g. the base
plate thickness tb, the cross plate thickness tc, the
weld length and height lw and hw and the notch
radius ρ.
A weld notch stress distribution σw (r/tb) is
assumed to be a linear superposition of an equi-
librium equivalent stress part (the linear far field
stress) and a self-equilibrating stress part (consist-
ing of a non-linear notch stress: Williams’ asymp-
totic solution and a linear weld geometry induced
bending contribution), as shown in (Figure 2) for
a Full Penetration (FP), Double Sided (DS) cruci-
form joint.
The far field stress distribution σf (Equation 1) is
the linear stress distribution in the cross-section at
the weld toe. It is consistent with the far field stress
definition in the fracture mechanics context and
characterised by the structural stress amplitude
σs as already introduced by (Niemi et al., 2004) and
the bending stress ratio R.
⎧ ⎛ r ⎞⎫
σ f = σ s ⎨1 + 2 ⋅ R ⋅ ⎜ ⎟ ⎬ (1)
⎩ ⎝ tb ⎠ ⎭
For shell plated (ship) structures, the load-
ing is predominantly membrane and bending
induced. Hence, in a Finite Element (FE) envi-
ronment, σs is determined using the shell element
m b
+ change of stiffness, a counter clockwise bending
=
s sition of its membrane force and bending moment
r
tb
Fm = +
Mb
Figure 2. Weld notch stress distribution.
268
MARSTRUCT.indb 268
nodal force related membrane stress and bending
stress amplitudes σm and σb and signs sgn(Fm)
and sgn(Mb), as shown in (Equation 2). Note that
monotonic as well as non-monotonic weld notch
stress distributions are considered.
σs = sgn(Fm) · σm − sgn(Mb) · σb (2)
The minus sign in (Equation 2) is the result of
the coordinate system used: counter-clockwise
positive.
The bending stress ratio R (Equation 3) describes
the amount of bending relative to σs. A positive R
indicates that the notch stress distribution will be
non-monotonic; a monotonic weld notch stress
distribution appears in case of a negative R.
σb
R sg (M b ) ⋅ (3)
σs
For Williams’ asymptotic solution (Equation 4),
the normal stress component σθθ with particular
stress angle β has been used (Williams 1952) which
denotes for arc-welded geometries:
λ s −1
⎛ r⎞
σ θθ μ s λ s ⋅ (λ + ) ⋅ ⎜ ⎟
⎝ tb ⎠
⋅ ⎣cos {(λ
⎡ ) } χs ⋅ cos {(λ − ) ⋅ β }⎤⎦
λ a −1
⎛ r⎞
+ μa ⋅ λ a ⋅ ( + )⋅⎜ ⎟
⎝ tb ⎠
⋅ ⎡⎣si {( } {
+ ) ⋅ β − χa ⋅ sin (λ a − ) ⋅ }⎤⎦ (4)
The symmetric and anti-symmetric parameters
λs and λa are the eigenvalues; χs and χa the corre-
sponding eigenvalue coefficients. The parameters
μs and μa can be solved for using force and moment
equilibrium. For joints like the FP DS cruciform
joint, symmetric with respect to (tb/2), these con-
ditions are not sufficient and an additional sym-
metry condition is required: dσw{(r/tb) = ½}/dr = 0.
Finally, a least squares solution is obtained.
As already shown in (Figure 2) for the FP DS
cruciform joint, the weld geometry introduces a
moment; a weld geometry induced bending stress
σgb, (Equation 5) for this type of joint. A superpo-
contribution is obtained using a basic weld geom-
etry beam FE model. The weld geometry induced
bending factor Cgb relates σgb to σs.
⎧ ⎛ r⎞ ⎫ t
σ gb σ s ⋅ C ggb ⋅ ⎨4 ⋅ ⎜ ⎟ − 1⎬ for 0 < r ≤ b (5)
⎩ ⎝ tb ⎠ ⎭ 2
Note that this stress component is not part of
the equilibrating far field stress.
2/18/2011 5:47:29 PM
 tc/t b = 1.0; hw/lw = 1.0; lw/t b = 1.0; ρ /t b = 0.0 stress part related handbook solutions for
0.0
standard fracture mechanics specimen (remember
0.1 its consistency with the far field stress definition in
0.2 fracture mechanics context) and a self-equilibrating
0.3 unit stress part induced correction to cover the
weld geometry induced (singularity) effects, i.e. to
0.4
cover the micro-crack effects.
r/t b [ - ]

0.5 weld notch stress σ w /σ s

The crack opening mode, mode I, is consid-
far field stress σ f /σ s
0.6
FE result
ered to be dominant for shell plated (ship) struc-
0.7 tures, since the base plate thickness tb is the critical
0.8
parameter and the loading is predominantly mem-
brane and bending induced. The tearing mode,
0.9
mode III, which can be important at locations of
1.0
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
high in-plane shear stresses, is ignored for now.
σw /σs, σf /σs [ - ] Using LEFM, according to (Albrecht et al.,
1977) the weld SIF has been written as a product
Figure 3. Notch stress distribution FP DS cruciform of crack size dependent correction factors Ym and
joint (monotonic). Yg and the SIF for the basic case: K = σ√ (πa).

K σs a ⋅Ym
⋅ ⎡⎣sgn ( Fm ) Ygm
tc/t b = 1.0; hw/lw = 1.0; lw/t b = 1.0; ρ /t b = 0.0
0.0

0.1 +R ⋅ { (Fm ) ⋅YYggm (Mb ) ⋅Ygb }⎤⎦ (6)

0.2

0.3
The geometry factor Yg includes a linear super-
position of the membrane and bending contri-
0.4
bution, Ygm and Ygb, obtained from (Tada et al.,
r/t b [ - ]

weld notch stress σ w /σ s

0.5
far field stress σ f /σ s
0.6
FE result
0.7
0.8
0.9
1.0
-1.5 -1.0 -0.5 0.0
σw /σs, σf /σs [ - ]
Figure 4. Notch stress distribution FP DS cruciform
joint (non-monotonic).
Two example weld notch stress distributions for
a FP DS cruciform joint are shown in (Figures 3–4)
and present a monotonic and a non-monotonic
case, a pure bending and combined membrane
and bending case respectively. Note that the notch
effect in the symmetric part is ignored since it is
assumed only to be important for a micro-crack
and its propagation. The analytical notch stress
formulations and FE results match quite well and
seem to be a sound basis to define the weld toe fail-
ure related SIF’s.
3 WELD STRESS INTENSITY FACTORS
Generalised K solutions for welded joints are not
available and therefore determined for weld toe
induced failure cases using equilibrium equivalent
2000), Murakami 1987) or (Newman et al., 1981)
for edge crack or semi-elliptical crack configura-
tions and accounts for the finite thickness as well
as free surface effects. The latter only in case of
weld toe failure. It covers the macro-crack part.
The magnification factor Ym takes the micro-
0.5 1.0 1.5 2.0 2.5 3.0 3.5
crack part into account. The self-equilibrating
unit stress σ(r), an uncracked geometry parameter
along the (virtual) crack path, is applied as crack
face pressure:
⎛ 2⎞
a
σ (r )
Ym = ⎜ ⎟ ⋅ ∫ dr (7)
⎝ π⎠ 2
r2
0 a
Weld SIF’s for a FP DS cruciform joint are
determined and compared to FE solutions, pre-
sented in (Figures 5–6) using the correction factors
Ym, Yg and Ym·Yg for two examples in edge crack
configuration; cases with a monotonic and non-
monotonic stress distribution. The equilibrium
equivalent stress amplitudes are the same; the bend-
ing stress amplitude σb is half the membrane stress
amplitude σm and sgn(Mb) is−1 for the monotonic
case and +1 for the non-monotonic one.
For both cases, it turns out that Ym dominates
for {(a/tb) ≤ 0.2}, with the notch effect, Williams’
asymptotic solution, governing {(a/tb) ≤ 0.1} and
the weld geometry induced bending dominant at
{0.1 < (a/tb) ≤ 0.2}; Yg dominates {0.2 < (a/tb) ≤ 1.0}.
Knowing that the area below the curves Ym·Yg
influences the fatigue life, the difference between

269

MARSTRUCT.indb 269 2/18/2011 5:47:31 PM

 tc/t b = 1.0; hw/lw = 1.0; lw/t b = 1.0; ρ /t b = 0.0 10-4
5.0 III
4.5 Yg
Ym
4.0
YmYg
10-5
3.5 FEM
Yg, Ym, YmYg [ - ]

da/dn
3.0

2.5
II
10-6
2.0

1.5

1.0

0.5 10-7

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
a/tb [ - ] I
-8
10
Figure 5. SIF correction factors, monotonic case. 100 K th 101 102 K c 10
K

tc/t b = 1.0; hw/lw = 1.0; lw/t b = 1.0; ρ /t b = 0.0

Figure 7. Characteristic crack propagation rate curve.
5.0

4.5

4.0 macro crack propagation region) and region III

3.5
(unstable crack propagation, up to final fracture).
Because of the high crack propagation rate in
Yg, Ym, YmYg [ - ]

3.0 Yg
region III, the number of cycles consumed in this
Ym
2.5
YmYg
region is quite small and it is considered to be not
2.0 FEM
that important from fatigue (design) point of view.
1.5
Hence, a well-known Paris based two-stage, region
I and II, crack propagation model is introduced:
1.0

γ
⎧ ⎛ ΔKth ⎞ ⎫
0.5
da
⎟ ⎬ ⋅( K)
m
= C ⋅ ⎨1 − ⎜ (8)
⎩ ⎝ ΔK ⎠ ⎭
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 dn
a/tb [ - ]

In (Equation 8), the constants C, γ and m have

Figure 6. SIF correction factors, non-monotonic case. to be determined by experiment. The crack prop-
agation rate (da/dn) is fully determined by (ΔKth,
ΔK), based on the hypothesis of similitude: similar
the monotonic and non-monotonic case is quite
conditions applied to the system will have similar
clear and its shows the importance of including
consequences. In other words: two cracks with dif-
the signs sgn(Fm) and sgn(Mb). Please note that the
ferent crack length but the same SIF range ΔK,
monotonic case is still the worst case scenario since
propagate at the same rate.
calculation of K requires σs as well, which is three
However, it has been shown by (Jones et al.,
times as high for the monotonic case compared to
2007) and (Molent et al., 2008) that similitude in
the non-monotonic one.
region I, even in the lower part of region II, can-
not generally be assumed to be true: it brakes
down for micro-cracks. The crack growth model
4 NON-SIMILITUDE CRACK
of (Frost & Dugdale, 1958), assuming that the
PROPAGATION
crack propagation rate is proportional to the crack
A cyclic loading introduces a cyclic stress inten- length: da/dn = C · a (verified by experiment), is
sity ΔK and may initiate crack propagation. The generalised by (Jones et al., 2008) to:
well-known characteristic crack propagation rate m
curve, (da/dn – ΔK) with n the number of cycles, da 1 −
⋅( K)
m
=C a 2 (9)
shown in (Figure 7), is divided into 3 regions: region dn
I (near-threshold, slow micro crack propagation
region; below the threshold value ΔKth crack propa- In (Equation 9), a−m/2, a modification of ΔK
gation does not occur), region II (linear, steady state for the lower part of region II, neutralizes the

270

MARSTRUCT.indb 270 2/18/2011 5:47:33 PM

 √a dependency in the macro-crack propagation
region to keep (da/dn) proportional to a.
In (den Besten & Huijsmans 2010), the region
I factor {1−(ΔKth/ΔK)}γ in (Equation 8) was pro-
posed to be replaced by Ymγ: both factors show
similar behaviour and dominate the micro-crack
(propagation) region, region I. Application to Sin-
gle Edge Notch Tension (SENT) aluminium speci-
men experimental (da/dn)−ΔK data demonstrated
that γ = 1.
Comparing (Equations 8–9), a modification
for the lower part of region II is proposed to be
included in (Equation 8): replace {1−(ΔKth/ΔK)}γ
by Ym1–2 m/3:
2m
( )
da 1 − m
= C Ym 3 ⋅ Kg
dn
In (Equation 10), Ym with γ = 1 as mentioned
γ = ∑ ⎨ μi ⋅ λ i ( λ i + ) ⋅ ( − χi ) ⋅ ⎜ ⎟
before is naturally included and a modification for
the lower part of region II is added. Note that ΔKg
only includes Yg as correction factor.
The proposed model has been verified using
experimental data of aluminium Centre Cracked
Tension (CCT) specimens, shown in (Figure 8).
4.1 CCT specimen—analysis
A solution for Kg of this specimen is available from
handbook solutions. However, to achieve a consist-
ent approach between the uncracked geometries of
arc-welded joints which are assumed to fail at the
weld toe and geometries with initial gap (crack)
2 · ag, like CCT specimen and arc-welded joints
which are assumed to fail at the weld root, the coor-
dinate system has been shifted for a to consider the
remaining uncracked geometry, as already shown
in (Figure 8).
The far field stress denotes: σs = σm. In analogy
to the procedure previously described, the notch
stress distribution at the cross-section of the
gap (crack) is required to cover the micro-crack
propagation effects. The linear stress components
σsc, σmc and σbc are obtained using a coarse (8 node
Lagrange element) meshed plain strain FE model.
Solving the eigenvalue problem that corresponds
to Williams’ asymptotic solution for 2α = 2π, the
crack configuration, it turns out that λs = λa = λ
and multiple feasible solutions has been found.
⎛ 1⎞ ⎛ 3⎞ ⎛ 5⎞ ⎛ 7⎞ ⎛ 9⎞
λ i = ⎜ ⎟ , ⎜ ⎟ , ⎜ ⎟ , ⎜ ⎟ , ⎜ ⎟ , ...
⎝ 2⎠ ⎝ 2⎠ ⎝ 2⎠ ⎝ 2⎠ ⎝ 2⎠
Note that only the first eigenvalue introduces
singular behaviour (Equati`on 4). Because of the
multiplicity of solutions, only 5 terms are included.
The notch stress distribution σc (r/tb′) becomes
for β = 0:
(10)
5 ⎧ λ −1
σc ⎪ ⎛ r ⎞ i ⎫⎪
⎬ (11)
σ s i =1 ⎪ ⎝ tb ′ ⎠ ⎪⎭
⎩
A weld geometry induced bending term is not
required, just like a far field bending stress projec-
tion σf (Equation 1); bending is incorporated using
the non-singular terms provided by Williams’
asymptotic solution. The coefficients μi have been
solved for using equilibrium conditions and 3
additional relations: 3 stress points σc(i) at {r(i)/tb′},
obtained using the FE model mentioned before.
An example is shown in (Figure 9).
The magnification factor Ym requires particular
attention. For geometries with an initial gap, Ym is
defined as:
Yn
Ym = (12)
Yl
The correction factor Yn is obtained using the
non-linear self-equilibrating unit stress and (Equa-
tion 7) taking the shift in coordinate system into
ag / b = 0.500
0.0

0.1

0.2

0.3

b 0.4
r/t b' [ - ]

0.5
ag 0.6
ag r sc, mc , bc m 0.7 crack notch stress σ c /σ sc

a 0.8
far field stress σ f /σ sc
FE result
b t b'
0.9

1.0
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
σc /σsc, σf /σsc [ - ]

Figure 8. Centre Cracked Tension (CCT) specimen. Figure 9. CCT specimen crack notch stress distribution.

271

MARSTRUCT.indb 271 2/18/2011 5:47:35 PM

 ag / b = 0.500 been conducted in air. The original data contains
5.0
some non-similitude behaviour, as shown in
4.5 Yg
(Figure 11 & 13). The (Figures 12 & 14) show the
Yn/Yl
4.0
-2/3
corresponding results using (Equation 10). The
Yg, Yn / Yl , ( Yn/Yl ) -2/3 [ - ]

( Yn/Yl )
3.5
FEM
factor (Yn/Yl) has been moved to the left-hand side
3.0
of this equation to achieve single-slope, region II
behaviour.
2.5

2.0 Al 5083, T-L config.

10-2
1.5

1.0 10-3
40 10 -5-

rolling direction
0.5 -4
10

(da/dn) [mm/cycle]
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
10-5
a/t b' [ - ]

10-6
Figure 10. CCT specimen correction factors.
base material; Rs = -1.0
10-7
base material; Rs = 0.0

account. Note that with respect to (Equation 11) 10-8

101 102 103 104
the linear bending stress component has to be ΔK [MPa √ mm]
subtracted. The linear unit stress correction factor
Yl denotes: Figure 11. (da/dn)—ΔK data, CCT specimen, base
material, (Sonsino et al., 2003).
a
⎛ 2⎞ 1
Yl = ⎜ ⎟ ⋅ ∫ dr (13) -2 Al 5083, T-L config.

⎝ π⎠
10

(ag + a)
2
ag −r 2
-3
10
(da/dn) . (Yl / Yn) [mm/cycle]

-4
10

⎛ 2⎞ ⎛ ag ⎞
Yl = 1 − ⎜ ⎟ ⋅ arcsin ⎜ ⎟ 10
-5

⎝ π⎠ ⎝ ag + a ⎠
-6
10

In agreement with the uncracked geometry for- -7 base material; Rs = -1.0

10
mulation for arc-welded joints, assumed to fail at base material; Rs = 0.0

the weld toe, Yl = 1.0 for ag = 0 and consequently 10

-8

Ym = Yn. An example for the correction factors is 10

1
10
2
10
3
10
4

ΔK . (Yl / Yn) 2/3 [MPa √ mm]

shown in (Figure 10).
Note that in comparison with (Figures 5–6),
Figure 12. (da/dn)—ΔK data, CCT specimen, base
the SIF is fully determined by the geometry fac-
material, two-stage non-similitude model.
tor Yg, since the gap (crack) effect is incorporated
in the SIF by definition. The magnification factor -2 Al 5083, T-L config.
Ym is only used to describe non-similitude crack 10

propagation behaviour. For increasing crack size -3

(a/tb′), the factors (Yn/Yl) and (Yn/Yl) equilibrate 10

40 10 -5-
rolling
around 1.0 as a result of using the self-equilibrating -4
direction
10
unit stress part.
(da/dn) [mm/cycle]

-5
10

4.2 CCT specimen—experimental data -6

10
Crack propagation data for aluminium 5083 CCT
weld material; Rs = -1.0
specimens in T-L configuration, published by 10
-7

weld material; Rs = 0.0

(Sonsino et al., 1999), has been used to validate
-8
the two-stage non-similitude crack propagation 10 1 2 3 4
10 10 10 10
model. Two data sets were provided, including ΔK [MPa √ mm]
data in base material and in weld material. The
constant amplitude tests are applied at 2 different Figure 13. (da/dn)—ΔK data, CCT specimen, weld
stress ratios Rs = 0.0 [−] and Rs = −1.0 [−] and have material, (Sonsino et al. 2003).

272

MARSTRUCT.indb 272 2/18/2011 5:47:37 PM

 -2 Al 5083, T-L config. nominal stress formulation; Tσ = 1 : 2.41
3
10 10

-3

Δσnom
10
(da/dn).. (Yl / Yn) [mm/cycle]

-4
10 2
10

-5
10

-6
10 1
10
T-joint
-7 weld material; Rs = -1.0
10 Butt joint
weld material; Rs = 0.0 Cruciform joint
Longitudinal stiffener joint
-8
10 1 2 3 4 0
10 10 10 10 10 3 4 5 6 7 8
ΔK . (Yl / Yn) 2/3 [MPa √ mm] 10 10 10 10 10 10
N

Figure 14. (da/dn)—ΔK data, CCT specimen, weld

material, two-stage non-similitude model. Figure 15. Fatigue test data: nominal stress formu-
lation.

5 MASTER CURVE FORMULATION Tσ (Sonsino 2007) has been adopted to quantify

Using (Equation 10), separating variables and inte-
gration of both sides yields an explicit formulation
for the number of cycles to failure N:
m
1-
2
tb
N=
C ⋅ ( Δσs )
m
The crack propagation integral I(R) is a function
of the bending stress ratio R considering the
membrane and bending contribution in Yg.
af
tb
I (R) = ∫ d⎜ ⎟
⎧ m⎫
ai
⎪ 1
tb
⎨Ym
⎪ ⎝ b⎠ ⎪
⎩ ⎭
To validate (Equation 14), some fatigue test data
series for basic welded geometries: T-joints, butt
joints, cruciform joints and longitudinal stiffener
joints—material: aluminium 5000 and 6000 series—
have been used. For now, only weld toe failure cases
are considered. The data is kindly provided by (van
Straalen et al., 1994a, 1994b, 1994c), (Maddox
1995), (Meneghetti 1997) and (Morgenstern 2006).
The stress ratio Rs for all specimens is in the range
0.0 … 0.1 [−], comparable to an important type of
loading for high-speed ship structures: hydrody-
namic impact. The base plate thickness tb is in the
range 5.0 … 25.0 [mm]. The nominal stress data
Δσnom−N is shown in (Figure 15).
Using a linear regression analysis, the mean
line and some survivability lines: 2.5 [%], 10.0 [%],
90.0 [%] and 97.5 [%] are determined. The parameter
the scatter:
σ a ( Ps = )
Tσ = 1:
σ a ( Ps = ) (16)
It turns out that for the nominal stress related
⋅ I (R ) (14) Tσ, the development of a fatigue master curve for-
mulation seems to be quite a heavy task.
One of the major causes in scatter is the weld-
ing induced second order bending stress, which is
not the same for the different series. For almost all
series some information is available and the next
step is to replace the nominal stress by the far field
stress. Results are shown in (Figure 16) and the
1 ⎛ a⎞ reduction for Tσ is already considerable.
⎝ tb ⎠ (15) Application of the master curve approach
m
m ⎛ a⎞ ⎪
− 2 requires information about initial crack size, which
2 Yg ⋅ ⎜ ⎟ ⎬ is not available. From design point of view, it is not
t
that important which value will be used for fatigue
life calculation, provided that it is same as used
for the master curve parameter calculation and
the joint quality under design will be comparable
to the fatigue test data. Its uncertainty is included
in the master curve. The chosen initial crack size:
(ai/tb) = 0.02 [−]. For the 5 [mm] joints, the smallest
tb value, ai = 0.1 [mm] and considered to be at the
lower boundary for application of LEFM.
Another point of interest is that the specimens
are tested until complete fracture. However, the
specimen width differs from one series to another,
a 3D effect. This is included for the T-joints, butt
joints and cruciform joints using the semi-elliptical
crack Yg formulations; the base plate thickness to
base plate width ratio is considered to define the
semi-elliptical crack aspect ratio (a/c). For the lon-
gitudinal stiffener joints, the edge crack Yg formu-
lations are suitable.

273

MARSTRUCT.indb 273 2/18/2011 5:47:40 PM

 3
structural stress formulation; Tσ = 1 : 1.78 The non-similitude generalised Frost & Dugdale
10
crack propagation model has been modified, correct-
ing the macro-crack propagation governing LEFM
Δσs

parameter, the SIF, using the micro-crack propaga-

2
tion dominating SCF related self-equilibrating unit
10 stress part along the virtual crack. Promising results
are obtained in comparison with experimental crack
propagation data of aluminium CCT specimens for
base and weld material.
10
1
A master curve formulation to cover differ-
T-joint
Butt joint ent type of welded joints has been developed.
Cruciform joint Validation using fatigue test data show satisfying
Longitudinal stiffener joint
results.
0
10 3 4 5 6 7 8
10 10 10 10 10 10
N REFERENCES
Figure 16. Fatigue test data: structural stress Albrecht, P. & Yamada, K. 1977. Rapid Calculation
formulation. of Stress Intensity Factors. Journal of the Structural
Division 103, American Society of Civil Engineers
master curve formulation; Tσ = 1 : 1.50 (ASCE): 377–389.
3
10 Atzori, B., Lazzarin, P. & Tovo, R. 1997. Stress Distribu-
tions for V-shaped Notches Under Tensile and Bend-
m
1
tb 2m I(R)

ing Loads. Fatigue & Fracture of Eng. Materials 20 (8):

Δσs
.

1083–1092.
2-m

2 Atzori, B., Filippi, S., Lazzarin, P. & Berto, F. 2005.

10
Stress Distributions in Notched Structural Compo-
nents under Pure Bending and Combined Traction
and Bending. Fatigue & Fracture of Engineering Mate-
rials 28: 13–23.
1
10 Besten, J.H. den & Huijsmans, R.H.M. 2010. Fatigue in
T-joint
Butt joint
High-Speed Ships: Crack Propagation in Aluminum.
Cruciform joint Proc. 11th Int. Symp. on Practical Design of Ships
Longitudinal stiffener joint and Other Floating Structures, PRADS 2010, Rio de
0 Janeiro, RJ, Brazil.
10
10
3
10
4
10
5
10
6
10
7
10
8 Dong, P. 2001. A Structural Stress Definition and Numer-
N ical Implementation for Fatigue Analysis of Welded
Joints. International Journal of Fatigue 23: 865–876.
Figure 17. Fatigue test data: master curve formulation. Dong, P. 2003a. A Robust Structural Stress Method for
Fatigue Analysis of Ship Structures. Proc. 22nd Int.
Conf. on Offshore Mechanics and Arctic Engineering,
Applying a non-linear regression analysis for OMAE 2003.
the master curve formulation using Δσs as predic- Dong, P., Hong, J.K. & Cao, Z. 2003b. Stresses and
tor and N as response, its parameters are deter- Stress Intensities at Notches: ‘Anomalous Crack
mined. Results are shown in (Figure 17); the slope Growth’ Revisited. International Journal of Fatigue
m = 3.46. 25: 811–825.
Frost, N.E. & Dugdale, D.S. 1958. The propagation
A scatter parameter Tσ not exceeding 1:1.50 is of fatigue cracks in sheet specimens. Journal of the
a value which can be found for good workman- Mechanics and Physics of Solids 6: 92–110.
ship welds. In this case, it might be a result of an Jones, R., Chen, B. & Pitt, S. 2007. Similitude: Fatigue
improved fatigue model. cracking in steels. Theor. & Appl. Fract. Mech. 48:
161–168.
Jones, R., Pitt, S. & Peng, D. 2008. The generalized Frost-
6. CONCLUSIONS Dugdale approach to modelling fatigue crack growth.
Engineering Failure Analysis 15: 1130–1149.
Analytical, parametric weld notch stress and Lazzarin, P. & Tovo, R. 1996. A Unified Approach to
the Evaluation of Linear Elastic Stress Fields in the
SIF formulations have been developed for Neighborhood of Cracks and Notches. Int. J. of
basic welded joints. Geometry parameters are Fracture 78: 3–19.
(implicitly) included as variables in order to allow Maddox, S.J. 1995. Scale Effect in Fatigue of Filled
to study the influence of joint modifications on Welded Aluminium Alloys. Proc. 6th Int. Conf. on
the fatigue life. Aluminium Weldments. Miami, FL, USA.

274

MARSTRUCT.indb 274 2/18/2011 5:47:42 PM

 Meneghetti, G., Tovo, R., Patricolo, M. & Volpone, M.
1997. Experimental analysis of the fatigue strength of
welded structural details in aluminium alloy. Proc. of
the 26th AIAS Nat. Conf., Catania, Italy.
Molent, L., Jones, R. & Pitt, S. 2008. Similitude and the
Paris crack growth law. Int. Journal of Fatigue 30:
1873–1880.
Morgenstern, C. 2006. Kerbgrundkonzepte für die
schwingfeste Auslegung von Aluminium schwei-
βverbindungen am Beispiel der naturhärteten Legierung
AlMg4,5Mn (AW-5083) und der warmausgehärteten
Legierung AlMgSi1 T6 (AW-6082 T6). Fraunhofer-
Institut für Betriebsfestigkeit (LBF), Darmstadt.
LBF-report No. FB-231.
Murakami, Y. 1987. Stress Intensity Factors Handbook
Vol. 1. 1st ed., Pergamon Press.
Newman, J.C. & Raju, I.S. 1981. An Emperical Stress
Intensity Factor Equation for the Surface Crack.
Engineering Fracture. Mechanics 15: 185–192.
Niemi, E., Fricke, W. & Maddox, S. 2004. Structural
Stress Approach to Fatigue Analysis of Welded
Components—Designer’s Guide, Final Draft. IIW
Doc. XIII-1819-00. The Int. Institute of Welding.
Schijve, J. 2001. Fatigue of Structures and Materials, 1st
ed., Kluwer Academic Publishers.
Sonsino, C.M. 2007. Course of SN-curves especially
in the high-cycle fatigue region with regard to com-
ponent design and safety. International Journal of
Fatigue 29: 2246–2258.
275
MARSTRUCT.indb 275
Sonsino, C.M., Brandt, U., Lehrke, H.P. & Radaj, D. 1999.
Fatigue Assessment of Welded Joints in AlMg 4.5Mn
Aluminium Alloy (AA 5083) by Local Approaches.
International Journal of Fatigue 21: 985–999.
Straalen, IJ.J. van, Soetens, F. & Dijkstra, O.D. 1994a.
EUREKA 269—Fatigue Tests on Transverse Butt
Welds. TNO-report 94-CON-R1563. TNO Building
and Construction Research.
Straalen, IJ.J. van, Soetens, F. & Dijkstra, O.D. 1994b.
EUREKA 269—Fatigue Tests on Transverse Non-Load
Carrying Fillet Welds. TNO-report 94-CON-R1565.
TNO Building and Construction Research.
Straalen, IJ.J. van, Soetens, F. & Dijkstra, O.D. 1994c.
EUREKA 269—Fatigue Tests on Longitudinal
Non-Load Carrying Fillet Welds. TNO-report
94-CON-R1566. TNO Building and Construction
Research.
Tada, H., Paris, P.C. & Irwin, G.R. 2000. The Stress
Analysis of Cracks Handbook, 3rd ed., Professional
Eng. Publishing.
Williams, M.L. 1952. Stress Singularities Resulting from
Various Boundary Conditions in Angular Corners of
Plates in Extension. ASME Journal of Applied Mech.
19: 526–528.
2/18/2011 5:47:43 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Stress and strain-based approaches for fatigue life evaluation

of complex structural details

M. Biot & L. Moro

Division of Naval Architecture and Marine Technology, Department of Mechanical Engineering
and Naval Architecture, University of Trieste, Trieste, Italy

ABSTRACT: The present work starts from the idea to improve fatigue strength of an actual struc-
tural detail where welded joints are subject to proportional biaxial variable-amplitude loadings leading
to elastic-plastic strains. After a brief discussion on how to manage low-cycle fatigue in ship structural
design, the case study is presented. Basing on outcomes drawn from the analyses performed on the con-
sidered structural detail, a critical examination is carried out on the procedures applied to check fatigue
strength. A comparison of fatigue damages obtained from application of linear elastic approach and
elastic-plastic approach is outlined in order to analyze advantages and drawbacks of the two approaches
and their potential to be implemented in a design-oriented practical assessment procedure. Investigations
point out that nonlinear approach may be a valid alternative to linear one only if designer can control
safety factors relating to fatigue capability and has available deeper information on material behaviour
and load history experienced by the structure.

1 INTRODUCTION stress-flow along a structural discontinuity, multi-

In the pragmatic approach to fatigue design of ship
structural details, fully linear methods based on the
additive process of partial damage contributions
due to given cyclic load blocks, are today widely
implemented. The essential ingredients of such
methods are provided by the Palmgren-Miner’s
rule to summing up the partial damages and by the
conventional S-N curves to completely describe the
fatigue capability in terms of endurable stresses
at given load cycles (Radaj et al., 2006). Fatigue
assessments performed according to that principles
should guarantee from incipient technical cracks
and are the premises for the most up-to-date clas-
sification societies’ rules and guidelines for fatigue
strength checking (DNV 2005, LR 2001).
A lot of works gives emphasis on different ways
to apply the cumulative fatigue damage approach
based on S-N curves. The core of each proposed
method is the distinctive technique set up to iden-
tify a fatigue effective stress by a proper FE anal-
ysis and to define a peculiar stress manipulation
(ISSC 2009) to consider ancillary effects. Among
the familiar techniques, those based on the struc-
tural stress concept are the powerful ones for map-
ping hot spot areas and checking performance of
a variety of differently complex structural details
(Hobbacher 2007).
In very complex structural details where stress
concentrations are emphasized by a tortuous
axiality fatigue concepts need to be put to use,
along with a proper scheme to handle the large
elastic-plastic strain field which may arise around
the crack tip. Adjustments have been made to adapt
standard linear methods to cases where very-high
stress concentrations exist and the material local
response is a repeated plastic deformation. Nev-
ertheless, to deal with that kind of fatigue critical
sites in ship structures, conventional methods are
hardly valid unless high safety factors have been
applied. Therefore, the strain-based approach,
which considers the elastic-plastic strain range as
the governing load parameter, becomes the refer-
ence design method when fatigue failure is expected
to occur after relative low number of load cycles
(Wang et al., 2006).
Even if a well sound methodology is available
to perform nonlinear fatigue analyses, the guide-
line framework in which designers are moving is
actually very poor. In effect, standards and rules
appear not so structured to effectively support
fatigue design process in the cases when a nonlin-
ear approach needs to be implemented, such as in
complex structural details. In general, the lack of
both a standard procedure and common fatigue
endurance limits is being decisive to lead designer
to pass over and use traditional linear methods.
In the work here presented, an attempt for
application of a practical method to a case study is
outlined, with reference to a structural detail subject

277

MARSTRUCT.indb 277 2/18/2011 5:47:44 PM

 to a low cycle fatigue regime. Comparison between
outcomes of standard linear elastic approach and
elastic-plastic approach is performed in order to
highlight potential of the two methods in fatigue
design of ship structural detail.
2 THE ELASTIC-PLASTIC APPROACH
In the standard cumulative damage procedure, the
fatigue design of any structural detail is performed
by making use of a reference parameter s derived
by the stress-strain field measured or calculated in
close vicinity of the crack site.
The fatigue check is performed by comparing
s with the strength value S, which is endurable at
the same number of cycles N. Fatigue capability
is given in form of S-N curve. Depending on the
chosen reference parameter, fatigue assessment is
carried out by different methods.
In the basic “nominal stress approach”, the
far-field stress due to the forces and moments at the
potential site of cracking is regarded as the govern-
ing parameter (s = σn being σn the nominal stress).
By such an approach only classified structural details
may be investigated. The nominal stress approach is
replaced by a local approach when a high demand
of accuracy is needed, and, first of all, when com-
plex and non conventional structural details have to
be fatigue checked. In effect, in such case the con-
cept of nominal stress could lose meaning or to be
difficult to be unambiguously identified.
In the local approaches, the fatigue strength of
any structural detail is assessed by making reference
to the intensity of the stress or strain field meas-
ured or calculated just in the area at risk of crack
initiation—such a critical location is usually called
“hot spot”.
As analyses of structural discontinuities are
in general not affordable by analytical methods,
numerical analyses come out to be mandatory, in
routine verifications too. FE inherent uncertainties
reflect on the computed stresses and strains.
So, different methods have been conceived for
getting the more appropriate FE-calculated
fatigue-effective parameters.
In the “structural stress approach” only that
part of the local stress concentration is taken
into account which is related to the structural
geometry, excluding the local nonlinear stress peak
at the notch (s = σs being σs the structural stress).
The decisive issue in any structural-stress based
approach is how to obtain a significant structural
stress value, that is a stress neglecting the notch
effect but, at same time, effective in describing
the macrostructural behaviour of the structural
component, and so, reactive to very slight changes
in the local configuration.
278
MARSTRUCT.indb 278
The most common method to process FE outputs
in order to obtain a sound structural stress is to
resort to the concept of hot spot stress (s = σhs),
conventionally determined by performing a surface
stress extrapolation. The ratio between the hot spot
stress and the nominal stress is called Khs.
A more comprehensive and general method is
the one based on the concept of the notch stress
(s = σk), which is the total local stress at the root of
a notch, taking into account the stress concentra-
tion caused by both the component geometry and
the local notch. In the notch-stress approach, the
elastic microstructural notch support hypothesis
is accounted by proper manipulation of the FE
calculated elastic stresses.
Fatigue-effective (elastic-plastic) notch stress is
calculated by averaging the elastic stresses obtained
on a small area in front of the notch site. Methods
for lowering the theoretical FE-calculated elastic
stress at the notch are of major concern in the more
recent investigations. The result of the application
of the different procedures is generally expressed
in term of “fatigue notch factor” Kf, which may
be defined as the ratio between the fatigue-effective
notch stress and the nominal stress.
A more straightforward (and more conservative)
approach proceeding from the effective notch
approach is also used. In this method the fatigue
notch factor is replaced by the elastic “notch
stress concentration factor” Kt, defined referring
the theoretical elastic notch stress to the nominal
stress. According to this simplification, the notch
stress based method becomes a purely geo-
metric stress approach, where all geometric effects
are considered simultaneously.
In the “notch strain approach” the controlling
parameter is the strain (s = ε). Like the notch stress
approach, the notch strain approach puts the fatigue
life in explicit relation to structural geometry, weld
shape and material parameters. However, unlike
the former, it is able to take into account also the
macrostructural support effect which arises when
an appreciable local plastic deformation occurs at
the notch. Basically, strain based procedures have a
better formal consistency and very clearly explain
the elastic-plastic mechanism which takes place
around the notch root, i.e., the interaction between
the notch plastic zone and the elastic surrounding
material which constraints the deformations inside
the plastic zone.
To carry out a notch-strain based fatigue
assessment, the elastic-plastic strain S-N curve is
necessary, along with the cyclic stress-strain curve.
The cyclic stress-strain behaviour at the notch of a
structural component is obtained by cyclic strain-
controlled tests performed on a small unnotched
comparison specimen. Fatigue life of the structural
component is therefore closely connected with that
2/18/2011 5:47:44 PM
 of the specimen (they should be as soon as possible
consistent with each other, in terms of material
properties, surface condition, and size of the highly
stressed material volume at the notch root).
It is widely accepted to express the cyclic stress-
strain curve according to the Ramberg and Osgood
formulation:
1/ n ′
σa ⎛ σa ⎞
εa ε ael + εapl = +⎜ ⎟
E ⎝ K′⎠
where εa is the total local strain amplitude at the
notch in a cycle (i.e., the sum of the elastic and
plastic components), σa the local stress amplitude
at the notch, E is the elastic modulus and K′ and n′
are material parameters.
The “strain S-N curve” (also know as ε-N curve)
is a four parameter relationship which may be writ-
ten according to Manson and Coffin as:
σ ′f
εa = εael + εapl = ( )b + ε′f ( )c
E
where εf′ is the fatigue ductility coefficient, σf′ the
fatigue strength coefficient, N is the number of
cycles to crack initiation and the two parameters b
and c are called fatigue strength and fatigue ductil-
ity exponents, respectively.
The first part of Equation 2 is the so called
Basquin’s equation and refers to the elastic strain
related fatigue. The second part stands for the
dependency of fatigue endurance from the plastic
strains. The four parameters are specific for the
different materials. As for steel, exponent c ranges
from about −0.5 to −0.7 and exponent b ranges
from about −0.06 to −0.14. The terms εf′ and σf′
are somewhat related to the true fracture strain
and stress (εf and σf respectively) in a monotonic
tensile stress (Fuchs & Stephens 1980).
The strain S-N curve inclusive of the mean stress
effect according to Morrow is given by:
σ ′f σm
εa = εael + εapl = ( )b + ε′f (
E
where σm is the mean stress. Both formulations
describe crack initiation and are referred to a
technical crack ai of about 1 millimetre.
As elastic-plastic stress-strain field is complex to
be mapped by numerical FE procedures too, proper
methods to derive that by linear elastic FE analyses
have been set. In plate-type structures, Neuber’s
rule may be appropriate to convert an elastically
computed stress-strain pair into the corresponding
actual stress-strain pair.
Unlike in the case of thick plates (plane strain
condition), where local plastic deformation may
be considered as fully governed by the external
MARSTRUCT.indb 279
strain field, in thin plates in tension (plane stress
condition) the strain field which takes place through
most of the crack tip region is not considered to be
able to effectively restrain the deformation inside
the local plastic zone. Accordingly to that, in the
latter case, relationship between true strain εa and
linearly computed strain is derived by imposing
equal deformation energy. This is the base of the
Neuber’s rule, by which the conversion from the
(1) elastically computed to the actual elastic-plastic
stress-strain field is given by the simultaneous solu-
tion of Equation 1 and the following equation:
σ a,k
2
σ a,n
2
K t2 σ a,hs
2
εa σ a = = K t2 = 2
(4)
E E K hhs E
True strain εa may be calculated in thick plates
complying with the hypothesis of an effective con-
straint due to the external linear-elastic field. In
such case, elastically computed and actual stress-
strain field at the notch tip may be assumed exactly
(2) alike:
Kt
εa εa,kk = K t εa,n εa,hs (5)
K hhs
Elastically computed nominal pair to actual
notch pair conversion based on Equation 4 agrees
with measurements in plane stress cases (σaS, εaS),
while Equation 5 is properly applied in plane strain
cases (σaN, εaN). It is worth pointing out that Equa-
tion 4 has a general application, provided that a
proper effective stress σhs has been derived.
Fatigue checking based on the elastic-plastic
stress-strain field computed at the potential site
of cracking is usually performed according to
the Palmgren-Miner’s rule, which turns out to be
expressed as follows:
εa,max p( a )NL
DNL ( a) =∫ dεa (6)
0 N( a )
where load history is expressed by the long term
)c (3)
distribution of the cyclic loads (wave loads), given
by product between the probability density func-
tion p(•) of the independent statistical variable ε
and the total number of load cycles NL acting on
the structure during the entire lifetime of the ship.
Above outlined formulation to assess cumula-
tive damage is generally valid as far as cyclic loads
act at zero mean stress (R-value equal to −1), even
though mean stress effect may be embedded in the
ε-N capacity curve (Equation 3). The most suitable
procedure to be applied in case of different R-values
is based on the damage parameter concept.
The damage parameter P is a stress-strain field
identifier capable to define cycles of equal damage
regardless of the mean stress at which they are
279
2/18/2011 5:47:44 PM
 taking place. The basic expression for the damage
parameter P has been given by Smith, Watson and
Topper (Smith et al., 1970, as cited in Radaj et al.,
2006):
P ( a m ) εa E (7)
Among the various damage parameters
defined by different authors, that proposed in
Equation 7 implies the inconvenience of not cap-
turing the crack closure effect but, on the other
hand, is based on very simple concept and is of
straightforward use.
According to the damage parameter P defini-
tion, the cumulated damage is given by:
Pmax p( P ) N L
DNL ( P ) = ∫ dP (8)
0 N ( P )σ m = 0
In fatigue design of ship structures, load history
is first calculated in terms of stress ranges Δσ under
the hypothesis of linear elastic behaviour of mate-
rial and structure.
The Weibull probability density function pW(Δσ)
is referred to as a sound model for describing
the load history experienced by the structure in
the hot spot, which is expressed by the product
pW(Δσ) NL.
This approach is enforceable on condition that
the Weibull shape parameter kW could be deduced
by proper considerations on type and location of
the structural detail. On the other hand, the Weibull
scale parameter λW may be derived as function of
a known (Δσ0, Q(Δσ0)) pair, where Q(•) is the prob-
ability of exceedance associated to a significant
value of Δσ. In general, reference load cycle used
as input in FE calculation of Δσ0 has a probability
of exceedance equal to 10−4.
After applying linear to nonlinear conversion
on each stress-strain pair, a proper procedure to
reduce stress range load history to strain ampli-
tude load history has to be implemented. That
is made by resorting to the standard method for
variable changing in probability density functions.
In the present case, being relationship εa = g(σa,k),
where σa,k = Δσ/2, the monotonic function derived
by the simultaneous application of Equation 1 and
4 (or 5), the density function of the statistical vari-
able εa comes out to be:
1 3
p( ) = pW ( g 1( ε )) (9)
g ′( g 1( ε ))
The same applies to calculation of the p(P) den-
sity function, where the g(•) relationship is now
the expression of the damage parameter P = g(εa,
σa,k(εa)) for any given σm:
280
MARSTRUCT.indb 280
1
p( P ) = pW ( g 1( P )) (10)
g ′( g 1( P ))
Which σm mean stress level has to be considered
as a typical value for the entire load history, is left
to designer. In case of load histories which may
be directly related to the global deformation of
the ship, σm may be derived on the basis of the
σ0,max and σ0,min stresses obtained at the structural
hot spot by applying to hull girder the reference
load cycle complying with rules specifications.
Accordingly to that, if F is the generalized hull
girder section force supposed to govern the stress
level on the structural detail under consideration,
the σm mean stress may be roughly defined as the
local structural stress due to the sectional force
Fm = ½(F0,max + F0,min), where F0,min and F0,max are
the sectional forces given by application to the
hull girder of the reference load cycle. In detail,
values for σ0,max and σ0,min may be derived in terms
of hot spot stresses by applying a linear extrapo-
lation procedure on the FE calculated stresses
read out close to the hot spot. Then, a conversion
need to be performed from structural to notch
stresses, by resorting to a proper weld shape factor
Kw = Kt / Khs, usually defined in the classification
societies guidelines.
Under assumption of a linear response of hull
girder to the design wave loads, any FE calculated
stress may be regarded as linearly dependent from
primary loads. Therefore, a global FE analysis car-
ried out on the ship’s hull under a given primary
load, leads to a FFEA value and, in turn, to relevant
σFEA,hs values which all may be used for a direct
evaluation of Δσ0:
ΔF0
Δσ 0 = K w (σ 00,max
max − σ 0,min )hs = K w σ FEA,hs
(10)
FFFEA
where ΔF0 = (F0,max−F0,min), that is the double
amplitude of sectional force F due to application
of the reference primary cyclic load.
By summing up, according to prescription of
classification societies rules, both Δσ0 at a given
probability of exceedance and σm should be derived
together with a proper kW value for the Weibull
load history model. Those parameters are the very
basis to define the load history.
THE CASE STUDY
Case study concerns fatigue analysis of the
structural details at the openings which, verti-
cally lined up on a longitudinal bulkhead, lead
from each deck to a lift and stairs trunk (Biot
et al., 2009). At the corners of the openings high
2/18/2011 5:47:47 PM
 stress concentrations arise due to the high shear
deformation of the bulkhead caused by the bending
of the ship hull in waves (Figure 1).
A first analysis carried out on the whole ship
structure shows large critical stress fields at the
vertically aligned door openings. At the opening
corners a biaxial plane stress state takes place,
associated to a proportional variable-amplitude
loading.
In a first phase, a new configuration for the
main supporting structures has been looked for.
Alternative structural solutions have been con-
ceived for reducing the loading on the details,
where dynamic structural stresses localized at the
opening edge approach the yield stress of the mate-
rial, an AH 36 steel. Analysis has been carried out
on the FE model of the entire ship.
In order to increase the shear stiffness of the
hull girder in way of the openings, two directions
have been followed: on one hand, a local reinforce-
ment of longitudinal bulkheads and deck girders;
on the other hand, an alternate aft-forward longi-
tudinal shifting of the openings at the level of the
lower decks or at the level of each accommodation
deck. Good results have been obtained with the
latter approach, which guarantees the presence of
a double access from each deck to the stairs trunk,
in accordance to the SOLAS safety rules.
In a second phase, in consideration of the heavy
modification of the general arrangement that the
optimized configuration implies, solutions aimed
to the reduction of the stress level just in way of
the opening’s hot spots have been also studied.
The study has been developed by focussing on
a new design of the opening corners. FE Analyses
have been carried out by making use of models hav-
ing a mesh refinement at the opening edges. Two
series of studies have been performed. Results have
been compared on the basis of the stress field arising
Figure 1. Stress distribution on the longitudinal bulk-
head in way of the openings (from an elastic-plastic FE
analysis).
281
MARSTRUCT.indb 281
in the detail as consequence of the hull girder shear
force due to the vertical bending of the ship.
In the first study, the relationship has been
analysed between the corner radius r of the open-
ing and the level of the local stresses which arise
along the weld line between the bulkhead plate and
the door frame. Analyses show that the maximum
stress parallel to the weld line σ , calculated in
accordance with LR guidelines (LR 2001), is low-
ered to an acceptable level if the corner radius is
very large: σ /σper = 1.10 for r = 250 mm, where σper
is the LR permissible stress level. Unfortunately,
such a solution is incompatible with the standard
casing configuration, which requires a corner radius
on the door frame of no more than 100 mm.
Therefore, a further study has been performed
in order to propose an enhanced solution. While in
the basic solution, the edge of the plate cut-out in
view of the opening corner is reinforced by a ring
to which the door frame is welded, in the enhanced
solution a larger cut-out having a radius of 200 mm
is created at the opening corner and the bulkhead
plate edge is reinforced by a ring (outer ring). The
gap between the door frame and the bulkhead plat-
ing edge is sealed with a small thickness plate just
to restore the A-class fire barrier.
Such a possible alternative solution for the struc-
tural detail has been studied in two configurations:
door frame with beam and stanchion joined with
an outer rounded edge (A-type) and door frame
with right-angle corners (B-type). The first one is
shown in Figure 2 and Figure 3, the latter giving
a partial view of the door frame and sealing plate
(Von Mises stress mapping is shown too).
Maximum stresses calculated on both A- and
B-type configuration are much below the highest
stress measured on the base configuration with
r = 100 mm, in the ratio of about 1:2. Since in
the A- and B-type structural details peak stresses
are very similar (i.e., in the ratio of 1:1.2), a com-
plete fatigue analysis on all the hot spots of the two
configurations has been carried out. Each struc-
tural configuration has been analysed by using
a shell-element FE model with fine mesh in the
hot-spot area (8-node elements, side length equals
to the plate thickness t) and coarse mesh out-
side. Elements have both membrane and bending
capability. Welds are not modelled. Structural
members are idealised, i.e., welded joints are per-
fectly aligned. No submodels have been created
and all the FE analyses have been performed on
the entire hull with a local mesh refinement.
The most critical hot spots have been located
by means of the von Mises stresses. Four different
hot-spot types have been defined into the enhanced
alternative solution.
Figure 2 shows the location of the fatigue crack
critical sites: HS1 identifies the hot spot at the weld
2/18/2011 5:47:49 PM

HS1
HS2
outer ring
inner ring
sealing plate
Figure 2. Enhanced structural detail and hot-spot loca-
tions HS1 (outer ring weld line) and HS2 (inner ring weld
line).
Figure 3. A partial view of the FE model of the
enhanced structural detail.
line between ring reinforcement and bulkhead plate;
HS2 localises the hot spot at the weld line between
door frame and sealing plate in the A-type detail
configuration. In the B-type detail configuration a
further hot spot is the weld line between the stan-
chion and the beam of the door frame (HS3).
The single components of the stress tensor have
been linearly extrapolated to the hot spot accord-
ing to standard procedure, by which the hot-spot
stress is obtained with a linear extrapolation based
on the FE stresses read out at 0.5 t and 1.5 t from
the weld toe.
Then, the principal stresses have been calcu-
lated by the matrix diagonalization of the local
stress tensor. The so obtained structural stress pair
(σ1, σ2) has been transformed in the (σ⊥, σ ) pair
according to the DNV procedure (DNV 2005) and
282
S2-SS2-C29.indd 282
it has been found that in most hot spots, parallel
stress is greater than normal stress. DNV stress
concentration factors have been applied for evalu-
ating the notch stresses to be used in accordance
to the DNV I-b S-N curve. NL has been derived
by an empirical formula on the basis of a 20 year
expected life. Finally, the fatigue damage D has
been evaluated by summing up D(σ⊥) and D(σ )
to obtain the total damage, and all hot spots have
been confirmed to be very critical.
Alternative procedures (Bäckström 2003) for
biaxial fatigue have been implemented for evalu-
ating the fatigue damage at the whole set of hot
spots. The maximum principal stress approach, the
criterion of the distortion strain energy, the crite-
rion of maximum shear stress and the approach
based on the linear superposition of D(σ⊥) and
D(τ ) have been also applied (Biot et al., 2009).
All applied criteria confirm critical state in respect
to fatigue strength.
So, a further fatigue check has been developed
on the basis of the notch-strain approach in order
to evaluate the role played by the high-stress cycles
measured on most of the hot spots. The nonlinear
fatigue analysis has been carried out on the HS1 hot
spot, as it is considered the most important part of
the structural detail. Elastic-plastic analyses car-
ried out to get the fatigue damage have been based
on cyclic stress-strain curves and elastic-plastic ε-N
curves as defined by different authors.
To apply nonlinear approach, some notes are
necessary to completely define the load history
scenery.
The two extreme hull girder load conditions for
determining the maximum stress range on the con-
sidered hot spot are the maximum and minimum
hogging of the ship’s hull. Considering the type of
ship, a large cruise vessel, expected average value
for σm ranges from 0 to Δσ0/2. Preliminary analyses
pointed out that the mean stress σm may be assumed
of about Δσ0/6, that is equal to σm /Δσ0 = 16.7% or
a σ0,min to σ0,max ratio equal to −0.5.
Outputs of the elastic-plastic approach give evi-
dence one more time of the lack in fatigue capabil-
ity of analysed structural detail, which experiences
a cumulated damage much higher than unity. Prac-
tical usage of elastic-plastic approach is anyway
not so straightforward and a deeper discussion is
necessary on how to apply and interpret elastic-
plastic concepts in fatigue design. That is made in
the following.
4 THE ELASTIC-PLASTIC APPROACH
AND THE FATIGUE DESIGN
Implementation of elastic-plastic methods in
fatigue design requires to have available a wider
2/18/2011 6:33:29 PM
 series of data than in the traditional linear
approach. First of all, material parameters drawn
from comparison specimen cyclic stress-strain tests
are at the basis of nonlinear formulation. On the
other hand, load history in any case proceeds
from a linear analysis carried out on the studied
structure by FEA and a successive stress manipu-
lation in order to obtain a significant structural
stress at the hot spot.
The conversion of linear elastic calculated
stresses to elastic plastic notch stresses is obtained
in a first approximation by using parameters given
in technical literature for structural metals. In the
relationship between the true plastic strain ampli-
tude and the true stress given by the plastic compo-
nent of Equation 1
σa K ′ ( ε apl )n ′ (12)
the exponent is found to be between 0.1 and 0.2
for most metals, including steel. The K′ parameter
is usually given as function of the ultimate tensile
strength of the material σu. The basic parameters
used in the present study are those given by Bäumel
and Seeger, which are summed up in Table 1 for
AH 36 steel together with parameters proposed by
same authors for the capability curve (Bäumel &
Seeger 1990, as cited in Radaj et al., 2006).
As stated above, fatigue capability of a given
terms of damage parameter P so obtaining the
DNL(P) damages. Comparison of that whole set
of results with the fatigue damage DL(Δσ) based
on the DNV rule’s S-N curve, gives a clear indica-
tion of the very different target of the two types
of capability curves. Relevant material dependent
parameters are given in Table 2, where data given
by Bohlmann (Bohlmann 1995, as cited in Radaj
et al., 2006) are relevant to Parent Material (PM)
and Heat-Affected Zone (HAZ) at the weld line.
Values are properly modified according to present
material ultimate tensile strength (AH 36 steel).
In order to solve the outlined inconsistency
on the capability data, in what follows the basic
hypothesis has been conceived, to make equal to
the DNV’s S-N curve the fatigue capability relevant
to the linear elastic component of each consid-
ered ε-N curve. That is supported by interpreta-
tion of the elastic-plastic approach as an intrinsic
enhancement of the basic linear elastic approach.
In other words, slope and axis intercept of the one-
slope S-N curve used in practical fatigue design has
been transferred to the linear elastic component of
the ε-N curve by simple formulae.
10
DL(Δσ)
DNL(P)−B&S
DNL(P)−B(PM)
Fatigue damage D

material is fully described by the four parameters

of the ε-N curve (see Equation 2 and 3), which DNL(P)−B(HAZ)
describes crack initiation, that is crack growth up DNL(P)−M&L
1
to a technical crack length. While ε-N curves are
generally given for an R-value of −1 and refers to
σm equal to zero, S-N curves implemented in ship
structure fatigue design, and particularly those
applied by classification societies, are set for the
specific purpose to consider randomness and
uncertainty in typical load histories. 0.1
Being different the principles governing the set- 200 300 400 500 700 900
ting of the two types of capability curves, a system- Δσ0, MPa
atic comparison between the fully elastic and the
elastic-plastic methodology cannot be performed Figure 4. Fatigue damage DL(Δσ) and DNL(P) (original
unless proper considerations have been made. ε-N curves) as function of the reference load (i.e., stress
Fatigue damage calculations performed by range Δσ0).
taking into consideration the original set of values
proposed for the different ε-N curves are shown Table 2. Material parameters for the elastic plastic anal-
in Figure 4. Calculations have been performed in ysis: values from Bohlmann (B) and Mattos & Laurence
(M&L) (Mattos & Lawrence 1977, as cited in Radaj
et al., 2006).
Table 1. Material parameters for the elastic plastic
analysis: basic reference values from Bäumel and Author* b c σf′ εf′ n′ K′
Seeger (B&S).
B(PM) −0.095 −0.66 891 1.926 0.144 776
Author* b c σf′ εf′ n′ K′ B(HAZ) −0.098 −0.781 1167 1.224 0.126 1089
M&L −0.132 −0.45 1200 0.27 0.249 1097
B&amp;S −0.087 −0.58 735 0.59 0.15 808
*Values are given with reference to the SI system of
*Values are given with reference to the SI system of units. units.

283

MARSTRUCT.indb 283 2/18/2011 5:47:50 PM

 Assuming that m and C are the two material pdf( ) cdf( )
dependent parameters of the S-N curve written 20 1.0
in short as NSm = C, and according to S-N curve
given by DNV rules (m = 3 and log10 C = 12.76),
16 0.8
transformation formulae are:
(σ/σmax) functions
1 1 (13) 12 0.6
σ ′f = (2 )1 / m and b = − (εa /εa,max) functions
2 m
The linear elastic DNV rule S-N curve is shown 8 0.4
in Figure 5 together with the original form of the
basic fatigue capability ε-N curve whose param- 4 0.2
eters are given in Table 1.
Fatigue load history considered in calculation has
been derived by applying the Weibull model. Then, 0 0.0
a set of differently severe load histories has been 0 0.05 0.1 0.15 0.2 0.25 0.95 1
obtained by varying the Δσ0 value. Different values εa /εa,max and Δσ /Δσmax
of the mean stress have been considered too. So,
a trend for the fatigue damage has been determined Figure 6. Statistical functions of the same load his-
as function of the reference stress range Δσ0. tory expressed in terms of linear elastic stresses Δσ and
The whole single procedure for fatigue damage elastic-plastic strains ε (Δσ0 = 627 MPa; DNV and B&S
capability curves; ‘max’ refers to extreme values).
calculation for any given (Δσ0, σm) pair is hereun-
der explained in its main steps. As for the load his-
tory, distributions of the Δσ and ε variables used in dD ( ) DD ( )
the linear elastic and in the elastic-plastic analysis 16 5
are shown in Figure 6, where pdf(•) means prob-
ability density function and cdf(•) means distribu-
tion function. There it is clear the shifting of the 4
12
strain based curves due to the nonlinear conver-
sion to the elastic plastic stress-strain field worked 3
out according to the Neuber’s rule. This fact leads, 8
within the nonlinear approach, to a damage dis-
tribution which gives a greater weight to the lower 2
values of the variable’s range. DL(Δσ)
4
Density function dD of damage D and its DNL(εa)
1
cumulating curve DD both calculated according DNL(P)
to the different applied methodologies are given
in Figure 7. Final cumulated values apart, dam- 0 0
0 0.2 0.4 0.6 0.8 1
age evaluated as function of the elastic-plastic
Δσ/Δσmax , εa /εa,max and P/Pmax

1 Figure 7. Density function and cumulative function of

damage calculated as function of different stress-strain
field identifiers (Δσ0 = 627 MPa, σm = 0).

strain confirms the major role played by lower

εa %

0.1 stress-strain values of the load history. In the

figure, the case is shown of nonlinear calculation
carried out by setting σm equal to zero, so DNL(P)
and DNL(ε) come out to have the same value.
linear approach, DNV As above mentioned, fatigue damage checking
nonlinear approach, B&S has been carried out according to the modified capa-
0.01 bility curves. Results obtained equalizing the linear
103 104 105 106 107 components of the ε-N curves are discussed in the
N(εa) following, where reference is made to load histo-
ries expressed in terms of elastic stress range (linear
Figure 5. Basic capability curves used in the linear elas- approach) and in terms of elastic-plastic strains or
tic approach and in the elastic-plastic approach. P-parameter (nonlinear approach). Figure 8 shows

284

MARSTRUCT.indb 284 2/18/2011 5:47:51 PM

 the modification imposed on the linear part of
any ε-N curve to match the S-N curve determined
in compliance with the DNV rules (the dashed
straight curve in the graph). Structural demand
n(εa) is also shown, represented in two equivalent
forms for linear and nonlinear calculations.
Damages resulting from the elastic plastic anal-
yses based on the P-parameter concept and car-
ried out for different mean values come out to
be very different. The role of the mean stress is
prominent especially at lower σm to which DNL(P)
is very sensitive. The comparative results shown in
Figure 9 give evidence of that fact. The relation-
ship between the DNL(P) trend and that of DL(Δσ)
is illustrated too. Curves show that fatigue check-
ing performed according to the linear approach is
in the middle between the damage values achieved
1 ity analysis on all those parameters is very complex
linear approach, DNV
nonlinear approach, B&S
Capability
εa %
for σm /Δσ0 value varying from 0% to 16.7%, that is
a σ0,min to σ0,max ratio ranging from −1 to −0.5.
According to DNL(P) values, and considering the
most probable case of σm /Δσ0 = 16.7%, the linear
elastic approach appears to be very unsatisfac-
tory, while referring to the generic case of alternate
cyclic loading (σm /Δσ0 = 0), the linear approach is
slightly conservative in the whole range of possible
Δσ0 values. That establishes, at least in the consid-
ered case study, a rather relevant result as elastic
plastic approach appears to be able to give deeper
information when data on expected mean stresses
are available. The more sophisticated analysis is
indeed able to properly take into account all the
main parameters of the cyclic load history.
Reliability level of nonlinear methodology is far
from to be fully ascertained, being final outcome
in term of P-based fatigue damage very dependent
on a lot of variables. On the other hand, a sensitiv-
and the few data available do not allow carrying
out extensive analyses. What appears practicable
is a systematic investigation on the effects due to
change in the value of the most significant param-
eters. Similarly, hypotheses on which procedure is

0.1 based should be examined too. In the following,

sensitivity of the nonlinear procedure to mate-
rial parameters and to some basic assumptions is
checked, with the aim to rank uncertainties.
Load history The first study has been carried out on the
material parameters’ effect on cumulated fatigue
0.01 damage. Two alternatives have been selected. Com-
103 104 105 106 107 parison has been made with data published by
N(εa) and n(εa) Bohlmann, that appears to be both suitable for the
present case study and very interesting for giving
Figure 8. Capability curves used in the analysis for the information on both parent material and material
linear elastic and the elastic-plastic calculation; the struc-
tural demand referred to Δσ0 = 627 MPa.
within the heat-affected zone.
Figure 10 illustrates such a comparative study,
taking the linear damage trend as common
2.5
DL(Δσ) 2.5
2 DNL(P)−B&S DNL(P)−B&S
DL
Fatigue damage D

DNL(P)−B(PM) nc
e
Nonlinear damage DNL
50 % 0

2
σ

ere
DNL(P)−B(HAZ)
σm /Δ

e f
30 %

r
7%

1.5
16, %
25

1.5
1 σm/Δσ0 % = 16.7 σm/Δσ0 % = 0
0% 1
0.5
0.5
0
0 100 200 300 400 500 0
Δσ0, MPa 0 0.5 1 1.5 2 2.5
Linear damage DL
Figure 9. Comparison between fatigue damage DNL(P)
calculated for given values of the mean stress, and linear Figure 10. Damage comparison based on different data
elastic based fatigue damage DL(Δσ). for the material parameters.

285

MARSTRUCT.indb 285 2/18/2011 5:47:53 PM

 reference. P-based damage has been evaluated for
σm /Δσ0 values of 0% and 16.7%, being those the
main significant for the present case study. For
both σm /Δσ0 values changes in the material data
are not very tangible, as far as the parent material
is considered. On the other hand, changes appear
more evident if data are referred to the material
heat-affected by welding procedure. Moreover,
the general trend of damage parameter P is con-
firmed to have a greater sensitivity when calcula-
tions are carried out at lower values of σm /Δσ0,
as could also be deduced by the rate by which
DNL(P) damage changes as function of σm /Δσ0 in
Figure 9.
A further analysis has been performed about the
validity of the basic hypotheses made in defining
the methodology. First of all, the conversion from
the elastically computed to the actual elastic-plastic
stress-strain field has been carried out under both
the hypothesis of plane stress (Equation 4) and
plane strain (Equation 5). Figure 11 shows the con-
version from the calculated linear stress-strain field
pair (point K) to the elastic plastic pair according
to Neuber’s rule (point S) or according to the plane
strain hypothesis (point N).
Comparative results are shown in Figure 12
with reference to the same mean stress levels above
referred to explaining Figure 10. As the graph
clearly illustrates, damage is almost independent
from the rule by which conversion is made. So,
conversion method appears to be ineffective as far
as Equation 4 and 5 are considered.
The second basic hypothesis analysed is that
referring to possibility to adopt a mean stress cor-
rection embedded in the fatigue capability ε-N
curve, as stated in Equation 3. Analyses have been
accomplished on the whole range of considered
400
σa,k K
S
σa S
σa N
σa, MPa
N
cyclic σa−εa curve
σa S εa S = (Ktσa,hs/Khs)2 /E
σa /εa= E
0 Then, a case study has been discussed in order to
0 εa N εa S
εa %
Figure 11. Methods to convert FE calculated stress-
strain field to elastic-plastic stress-strain field (E = 206.85
GPa and Δσ0 = 627 MPa).
MARSTRUCT.indb 286
10
DL(Δσ)
DNL(P)−B&S
Fatigue damage D
plane strain rule (Eq. 5)
σm effect (Eq. 3)
1 σm/Δσ0 % = 16.7
σm/Δσ0 % = 0
0.1
100 200 300 400 500 600
Δσ0, MPa
Figure 12. Fatigue damage DNL(P) calculated under dif-
ferent hypotheses within the elastic-plastic procedure.
σm to Δσ0 ratios. They all agree in confirm that
formulation of Equation 3 appears ineffective to
treat mean stress effect, being DNL(ε) almost the
same for the different σm values. Figure 12 shows
the results of the investigation: values calculated
with σm /Δσ0 ranging from 0 to 50% are all aligned
with the curve of DNL(P) calculated for zero mean
stress.
By summing up, systematic investigations point
out that the most important factor influencing out-
comes in elastic-plastic approach is the mean stress
value. Secondly, set of material data to be used for
specifying stress-strain curve (two parameters) and
capability curve (four parameters) has an impor-
tant role just if distinction is made in welded joints
between parent material and material lying in the
heat altered zone along the weld line. Finally, alter-
native hypotheses about mean stress correction on
ε-N curve end about conversion from elastic to
elastic-plastic stress-strain field have no effect to all
practical purposes.
5 CONCLUSIONS
This work deals with the improvement of the
method for fatigue checking ship structural details
during design stage. First, theory and practice
related to a basic nonlinear procedure has been
outlined, within the cumulative damage approach.
0.3 analyse advantages and drawbacks of the method
proposed to account for large plastic zone at crack
tip, like in case of a low cycle fatigue scenario.
Finally a comparison has been made between lin-
ear elastic and elastic-plastic approach, that shows
a general agreement between the two methods, on
286
2/18/2011 5:47:55 PM
 condition that elastic component of elastic-plastic
capability has been properly modified to account
for safety factors applied by classification
societies.
Role of different parameters necessary to set
material strength has been examined too, together
with influence of alternative methods for consider-
ing plasticity in basic nonlinear fatigue analysis.
It has been shown that notch stress factor Kt
based on hot spot stress calculations is not fit for
use in strain-based approach if weld shape factor
Kw given by rules has not been suitably modified.
To do this, rule’s safety factors should be known.
Otherwise notch strain approach may not be
considered to support decisions during an actual
design stage.
In conclusion, nonlinear approach may be a
valid alternative to the linear one only if a strict
control on the safety factors relating to fatigue
capability is practicable. Moreover, in order to
properly apply the non linear procedure, the mean
stress value of the load history together with the
mean stress effect due to residual stresses should be
known, otherwise no further valuable information
may be drawn in respect to the application of the
linear approach.
REFERENCES
Bäckström M., 2003. Multiaxial Fatigue Life Assessment
of Welds Based on Nominal and Hot-spot Stresses.
VTT Publications 502, ISBN 951 38 6234 8.
Bäumel A. & Seeger T. 1990. Materials Data for Cyclic
Loading. Suppl. 1, Amsterdam, Elsevier Science,
1990.
287
MARSTRUCT.indb 287
Biot M., Calcinotto P. & D’Este F. 2009. Strain-based
Approach for Fatigue Life Evaluation: Analysis of
Complex Structural Details on Cruise Ships. 16th Inter-
national Conference of Ship and Shipping Research,
Messina, Italy.
Bohlmann B., 1995. Applications of local approaches to
the fatigue assessment of welded tubular joints (actu-
ally a welded pipe branching). Materials. International
Institute of Welding IIW, Doc. XIII-1591–95.
Det Norske Veritas, 2005. Fatigue Assessment of Ship
Structures. DNV Classification Notes, N.30–7, July
2005.
Fuchs H.O. & Stephens R.I. 1980. Metal Fatigue in Engi-
neering. John Wiley & Sons, New York NY, ISBN
0-471-05264-7.
Hobbacher A. 2007. Recommendations for Fatigue Design
of Welded Joints and Components. International Insti-
tute of Welding IIW, Doc. XIII-2151-07/XV-1254-07.
ISSC, 2009. Report of Technical Committee III.2: Fatigue
and Fracture. 17th International Ship and Offshore
Structures Congress, Seoul, Korea.
Lloyd’s Register, 2001. ShipRight Structural Design
Assessment Procedure, Primary Structure of Passenger
Ships, Direct calculations. November 2001.
Mattos R.J. & Lawrence F.V. 1977. Estimation of the
fatigue crack initiation life in welds using low cycle
fatigue concepts. SAE SP-424,Warrendale PA, SAE.
Radaj D., Sonsino C.M. & Fricke W. 2006. Fatigue
Assessment of Welded Joints by Local Approaches. 2nd
Edition, Woodhead Publishing, Abington, England,
ISBN 1 85573 948 8.
Smith K.N., Watson R. & Topper T.H. 1970. A stress-
strain function for the fatigue of metals. J Mater,
JMLSA, 5(4), 767–778.
Wang X., Kang J.K., Kim Y. & Wirsching P.H. 2006.
Low Cycle Fatigue Analysis of Marine Structures.
25th International Conference on Offshore Mechan-
ics and Arctic Engineering OMAE2006, Hamburg,
Germany.
2/18/2011 5:47:56 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Different finite element refinement strategies for the computation

of the strain energy density in a welded joint

C. Fischer, A. Düster & W. Fricke

Hamburg University of Technology (TUHH), Hamburg, Germany

ABSTRACT: Structural failure due to a lack of fatigue strength and crack initiation is often caused
by local stress concentrations at welded joints. By decreasing the weld notch radius to small values, the
magnitude of stress concentration no longer affects the fatigue strength but a mean value of stress deter-
mined around the notch tip. Moreover, in linear elasticity the local stress distribution becomes singular
with vanishing radius. Besides the common methods for fatigue assessment, the utilisation of an averaged
value of strain energy, computed for a small volume of material, is an alternative approach. The method
shows promise with respect to finite element analysis of complex problems. In most of the previous work
the quality of the numerical approximation was controlled by mesh refinement. In this paper, alternative
refinement techniques are investigated, accounting for the singular behaviour of the exact solution in the
vicinity of the re-entrant corner. After the introduction of the concept based on the averaged strain energy
density, different refinement strategies improving the quality of the finite element approximation are
described and applied to a selected example. The results of the different approaches related to both h and
p-refinement are presented and discussed also in view of application to more complex three-dimensional
problems.

1 INTRODUCTION

In the theory of linear elasticity, a vanishing notch

radius ρ results in a singular stress distribution.
However, the fatigue strength is controlled by a
mean stress value averaged in a small volume of
material. The different common fatigue approaches,
which have been summarised by Radaj et al. (2006),
estimate the fatigue strength in various ways.
For instance, the effective notch stress approach
applies a fictitious notch radius of ρref = 1 mm to
account for the effect of microstructural support at
the notch root (Hobbacher, 2009).
Alternatively, the stress intensity factors describe
the local asymptotic stress distribution at the weld Figure 1. Control sector surrounding the weld toe.
toe in a similar way as for crack tips (Lazzarin &
Tovo, 1996), with an opening angle 2α differing
from zero, Figure 1. The stress intensity factors a small defined region close to the singular notch
have been proven to be suitable parameters con- root the so-called Strain Energy Density (SED) is
cerning fatigue assessment of welds (Lazzarin & computed. The SED indicates fracture of brittle
Tovo, 1998). However, the demand for very fine materials weakened by a sharp V-notch under
finite element meshes in highly stressed regions is static loads (e.g. Yosibash et al., 2004) and of
one of the main disadvantages of the approach. cracked components under mixed mode loading
Besides evaluation of stresses, the strain energy by estimating the product of the SED and a criti-
is utilized by different failure assessment methods. cal distance from the crack tip (Sih, 1974).
For the prediction of fatigue life under multiaxial For high-cycle fatigue of welded joints assum-
loading conditions Glinka et al. (1995) defined ing a V-notch with ρ = 0, Lazzarin & Zambardi
a parameter, which is based on the strain energy (2001) used a geometrically based SED parameter,
occuring inside the critical plane. Alternatively, in which is averaged over a small cylindrical volume.

289

MARSTRUCT.indb 289 2/18/2011 5:47:57 PM

 Based on the relationship between the SED and
the notch stress intensity factors, a unified S-N
curve independent of the welded detail has been
developed. The approach shows promise because a
relative coarse finite element mesh yields satisfac-
tory results.
When discretizing problems of linear elasticity
with singular behaviour, the refinement strategy
has a very important influence on the quality and
efficiency of the computation. In this paper the
influence of different refinement techniques com-
puting the averaged SED has been investigated for
a doubler plate with fillet-welded joints. After a
brief introduction of the SED approach the differ-
ent refinement strategies will be described and the
numerical results will be presented.
2 STRAIN ENERGY DENSITY APPROACH
The application of the SED approach for fatigue
assessment of welded joints with a sharp V-shaped
notch including an opening angle of 2α, has been
recommended by Livieri & Lazzarin (2005) and
Lazzarin et al. (2008).
The local region, where the total strain energy
is averaged, is defined as a cylindrical volume sur-
rounding the notch root. The weld toe considered
as re-entrant corner typically exhibits an opening
angle of 2α = 135° and the weld root at slit ends
2α = 0°. In case of two-dimensional problems the
volume degenerates to a sector with a characteris-
tic radius R0, Figure 1.
The size of the radius R0 over which the SED
needs to be computed depends on the material and
has been derived from Beltrami’s failure criterion
by Lazzarin & Zambardi (2001) considering stress
intensity factors for welded joints under mode
I loading and high cycle fatigue data (N = 5 × 106).
Choosing R0 = 0.28 mm results in a conservative
approach for welds made of steel and arc welding
technology with failure from weld toe and root
(Livieri & Lazzarin, 2005). For aluminium alloys
a value of 0.12 mm has been suggested. It should
be noted that these considerations are solely based
on continuum mechanics. The material effects
are taken into account by the high cycle fatigue
failure data.
Based on an evaluation of about 650 fatigue
tests, which cover both non-load carrying and
load-carrying fillet cruciform welded joints with
various plate thicknesses, a uniform S-N curve
with reference to the range of the averaged SED
ΔW has been obtained by Lazzarin et al. (2008).
Considering probabilities of survival Ps = 2.3% and
97.7%, Figure 2 shows the corresponding scatter
band involving fractures initiated from weld toe as
well as weld root.
290
MARSTRUCT.indb 290
Figure 2. Fatigue strength depending on range of the
averaged SED ΔW (Lazzarin et al., 2008).
Furthermore, a relationship between the range
of the averaged SED ΔW and the equivalent local
stress range Δσeq is given under plane strain condi-
tions by:
2 E ⋅ ΔW
Δσ eq = (1)
1 − v2
where E is Young’s modulus and ν is Poisson’s
ratio.
3 EXAMPLE
Within various types of welded joints, Fricke &
Feltz (2009) investigated a partial-load carrying
doubler plate on both sides of a continuous main
plate. Here two different weld throat thicknesses a
were considered resulting in two types of specimen,
which are denoted as D.3 and D.7. Figure 3 shows
the geometry of the specimens.
Besides statistical evaluation of the fatigue tests,
the specimens were numerically investigated by dif-
ferent approaches of fatigue assessment (Fricke &
Feltz, 2009). The computations were conservative,
but some restrictions and shortcomings con-
cerning the estimation of the critical notch were
discovered.
Furthermore, the notch stress intensity fac-
tor approach was applied by Fischer et al. (2010)
analysing the averaged SED in the control sector
at both the weld toe and the weld root. In addi-
tion to the mesh layouts suggested by Lazzarin
et al. (2008) two different uniform meshes, which
use both elements with a quadratic shape function,
were applied: a rather coarse mesh with triangu-
lar elements having a size of 0.28 mm and a finer
mesh, see Figure 4.
The computed values of the averaged SED
ΔW for the two specimen types are conservative
and are located within the proposed scatter band.
2/18/2011 5:47:57 PM
 Figure 3. Geometry of the specimens (Fricke & Feltz,
2009).
Figure 4. Meshes used for the computation of the
mean SED.
Moreover, the predicted crack locations, which are
identified by means of a higher SED-value, agree
with the tests.
4 CONTROL OF THE DISCRETIZATION
ERROR
The quality of a finite element approximation
can be controlled either by reducing the size of
the elements (h-refinement) or by increasing the
polynomial degree p of the shape functions of
elements (p-refinement), see Szabo & Babuska
(1991). A combination of both approaches yields
the hp-version of the finite element method. A uni-
form h or p-refinement corresponds to a global
reduction of the element size or an increase of the
polynomial degree of all elements, respectively.
The convergence rate of the refinement proc-
ess depends on the strength of the singularity
λ controlled by the angle of the V-notch and the
chosen refinement strategy. In Figure 5 the error
in energy norm is plotted in a double logarithmic
291
MARSTRUCT.indb 291
Figure 5. Convergence rates of different refinement
strategies applied to a 2D problem of linear elasticity
with singularities.
style against the number of degrees of freedom
N for different refinement strategies applied to a
two-dimensional linear elastic problem.
In absence of singularities a uniform p-extension
yields an exponential rate of convergence, which is
characterized in Figure 5 by a curve with increas-
ing (negative) slope resulting in a highly efficient
discretization. Unfortunately, in almost all prob-
lems of linear elasticity, singularities are present
and a p-extension on a uniform mesh will result in
an algebraic rate of convergence, being still twice
as high as the one of an h-extension based on uni-
form meshes.
More efficient finite element refinements can
be obtained when combining the increase of the
polynomial degree with a local mesh refinement.
In the hp-extension the mesh is refined in a geo-
metric progression with a grading factor q = 0.15
towards the point of singularity and the polyno-
mial degree is increased simultaneously. Small
elements close to the corner are assigned to a low
polynomial degree and with increasing distance
from the singularity the polynomial degree is ele-
vated linearly. The layout of these so-called geo-
metric meshes is depicted in Figure 6 using R0 as
reference parameter.
In practice it is sufficient to carry out a uniform
p-extension on a strongly graded geometric mesh.
This extension process yields an exponential rate
of convergence in the pre-asymptotic range, which
asymptotically slows down to an algebraic rate still
twice as high as the one of a uniform h-refinement,
see Figure 5.
When computations need to be performed with
finite elements fixed in polynomial degree of low
order (e.g. p = 2), an optimal convergence rate can
be obtained applying radical meshes, see Figure 7,
which are strongly refined towards the singularity
(Szabo & Babuska, 2011). For the radical mesh the
distance of the nodes with respect to the singular
point are set to
2/18/2011 5:47:58 PM
 toe as well as the weld root are performed under
plane strain conditions with Young’s modulus
E = 2.06 × 105 MPa and Poisson’s ratio ν = 0.3.
While considering frictionless contact conditions
between doubler and main plate with zero gap,
both p-version and h-version refinements were
analysed.
In the following, results are only described
for the weld toe notch. Thereby, the convergence
behaviour of the SED is studied based on the rela-
tive error e, which is defined similar to the error in
energy norm by
Figure 6. Geometric mesh with three layers at weld toe
(not to scale). | Wex W |
e= (3)
Wex

where Wex is obtained by the most accurate numeri-

cal approximation using finite elements with a pol-
ynomial degree of p = 8. It was decided to estimate
the relative error on the basis of the averaged SED
instead of the equivalent stress Δσeq mentioned in
equation (1).

5.2 p-version refinement

Computations increasing the polynomial degree
of the finite elements from five up to eight on a
fixed mesh were carried out by Actis applying the
Figure 7. Nodal distances for radical mesh three layers finite element software StressCheck® (2010). A rel-
applied (not to scale). atively coarse geometric mesh with 121 elements
graded towards the notch root has been utilized,
see Figure 8.
⎛ p + 1⎞ Table 1 lists the number of degrees of freedom
⎛ k ⎞⎝ λ ⎠
(2) (DOF) for the global model and elements inside
dk R0 ⎜ ⎟ ,k , , ,M
⎝M⎠ the sector as well as the computed mean SED ΔW
and the error e according to equation (3).
where M is the number of layers of elements, k is
the number of the current layer and R0 represents 5.3 h-version refinement
the reference length.
For further details on h, p and hp-refinement we In this section the results obtained by an
refer to Szabo & Bakuska (1991) and Szabo et al. h-refinement will be considered. The correspond-
(2004). A detailed description of radical meshes ing computations were performed under plane
can be found in Szabo & Babuska (2011). strain conditions with ANSYS® 11.0 using element

5 NUMERICAL ANALYSIS
OF THE DOUBLER PLATE

5.1 Problem description

In order to study the influence of the different
refinement strategies the example of the doubler
plate with a small weld throat thickness a = 3 mm
(D.3) is considered. Due to symmetry only a quar-
ter of the doubler plate has to be discretized.
The structure is subjected to a nominal axial stress
range Δσn = 150 MPa. The computations of the Figure 8. Finite element model used for p-version
SED within a sector of R0 = 0.28 mm at the weld refinement computed with StressCheck®.

292

MARSTRUCT.indb 292 2/18/2011 5:48:00 PM

 Table 1. Computed ΔW and relative error e for Table 3. Computed ΔW and relative error e for
R0 = 0.28 mm and Δσn = 150 MPa at weld toe using R0 = 0.28 mm and Δσn = 150 MPa at weld toe for a geo-
p-refinement. metric mesh.

Polynomial DOF DOF ΔW [Nmm/mm3] DOF DOF ΔW [Nmm/mm3]

degree p (model) (sector) at weld toe e [%] (model) (sector) at weld toe e [%]

5 3,163 670 3.52732 × 10 −1 1.5 28,629 36 3.50250 × 10−1 8.3

6 4,441 880 3.52667 × 10−1 0.6 28,749 70 3.51275 × 10−1 6.3
7 5,961 1,120 3.52656 × 10−1 0.2 28,985 138 3.51424 × 10−1 5.9
8 7,723 1,390 3.52654 × 10−1 0.0 29,229 206 3.51424 × 10−1 5.9

Table 2. Computed ΔW and relative error e for Table 4. Computed ΔW and relative error e for
R0 = 0.28 mm and Δσn = 150 MPa at weld toe for a uni- R0 = 0.28 mm and Δσn = 150 MPa at weld toe for a radi-
form mesh. cal mesh.

DOF DOF ΔW [Nmm/mm3] DOF DOF ΔW [Nmm/mm3]

(model) (sector) at weld toe e [%] (model) (sector) at weld toe e [%]

28,629 36 3.50250 × 10−1 8.3 28,629 36 3.50250 × 10−1 8.3

32,619 120 3.50920 × 10−1 7.0 28,749 70 3.51473 × 10−1 5.8
122,451 418 3.51468 × 10−1 5.8 28,985 138 3.51460 × 10−1 5.8
477,611 1554 3.51738 × 10−1 5.1 29,221 172 3.51463 × 10−1 5.8

Figure 9. Coarsest and finest mesh utilised by uniform

h-refinement at weld toe.

Figure 10. Error curves plotted against DOF of whole

type PLANE82, which is an eight-node element
utilizing quadratic shape functions.
Applying a uniform mesh refinement within the
sector results in a noticeable growth of the glo-
bal number of degrees of freedom and therefore
computational time, since the refinement spreads
globally. The computed values of the SED ΔW for
the four different levels of mesh refinement and
the corresponding error e based on the most accu-
rate approximation obtained by the p-version are
shown in Table 2.
Furthermore, the geometric and the radical
meshes, which are geometrically graded towards
the singular point, are considered. An example of
each mesh with three layers of elements is shown
in Figures 6 and 7, taking the size of parameter
R0 as basis. Both refinement strategies increase the
number of finite elements in the vicinity of the sin-
gular point, so that the surrounding mesh does not
change. The corresponding results are shown in
Tables 3 and 4, respectively.
model.
5.4 Comparison of results
In order to compare the results of the different
refinement strategies and to assess the conver-
gence rate, the relative error e is plotted against
the global degrees of freedom in a double logarith-
mic style, Figure 10. From this, it is evident that
the p-extension on the geometric mesh yields the
most efficient approach, which is due to its high
convergence rate. Furthermore it can be observed
that when being restricted to finite elements of low
order, mesh refinement in a radical or geometric
fashion improves the efficiency.
In order to illustrate the effect of the geomet-
ric and radical mesh refinement more clearly, in
Figure 11 the relative error e is plotted against the
degrees of freedom related to the sector only. From
this it can be seen that the application of geomet-
ric and radical meshes yields an improvement as

293

MARSTRUCT.indb 293 2/18/2011 5:48:01 PM


Figure 11. Error curves plotted against DOF inside
sector.
compared to a uniform h-refinement, modifying
the mesh only very locally.
When comparing Figures 10 and 11 with respect
to the efficiency of the computations, it should be
kept in mind that the total number of degrees of
freedom is relevant for the computational time.
Hence, the p-refinement on a geometric mesh
yields the best performance.
6 SUMMARY AND CONCLUSIONS
In estimating the fatigue strength of welded joints,
the computation of the averaged strain energy
in a cylindrical volume represents a promising
approach. In order to compute the SED efficiently
different finite element refinement strategies have
been applied and compared to each other. To this
end, a doubler plate has been chosen as a bench-
mark problem. The most efficient approach to
compute the SED is to perform a p-extension on
a geometric mesh. If finite element computations
are limited to low-order elements, geometric or
radical mesh increase the efficiency significantly.
Although in two dimensions the SED can be also
computed with h-refinement strategies, the effi-
ciency of high-order finite elements is very impor-
tant since the numerical effort in three dimensions
increases dramatically, calling for efficient strate-
gies such as the p-version of the finite element
method. Further research will focus on the SED
approach in three dimensions applied to welded
joints which will be discretized with the efficient
discretisation strategies studied in this paper for
two-dimensional problems.
ACKNOWLEDGEMENTS
The authors would like to thank Prof. Szabo and
Dr. Actis from Engineering Software Research and
Development, Inc., St. Louis, U.S.A., who carried
294
MARSTRUCT.indb 294
out the p-version finite element analysis of the
example presented in this paper applying the
software StressCheck®. We would also like to thank
both of them for fruitful discussions and access to
their unpublished data and literature.
REFERENCES
Fischer, C., Feltz, O., Fricke, W. & Lazzarin, P. 2010.
Application of the notch stress intensity and crack
propagation approaches to weld toe and root fatigue.
IIW-Doc. XIII-2337-10/XV-1354-10, Int. Inst. of
Welding.
Fricke, W. & Feltz, O. 2009. Fatigue tests and numeri-
cal analyses of partial-load and full-load carrying fil-
let welds at cover plates and lap joints. Welding in the
World 54, No. 7/8, R225–233.
Glinka, G., Shen, G. & Plumtree, A. 1995. A multiaxial
fatigue strain energy parameter related to the critical
fracture plane. Fatigue Fract. Eng. Mater. Struct. 18,
37–46.
Hobbacher, A. 2009. Recommendations for Fatigue
Design of Welded Joints and Components. IIW-Doc.
1823-07, New York: Welding Research Council Bul-
letin 520.
Lazzarin, P., Berto, F., Gomez, F.J. & Zappalorto, M.
2008. Some advantages derived from the use of
the strain energy density over a control volume in
fatigue strength assessments of welded joints. Int. J.
Fatigue 30, 1345–57.
Lazzarin, P. & Tovo, R. 1996. A unified approach to the
evaluation of linear elastic stress fields in the neighbour-
hood of cracks and notches. Int. J. Fract. 78, 3–19.
Lazzarin, P. & Tovo, R. 1998. A notch intensity factor
approach to the stress analysis of welds. Fatigue Fract.
Eng. Mater. Struct. 21, 1089–103.
Lazzarin, P. & Zambardi, R. 2001. A finite-volume-energy-
based approach to predict the static and fatigue behav-
iour of components with sharp V-shaped notches. Int.
J. Fract. 112, 275–298.
Livieri, P. & Lazzarin, P. 2005. Fatigue strength of steel
and aluminium welded joints based on generalised
stress intensity factors and local strain energy values.
Int. J. Fract. 133, 247–276.
Radaj, D., Sonsino, C.M. & Fricke, W. 2006. Fatigue
Assessment of Welded Joints by Local Approaches.
Cambridge: Woodhead Publishing (2nd Edition).
Sih, G.C. 1974. Strain-energy-density factor applied to
mixed mode crack problems. Int. J. Fract. 10, 305–321.
StressCheck 2010. http://www.esrd.com.
Szabo, B. & Babuska, I. 1991. Finite Element Analysis.
New York: John Wiley & Sons.
Szabo, B. & Babuska, I. 2011. Introduction to Finite Ele-
ment Analysis: Formulation, Verification and Valida-
tion, to be published by John Wiley & Sons.
Szabo, B., Düster, A. & Rank, E. 2004. The p-version of
the finite element method. In E. Stein, R. de Borst,
T.J.R. Hughes (ed.), Encyclopedia of Computational
Mechanics, Volume 1: Fundamentals, Chapter 5,
pp. 119–139, New York: John Wiley & Sons.
Yosibash, Z., Bussiba, A.R. & Gilad, I. 2004. Failure cri-
teria for brittle materials. Int. J. Fract. 125, 307–333.
2/18/2011 5:48:02 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Round-robin on local stress determination and fatigue assessment

of load-carrying fillet-welded joints

Wolfgang Fricke
Hamburg University of Technology (TUHH), Hamburg, Germany

Matteo Codda
CETENA—Centro Tecnico Navale, Genova, Italy

Olav Feltz
Hamburg University of Technology (TUHH), Hamburg, Germany

Yordan Garbatov
Instituto Superior Técnico (IST), Lisbon, Portugal

Heikki Remes
Aalto University, Helsinki, Finland

Giovanni Risso
CETENA—Centro Tecnico Navale, Genova, Italy

Cesare Rizzo
Università di Genova (DINAEL), Genova, Italy

Jani Romanoff
Aalto University, Helsinki, Finland

ABSTRACT: The fatigue assessment of welded ship structures includes high uncertainties. One major
source of uncertainty is the determination of relevant local stresses based on the structural hot-spot
or notch stress approach. To quantify the uncertainties and to improve the analysis guidelines, round-
robin studies were performed within the MARSTRUCT project. One recent study concerned load-
carrying fillet welds which are treated in different ways in structural stress approaches. In total, five
partners participated in the work. Two structural configurations at 12 mm thick plates were analysed,
i.e., double-sided lap-joints and doubler plates, each with two different weld throat thicknesses (2.5 mm
and 7 mm). After shortly describing the fatigue assessment approaches, the models of the partners and
their results are described in more detail. Differences in the results are identified and conclusions drawn
with respect to modelling guidelines and typical scatter of computed fatigue lives.

1 INTRODUCTION partly related to the stress analysis of the structural

Fatigue strength is still an important design cri-
terion during ship structural design. Although
corresponding rules have been introduced and har-
monised in the common rules for tanker and bulker
structures (IACS, 2010), there are still many uncer-
tainties about the quality of predicting the fatigue
life of ship structural details. Round robin studies
performed by the Committee ‘Fatigue and Frac-
ture’ of the International Ship Structures Congress
(ISSC) have quantified the uncertainties. These are
details (Fricke et al., 2002). In several cases, the nom-
inal stress approach is no more applicable due to
their geometrical complexity, and local approaches
based on the structural hot-spot stress or the effec-
tive notch stress need to be applied (Radaj et al.,
2006). These approaches are able to consider the
local geometry more rationally, however, different
guidelines for modelling structural details and for
stress evaluation exist. In addition to the usage of
different finite element types and software, this is
one of the major causes for the uncertainties.

295

MARSTRUCT.indb 295 2/18/2011 5:48:02 PM

 In order to further quantify the modelling In the following, the round-robin is described,
uncertainties and to identify the sources of scat- showing the welded joint chosen together with typ-
ter of stress results, partners of the Network of ical finite element models, results and conclusions
Excellence on Marine Structures (MARSTRUCT) drawn from the results. In total, three to five mem-
decided to perform round-robin studies on some ber institutions of MARSTRUCT participated
typical ship structural details. The first round-robin in the round-robin. It has to be emphasized that
on three different details was performed between the focus is on the stress analysis. A more detailed
2004 and 2006 and published by Fricke et al. description can be found in the internal deliverable
(2007), showing interesting results regarding the report (Fricke et al., 2010).
structural hot-spot stress and the effective notch
stress approach. The results were partly taken into
consideration in IIW guidelines on the notch stress 2 SPECIFICATION OF THE
approach (Fricke, 2008). ROUND-ROBIN
It was decided in MARSTRUCT to continue the
round-robin with another structural detail which Two types of test specimens were investigated, each
was experimentally investigated, i.e., a plate with with two different throat thicknesses a, see Figure 1:
additional doubler plates and a lap joint where the
L Test specimen with non-continuous parent plate
fillet welds show different grades of load carriage
and two-sided doubler pates, i.e., a lap joint
(Feltz and Fricke, 2009). In a fatigue assessment,
D Test specimen with continuous parent plate and
this is taken into account by the S-N approaches
two-sided doubler plates
available in different ways:
The loading is uni-axial at the end of the par-
• As the traditional structural hot-spot stress
ent plate (actually, the test specimens are longer
approach (Niemi et al., 2006) cannot consider
to allow space for the grips). The welds carry the
this in the structural stress, a decreased FAT
full load in specimen L and part of the load in
class is assumed (FAT 90 instead of 100),
specimen D.
• The ‘1 mm stress approach’ by Xiao and Yamada
All plate thicknesses are 12 mm. The nominal
(2004) accounts for the increased stress at load
throat thicknesses are a = 2.5 and 7 mm. The cor-
carrying fillet welds because the relevant stress is
responding leg lengths are 3.5 and 9.9 mm.
located close to the weld toe,
A 2D model using plane strain conditions was
• The structural stress approach by Dong (2001)
considered to be sufficient. Symmetry conditions
claims to consider the effect by assuming a bi-
could have been used in two planes so that the
linear stress distribution at double-sided fillet
model is relatively simple. One general question
welds,
concerned the problem of contact between the
• The more recently developed structural stress
doubler and parent plates. It was decided to con-
approach by Poutiainen (2006) assumes a bi- or
sider both alternatives: without and with contact,
multi-linear stress distribution in plate thickness
neglecting any friction effect, although contact
direction; the stress increase towards the weld
was regarded as the more probable alternative.
toe is governed by the stress in the weld.
• The effective notch stress approach assumes a
fictitious radius of 1 mm (Hobbacher, 2009) and
thus considers the increased local stress at load-
carrying fillet welds.
Poutiainen (2006) applied four of the five
approaches to a doubler plate on an I-beam and
found large differences between them when consid-
ering different weld throat thicknesses. However,
no fatigue test results are available indicating which
result is superior. Therefore, also fatigue tests have
been performed in MARSTRUCT, showing how-
ever no significant differences in fatigue behaviour
in those cases where the crack initiation occurred
at the weld toe (Feltz and Fricke, 2009). It should
be mentioned that cracks may initiate also from the
weld root. From the approaches mentioned, only
the effective notch stress approach is able to ana-
lyse weld root failure, however, with some restric- Figure 1. Geometry of the specimens for the fatigue
tions (Feltz and Fricke, 2009). tests.

296

MARSTRUCT.indb 296 2/18/2011 5:48:02 PM

 Guidelines given for the different approaches
should have to be observed. It is worth mentioning
that none of the guidelines account for potential
contact between surfaces of plates.
For the notch stress approach, also root failure
should have been considered, either with a U- or
a keyhole shape. The tests have shown whether
the weld toe or the weld root is more critical.
From a total of 80 specimens, 57 showed failures
from the weld toe, with a characteristic fatigue
strength of FAT 73 if evaluated altogether (Feltz
et al., 2010). The FAT class corresponds to the
fatigue strength in [MPa] at 2 ⋅ 106 cycles for a
probability of survival Ps = 97.7%. The remain-
ing 23 specimens showed root failure, including
all 20 specimens with lap joints and small weld
throat thickness of 2.5 mm, yielding characteris-
tic fatigue strength of FAT 47 based on the nomi-
nal stress in the parent plate.
The models were designated L-2.5, L-7, D-2.5
and D-7 according to the type of specimen and
weld throat thickness.
Participant 1 did not perform a finite element
analysis, but concluded that from equilibrium
conditions the structural hot-spot stress is per
definition equal to the nominal stress—as can
directly be seen from the through-thickness lin-
earization (Fig. 3a). The numerical result from a
very coarse mesh (one element over the thickness)
would also yield this.
The other three participants applied the surface
stress extrapolation over reference points at 0.4 t
and 1.0 t in front of the weld toe.
Participant 2 used a fairly fine mesh and
achieved results close to unity (±5%). Participant
3 created a coarser mesh with four elements in
thickness direction, showing results below unity
particularly for the lap joints. Participant 4 estab-
lished a 3D model with elements having a depth
of t/2 in thickness direction. The results obtained
for the small throat thickness are significantly
above unity.
The scatter of results seems to be quite large
for this simple example and deserves further

3 APPLICATION OF THE STRUCTURAL

HOT-SPOT STRESS APPROACH

The structural hot-spot stress approach, which is

applicable only to weld toe failures, is based on
the stress component at the weld toe which dis-
regards the local stress increase due to the local
weld toe notch. The structural hot-spot stress can
be computed by surface stress extrapolation using
reference points at locations depending of plate
thickness or by linearization of the stress through Figure 2. FE model for the structural hot-spot stress
the plate thickness (Niemi et al., 2006). analysis (1/8 model; participant 4).
The results in terms of structural stress concen-
tration factors, i.e., structural hot-spot stress divided
by nominal stress, are summarized in Table 1.

Table 1. Computed structural stress concentration

factors acc. to the structural hot-spot stress approach.

Participant FE mesh L-2.5 D-2.5 L-7 D-7

No contact:
1 None (theory) 1.00 1.00 1.00 1.00
2 Very fine 1.04 1.03 0.95 0.98
(t/10xt/10)
3 Fine (0.4txt/4) 0.87 0.95 0.81 0.90
4 Fine (0.4txt/2) 1.20 1.12 – –
Contact:
1 None (theory) 1.00 1.00 1.00 1.00
2 Very fine 0.98 1.00 0.95 0.98
(t/10xt/10)
3 Fine (0.4txt/4) 0.99 1.02 0.99 1.02 Figure 3. Nominal and structural hot-spot stress in
4 Fine (0.4txt/2) 1.18 1.12 – – the lap joint (a) and meshing effects on the extrapolated
structural stress (b, c).

297

MARSTRUCT.indb 297 2/18/2011 5:48:03 PM

 discussion. Three factors are considered to be
responsible for this:
• The 3D model used by participant 4 shows a
slightly uneven stress distribution over the speci-
men width (60 mm) resulting in a small stress
increase if the max. stress is evaluated.
• A further stress increase in the 3D model is
caused by the fact that the element depth is t/2,
whereas no further subdivision is performed
in loading direction. The result is a bi-linear
stress distribution 0.4 t away from the hot spot
(Fig. 3b) yielding a higher extrapolated stress.
This effect, which has been briefly discussed by Figure 4. Finite element model (1/4 model) for the
Niemi et al. (2006), can be avoided by using only ‘1 mm stress approach’ of Xiao/Yamada (participant 1).
one element in thickness direction or refining
the mesh in both directions. Table 2. Computed structural stress concentration
• The small stresses found by participant 3 in factors for σx acc. to the ‘1 mm stress approach’ by Xiao
meshes having an element depth of t/4, seem to and Yamada (2004).
be created by the effect illustrated in Figure 3c.
Particularly at full-load carrying fillet welds, the Participant Element size L-2.5 D-2.5 L-7 D-7
force flow creates a zone with reduced stresses in
front of the weld toe which seems to affect the No contact:
stress at a distance of 0.4 t particularly in this 1 0.5 mm 1.43 1.12 1.30 1.17
type of model. Again, a very coarse or a refined 2 1.0 mm 1.25 1.08 1.13 1.07
model seems to be less affected by this. The bet- 3 1.0 mm 1.19 1.03 1.18 1.09
ter alternative would be the through-thickness 4 1.0 mm 1.21 1.06 – –
linearization in such cases although Table 1 of Contact:
1 0.5 mm 1.43 1.20 1.20 1.13
the IIW guideline (Niemi et al., 2006) allows
2 1.0 mm 1.29 1.04 1.10 1.06
different mesh refinements and extrapolation
3 1.0 mm 1.41 1.20 1.12 1.11
strategies.
4 1.0 mm 1.19 1.05 – –
Finally it should be remarked that the char-
acteristic fatigue strength FAT 90 for partial-
and full-load carrying fillet welds is slightly observed in previous analyses with 1 mm element
non-conservative with respect to the test results size (Fricke and Kahl, 2005).
achieved. One reason is the neglect of angular The results are summarized in Table 2, show-
misalignment effects on the structural stress which ing a quite large scatter which is due to the differ-
were found to increase the structural stresses on ent element sizes as well as finite element codes
average by 16%. This might also affect the other and element formulations. It can be seen that the
results described below. finer mesh with 0.5 mm long elements yields the
highest SCFs which are considered to be more
reliable.
4 APPLICATION OF THE ‘1 MM Table 2 gives the directional stress σx in lon-
APPROACH’ BY XIAO/YAMADA gitudinal direction. The max. principal stress σ1
might be a more relevant parameter, matching
The structural stress approach of Xiao and better the fatigue test results (Feltz and Fricke,
Yamada (2004) applies the stress at a location 2009). The principal stress is generally larger
1 mm below the weld toe on the anticipated crack than the directional stress particularly in the case
path as the relevant stress parameter for the of full-load carrying fillet welds with inclined
fatigue assessment using an S-N curve according force flow.
to FAT 100. A finite element analysis is neces-
sary with a mesh having element sizes of 1 mm
or less. 5 APPLICATION OF THE STRUCTURAL
The participants of the round-robin chose STRESS APPROACH BY DONG
partly 1 mm and partly 0.5 mm for the element size
(Fig. 4). The latter had been found to be advanta- Dong (2001) proposed a modified structural stress
geous to avoid stress underestimation which was approach which explicitly takes into account the

298

MARSTRUCT.indb 298 2/18/2011 5:48:03 PM

 effects of the plate thickness and stress gradient at
the hot spot. For two-sided fillet welds subjected to
symmetrical loading, a bi-linear stress distribution
is assumed, see Figure 5. This considers the local
force flow into load carrying fillet welds by linear-
izing the stress distribution over half thickness.
Dong proposed also a procedure to derive the
stress distribution at the hot spot from the stresses
acting in a section at a distance δ apart from the
hot spot, using equilibrium conditions between
the two sections. As the stresses in the distance δ
are less affected by the stress singularity, the proce-
dure is claimed to be mesh-insensitive.
Two participants applied the approach using
relatively fine meshes. The results are summa-
rized in Table 3. Same results were found by par-
ticipant 2 setting a distance δ = 2.4 mm and using
a 1.2 × 1.2 mm mesh size. It can be seen that the
assumption of a bi-linear stress distribution results
in rather large structural stresses. Some differences
between the results are observed particularly for the
lap joints, the reason of which remains unclear.
It should be noted that the fatigue assessment
is performed with a special stress parameter which
takes into account the stress gradient as well as
the plate thickness. Insofar, the structural stresses
obtained from the analysis cannot directly be com-
pared with the other results.
6 APPLICATION OF THE MODIFIED
STRUCTURAL STRESS APPROACH
BY POUTIAINEN
The modified through-thickness structural stress
proposed by Poutiainen (2006) takes into account
the local stress increase due to the force transferred
by the weld. Figure 6 shows the tri-linear stress
distribution for a plate with two-side fillet welds.
The weld stress σw, which can be calculated from
the forces F1 in the attached plates, determines the
stress increase at the weld toe.
Table 4 summarizes the analysis results of three
participants. The differences are very small. It can
be seen that the small throat thickness increases
considerably the structural stress at the weld toe,
particularly if it is full-load carrying (lap joint). On
the other hand it should be noted that the charac-
teristic fatigue strength is decreased in Poutiainen’s
approach.

Figure 5. Assumptions in the structural stress approach

after Dong (2001) for the stress analysis at two-sided fil- Figure 6. Modified through-thickness structural stress
let welds. for axial loading according to Poutiainen (2006).

Table 3. Computed structural stress concentration Table 4. Computed structural stress concentration
factors acc. to the approach after Dong (2001). factors acc. to the approach after Poutiainen (2006).

Element Participant Element size L-2.5 D-2.5 L-7 D-7

Participant size δ L-2.5 D-2.5 L-7 D-7
No contact:
No contact: 2 1.0 mm 2.21 1.64 1.18 1.11
2 1.0 mm 2 mm 2.09 1.56 1.59 1.43 3 1.0 mm 2.21 1.58 1.18 1.09
3 1.0 mm 1 mm 2.40 1.55 1.80 1.53 Contact:
Contact: 1 0.5 mm 2.20 1.62 1.18 1.11
2 1.0 mm 2 mm 2.06 1.59 1.42 1.40 2 1.0 mm 2.21 1.68 1.18 1.12
3 1.0 mm 1 mm 2.05 1.59 1.44 1.41 3 1.0 mm 2.21 1.65 1.18 1.11

299

MARSTRUCT.indb 299 2/18/2011 5:48:04 PM

 7 APPLICATION OF THE EFFECTIVE Table 5. Computed notch stress concentration factors
NOTCH STRESS APPROACH for the weld toe.

The effective notch stress approach, using the elas- Participant Element size L-2.5 D-2.5 L-7 D-7
tic stress in the notches rounded by a reference No contact:
radius rref = 1 mm, is able to assess both, weld toes 1 0.2 mm 8.38 3.54 3.97 3.03
and weld roots (Radaj et al., 2006; Hobbacher, 2 0.2 mm 8.23 4.01 3.66 2.93
2009). Weld roots at the end of non-fused plate 3 0.05 mm 9.32 3.71 4.17 3.09
surfaces as occurring in the test specimens are usu- 4 0.2 mm 6.62 3.69 – –
ally rounded with a so-called keyhole shape. Contact:
Relatively fine-meshed finite element models 1 0.2 mm 6.62 3.77 3.35 2.82
have been created for the numerical analyses (Fig. 7) 2 0.2 mm 7.42 4.01 3.41 2.84
considering the recommendations by Fricke (2008). 3 0.05 mm 6.89 3.86 3.24 2.73
Participant 4 used again a 3D model. 5 0.2 mm 7.14 3.97 3.42 2.84
As the throat thicknesses is partly only 2.5 mm
and the rounded keyhole notch reduces the throat
thickness by approx. 0.3 mm when placed at the Table 6. Computed notch stress concentration factors
end of the slit, two participants (1 and 5) modi- for the weld root.
fied the model by shifting the circle in horizontal
and/or vertical direction so that the nominal throat Participant Element size L-2.5 D-2.5 L-7 D-7
thickness is maintained. Figure 8 illustrates the
No contact:
modification made by the participants.
1 0.2 mm 7.61 2.59 2.73 1.38
Tables 5 and 6 summarize the results at the weld
2 0.2 mm 10.38 4.43 4.17 2.23
toe and weld root, respectively. The SCFs were
3 0.05 mm 9.01 2.94 2.87 1.40
computed using the max. principal stress at the
4 0.2 mm 10.37 4.55 – –
rounded surface, according to Fricke (2008).
Contact:
Some scatter in the results is observed. For 1 0.2 mm 8.38 4.04 4.12 2.24
instance, the models without shifted keyhole notch 2 0.2 mm 9.28 4.38 4.30 2.36
3 0.05 mm 8.66 4.20 4.05 2.24
5 0.2 mm 7.89 3.90 3.89 2.13

yield larger SCFs at the weld root than those with

Figure 7. Typical mesh (1/4 model) with keyhole notch
and equivalent stresses for the effective notch stress anal-
ysis (model D-2.5, σn = 200 MPa, participant 3).
Figure 8. Horizontal shift (a, participant 1) and diago-
nal shift (b, participant 5) of the keyhole notch to main-
tain the weld throat thickness a.
shifted circles, see e.g., the results for L-2.5 in Table 6
(contact: participants 2/3 vs. 1/5). Participant 5
found the difference between the non-shifted and
shifted keyhole (Fig. 8b) to be 12–18% depending
on the throat thickness and grade of load-carriage.
This explains part of the differences in the tables,
but not all.
According to the results, most specimens should
have failed from the weld root showing the larger
SCF (except for D-7). The reason is the overesti-
mation of notch stresses by the keyhole notch at
the weld root.
Therefore, an additional analysis assuming a
U-notch was performed by participant 3. As the
U-notch creates a wide gap, no contact was assumed.
Figure 8 gives an impression of the mesh.
Table 7 summarizes the results. The U-notch
slightly increases the SCF at the weld toe and con-
siderably decreases it at the weld root. The weld toe
is critical in all cases which might to be expected
also if contact had been assumed. Therefore it can
be concluded that the assumption of the U-notch
is non-conservative as already shown in previous
analyses (Fricke et al. 2009).

300

MARSTRUCT.indb 300 2/18/2011 5:48:05 PM

 Table 7. Computed notch stress concentration factors
with weld root modeled with keyhole and U-notch (par-
ticipant 3, no contact).
Stress location Root notch L-2.5 D-2.5 L-7 D-7
Weld toe Keyhole 9.32 3.71 4.17 3.09
Weld toe U-notch 10.07 3.79 4.39 3.20
Weld root Keyhole 9.01 2.94 2.87 1.40
Weld root U-notch 6.62 2.07 1.39 0.71
Figure 9. Typical mesh (1/4 model) with U-notch
and equivalent stresses (model L-7, σn = 200 MPa,
participant 3).
8 SUMMARY AND CONCLUSIONS
Finite element calculations were performed in a
round-robin within the MARSTRUCT project
aiming at the determination of relevant structural
and notch stresses in load-carrying fillet welds at
lap joints and doubler plates with 12 mm thick-
ness. Both weld toe and weld root cracks are prob-
able, the latter particularly in welds with small weld
throat.
The results obtained by up to five participants
showed some unexpected scatter when applying
the different fatigue approaches. The following
causes were identified and conclusions drawn:
• In the structural hot-spot approach, the mesh
refinement solely in thickness direction affects
the structural stress in a distance 0.4 t from the
hot spot and, hence, the extrapolated stress. Two
elements result in increased stresses and four ele-
ments in decreased stresses, both particularly
at load-carrying welds. The mesh refinement
should be performed in both directions. Alter-
natively the through-thickness linearization is
recommended. The fatigue assessment is slightly
non-conservative when FAT 90 is applied, if
misalignments are not taken in due account.
• In the ‘1 mm stress approach’ by Xiao/Yamada,
the local stress varies if elements with 1 mm
length are chosen. A refined mesh (0.5 mm)
and the evaluation of the principal stress are
301
MARSTRUCT.indb 301
recommended to achieve conservative results.
However, this is not clearly specified in Xiao and
Yamada (2004).
• The structural stress approaches by Dong and
Poutiainen consider the stress increase due to
load-carrying welds in a simplified way, which
could not be validated by the tests. However, the
results are more on the conservative side.
• The effective notch stress approach, which is
able to assess both weld toe and weld root fail-
ure, overestimates the local stress at the weld
root when the notch is rounded with a key-
hole shape. In addition to the increased stress
concentration factor by the keyhole notch, the
weld throat thickness is slightly reduced which
increases the local stress by 12–18% compared
to a shifted notch maintaining the throat thick-
ness. A U-notch is obviously non-conservative
as the weld root stress is smaller than the weld
toe stress even in case of root failure.
In summary, the investigation confirms the
statement of a preceding paper (Feltz and Fricke,
2009) that the fatigue assessment with current
approaches is not satisfactory in joints with non-
welded root faces oriented parallel to the load
direction. Here, the crack propagation approach as
well as the notch stress intensity factor approach
could be more advantageous (Feltz et al., 2010).
ACKNOWLEDGEMENTS
The investigations were performed within the Net-
work of Excellence on Marine Structures MAR-
STRUCT (http://www.mar.ist.utl.pt/marstruct/),
funded by the European Union through the
Growth programme under contract TNE3-CT-
2003-506141.
REFERENCES
Dong, P. 2001. A structural stress definition and numeri-
cal implementation for fatigue analyses. Int. J. Fatigue,
23:10, 865–876.
Feltz, O. & Fricke, W. 2009. Experimental and numerical
fatigue analysis of partial-load and full-load carrying
fillet welds at doubler plates and lap joints. In: Analy-
sis and Design of Marine Structures (Ed. Guedes
Soares & Das), Taylor & Francis, London.
Feltz, O., Fischer, C. & Fricke, W. 2010. Fatigue Assess-
ment of Weld Toe and Root Cracks with the Notch
Stress Intensity Factor and Crack Propagation
Approach. Proc. of 11th Int. Symp. on Practical
Design of Ships & Other Floating Struct., COPPE,
Rio de Janeiro.
Fricke, W. 2008. Guideline for the Fatigue Assessment by
Notch Stress Analysis for Welded Structures. IIW-Doc.
XIII-2240r1-08/XV-1289r1-08, Int. Inst. of Welding.
2/18/2011 5:48:06 PM
 Fricke, W., Cui, W., Kierkegaard, H., Kihl, D., Koval, M.,
Mikkola, T., Parmentier, G., Toyosada, M. & Yoon, J.-H.
2002. Comparative Fatigue Strength Assessment of
a Structural Detail in a Containership using Various
Approaches of Classification Societies. Marine Struc-
tures 15:1–13.
Fricke, W. & Kahl, A. 2005. Comparison of different struc-
tural stress approaches for fatigue assessment of welded
ship structures. Marine Structures 18:473–488.
Fricke, W., Bollero, A., Chirica, I., Garbatov, Y.,
Jancart, F., Kahl, A., Remes, H., Rizzo, C.M., von
Selle, H., Urban, A. & Wei, L. 2007. Round Robin
study on structural hot-spot and effective notch stress
analysis. In: Advancements in Marine Structures
(Ed. C. Guedes Soares & P.K. Das), 169–176, Taylor &
Francis, London.
Fricke, W., Paetzold, H. & Zipfel, B. 2009. Fatigue tests
and numerical analyses of a connection of steel sand-
wich plates. Welding in the World 53, No. 7/8, S.
R151–R157.
Fricke, W., Feltz, O., Gritl, D., Romanoff, J., Remes, H.,
Rizzo, C., Risso, G., Codda, M., Casuscelli, F., Garbatov,
Y., Bäckström, M. & Kukkanen, T. 2010. Round Robin
of Fatigue Analyses (Deliverable D2.4). Report MAR-
D2-4-TUHH-08(1), Network of Excellence on Marine
Structures (MARSTRUCT).
302
MARSTRUCT.indb 302
Hobbacher, A. 2007. Recommendations for Fatigue
Design of Welded Joints and Components, Final Draft,
IIW-Doc. XIII-2151r1-07/XV-1254r1-07, Int. Inst. of
Welding.
IACS (2010): Common Structural Rules, http://www.iacs.
org.uk/publications/default.aspx
Niemi, E., Fricke, W. & Maddox, S.J. 2006. Fatigue
Analysis of Welded Components—Designer’s Guide
to Structural Hot-Spot Approach, Cambridge: Wood-
head Publ.
Poutiainen, I. 2006. A modified structural stress method
for fatigue assessment of welded structures. Doctoral
Thesis 251, Lappeenranta University of Technology.
Poutiainen, I. & Marquis, G. 2006. Comparison of local
approaches in fatigue analysis of welded structures.
IIW-Doc. XIII-2105-06, Int. Institute of Welding.
Radaj, D., Sonsino, C.M. & Fricke, W. 2006. Fatigue
Assessment of Welded Joints by Local Approaches.
Cambridge: Woodhead Publishing (2nd Edition).
Xiao, Z.G. & Yamada, K. 2004. A method of determin-
ing geometric stress for fatigue strength evaluation of
steel welded joints. Int. J. Fatigue 26:1277–1293.
2/18/2011 5:48:07 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Fatigue estimation of a ship structural detail

S. Giuglea & A. Chirica

Ship Design Group Galati, Romania

I. Chirica & E.F. Beznea

University “Dunarea de Jos” of Galati, Romania

ABSTRACT: A FEM analysis used for fatigue assessment, based on the rules, was done in the paper.
The approach for fatigue assessment in the early design stage has been developed. To overcome the
challenges due to limited information in the early design stage, generic structural elements and predefined
fatigue-critical details were chosen. This allows the development of a common approach for different ship
types, which is also applicable for optimization purpose. The aim of the sensitivity analysis is to study the
effect of the structural scantlings on the structural hotspot stress factor. The analysis is focused on a pillar
and girder connection of the ship hull structure of a Ropax vessel.

1 INTRODUCTION Structure Conference). In reality, fatigue problems

The fatigue life of a structure under repeated
loading can be divided into the crack initiation life
and the crack propagation life where the initiated
crack grows to a certain point where it affects the
safety of structure.
The surface cracks that initiate and propagate
at the toe of the welding area are affected by the
residual stresses that are created during the welding
process. In this experiment, an effect of these
residual stresses was reflected on the analysis of
the crack propagation, and it was compared with
the actual raw data.
Fatigue is responsible for a large proportion of
cracks occurring in welded ship structural details.
For many years fatigue related failure has become
a major concern in the maintenance of existing
vessels and the design of new vessels.
Ships are usually built to a defined set of
details which are documented by classification
societies, owners (such as navies), or the builders.
Details include stringer and frame intersections,
bulkhead and stiffener connections (watertight
and non-watertight), penetrations, cutouts, etc.
The increasing importance of fatigue strength
for maritime structures has resulted corresponding
research activities and recommendations for design,
fabrication and operation of maritime structures
to counter the increased risks described above.
These worldwide research efforts have produced an
amount of publications that it is impossible to give
a comprehensive overview. The best starting point
for at least extensive literature surveys remains
the proceedings of the ISSC (International Ship
appear after a certain number of load cycles, often
only after months or even years of operation. The
feedback for design is then often slow. Experiments
allow to accelerating the time scale in applying
realistic loads at a much higher frequency and
observing then fatigue problems (after the same
number of load cycles) in much shorter time,
namely hours or days.
Over several decades, test data were accumulated
for many typical ship structures and these data
have been used by classification societies (along
with feedback from fatigue damages on actual
ships) to compile catalogues of stress increasing
factors. These catalogues allow a simple, pragmatic
approach to structural design. The structural
designer can compute nominal stress using long
established and widely available standard tools
for structural analysis. The catalogues then give
a correction for the influence of a discontinuity
(like a weld or a corner) allowing to transform
fatigue strength into a changed upper limit for the
static stresses. The approach is pragmatic, but not
applicable to structures which are not (yet) found
in the catalogues. This poses problems each time
new structural details or new materials are used.
Design of ships is an interactive process, where
major decisions are made in an early design stage
covering for instance main dimensions and general
arrangements. Information of structural details,
which are basic requirements for reliable fatigue
assessment, is available in the following design
stages. This is a significant obstacle for the early
design stage, because the decisions done in this
stage have a strong influence on the fatigue life

303

MARSTRUCT.indb 303 2/18/2011 5:48:07 PM

 of the hull girder. Structural modifications done
after the early design stage are usually limited
and expensive for production. At present, there
is no common approach for fatigue assessment
which can fulfill requirements of the early fatigue
design.
Traditional dimensioning of ship structures
followed simple semi-empirical formulae giving
directly the thickness of plates of stiffeners or
a required section modulus. Fatigue strength,
however, requires a more detailed stress analysis.
Complex ship structures are increasingly analyzed
using 3D finite elements models. These models
allow a realistic distribution of loads and capture
the interaction between the main structures.
Usually the whole ship hull with its main structures
is modeled using plate elements. Secondary
structures like stiffeners are modeled simplified
using truss elements. The analysis gives global
nominal stress distributions for coarse grids. For
fine grids, effects of effective width and geometry
of the structure are also captured.
Local finite-element models serve to determine
the stress increase due to geometry of the
structure.
Usually plane-strain plate elements suffice to
determine the notch stress at plate edges of holes.
For hot-spot stress at weld toes and plate structures,
either plate elements of volume elements are
employed.
Volume elements require more effort, but consider
the stiffness and load distributing effect of the weld
better. The definition of the hot-spot stress requires
the evaluation of the linear stress component
over the plate thickness. This is automatically
given for plate elements. For volume elements,
an elegant solution is arranging only one element
over the plate thickness. Then intermediate nodes
are necessary at the element edges to capture the
bending properly. Using only 2 integration points
over the thickness yields directly the linear stress
component. This can then be extrapolated to the
plate edge to give the hot-spot stress. The loads for
the local finite-element models come either from
prescribed external stresses or deformations. These
are taken either from a defined initial state or from
a global analysis.
In the paper, an approach for fatigue assessment
in the early design stage has been developed. To
overcome the challenges due to limited information
in the early design stage, generic structural
elements and predefined fatigue-critical details
are applied. This allows the development of a com-
mon approach for different ship types, which is
also applicable for optimization purposes.
Based on a scientific and engineering approach
fatigue-critical structural details and important char-
acteristics of the ships are assessed. The scientific
304
MARSTRUCT.indb 304
approach is based on damage statistics of fatigue
failures in ship structures, and aims to identify
critical structural details and corresponding load-
ing modes. In the European Project IMPROVE the
review is concentrated especially on tankers, and is
extended to cover Ropax and LNG ships based on
an engineering approach. The engineering approach
is based on pre-existing know-how and knowledge.
The study is focused on special features of differ-
ent ship types which are further developed in the
Improve project. The main results of this study are
the identification of generic and ship-type-dependent
features in fatigue assessment. This is the basis for
the development of the fatigue approach for early
design stage.
The important calculus is focused on the hot-
spot and notch stress level, and it is based on the
results of extensive FE-analysis of typical structural
detail, performed as sensitivity analysis.
In general, there are several approaches for
fatigue assessment. The commonly used methods
can be divided into different groups according to
the applied strength parameters and corresponding
response analysis, (DNV 2005). The global
approaches, such as nominal stress approach, are
based on main dimension of the structure. These
approaches are easy to apply, but their practical
applications for complex ship structures are limited.
Therefore, the use of more advance approaches
taking into account local parameters are preferred
and applied (ISSC 2003, IACS 2008a, IACS 2008b).
These methods in ship design are structural hot-spot
stress and notch stress approaches. The fracture
mechanics approaches with J-integral or stress
intensity factors are not commonly used in the
design stage due to the extremely time-consuming
structural analysis.
Weakness of the structural hot-spot stress
and notch stress approaches is that they require
structural analysis in a detailed level. This can be
obtained by applying the finite element method,
which is however time consuming and, thus, not
suitable for the early design stage. At present there
is no common and simplified approach for fatigue
assessment which can fulfill the requirements of
early fatigue design.
Estimating the production cost is a fundamental
part of ship design. Traditionally, shipyards employ
empirical methods to estimate the cost of a new
ship, as ships typically are one-of-a-kind products
and orders are won based on early bids, i.e. naval
architects must estimate costs a priori and within
relatively short time (order of several days). Most
of these traditional cost estimate approaches
are related to the ship weight, which in turn is
estimated based on ship type and main dimen-
sions as well as installed power and equipment
and outfit.
2/18/2011 5:48:07 PM
 Design for fatigue leads to much more elaborate
structural design and assembly procedures, requir-
ing grinding welds to reduce notch stresses. The
structural weight is virtually unaffected, but the
construction costs are naturally significantly higher.
Thus a cost estimate following the traditional
approach with traditional empirical coefficients
based on yesterday’s practice underestimates costs.
New approaches to cost estimates need to be
developed to reflect modern design—for fatigue,
design—for—production approaches in ship
building. Such approaches are pursued worldwide
at a few places and also subject to a research and
development in the Project FP6 IMPROVE.
The development of the approach for fatigue
assessment requires a balanced approach for load,
response and strength with sufficient accuracy.
However, the approach should overcome the
challenge of limited information of structural
details in the early design stage. Simplification is
also needed to obtain a generic approach, which
is applicable for optimization purposes and can Figure 1. Structural model of the pillar and web
be linked to the existing design tools. The existing connection.
design tools have suitable databases of general
geometry of structure and load specifications. These ship types. However, to obtain optimum space for
design tools can be also applied to analyze primary cargo transporting, special structural elements such
and secondary stresses of hull girder, and they as pillars have been applied in some ship types.
should be exploited to fatigue analysis. Therefore, A Ropax transports cars and passengers requir-
this study is focused especially on the analysis of ing large open spaces. Large open spaces are
local stresses. The main challenges are the transfor- obtained by pillars and large frame structures.
mation of the response from existing design tools Thus, the general arrangement and geometry of the
and the calculation of fatigue effective stresses main frame results in hull girder deformations. Due
in critical structural details with sophisticated to large openings Ropax ships have a relative low
assumptions. This requires pre-defined structural stiffness and strength against transverse loading,
details, which are generic, but however define the so called racking. Therefore, the fatigue assessment
fatigue strength of the hull girder. of the connections of stiffened plates and longi-
Fatigue failures are usually detected in inspec- tudinals are not sufficient, and the analysis has to
tions of classification societies. While fatigue cover connections of pillars and web frames under
problems are frequent, catastrophes due to fatigue racking (see Fig. 1).
are quite rare, because micro-cracks take usually
a long time (sometimes months to years) before
progressing to structure failure. They are then 2 STRUCTURAL MODEL
typically detected and rectified at an early stage.
The inspection requires surveyors (representatives The aim of the sensitivity analysis is to study the
of the classification society inspecting on-board) effect of the structural scantlings on the structural
to look at the structures up close. This poses hotspot stress factor. The analysis is focused on the
problems in narrow spaces. pillar and girder connections marked by the red
Some characteristics affecting the fatigue circles in Figure 1. The modeled structure is part
strength of the hull girder are strongly depended of deck structure of ROPAX ship including pillars,
on the ship type. The use of the ship is defining transverse web frame and longitudinal girder. The
the main dimensions of the ship, shape of the hull model includes, in the longitudinal direction, the
girder, geometry of the main frame and the steel web frame spacing, and in the transverse direction,
arrangement. Sailing routes i.e. weather condition some part of the deck. In the analysis the scantlings
and service speed can be different between different such as thickness of deck plate, pillar, web frame
ships. These differences affect mainly wave and and girder are varied. However, the topology of
cargo induced fatigue loading and response of hull the structure is fixed to simplify the analysis.
girder in nominal stress level. Structural details Calculus was done with COSMOS/M package
and connections are quite similar between different soft. The FEM model has almost 1 million degrees

305

MARSTRUCT.indb 305 2/18/2011 5:48:07 PM

 of freedom. The mesh was done according to
Bertram & Moctar (2002). The 8 nodes quadrilateral
elements were used. In the interested area, the length
of the elements is 0.5 of the thickness.
The structure is modeled using shell elements.
The element type is parabolic quadrilateral shell
elements. The example of finite mesh at the hot-
spot area is shown in Figure 3. The deck stiffeners,
i.e. HP profiles, were modeled with L-profiles. The
L-profiles have same cross-sectional area, moment
of inertia and neutral axis position as the HP
profiles.
The displacement restraints are applied to the
selected boundaries of the model (see Fig. 1).
The plane A is clamped and the symmetry
boundary condition is applied to the planes
B and C (Table 1).
To cover different loading cases, which can exist
for this kind of ship structures under racking, FE Figure 3. FEM model.
analysis is done to the group of load cases instead
of one specified load case. The analyses of all FE
models should include totally nine load cases. Nine the plane. The forces and moments that are applied
different loading cases are applied for each of the in Figure 2 are unitary.
free boundary of the model, see Figure 2. The load
was modeled so that the loaded boundary remains
3 NUMERICAL RESULTS
Table 1. Boundary condition of FE model. Stress distributions at hotspot areas are determined
Displacement according to the paths presented in Figures 4
and 5.
Plane Condition ux uy uz θx θy θz The naval rules give a general background for the
rule requirements for fatigue control of ship struc-
A Clamped 0 0 0 0 0 0 tures, and to provide detailed recommendations for
B Symmetry 0 – – – 0 0 such a control. The aim of the fatigue control is to
C Symmetry 0 – – – 0 0 ensure that all parts of the hull structure subjected
to fatigue (dynamic) loading have an adequate
fatigue life.
Calculated fatigue lives, calibrated with the
relevant fatigue damage data, may give the basis
for the structural design (steel selection, scantlings
and local details).
Furthermore, they can form the basis for effi-
cient inspection programs during fabrication and
throughout the life of the structure.
To ensure that the structure will fulfil its intended
function, fatigue assessment, supported where
appropriate by a detailed fatigue analysis, should
be carried out for each individual type of struc-
tural detail which is subjected to extensive dynamic
loading. It should be noted that every welded joint
and attachment or other form of stress concentra-
tion is potentially a source of fatigue cracking and
should be individually considered.
Stress distributions at hotspot areas are deter-
mined according to the paths presented in Figure 4.
The naval rules give a general background for the
rule requirements for fatigue control of ship struc-
Figure 2. Load cases for FE models. tures, and to provide detailed recommendations for

306

MARSTRUCT.indb 306 2/18/2011 5:48:08 PM

 Figure 4. Directions for hot spot stress calculus in the Figure 5. Directions for hot spot stress calculus in the
upper part of the connection. lower part of the connection.

Figure 6. FEM model detail of hot spot area.

307

MARSTRUCT.indb 307 2/18/2011 5:48:09 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Adjoint design sensitivity analysis of crack propagation

using molecular-continuum multiscale approach

Hong-Lae Jang, Min-Geun Kim & Seonho Cho

National Creative Research Initiatives Center for Isogeometric Optimal Design and Department of Naval
Architecture and Ocean Engineering, Seoul National University, Seoul, Korea

ABSTRACT: We have developed a multi-scale design sensitivity analysis method for transient dynamics
using a bridging scale method by a projection operator for scale decomposition. Employing a mass-weighted
projection operator, we can fully decouple the equations of motion into fine and coarse scales using the
orthogonal property of complimentary projector to the mass matrix. Therefore, independent solvers in
response analysis can be utilized for the fine scale analysis of molecular dynamics and the coarse scale
analysis of finite element analysis. To reduce the size of problems and to improve the computational
efficiency, a Generalized Langevin Equation (GLE) is used for a localized Molecular Dynamics (MD)
analysis. Through demonstrative numerical examples, it turns out that the derived sensitivity analysis
method is accurate and efficient compared with finite difference sensitivity.

1 INTRODUCTION in the response analysis, the computation cost of

Due to the rapid development of nano-technologies,
the necessity of nano-scale analysis is continuously
increasing. We present a Design Sensitivity
Analysis (DSA) method for multi-scale problems,
combining molecule and continuum scales, based
on bridging scale approach (Wagner, G.J. and
Liu, W.K., 2003) which is a concurrent coupling
method through the GLE at the confined region
of interest. By the decomposition of total solution
into fine and coarse scales using a mass-weighted
projection operator, each scale can be consid-
ered independently; fine scale solution can be
obtained by MD simulation and coarse scale
solution by Finite Element Analysis (FEA). To
reduce the computation cost, the fine scale MD
region is confined in a local domain while the
coarse scale FEA in the whole. For a localized
MD simulation, a generalized Langevin equation
is introduced for the reduced MD system and the
GLE force using a time-history kernel is applied at
the boundary atoms in the MD system.
Transient dynamic sensitivity analysis is
carried out for the macro scale performance of
system considering micro scale effects using the
adjoint variable method (Choi, K.K. and Kim,
N.H., 2004). The sensitivity decomposition is also
performed by the same projection operator used
in the response analysis, which gives the sensitivity
equation for fine and coarse scales independently
without any coupling. Since we can use the time
history kernel function that is already computed
sensitivity is much less than that of the response
analysis. In this paper, we first decompose the
sensitivity into fine and coarse scales and derive
the design sensitivity expressions for the transient
dynamic problems using the Adjoint Variable
Method (AVM). The accuracy of the derived
analytical shape sensitivity is compared with that
from Finite Difference Method (FDM) and Direct
Differentiation Method (DDM).
2 BRIDGING SCALE METHOD
2.1 Scale decomposition
In the bridging scale method, the total solution z
is decomposed into the fine scale solution w and
coarse scale solution v.
z(X, t) = v(X, t) + w(X, t), (1)
where X is the undeformed coordinate; and t is time
variable. By the decomposition of total solution
into fine and coarse scale using a mass-weighted
projection operator that possesses the orthogo-
nal property of complimentary projector to mass
matrix, each scale can be written as Eq.(2).
z = Nd + Qq, (2)
where N is the coarse scale finite element shape
function matrix; d is the coarse scale nodal solution;

311

MARSTRUCT.indb 311 2/18/2011 5:48:13 PM

 q is the fine scale solution; and Q is complimentary
∫o
t
a ( ) = fa ( q a ,v b )
M Aaq (t − τ )
projection matrix which is expressed as Eq.(3). (9)
{qa ( ) v a ( )} d ext
a (t ),
Q = I − NM−1NT MA, (3)

where MA is mass matrix of atoms; and M is the  = NT f MD ( Nd + Q q ) + NT f CB (d ),

Md (10)
a a
coarse scale matrix which is expressed as Eq.(4).

M = NT MAN. (4) where the time history kernel function θ(t) is

obtained by Eq.(11).
The Lagrangian of multi-scale system is assumed
{s }
−1 ⎡ −1
to be given as Eq.(5). θ( ) L−1 ab − Ab bb M −Ab1 Kba ⎤⎥ . (11)
⎣ ⎦
1
L( ,  ) T
A U (z) T ext
, (5) The second term of Eq.(9) which includes the
2
time history kernel function θ(t) is called GLE force
where U(z) is interatomic potential and f ext is external and means the impedance force which is equivalent
forces. We obtain the decomposed multi-scale with the influence of eliminated atom.
equations of motion by substituting Eqs.(2) into (5).
2.3 Time integration
M A
q = f(
f ( q ) + f ext , (6)
In this paper, we integrate Eqs.(9) and (10)

{ },
concurrently using the velocity Verlet method for
 = NT f z
Md f (7) MD and explicit central difference for FE. The
MD simulation is advanced by m steps of size
where the internal force f(q) can be obtained as Δtm = Δt/m while the FE simulation is advanced
Eq.(8). through a single time step of size Δt. The more
detailed process can be referred in the reference
∂U ( q ) (Wagner, G.J. and Liu, W.K., 2003).
f (q) = − (8)
∂q
3 DESIGN SENSITIVITY ANALYSIS (DSA)
2.2 Generalized Langevin equation
Most of the multi-scale problems are inter- The first order variation the total solution z can be
ested in the fine scale effects only on locally con- expressed as,
fined regions, not the whole domain so that we
introduce the bridging scale approach by reducing z′ Nd Qq ′ . (12)
the MD domain to analyze a locally confined fine
scale region in which we are interested as described We decomposed the total adjoint variable λ into
in Figure 1. fine scale ξ and coarse scale η like the total solution
To prevent fine scale wave reflection on the decomposition in the response analysis.
MD/FE boundary, a GLE force f imp is introduced.
The multi-scale equations of motion is written as, λ η ξ. (13)

Interested atoms Ωa
We can obtain the multiscale adjoint equations
fimp (GLE Force)
Eliminated atoms Ωb using the decomposed adjoint variables.

 ∂fa ( q a , v b , b )
ζTa Aa ζTa (t )
∂q a
Ωa tT
Ωa Ωc
+ ∫ ζTa ( )θ( t )dτ
t
∂f MD ( + a a)
Ωb + η(t )T NT Qa
∂( + a a)
Full Domanin (MD) Fine scale (GLE) Coarse scale (FE) ⎛ ∂G d ∂G ⎞
+⎜ − , (14)
Figure 1. Bridging scale method. ⎝ ∂q a dt ∂q a ⎟⎠

312

MARSTRUCT.indb 312 2/18/2011 5:48:13 PM

 ∂fa ( + a a , b) mA b1mA0 , b2ε 0 , σ = b3σ 0 (20)

η( )T ( ) η( )T NT N
∂( + a a)

+ η( )T
∂NT f CB
C
{
(d,, ) } In the Figure 2, the red line denotes the initial
crack line. In this model, uniaxial tensions are
∂d imposed on both the top and bottom of the struc-
∂ a f ( q a ,v
, vb , ) ture with the velocity profile given in the right side
+ ζTa ( ) N(( b)
∂v b of Figure 2. The analysis results at each time step
tT are shown in Figure 3 and Figure 4. The contours
− ∫ ζTa ( )θ( , b)) (Xa ) in Figure 3 are L-J potential values and the ones in
t
⎛ ∂G d ∂G ⎞ Figure 4 are y-directional displacement.
+⎜ − . (15)
⎝ ∂d ddt ∂d ⎟⎠ Performance measure is the total displacement
on the point XP at each time step. Table 1 shows the
The adjoint system is a terminal value problem design sensitivity obtained by FDM, DDM AVM,
in transient dynamics. The corresponding terminal respectively. Compared with the results of FDM
conditions are derived as and DDM, that of AVM is very accurate.
Figure 5 shows the comparison of computational
∂g costs of the FDM, DDM, and AVM. The computa-
ζTa (t )MAa = , (16) tion cost for the FDM and DDM linearly increases
t tT
∂q a t tT as the number of design variables increases.
However, since the AVM needs only one more time
⎛ ∂g ∂G ⎞ integration for the adjoint equations regardless of
ζ Ta = −⎜ + , (17)
Aa
t tT
⎝ ∂q a ∂q a ⎟⎠ t tT
V

∂g
η( )T M =  , (18) FE region

t tT ∂d t tT V
b/2
XP
Initial crack
h/7
h
h/2
4 / 7b Vmax
h/4
(19) MD+FE region

FE region

We can use the same solver as used in the V

t1 tT t
response analysis since the derived adjoint system b

written as Eqs.(14) and (15) has the similar form

Figure 2. Model and prescribed velocity profile.
as the multiscale equations of motion, Eqs.(9)
and (10). By using the adjoint variables obtained
by the adjoint system and derived terminal con-
ditions, we can calculate the design sensitivity of
multiscale system easily.

4 NUMERICAL EXAMPLES

4.1 Crack propagation problem (1)

Figure 3. L-J potential value (0Δt, 2000Δt, 4000Δt,
A flat plate with initial crack is subjected to the 6500Δt).
tensile force shown in Figure 2. We performed
the bridging scale analysis and multiscale design
sensitivity analysis to the model problem. The
MD simulation is performed using in-house code.
To describe the atomic breaking on the crack
tip, we use L-J potential for interatomic force
(Park, H.S., Karpov, E.G., and Liu, W.K. 2005).
To verify the accuracy of design sensitivities,
we take the design parameterization for mass of
atoms mA, energy depth σ and collision diameter Figure 4. Total y-displacement (0Δt, 2000Δt, 4000Δt,
σ in L-J potential as 6500Δt).

313

MARSTRUCT.indb 313 2/18/2011 5:48:21 PM

 Table 1. Comparison of design sensitivity results.

Time step (×Δt) DV Δψ (a) δψ DDM(b) δψ AVM (c) (c)/(a) * 100

b1 1.676E-06 1.676E-06 1.698E-06 101.32

4,000 b2 −1.677E-06 −1.676E-06 −1.699E-06 101.33
b3 7.711E-05 7.705E-05 7.967E-05 103.32
b1 −3.978E-06 −3.978E-06 −4.102E-06 103.13
6,500 b2 3.978E-06 3.978E-06 4.102E-06 103.13
b3 −4.756E-04 −4.757E-04 −5.257E-04 110.53

Table 2. Comparison of design sensitivity results.

Time step (×Δt) DV Δψ (a) δψ coarse

AVM
( b) (b)/(a) * 100

t2 −2.000E-10 −2.009E-10 100.43

t3 −2.011E-10 −2.011E-10 99.98
1,200 t4 5.769E-10 5.769E-10 100.00
t5 −1.345E-08 −1.345E-08 100.00
t6 −2.089E-08 −2.089E-08 100.00
t7 −2.086E-08 −2.086E-08 100.00
t8 −2.069E-08 −2.076E-08 100.35

V
Non-designable
thickness
t8
9
h FE 1
20 b t7
2
t6
t5
h Xd
t4
Initial crack t3
t2
MD+FE Xq t1
Non-designable
b/8 FE b/8 thickness

V
Figure 5. Comparison of computational costs. b

Figure 6. Model and design variables for crack

the number of design variables, it turns out to be propagation.
the most efficient method as the design variables
increase.

4.2 Crack propagation problem (2)

Next we consider the crack propagation problem
with different type of initial crack. The initial
crack lines are at the end of the flat plate shown in
the Figure 6. The same velocity profile and intera-
tomic potential of the previous example is used.
The design variables are the thickness at each Figure 7. Analysis results at terminal time step 2000Δt;
section of the plate and the performance measure L-J potential values and y-directional displacement
is the y-directional displacement at the point Xd. contour.

314

MARSTRUCT.indb 314 2/18/2011 5:48:26 PM

 5 CONCLUSIONS
A multi-scale DSA method in a bridging scale
approach is developed using the AVM. To avoid any
iterative computations between the scales, we use a
fully decoupled equation for each scale in original
response as well as sensitivity analyses. Through
a GLE, a locally confined MD region instead of
whole MD system is considered for the fine scale
solution whereas a FE analysis for the coarse scale
solution is performed on the whole region. The
efficiency of the developed method is achieved
due to the fully decoupled multi-scale equations
in these scales, which are derived by the use of
identical mass-weighted projection in the response
as well as adjoint analyses. Numerical implementa-
tions demonstrate the accuracy of the developed
DSA method for various design variables. The
developed method turns out to work very well
for any design variable in both scales. The devel-
oped bridging scale DSA method can be utilized
in industrial problems if it is extended to parallel
computation and additional considerations. Also
realistic simulation of crack propagation problems
should include thermal effects due to the breaking
of an atomic bond. That is a future work for our
research group.
315
MARSTRUCT.indb 315
ACKNOWLEDGEMENTS
This research was supported by Basic Science
Research Program through the National Research
Foundation of Korea (NRF) funded by the
Ministry of Education, Science and Technology
(Grant Number 2010-18282). The support is grate-
fully acknowledged.
REFERENCES
Choi, K.K. & Kim, N.H. 2004. Structural Sensitivity
Analysis and Optimization 1: Linear Systems. New York:
Splinger.
Kadowaki, H. & Liu, W.K. 2004. Bridging multi-scale
method for localization problems. Computer Methods in
Applied Mechanics and Engineering. 193: 1733–1772.
Liu, W.K., Karpov, E.G. & Park, H.S. 2004. An intro-
duction to computational nanomechanics and
materials. Computer Methods in Applied Mechanics
and Engineering. 193: 1529–1578.
Park, H.S., Karpov, E.G. & Liu, W.K. 2005. The bridg-
ing scale for two-dimensional atomistic/continuum
coupling. Philosophical Magazine. 85(1): 79–113.
Wagner, G.J. & Liu, W.K. 2003. Coupling of atomistic
and continuum simulations using a bridging scale
decomposition. Journal of Computational Physics.
190: 249–274.
2/18/2011 5:48:28 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

A study of design loads for fatigue strength utilizing direct

calculation under real operational conditions

Masayoshi Oka, Tomoki Takami & Yoshitaka Ogawa

National Maritime Research Institute, Japan

Ken Takagi
The University of Tokyo, Japan

ABSTRACT: In this paper, in order to examine the wave loads for fatigue design, a tank test using
newly designed elastic model was performed, and a further detailed analysis of full scale measurement
on Post-panamax container ship was carried out. It was verified that the actual stress by the full scale
measurement is quite smaller than that by the direct computation, and the difference is mainly caused
by of environmental condition applied to the computation. Moreover, the probability of occurrence of
whipping during long term was estimated. It was found that whipping induced stress cumulates a mount
of fatigue damage, but the probability of occurrence of whipping is small in terms of the subjected ship.
Finally, the further advanced view for the fatigue design is proposed.

1 INTRODUCTION in order to realize the phenomena of real ship in

In terms of the increasing global demand for
transportation and transport efficiency from the
view point of reduction of GHG in recent years,
a demand for the scaling up of ships have been
growing. Moreover, the adoption of IMO/GBS was
promoted to clarify that the safety level of the hull
structure rules. For further ensuring the structural
safety, total hull steel weight maybe increase, while
in terms of transport efficiency, the steel weight is
demanded to be minimized. Therefore it is expected
a reasonable design to meet these needs.
Due to scaling up of a ship, a risk in relation to
loads, which have not been considered in the con-
ventional design, may arise. It is considered that
investigation of the characteristic of loads for the
large container ships is one of significant subject. In
particular, in terms of container ships, about 350 m
in length have been constructed and delivered, but
technical subjects regarding the characteristic of
loads act on such larger ships are remained.
With regard to large container ships, as a main
characteristic, the hull girder vibration is men-
tioned. To ensure the structural safety, the effects
of cyclic stress caused by hull girder vibration on
fatigue strength needs to be clarified. The effects
have been studied (Wang, 2009; Drummen, 2006;
Gu, 2002). However, they have not corresponded
with the actual damage on real ships.
In this study, a series of tank test utilizing the
backbone type elastic model was carried out,
rough seas, and to recognize the characteristic of
wave loads in various operational parameters such
as ship speeds and wave directions.
Further, in order to evaluate the long term
wave loads, the data of the full scale measure-
ment (Okada, 2006) were analyzed and actual
fatigue damage was estimated. In addition, the
direct load and strength calculation using full ship
finite element model was performed for verifying
the fatigue strength. Moreover, the probability
of occurrence of whipping in long term was esti-
mated. As a result, the effect of hull girder vibra-
tion and environmental/operational conditions on
long-term fatigue damage was examined.
2 TANK TEST
2.1 Summery of the tank test
Tank test was carried out by means of 12,000TEU
container ship (Oka, 2009). The model is 3 m in
length, 1/117 scaled version of the real ship.
The test was conducted at the Ocean Engineering
tank (40 m × 27 m × 2 m) of the National Maritime
Research Institute of Japan, which has X-Y carriage
and wave generator. The towing device, which had
been developed with ability for 6 degrees of freedom
of motion, was mounted on X-Y carriage and was
connected at the gravity centre of the model ship.
The course of the model was controlled by PD
control of yaw motion using a torque motor.

317

MARSTRUCT.indb 317 2/18/2011 5:48:28 PM

 2.2 Similarity law of hull rigidity VBM on hammerring (Component of 2-node vibration)
Fn = 0.0
2.5
The main purpose of the tank test is an evaluation Response β=0.067
2
of the wave-induced dynamic response. In order to

Vertical bending moment [kgf-m]

a0* exp(- β t/ T)
evaluate the dynamic response adequately, rigidity 1.5

of elastic model was determined to be equivalent to 1

the real ship in accordance with a following equa- 0.5
tion, which is a function of Froude number. 0
0 1 2 3 4 5
-0.5
5 (1)
Em ES I S -1
-1.5
α : Scale ratio -2
Em: Young's modulus of elastic model -2.5
IBB: Momentum of inertia of elastic model t (s)
ES: Young's modulus of hull girder of real ship
IS: Momentum of inertia of real ship Figure 2. Example of time history and logarithmic
damping ratio (δ = tanβ) by the hammering test in still
water (Oka, et al., 2009).
2.3 Backbone type elastic model
In this study, a backbone type elastic model (Oka, VBM at SS5.5
2009) was applied. Backbone model is easy to kgf-m 4
2
adjust their rigidity by choosing the appropriate 0
-2
shape, size and materials. More notable merits -4
-6
are that no stress concentration and less creep. 0 1 2 3 4 5 6 7 8 9 10
sec
Figure 1 shows the outline of the present back-
bone model. The model is composed of segmented Figure 3. Example of time history of VBM in irregular
ship hulls, which are connected to be continuous wave by the tank test (Fn = 0.219, χ = 135 deg., Tz/Hs in
by means of uniform beam backbone. real scale = 11.0 s/5.7 m).
The hull girder loads at each segmented section
can be obtained by converting measured stress on
backbone. Figure 2 shows an example of the result Hs=5.7m, T=11.0s
5.0E-05
Fatigue damage (in 20min.)

of hammering test in still water. It is clarified that 4.5E-05 RAW

this model is consistent with the real ship in terms 4.0E-05 LPF
3.5E-05
of the natural frequency and structural damping 3.0E-05
HPF
of 2-node vibration. 2.5E-05
2.0E-05
1.5E-05
1.0E-05
2.4 Elastic vibration in irregular wave 5.0E-06
0.0E+00
Figure 3 shows an example of time history of Ver- Fn= Fn= Fn= Fn= Fn= Fn= Fn=
tical Bending Moment (VBM) measured by the 0.164 0.164 0.164 0.164 0.219 0.219 0.219
tank test in irregular wave. The backbone model χ= χ= χ= χ= χ= χ= χ=
180deg 150deg 135deg 45deg 180deg 150deg 135deg
has good ability to realize the whipping induced
elastic vibration in rough seas.
Figure 4. Short term damage based on the tank test in
irregular wave.

2.5 Fatigue analysis based on the tank test

Cumulative fatigue damages in short sea state
(short term damage) were estimated based on the
tank test. Figure 4 shows the short term damage in
terms of operational parameters.
Fatigue damage was estimated by using the
vertical bending stress on upper deck at S.S.5.5 in
real ship converted from Vertical Bending Moment
(VBM). In order to evaluate the different compo-
nents superposed on the response, the stresses with
Figure 1. Outline of the backbone model. three different frequencies have been prepared.

318

MARSTRUCT.indb 318 2/18/2011 5:48:28 PM


Normalized long-term fatigue damage
1.2
1.0
0.8
0.6
0.4
0.2
0.0
Figure 5. Long-term (= 25 years) fatigue damages based
on the tank test (Oka, 2009).
• RAW: 0.2 ∼ 40.0 Hz;
• LPF (Low Pass Filtered): 0.2 ∼ 5.0 Hz;
• HPF (High Pass Filtered): 5.0 ∼ 40.0 Hz
LPF means the wave component. HPF means
the component of elastic vibrations. Fatigue dam-
ages based on each component were evaluated sep-
arately. Rain flow method was applied for counting
of stress range. For the S-N curve, the D curve
(IACS Recommendation No. 56) of UK-HSE for
the butt weld joint was applied. The ratio of the
damage estimated by RAW component to that by
LPF component was calculated to check the effect
The wave height and vertical bending stress
were measured on a post-panamax container-ship
LPF(without vibration)
during from 1999 to 2002 (approximately 2 and
RAW(with vibration)
half years). The ship navigated between Japan and
Europe through the Suez Canal. The stress was
measured on upper deck at midship. Data were
recorded for 20 minuts in every 2 hours, and 10 Hz
sampling rate. Figure 6 shows an example of time
history including the typical whipping response.
Fn=0.164 Fn=0.219
3.2 Environmental condition
In order to verify the effect of the environmen-
tal condition on the fatigue strength, the meas-
ured wave height was analyzed and compared
with that of ordinary scatter diagrams (Global
wave statistics; Hogben, 1986). Figure 7 shows
the probability of exceedance of encountered
wave height. These distributions are quite differ-
ent between measured wave and scatter diagrams.
Especially, a significant difference is confirmed
in the low probability of exceedance nearby 10−2,
and the stress occurring in such frequency level is
100 10
vertical bending stress
80 wave height 8
60 6

Wave height (m)

of hull girder vibration on fatigue damage.
Stress (N/mm2)

40 4
It is found that the short term damage includ- 20 2
ing the effect of hull girder vibration is sensitive to 0 0
the ship speed and wave direction (χ). The speed -20 280 290 300 310 320 330-2
reduction and/or course change generate lower -40 -4
damage. For the rational evaluation of the fatigue -60 -6
strength, it is necessary to evaluate wave laods tak- -80 -8
ing the real operation in rough seas, speed change -100 -10
Time (sec.}
and course change, into account.
Figure 5 shows the long-term fatigue damages Figure 6. Example of time history on full scale
based on the tank test. The IACS Recommenda- measurement.
tion No.34 was used for the wave scatter diagram.
Short term sea state excluded in the tank test, was
complemented by means of the response ampli- Long term distribution of wave height
tude operator measured in the regular wave test. 30
Full scale measurement
Under the assumption that the subjected ship is
JPN-EU (GWS)
operated at its service speed during 25 years, which 25
North Atrantic (GWS)
corresponds to a whole life, long-term fatigue
damage including the hull girder vibration (RAW) 20
Hw (m)

is double of that excluding the vibration (LPF). In

15
addition, consideration of the speed reduction set
to 3/4 of service speed, that rate reduces to 1.5. 10

5
3 FULL SCALE MEASUREMENT
0
0 1 2 3 4 5 6 7 8
3.1 Subjected ship and measured data
-LOG(Q)
In order to evaluate the characteristic of the long-
term fatigue damage, the data of full scale meas- Figure 7. Probability of exceedance of encountered
urement (Okada, 2006) were analyzed. wave height.

319

MARSTRUCT.indb 319 2/18/2011 5:48:29 PM

 Table 1. Parameter of distribution and probability of exceedances of encountered wave height.

Weibull parameter Wave height (m)

Shape Scale Q* = 10−2 Q = 10−4 Q = 10−8

Meas.** 0.64 0.47 4.0 10.9 30.9

JPN-EU*** 0.93 1.52 7.4 13.7 26.0
N.A.*** 1.14 2.51 9.4 16.4 29.2

*Probability of exceedance.
**Full scale measurement.
***Global wave statistics (Hogben, 1986).

Table 2. Long-term fatigue damage based on encoun- 7.00E-04

tered wave height.

Cumulative fatigue damage

6.00E-04
RAW
Normalized long-term 5.00E-04 LPF
fatigue damage
4.00E-04 HPF

Full scale measurement 1 3.00E-04

Japan-Europe (GWS) 4.6
North Atlantic (GWS) 17.9 2.00E-04

1.00E-04

dominant to long-term fatigue damage. The dif-
ferences can be quantitatively confirmed from the
Weibull shape parameter shown in Table 1. Assum-
ing that vertical bending stress and wave height
have linear correlation, the long-term fatigue dam-
age was able to be simply estimated. Table 2 shows
the normalized long-term fatigue damages calcu-
lated based on the each wave distributions. It is
found that the long-term fatigue damage based on
the actual wave is much lower than that based on
the scatter diagram, and the effect of the long term
distribution of encountered wave height on fatigue
strength is significant.
3.3 Fatigue analysis based on the full scale
measurement
Cumulative long-term fatigue damage of deck
structure was estimated based on the full scale
measurement. As with the tank test, three kind
of frequency component of stress response have
been prepared taking the natural frequency of the
2-node hull girder vibration of vertical bending
into account.
• RAW: 0.01 ∼ 1.0 Hz;
• LPF: 0.01 ∼ 0.5 Hz;
• HPF: 0.5 ∼ 1.0 Hz
Figure 8 shows the cumulative trend of fatigue
damages. Extrapolation to 25 years is necessary
in order to estimate the long-term fatigue damage
because the stress had been recorded in 20 minuts
in every 2 hours and the period two and half years.
Table 3 shows the predicted long-term fatigue
0.00E+00
0 2000 4000 6000 8000
Number of short term measurement
Figure 8. Transition of the cumulative damages based
on the actual stress by the full scale measurement.
Table 3. Fatigue damage extrapolated to 25 years based
on the actual stress by the full scale measurement.
Long-term fatigue damage
DRAW 0.065
DLPF 0.030
DHPF 0.005
damages. Rain flow method and D curve of UK-HSE
are applied. The long-term fatigue damages based
on the actual stress is quite low, even taking the hull
girder vibration into account (DRAW).
4 FATIGUE DESIGN ASSESSMENT
4.1 Direct calculation
A direct calculation of spectral-based fatigue
analysis (ABS, 2007; DNV, 2008; LRS, 2004) was
performed in order to verify the actual stress and
fatigue strength. As should be noted, the direct cal-
culation which is based on liner theory does not
take the effect of hull girder vibration into account.
The summery of the direct calculation in this study
is shown in following.

320

MARSTRUCT.indb 320 2/18/2011 5:48:31 PM

 1. External water pressure computed by strip
method. And the pressure acts on the shell plating
of full ship FE model, taking non-liner pressure
on side hull nearby free surface into account.
2. Inertia force induced by the acceleration of the
containers and heavy equipments e.g. engine
acts on inner hull, utilizing a function of MSC/
NASTRAN, Inertia relief, by means of mod-
eling the containers as solid element.
3. Static FE analysis are executed on 12 phases per
one period of encountered regular wave.
4. The Response Amplitude Operators (RAO) of
local stresses are obtained by series calculation
through FEA. Stress amplitudes are obtained
by half of the difference between maximum and
minimum value during one period consisting of
12 phases.
5. Standard deviations (R) of local stresses in
short states are obtained by the wave spectra
and stress RAOs based on liner theory.
6. Long term distribution of local stresses is
obtained by R (Fn, χ, T) and wave scatter
diagram.
7. Long-term fatigue damage is calculated in
accordance with the Palmgren-Miner cumula-
tive damage rule.
In this study, the operational condition was set
Fn = 0.179, which is 3/4 of service speed, and all
headings on probability of wave directions. Two
parameters Pierson Moskowiz spectrum (IACS
Recommendation No. 34), wave scatter diagram
(Hogben, 1986) and D curve of UK-HSE were
applied.
Figure 9 shows the computed stress comparing
with the actual stress excluding the effect of hull
girder vibration. The actual stress was analyzed
based on the two kinds of peak counting methods;
Zero-cross and Rain-flow. There is not remarkable
difference between these analysis methods.
Deck stress , Single Amp.
180
Measured (LPF ; ZeroCross)
160 Measured (LPF ; Rain Flow)
140 Calc.(GWS/JPN-EU)
Calc.(Actual encountered wave)
120
Stress (MPa)
100
80
60
40
20
0 The effect of hull girder vibration on long-term
0 1 2 3 4 5
-logQ
Figure 9. Long-term distribution of computed stress
(without considering hull girder vibration) and actual
stress by the full scale measurement.
MARSTRUCT.indb 321
Table 4. Long-term fatigue damage by means of direct
calculation and full scale measurement.
Long-term fatigue
damage (25 years)
DSIM (Scatter diagram 0.36
JPN-EU route)
DSIM* (considering the 0.072
actual wave)
DLPF (full scale measurement) 0.030
The actual stress is lower than the simulated stress.
The computation gives much safer estimation.
Figure 9 also shows a result of simulation
applying the actual encountered wave instead of
scatter diagram with dotted line. This enables more
accurate prediction.
4.2 Verification by the actual fatigue damage
The simulated fatigue damage based on the direct
calculation (DSIM) was compared with the actual
fatigue damage. Table 4 shows the result. DSIM were
corrected by the rate of service time (= 0.85 ). DSIM
should compare with DLPF because spectral fatigue
analysis can not be considered the hull girder
vibrations. The large difference between DLPF and
DSIM are confirmed.
Table 4 also shows the long-term fatigue damage
considering the actual wave. The simulated fatigue
damage (DSIM*) is closer to the actual damage (DLPF)
owing to applying the actual wave to the direct cal-
culation. However, substantial amount of aliena-
tion between the direct calculation and full scale
measurement remains. It is considered there are
variety of causes, operation in rough seas, property
of wave sensor including the conversion system of
from relative wave height to global wave height,
and tolerance of spectral analysis including loads
and strength computation. In future, in order
to achieve more accurate evaluation of fatigue
strength, it is necessary to solve the level of such
effect more quantitatively.
5 EFFECT OF THE VIBRATION STRESS
ON HULL GIRDER FATIGUE DESIGN
ASSESSMENT
5.1 Long-term fatigue damage
6 7 8 fatigue damage (CEL) is simply defined in follow-
ing equation.
DRAW
CE
EL = (2)
DLPF
321
2/18/2011 5:48:32 PM
 The difference between DLPF and DRAW is caused Whipping
Whipping

Vertical bending stress [N/mm2]

50
by the synergistic effect between peak stresses (S) 40 HPF
and number of cyclic stress (N). The peak stress 30
20
Threshold of peak stress

and cyclic stress are amplified due to hull girder 10

0
vibration, whipping and springing. CEL, which -10
is evaluated based on the full scale measurement, -20
-30 Threshold of peak stress
is approximately two. This is clearly derived by -40
-50
figure 8 and Table 3. In addition, CEL based on the 0 200 400 600 800 1000 1200
sec.
tank test using the elastic model is approximately
two. This is also derived by figure 5. From these Figure 11. Methodology to distinguish whether or not
result, CEL in terms of the longitudinal strength each incident wave induce whipping (The threshold of
member is supposed to be same level. peak stress is set to 20 N/mm2 of single amplitude).

5.2 Probability of occurrence Sea states

of whipping during ship’s life 4000 3659

3000

Frequency
In order to evaluate the effect of elastic vibration
2000
on fatigue strength, the probability of occurrence 1253
770
1000 465
of whipping during long term is estimated by the 267 125
57 51 41 19 20 10 3 6 2 5 2 3 2 1 1 1
0
full scale measurement. In this study, in order to

10.5
0.5

1.5

2.5

3.5

4.5

5.5

6.5

7.5

8.5

9.5
10

11
1

9
determine whether or not the whipping occurs, the Hs(m)

peak value of high pass filtered stress was utilized. Whipping occurance
40
Maximum or minimum HPF stress during zero-
30
cross period of LPF stress is cyclically counted, and
Frequency

Stress_trh=20MPa
the long term distribution of the slamming induced 20
Stress_trh=30MPa

and higher component stress is obtained. Here 10

the zero-cross period of LPF stress are regarded 0

10.5
0.5

1.5

2.5

3.5

4.5

5.5

6.5

7.5

8.5

9.5
10

11
1

as encounter wave period. Figure 10 shows the
probability of exceedance of LPF stress and
HPF stress. The HPF stress become larger under
Q = 10−3. However, for example assuming whipping
over 20 N/mm2 of deck stress single amplitude,
the probability exceedance is about 5 * 10−5 which
equal about 2,000 times during life (N = 108). It
is assumed that the probability of occurrence of
whipping during a whole life is small.
5.3 Frequency of whipping in each sea state
When the peak HPF stress during the encounter
wave period exceeds the threshold value of stress,
1.E+00
1.E-01
Probability of exceedance
9
Hs(m)
Figure 12. Frequency distribution of the sea states
during period of the full scale measurement (upper) and
number of whipping occurrence per sea state (lower).
the incident wave is judged as induced whipping.
Figure 11 shows a case of the threshold is set to
20 N/mm2 in single amplitude. The histogram of
whipping occurrence per sea states is obtained.
Figure 12 shows an example in case of threshold
setting to 20 N/mm2 and 30 N/mm2 in single
amplitude. Any whipping was not occurred in
less than of equal to in 2 m significant wave
height. Moreover, the sea state lower than 2 m
in significant wave height accounts for 95% of
the total number of sea states, as known in upper

graph of figure 12. The sea state whipping occurs

1.E-02
LPF is regarded to be limited.
HPF
1.E-03 For the rational assessment of structural relia-
1.E-04 bility, it is important to predict accurately the level
of the vibration stress in such sea state (Ogawa,
1.E-05
2009).
1.E-06

1.E-07
0 10 20 30 40 50 60
Stress Amplitude (MPa)
6 CONCLUSIONS

Figure 10. Probability of exceedance of deck stress The analysis of fatigue strength based on tank test
amplitude induced by the wave loads (LPF) and by the and full scale measurement were performed. The
hull girder vibration (HPF). conclusions are as follows.

322

MARSTRUCT.indb 322 2/18/2011 5:48:32 PM

 1. Ship’s response including hull girder vibration
in rough seas was realized by tank test utilizing
the newly designed backbone typed elastic
model.
2. As a result of the tank test, the effect of
operational parameter such as ship speeds and
wave directions on fatigue strength was indicated
quantitatively. It was confirmed that the
speed reduction and/or course change from head
sea contribute to a decrease in fatigue damage.
3. Long-term fatigue damage based on the analysis
of the full scale measurement is quite small com-
paring with the direct calculation. That is mainly
caused by the difference of wave condition thanks
to the adequate operation to avoid severe sea.
4. It was indicated that the probability of exceedance
of the slamming induced whipping vibration is
statistically small. For the rational assessment of
structural reliability, it is important to estimate
accurately not only the level of higher compo-
nent stress but also the probability of occurrence
of the vibration during a whole life.
ACKNOWLEDGEMENTS
The data of full scale measurement were provided
from IHI Marine United Inc. The authors wish to
gratefully acknowledge their confidence.
A part of the present study was supported by a
Grant-in Aid for Scientific Research of the Japan
Society for Promotion of Science (No. 20360400).
REFERENCES
American Bureau Shipping 2007. Guidance notes on
spectral-based fatigue analysis for vessels.
Det Norske Veritas 2008. Classification notes No. 30.7
Fatigue Assessment of ship structures.
323
MARSTRUCT.indb 323
Drummen, I., et al 2006. Experimental and full scale
investigation of the importance of fatigue damage due
to wave-induced vibration stress in a container vessel.
Design & Operation of container ships, London, UK.
Gu, X. & Moan, T. 2002. Long-Term Fatigue Damage of
Ship Structure Under Nonlinear Wave Loads. Marine
Technology, Vol. 39, No. 2 April pp. 95–104.
Hogben, N., et al 1986. British Maritime Technology
“Grobal Wave Statistics”, Unwin Brothers Limited,
London.
IACS Recommendation No. 34 Standard wave data.
IACS Recommendation No. 56 Fatigue assessment of
ship structures.
Lloyd’s Register 2004. Ship Right, Fatigue design assess-
ment, Guidance on direct calculations.
Ogawa, Y., Oka, M. & Takagi, K. 2009. A prediction
method of wave loads in rough seas taking hydroelas-
tic vibration into account. Proc. 4th Int. Conf. Hydroe-
lasticity in Maritime Technology: 373–382.
Oka, M., et al 2009. An experimental study on wave loads
of a large container ship and its hydroelastic vibra-
tion. Proc. 4th Int. Conf. Hydroelasticity in Maritime
Technology: 183–192.
Oka, M., et al 2009. Hull Fatigue Evaluation utilizing
the elastic model test. Conference Proceedings of The
Japan Society of Naval Architects and Ocean Engi-
neers, vol.9E: 19–22 in Japanese.
Okada, T., et al 2006. On board measurement of stresses
and deflections of a Post-Panamax containership and
its feedback to rational design. Marine Structures, No.
19 Elsevier Ltd: 141–172.
Wang, S., et al 2009. Springing induced fatigue load on
container carriers. Proc. 4th Int. Conf. Hydroelasticity
in Maritime Technology: 111–120.
2/18/2011 5:48:33 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

The estimation of stress range distribution due to wide banded

random loading obtained by rain-flow counting method

J.B. Park, K.S. Kim & J. Choung

Inha University, Incheon, South Korea

ABSTRACT: Structures installed on offshore field continuously experience wide-banded random

loadings such as the combination of long and short period wave loadings, and the interaction between
mooring line/riser tensioning and hull structure motion. Those kinds of loadings usually cause multi-peak
response spectrum which consequently influence the accumulation of fatigue damage. Most combina-
tion models to estimate the fatigue damage on the complex loadings have been proposed in the form of
a correction factor that correlates the fatigue damage from narrow-band assumption with the reference
fatigue damage from rain-flow counting method. It is well-known that excellent prediction of the stress
range distribution can lead precise fatigue damage estimation. Therefore, this study will suggest a model
for new stress range distribution based on the rain-flow counting method and fatigue damage calculated
from the new model will be compared with ones from different existing models.

1 INTRODUCTION an introduction of a new model for approximating

rain-flow range distribution and an associated
The characteristics of loading pattern for offshore fatigue damage equation. Figure 1 shows typi-
structures, that are to be encountered various cal examples of narrow banded and wide banded
dynamic loadings from swell, wind-driven sea wave process.
and the interaction between mooring line/riser ten-
sioning and hull structure motion, and high-speed
sea-going vessels, that experience hydro-elastic
behavior by whipping/springing, have a tendency
of two- or multi-peak spectrum and wide band
random loading. These kinds of dynamic ran-
dom loading can be occurred fatigue damage and
a model to estimate exact fatigue damage for the
wide-banded random loading is necessary. These
kinds of model to predict fatigue damage for
wide-banded loading have been proposed by sev-
eral researchers and can be classified as two types.
The first is to approximate the rain-flow range
distribution directly such as Zhao-Baker model
and Dirlik (1985) model and the second is to
evaluate a correction factor that correlates the
fatigue damage from narrow-band assumption
with the reference fatigue damage from rain-flow
counting method. There are Wirsching and Light
(1980) model, Tove-Benasciutti (2005) method,
Jiao-Moan (1990) method, Sakai-Okamura (1995)
method and Fu-Cebon method that proposed
a typical model using damage correction factor.
However the latter has less flexibility of the spec-
trum combination because each spectrum peak is
assumed as a narrow-banded in some cases and it
adopted only one-slope S-N curve. This study aims Figure 1. Examples of time series processes.

325

MARSTRUCT.indb 325 2/18/2011 5:48:34 PM

 2 TIME DOMAIN EXTRACTION The criteria were up to 98% for the following
PROCEDURE OF AN ARBITRARY equations:
SPECTRUM
Zero Crossing Rate(Time domain)
In order to get rain-flow range distribution relia- Zr = (9)
Zero Crossing Rate(Spectru
tr m)
ble time-domain stress data extracted from an
arbi-trary spectrum are needed. Those data can Peak Rate(Time domain)
be obtained by using inverse discrete Fourier Pr = (10)
Peak Rate(Spectrum)
transformation:
N Irregular Factor(Time domain)
x(t ) = ∑ 2S ( F = (11)
k ) k ( kt + k ), (1) IFr
Irregular Factor(Spectrum
u )
k =1
Based on the criteria, it was decided that the
where, S(w ) is a spectral density function, N is the number of frequency division was 20,000 for model
number of frequency division and Wk is random development.
frequency defined as below. As seen Figure 2, peak values are more impor-
tant than response frequency in the fatigue point
2π
0 = ω 0 < ω1 <⋅⋅⋅ < ω N = (2) of view. Therefore smaller time step should be
2 Δt required.
From many parametric study of time step, it
Δω k ω k ω k −1 (3) was decided that 0.1 second for a spectrum with
ω k + ω k +1 maximum 5 rad/sec frequency was sufficient to get
ωk = (4) the peak values and Dirlik (1985) mentioned about
2 this matter that there would be roughly 5 points
θk is a random phase angle.
A procedure to be used to get suitable time-
domain stress data is similar to one proposed by
Dirlik, that is,
1. Generate time-domain stress data x(t) from
S(w) using equation (1).
2. Repeat step (1) 20 times with new random fre-
quency and phase angle in order to obtain a
sufficient long random data, and call this data
‘one block’.
3. Step (2) is repeated 10 times for a spectrum.
The averaged values of 10 blocks for the number
of zero up-crossings, the number of peaks and
the irregularity factor are compared with ones
obtained from spectral moment equations.

1 λ4
υ+ = (5)
2π λ2

1 λ2
υ0 = (6)
2π λ0

υ0 λ2
IF = = , (7)
υ+ λ0 λ 4
where, υ+ is peak period, υ0 is zero up-crossing
period, IF is irregularity factor and λm is spectral
moment defined by:
∞
λm ∫ω S (ω )dω
m
m = 1, 2,... (8) Figure 2. Random responses of time domain corre-
0 sponding with sampling time step.

326

MARSTRUCT.indb 326 2/18/2011 5:48:34 PM

 Figure 3. Rain-flow counting and stress-strain hyster-
esis loops in the case when the load represents a repeating
time history.

between successive peaks and troughs yielding a

‘smooth’ stress-time history. Regarding the extrac-
tion time, 3 hours were taken because the short-
term investigation time of ocean field is 3 hours.

3 RAIN-FLOW COUNTING METHOD

In order to obtain stress range distribution from

time-domain data, reasonable range counting
method should be required. The most preferred
method is the rain-flow counting method that
has initially been pro-posed by M. Matsuiski and
T. Endo (1968) to count the cycles or the half
cycles of strain-time signals. Counting is carried
out on the basis of the stress-strain behavior of the
material.
Practical definition of the rain-flow cycle count-
ing can be explained according to the ASTM
E–1049 Standard Practices for Cycle Counting in
Fatigue Analysis.
In this study rain-flow counting was carried
on the 10 blocks defined in the (2) step of the
procedure.
Figure 4. Spectral densities used for simulating
Gaussian random processes.
Spectral densities S(w) having constant, linear,
double symmetric or anti-symmetric parabolic
shape were performed and named from type 1
to type 5 per each spectrum. All spectra have the
same variance, i.e. α0 and the range of α1 and α 2
of numerical simulations were from 0.1 to 0.8 and
from 0.2 to 0.9 respectively.
The αm is a bandwidth parameter and defined
as below:

λm
αm = (12)
4 NUMERICAL SIMULATIONS λ0 λ2 m

Spectrum examples performed in this study were Resultant time history and rain-flow range
used from Tove-Benasciutti (2005) like Figure 4. distribution are shown in Figure 5.

327

MARSTRUCT.indb 327 2/18/2011 5:48:39 PM

 Figure 6. Half Gaussian and Rayleigh probability
density functions.

6 THE MOMENTS OF RAIN-FLOW

RANGES (MRR)

The moments of rain-flow range densities are

calculated with standardized variable z = x/2σx,
where σx is standard deviation and λ0 σ x2 , and
are normalized with the moments of the Rayleigh
PDF as below equation.
∞

∫z
m
pRR ( z )ddz
M RR ( m ) = 0
(13)
∞ z2
−
∫z
m 2
ze ( z )dz
0

As the exact value of MRR(m) can be taken from

Figure 5. Spectral densities used for simulating
Gaussian random processes.
a rain-flow range probability distribution, this can
be approximated with a combination of spectral
moments calculated from the applied spectrum
and utilized to obtain coefficients of rain-flow
range distribution model to be developed.
From parametric study over 260 spectra with
same variance, approximated MRR(m) per each
negative S-N curve slope parameter was obtained.

5 PROBABILITY DENSITY FUNCTIONS M RR ( ) ≈ α 2 (14)

As seen Figure 5, there is a peak in way of zero- M RR ( ) ≈ α 0.95

95α1.95 (15)
stress range. In order to approximate this peak dis-
tribution a probability density functions having a M RR ( ) ≈ α 0.5α 0.95α1.95 (16)
finite value at zero-stress range is required. In this
study three probability density functions were M RR ( ) ≈ α 0.26
26α 0.58α 0.89
89α1.96 (17)
adopted, that is, two Rayleigh PDF (Probability
Density Function) and half Gaussian PDF which As shown in Figure 7, scatter was increased for
has zero mean and doubled area in the positive large m and lower values of MRR mean that the
to get unit area like Figure 6, so that it can give associated spectra are wide-banded and on the
non-zero probability density function around other hand higher MRR values are almost same as
zero-stress range. narrow banded spectra.

328

MARSTRUCT.indb 328 2/18/2011 5:48:41 PM

 ⎡ C1 m +1
⎛ m + 1⎞ ⎤
⎢ 2π σ ( x1 ) Γ ⎝ 2 ⎠ ⎥
2

⎢ x1
⎥
⎢ m⎞ ⎥
m
m
⎛ m⎞ ⎛1 +
⎢⎣ + C2 ( x 2 ) 2 ⎝ 1 + 2 ⎠ + C3 (
2
2 ⎠ ⎥⎦
)
JB ⎝
M RR ( m ) = m

( )
2 ⎛1 + m⎞
⎝ 2⎠

Γ
⎛ m + 1⎞
C1 ⎝ 2 ⎠
= ( σ x1 ) m
+ C2 ( )m + C 3 (19)
⎛ m⎞
x2
π
Γ 1+
⎝ 2⎠
4 equations to be used were made as below.
C1 C2 + C3 = 1 (20)
V1C1σ x1 + C x2 C3 = α 2 for m = 1 (21)

V2C1σ x12 + C x2
2
C3 = A fo m = 2 (22)

V3C1σ x13 + C x2
3
C3 = B fo m = 3 , (23)
where,
1 (1)
V1 =
π Γ (1.5)
1 (1.5)
V2 =
π Γ (2)
1 (2)
V3 =
π Γ (2.5)

Figure 7. Relationship between MRR(m) and multiplica- A = α 0.95

95α1.97
tion of spectral moments.
B = α 0.54
54α 0.93α1.95

7 DEVELOPMENT OF RAIN-FLOW It was assumed V2C1σ x12 ≈ 0 and V3C1σ x13 ≈ 0

RANGE DISTRIBUTION MODEL because half-Gaussian distribution is dominant in
way of small stress range region, which means it
The model is composed of one half-Gaussian prob- has a little fatigue damage for higher negative S-N
ability distribution with σx1 variance, two Rayleigh curve slope.
probability distributions with σx2 variance and unit From parametric study among several spectra it
variance as below equation. was founded that σx1 can be assumed as σ2 to gain
the most reasonable rain-flow range distribution
⎛ z2 ⎞
⎜− ⎟
and also means that second Rayleigh distribution
2 2σ 2
JB
pRFC C1 e ⎝ x1 ⎠ is determined by the irregularity factor.
2π σ x1 Therefore all unknown variable can be calcu-
⎛ z2 ⎞ ⎛ z2 ⎞ late. In case that A > B, α 2 < B/A and α2 > A +
z ⎜− 2⎟ ⎜− ⎟
⎝ 2σ ⎠ ⎝ 2⎠
+ C2 e C3 ze , (18) A2 4(A B ) /22 to avoid negative C2,C3, and σx1,
σ x 22
where C1,C2,C3 are weighting factors for each C1 1 C2 − C3 (24)
probability distribution.
As there are 5 unknown variables to be A B
C2 = (25)
obtained, i.e., C1,C2,C3,σx1,σx2, 5 relevant equations α 2 (1 − α 2 )
2
are required to make complete rain-flow range
distribution model. To do this, 3 equations were
−AAα 2 B
made from MRR(m), m = 1,2,3 and analytical equa- C3 = (26)
tion of MRR(m) for pJBRFC can be calculated as below.
(1 − α 2 )

329

MARSTRUCT.indb 329 2/18/2011 5:48:44 PM

 1 where, c, c ′, m, m′ is fatigue strength and negative
σ x1 (α 2 − C2α 2 C3 ) (27) slope for first and second slope of S-N curve
V1C1
respectively, Vp is associated peak period and Td
Otherwise, is design life. Γ(;) is complementary Incomplete
Gamma function using the first slope of S-N
C1 1 C3 (28) curve and γ (;) is incomplete Gamma function
using the second slope of S-N curve.
C2 = 0 (29)
8 THE FITTING RESULTS OF RAIN-FLOW
C3 B (30) RANGE DISTRIBUTION

1 In this chapter the rain-flow range distribution for

σ x1 (α 2 − B ) (31)
V1C1 the developed model is shown in some cases and
compared with ones from Dirlik model.
This case is occurred when a spectrum have From the Figures 8 and 9, it was be verified
extremely wide banded. that the accumulated distance for each simulated
The damage equation of the developed model spectrum between Rain-flow range distribution
for one-slope S-N curve is and the developed model has around 64% closer
than one between rain-flow range distribution
⎛ C1 ⎛ m + 1⎞ ⎞ and Dirlik model and it means that the developed
σ mΓ
⎜ π x1 ⎝ 2 ⎠ ⎟ model has reasonable probability distribution for
Tdν p m⎜ m ⎟ the rain-flow range distribution.
JB
DRFC ( m ) =
c ( ) ⎜ +C2σ x 2 m Γ ⎛ 1 + ⎞ ⎟
⎜ ⎝ 2⎠⎟ (32)
⎜ ⎛ m⎞ ⎟ 9 THE COMPARISON OF FATIGUE
⎜⎝ +C3Γ ⎝ 1 + ⎠ ⎟⎠
2 DAMAGE RESULTS

In case of two-slope S-N curve, the associated The damage ratio between the spectra in Figure 4
damage equation is as follows: using the developed model and damage from

⎡
⎢ ⎜ (
⎛ 2 2λ m
0
σ
) ⎛
m ⎜ m +1
Γ
⎛ S0 ⎞ ⎞
2

⎟
⎞ ⎤
⎟ ⎥
⎢ ⎜ c x1 ; ⎜ ⎟ ⎟ ⎥
⎜
⎝
2 ⎝ 2 2 λ σ ⎠ ⎟
⎠
⎢ C1 ⎜ 0 x1
⎟ ⎥
⎢ π⎜ 2⎞⎟ ⎥
( )
m′
⎛
⎢ ⎜ + 2 2 λ0 σ x1m ′γ ⎜
m′ + 1 ⎛ S0 ⎞ ⎟ ⎥
⎢ ⎜ ;⎜ ⎟ ⎟⎟ ⎥
⎢ c′ ⎜⎝ 2 ⎝ 2 2 λ0 σ 1 ⎠ ⎟⎠ ⎠ ⎥
⎝
⎢ ⎥
⎢
⎢
⎛ 2 2λ m
⎜ (0 ) ⎛
σ x 2 Γ ⎜1 + ; ⎜
m ⎛ S ⎞ ⎞
2 ⎞⎥
⎟⎥
⎟ ⎟
m 0
⎢ ⎜ c ⎜⎝ 2 ⎝ 2 2 λ0 σ x 2 ⎠ ⎟ ⎟⎥
⎠
DRFC ( m ) = Tdν p ⎢ + C2 ⎜
JB ⎟⎥ (33)
⎢ ⎜ m′ ⎛ ⎛ ⎞
2 ⎞⎟ ⎥
⎜+ ( 2 2 λ ) m ′ S ⎟
⎢ 0
σ x 2 m ′γ ⎜1 + ; ⎜ 0
⎟ ⎟ ⎥
⎢ ⎜ c′ ⎜⎝ 2 ⎝ 2 2 λ0 σ 2 ⎠ ⎟ ⎟ ⎥
⎢ ⎝ ⎠⎠ ⎥
⎢
⎢
⎛ 2 2λ m ⎛
⎜ (0
Γ ⎜ +
) m ⎛ S0 ⎞
2⎞

⎟
⎞
⎟
⎥
⎥
⎢ ⎜ c 1 ; ⎜ ⎟ ⎥
⎜ 2 ⎝ 2 2 λ0 ⎠ ⎟ ⎟
⎢ ⎜ ⎝ ⎠ ⎟ ⎥
⎢ + C ⎥
( )
3⎜ m′ 2⎞⎟
⎛
⎢ ⎜ + 2 2 λ0 γ ⎜1 + ; ⎜
m′ ⎛ S0 ⎞ ⎟ ⎥
⎢ ⎜ ⎟ ⎟⎟ ⎥
⎢ c′ ⎜⎝ 2 ⎝ 2 2 λ0 ⎠ ⎟ ⎥
⎣ ⎝ ⎠ ⎠ ⎦

330

MARSTRUCT.indb 330 2/18/2011 5:48:50 PM

 Figure 8. Rain-flow range distribution of type 1,
α1 = 0.4 and α2 = 0.3 spectrum for Dirlik and JB model
respectively.

Figure 10. Fatigue damage ratio for JB model and

Dirlik model according to Vanmarke parameter.

Figure 9. Rain-flow range distribution of type 5,

α1 = 0.9 and α 2 = 0.8 spectrum for Dirlik and JB model
respectively.

rain-flow distribution were compared with ones

with Dirlik model as seen Figure 10.
Fatigue damage comparison was carried out
with other existing models using ideal bi-modal
spectrum that was taken from Tove-Benasciutti like Figure 11. Ideal bi-modal spectrum used for fatigue
Figure 11 and results are shown in Figures 12–16. analyses.

331

MARSTRUCT.indb 331 2/18/2011 5:48:53 PM

 Figure 12. Fatigue damage ratio for Wirshing model Figure 14. Fatigue damage ratio for Tovo-Benasciutti
according to Vanmarcke parameter. model according to Vanmarcke parameter.

Figure 13. Fatigue damage ratio for Dirlik model Figure 15. Fatigue damage ratio for Jiao-Moan model
according to Vanmarcke parameter. according to Vanmarcke parameter.

332

MARSTRUCT.indb 332 2/18/2011 5:48:56 PM


Figure 16. Fatigue damage ratio for JB model accord-
ing to Vanmarcke parameter.
10 CONCLUSIONS
Fatigue damage estimation of offshore structure
and sea-going vessel experiencing wide-banded
random loading is very important.
The purpose of study is to propose new model
predicting rain-flow range distribution and relevant
fatigue damage equation applicable to two-slop
S-N curve as well as one-slope one for the practical
application.
This model was compared with other existing
models in terms of fatigue damage and rain-flow
range distribution and it can be easily verified that
the model gives more reasonable results than other
models.
333
MARSTRUCT.indb 333
It is necessary to study that this model can
also give reasonable results for multi-modal spec-
trum cases.
REFERENCES
Benasciutti D & Tovo R, 2005, “Spectral methods for
life times prediction under wide band stationary
random processes,” Int J of Fatigue, Vol. 27, No. 8,
pp. 867–877.
Denis Benasciutti, 2004, “Fatigue Analysis of Random
Loadings,” PhD Thesis, University of Ferrara.
Dirlik T, 1985, “Application of computers in fatigue,”
PhD Thesis, University of Warwick.
Jiao G & Moan T, 1990, “Probabilistic analysis of fatigue
due to Gaussian load processes,” Probabilistic analy-
sis of fatigue due to Gaussian load processes, Vol. 5,
No. 2, pp. 76–83.
Matsuishi M & Endo T, 1968, “Fatigue of metals sub-
jected to varying stress,” Japan Society of Mechanical
Engineers, Fukuoka, Japan.
Miner MA, 1945, “Cumulative damage in fatigue,”
Journal of Applied Mechanics, Vol. 12, pp. 159–64.
Rychlik I, 1987, “A new definition of the rainflow cycle
counting method,” Int J of Fatigue, Vol. 9, No. 2,
pp. 119–121.
Rychlik I, 1993, “On the narrow-band approximation
for expected fatigue damage,” Prob Eng Mech, Vol. 8,
pp. 1–4.
Sakai S & Okamura H, 1995, “On the distribution of
rainflow range for Gaussian random processes with
bimodal PSD,” JSME Int J Ser A, Vol. 38, No. 4,
pp. 440–445.
Samuel D Stearns & Ruth A David, “Signal Process-
ing Algorithms using Fortran and C,” Prentice-Hall
International, Inc.
SECIL ARIDURU, 2004, “Fatigue Life Calculation by
Rainflow Cycle Counting Method,” Master Degree
Thesis, Middle East Technical University.
Wirsching PH & Light MC, 1980, “Fatigue under wide
band random stresses,” Proceeding of the ASCE,
Journal of the Structural Division, Vol. 106, No. ST7,
pp. 1593–1607.
2/18/2011 5:49:00 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Three-dimensional fracture mechanics analyses of surface cracks

at welded joints in ship structure

S. Tanaka
Graduate School of Engineering, Hiroshima University, Japan

H. Okada
Faculty of Science and Technology, Tokyo University of Science, Japan

S. Okazawa
Graduate School of Engineering, Hiroshima University, Japan

ABSTRACT: The purpose of the present study is to solve surface crack in welded joint of ship structure.
To know the mechanical properties and the crack growth predictions of the surface crack are very
important from the point of view of fatigue strength evaluation. In this study, shell-solid mixed analysis
using Rigid Body Element 3 (RBE3) in MSC.Nastran and Virtual Crack Closure Method (VCCM) for
quadratic tetrahedral finite elements adopted to evaluate Stress Intensity Factors (SIFs) in the welded
joint of a ship. As a numerical example, the calculation of SIFs to the surface crack in welded joint of
three cargo hold carrier are demonstrated. The accuracy and effectiveness of the proposed method are
discussed.

1 INTRODUCTION modeled by triangular or quadratic shaped shell

In this study, an effective and accurate mechanical
evaluation approach to the surface cracks in
the structural members of a ship is presented.
Finite Element Analyses (FEA) are often used
in structural analysis of ships. The shell finite
elements are used to model the structures.
The shell element is so-called structural ele-
ment and the plate thickness is assumed in the
interpolation function. However, it is difficult to
take into account the three-dimensional surface
cracks in the shell element modeling. It is known
that solid elements are needed to solve the three-
dimensional crack problems with high accuracy.
Many degrees-of-freedoms are needed to analyze
the cracks in very large and very thin ship structural
members. This study presents a highly accurate and
efficient approach to calculate the Stress Intensity
Factors (SIFs) of the three-dimensional surface
crack in ship structural members. The Rigid Body
Element 3 (RBE3) in MSC.Nastran is adopted to
perform the shell-solid mixed analysis. The RBE3
is one of the so-called rigid elements. To effectively
employ the shell-solid mixed analysis to a three-
dimensional local structure such as welded joint,
the semi-auto shell-solid connecting technique
using the RBE3 was developed. In the shell-solid
mixed analysis, the whole structure of the ship was
elements and the local structure with welding and
the surface crack was modeled by the tetrahedral
solid elements. The geometry of the local structure
was created using commercial 3D-CAD software
and automatic tetrahedral mesh generation was
carried out. The use of the 3D-CAD and the
automatic mesh generation software enhanced the
efficiency of the FEA for the three-dimensional
complex shaped structure. The shell model and the
solid model were joined with the semi-auto RBE3
connecting technique to perform the shell-solid
mixed analysis with high accuracy and efficiency.
In addition, the Virtual Crack Closure-integral
Method (VCCM) for the tetrahedral finite elements
was adopted to calculate the SIFs (Okada et al.
2008). In the VCCM, the SIFs calculated only
the nodal forces and the opening displacements
of the FEA solutions. The use of the tetrahedral
finite elements made it possible to simulate three-
dimensional crack propagation analysis because
automatic mesh generation software was available.
The automated crack propagation analysis system
has already been proposed using tetrahedral finite
elements and the VCCM (Tokuda et al. 2010).
Although commercial FEA software such as
Abaqus and MSC.Marc can also calculate the SIFs
for the three-dimensional surface crack, it is diffi-
cult to model the three-dimensional surface crack in

335

MARSTRUCT.indb 335 2/18/2011 5:49:00 PM

 the complex shaped structure because hexahedral
elements are needed in the crack modeling and the
SIF calculation. For example, the local structure
model without the crack, and the detailed crack
model were made independently, and were
connected with the GLUE contact option in the
three-dimensional fracture mechanics analysis of
MSC.Marc. They were virtually a manual opera-
tion and were also complicated tasks. For the rea-
sons mentioned above, the proposed method has
the potential to solve the three-dimensional surface
crack with high efficiency and high accuracy.
In this paper, the detailed description of the
proposed approach and the three-dimensional
fracture mechanics analysis of a surface crack in
the structural members of a ship are presented
as numerical examples. The effectiveness of the
proposed method is discussed.
2 SHELL-SOLID MIXED MODELING
USING THE RBE3 CONNECTION
This section presents shell-solid mixed modeling
using RBE3. RBE3 is one of the so-called rigid
elements in MSC.Nastran and constructs the
constraint relationship between the reference nodes
and the slave nodes. The moment and force of the
reference nodes are distributed to the slave nodes.
As a detailed description of the RBE3 is given
in MSC.Nastran’s instruction manual, we only
present the application of the RBE3 connection
to the shell-solid mixed analysis for a thin plate
structure such as a ship’s structural members.
The schematic illustration for the shell-solid
mixed analysis using the RBE3 is shown in
Figure 1. In the shell-solid mixed analysis, the
linear triangle shell elements (CTRIA3) or the
linear quadrilateral shell element (CQUAD4)
are used in the shell model and the quadratic tetra-
hedral finite elements (CTETRA) are used for solid
modeling in MSC.Nastran. The plate thickness of
the shell model assumes the same thickness as that
of the solid model. In the shell-solid mixed analysis
using RBE3, the shell element nodes correspond
Figure 1. Schematic illustrations for shell-solid mixed
analysis using the RBE3 connection.
336
MARSTRUCT.indb 336
Figure 2. Semi-Auto RBE3 connection for shell-solid
mixed analysis.
to the reference nodes and the solid element nodes
correspond to the slave nodes. In the RBE3 con-
nection, the moment or the force of the shell ele-
ments are distributed to the solid nodes in the
shell-solid mixed analysis. The slave nodes within
a radius R from a master node are constructed by
the RBE3 with a linking relationship as shown in
Figure 1. The operation is adopted by all the shell
nodes located between the shell model and the
solid model. In addition, it is possible that the slave
nodes can construct a linking relationship with
some of the master nodes.
It is a very hard task to make the linking
relationship manually for all the shell nodes in
three-dimensional shell-solid mixed modeling
such as welded joints when the commercial FEA
pre-processor is adopted. In this study, we origi-
nally developed the semi-auto RBE3 connecting
technique. In this technique, a linking relationship
is made automatically, only selecting the face of
the solid model and the edge of the shell model.
The sample of the shell-solid mixed model and the
semi-auto RBE3 connecting technique is shown in
Figure 2. The three-dimensional shell-solid mixed
model can be generated very much faster than
using the conventional technique.
3 CALCULATION OF STRESS INTENSITY
FACTORS USING THE VCCM
EMPLOYING TETRAHEDRAL
FINITE ELEMENTS
This section presents the calculation of the SIFs
using the VCCM. The illustration for the two-
dimensional VCCM is shown in Figure 3a. The
SIFs are calculated based on Irwin’s crack closure-
integral (Irwin G. R., 1958) and the equation can
be represented as,
Δ
G Total = li ∫
Δ →0 0
σ3i (r ) v i ( Δ r )dr (1)
2/18/2011 5:49:01 PM

Figure 3. Schematic illustrations for the 2D VCCM
calculation [a) Irwin’s crack closure-integral, b) FEM
discretization of VCCM calculation].
where vi(Δ − r) are the crack opening displacements
for the ith (i = 1, 2, 3) direction and σ3i(r) are the
cohesive stresses to close the crack opening dis-
placement vi(Δ − r). Rybicki et al. (1977) proposed
the crack closure integral using FEA. For the eight-
node finite element analysis shown in Figure 3b,
eq. (1) can be discretized as,
2 2
G Total li
Δ→ 0
∑ vI PI li
Δ →0
∑ v1I P1I (2)
I =1 I =1
where v1I, v3I (I = 1, 2), and P1I, P3I (I = 1, 2) are the
crack opening displacements and nodal forces for
the x1 and x3 directions. The first term on the left
in eq. (2) corresponds to mode I energy release
rate GI and the second term on the right in eq. (2)
corresponds to mode II energy release rate GII. Then,
in the VCCM, it is easy to separate the energy release
rate into several modes of the SIFs, i.e. KI and KII.
Okada et al. (2008) propose the VCCM calcula-
tion technique for three-dimensional cracks using
quadratic tetrahedral finite elements. The quadratic
tetrahedral finite element modeling of the complex
shaped crack is relatively easy because automatic
mesh generation software is available. The detailed
descriptions for the VCCM for tetrahedral finite
elements are presented in Okada et al. (2008).
Finally, mode I, II, and III stress intensity factors
KI, KII and KIII are obtained as,
K I = E ′G IVCCM (3)
337
MARSTRUCT.indb 337
G IIVCCM
K II = E ′ G IIVCCM (4)
G IIVCCM
G IIIVCCM
K III = 2μ G III
VCCM
. (5)
G IIIVCCM
where μ is shear modulus. In eq. (3), (4) and (5),
GIVCCM, GIIVCCM and GIIIVCCM are energy release rates
for the mode I, II, and III. In this paper, we use the
technique to calculate the SIFs of the surface crack
in a welded joint structure.
4 APPLICATION TO A SHIP’S
STRUCTURAL MEMBERS
A practical application of the proposed technique
to a three-dimensional surface crack is presented.
The surface crack is located in a cruciform welded
joint in a cargo carrier. In this section, the shell-
solid mixed analysis using the semi-auto RBE3
connecting technique is performed on the ship’s
structure. The SIFs are calculated using the VCCM
for the quadratic tetrahedral finite elements. The
solutions are compared with the SIFs calculated
using the J-integral calculations of MSC.Marc.
Finally, the mechanical evaluations of the surface
crack are performed based on the SIFs.
A three hold shell element model of the
cargo carrier and the cross-section are shown in
Figures 4a and b. The average size of the shell
element in the whole structure is about 800 mm.
The surface crack is assumed at the weld toe of the
cruciform welded joint in the mid-hold. Figure 4c
shows a detailed representation of the cruciform
welded joint. A close-up illustration of the shell-
solid modeling region is shown in Figure 4d. The
shell-solid mixed modeling for the three-hold
model is made using the following procedures. The
re-meshing is performed to the cruciform welded
joint using small fine shell elements as shown in
Figure 4c. The mesh sizes are from 5 mm to 20 mm.
The solid model with the surface crack is generated
using the 3D-CAD and the automatic mesh gen-
eration is applied using quadratic tetrahedral finite
elements. The shell elements which are located in
the solid modeling region are removed and the solid
model is replaced. The solid model and the three
hold shell model are connected using the semi-auto
RBE3 connecting technique. Though manually
complicated tasks are needed in conventional tech-
niques to make the shell-solid mixed models, the
proposed method can make them in half an hour if
the three hold model is already arranged.
The cross-section of the cruciform welded
joint is shown in Figure 5a. The welded joint is
2/18/2011 5:49:02 PM

Figure 4. Cargo carrier example [a) Whole structure,
b) Cross-section of the analysis model, c) Local structure,
d) Shell-Solid mixed model].
Figure 5. Schematic illustrations for two-dimensional
VCCM analysis [a) Cross-section of welded joint (full
penetrated weld), b) Local coordinate of crack].
partially penetrated weld. Here, x1, x2 and x3 are
global coordinates. The crack surface and the local
coordinates x1’, x2’ and x3’ are shown in Figure 5b.
The crack has a semi-elliptical shape and the crack
sizes are the same as those used in the previous
section. The width is 2c = 10 mm and the depth is
a = 2 mm (The aspect ratio a/c = 0.4). Two types
of solid model were made. In Figure 5a, one sur-
face crack is located 3 mm from the weld toe
(Model A) and the other surface crack is located
at the weld toe (Model B). Both solid model sizes
are 400 × 400 × 400 mm. Total number of nodes
and elements for the Model A and the Model B are
shown in Table 1. Young’s modulus is 206,010 MPa
338
MARSTRUCT.indb 338
Table 1. Total number of nodes and elements.
Nodes Elements
Model A 1,770,837 1,213,832
Model B 2,038,356 1,385,521
and Poisson’s ratio is 0.3. The two loading condi-
tions are assumed.
Case 1 (HC): Head sea, Heavy Ballast Condition
and Hogging Condition (Wave Crest)
Case 2 (HT): Head sea, Heavy Ballast Condition
and Sagging Condition (Wave Trough)
The deformations of the ship under loading
conditions HC and HT are shown in Figures 6a
and b. Linear elastic analyses are performed on the
shell-solid mixed analysis models and the SIFs are
calculated using the VCCM as a post process.
To verify the proposed method, the SIFs obtained
by the VCCM calculations were compared with the
reference solutions. The reference solutions were
evaluated using the PSCM (Perpendicular Shell
Coupling Method) and the displacement-based
zooming method. The PSCM is a high accuracy
shell-solid mixed analysis technique proposed by
Osawa (Osawa et al. 2007, 2008). In this technique,
the perpendicular shell is placed on the original
shell plane and the perpendicular shell is connected
with hexahedral solid elements. The PSCM model
using linear hexahedral elements for the cruciform
welded joint without a surface crack is shown in
Figure 7a. Very small elements of about 0.5 mm
were used near the welding region. Thirty divi-
sions were performed using the hexahedrons along
the plate thickness direction. The solid model size
was 400 × 400 × 400 mm. Total number of nodes
was 1,264,487 and of elements are 1,256,970. The
triangle or the quadrilateral shell elements were
used to model the external region of the PSCM
model. The shell region and the solid region were
connected with the PSCM. The detailed surface
crack model is shown in Figure 7b. The crack
model was independently generated. Linear hexa-
hedral finite elements were also used to make the
detailed crack model. There were 230,481 nodes
and 218,440 elements. In the PSCM model, the
solid elements where the detailed crack model
would be located were removed, and the detailed
model was replaced. In addition, the PSCM model
and the detailed crack model were connected
with the GLUE contact option of MSC.Marc.
The surface region of the PSCM model is shown
in Figure 7c. J-integral calculation was used to
evaluate the SIFs. The comparisons of the SIFs
were only performed for Model A.
The calculation procedures were as follows. Linear
elastic analysis was performed using MSC.Nastran
2/18/2011 5:49:03 PM
 Figure 6. Loading Conditions [a) Case 1: Heavy ballast
crest (HC), b) Case 2: Heavy ballast trough (HT)].
Figure 7. Shell-Solid Zooming Analysis using PSCM
method [a) Whole structure, b) Detailed structure,
c) Crack model].
on the three hold shell model as shown in Figure 4a.
The nodal displacements were extracted and the
displacements were added to the PSCM model as
the displacement boundary condition. The zoom-
ing analysis was performed and the SIFs were cal-
culated using the J-integral from MSC.Marc.
The SIFs KI, KII and KIII for the VCCM calcu-
lations are shown in Figures 8a, b and c. In the
SIFs calculations, the plane strain condition was
assumed over the entire crack front. The SIFs
obtained by the J-integral of MSC.Marc for the
analyses in Model A are also listed in Figures 8a, b
and c. In the graph notations, HC and HT are the
loading conditions.
We now discuss the comparison of the SIFs
solutions. Although there were differences in the
solution processes between the semi-auto RBE3
connecting technique and the PSCM plus zooming
method, the SIFs KI, KII and KIII obtained by the
VCCM were in good agreement with the J-integral
solutions of MSC.Marc. It was established that
the proposed method was effective in solving the
three-dimensional fracture mechanics analyses of
surface cracks in large scale structures of ships.
In Figures 8a, b and c, the SIFs of the loading
condition HT are larger than those of the loading
339
MARSTRUCT.indb 339
Figure 8. Stress Intensity Factors [a) KI, b) KII, c) KIII].
condition HC. The trends were almost the same
for Model A and Model B. It is considered that the
loading condition HT was a more severe situation
than the loading condition HC for the surface
crack in this study. Furthermore, the SIFs KII and
KIII in Model A were small compared with the case
in Model B in both loading conditions. The surface
crack of Model A was almost a pure opening mode
and the surface crack of Model B become mixed
modes. Although the SIF KI of Model A was large
in the deepest region (Eccentric angle θ = 1) of the
crack, the KI of Model B was large at the edges
(Eccentric angle θ = 0, θ = 2). In addition, the KI of
Model B was larger for all angles θ in both loading
conditions. This explains why the surface crack of
Model B was located in the high stress concentra-
tion region at the weld toe. Large SIFs values were
observed in the surface crack of Model B. Finally,
we predicted the crack propagation directions
for the surface cracks of Model A and Model B.
According to the Erdogan-Sih maximum tangential
stress criterion (Erdogan and Sih, 1963), the crack
2/18/2011 5:49:04 PM

Figure 9. Direction of crack propagation [a) KII > 0,
b) KII < 0].
will be propagated as illustrated in Figure 9a and b.
Here, φ is crack direction angle. In this study, the
KII for model A is very small and the KII for Model
B is a positive value for all eccentric angles θ of the
surface crack. It is assumed that the crack of Model
A will propagate in the plate thickness direction and
the surface crack of Model B will propagate at the
bottom of the welding as shown in Figure 9a.
The crack size used in this study is relatively
large compare to actual initial cracks in welded
joints. And, large values of SIFs are also obtained.
My future study, we will consider the crack size
and the loading condition to fit actual situation of
ships.
5 CONCLUSIONS
This paper presented three-dimensional fracture
mechanics analyses using the semi-auto RBE3
and the VCCM for quadratic tetrahedral finite
elements. Three-dimensional surface crack analysis
for a welded joint was demonstrated as a practical
application of the proposed technique. The surface
crack was embedded in a cruciform welded joint
in a cargo carrier. The shell-solid mixed analysis
using the semi-auto RBE3 connecting technique
was performed on the ship’s structure. The SIFs
were calculated using the VCCM for the quadratic
tetrahedral finite elements. The solutions were
compared with the SIFs calculated by the J-integral
calculations of MSC.Marc. They were in good
agreement with the J-integral solution. Finally, the
mechanical evaluations of the surface crack were
performed based on the SIFs. For the future study,
we propose to carry out three-dimensional crack
propagation analysis using the automated system
(Tokuda et al. 2010).
ACKNOWLEDGEMENTS
We would like to thank Prof. Osawa (Osaka
University) for contributing the sample data of the
PSCM model.
340
MARSTRUCT.indb 340
REFERENCES
Erdogan F., & Sih G.C. (1963), On the crack extension
in plates under plane loading and transverse shear,
Transactions ASME Ser D, Journal of Basic Engineering,
85–4, pp. 519–527.
Irwin G.R. (1958), Fracture, “Handbuch der Physik”, 6, 551.
Irwin G.R. (1962), Crack-extension force for a part-
through crack in a plate, Transactions ASME Ser E,
Journal of Applied Mechanics, 29, pp. 651–654.
Newman Jr. J.C., & Raju I.S. (1981), An empirical
stress intensity factor equation for the surface crack,
Engineering Fracture Mechanics, Vol. 15, pp.185–192.
Okada H., Kawai H., & Araki K. (2008), A virtual crack
closure-integral method (VCCM) to compute the
energy release rates and stress intensity factors based
on quadratic tetrahedral finite elements, Engineering
Fracture Mechanics, 75, pp. 4466–4485.
Okamura H. (1976), Linear fracture mechanics, Baifukan,
Japan, (in Japanese).
Osawa N., Hashimoto K., Sawamura J., Nakai T., &
Suzuki T. (2007), Study on shell-solid coupling FE
analysis for fatigue assessment of ship structure,
Marine Structures, 20, pp. 143–163.
Osawa N., Sawamura J., Suzuki S., Fujiwara T., &
Kuwahara M. (2008), Study on the local stress
evaluation technique for web stiffened cruciform joint
subject to pressure loading, Proc. of the 22th Asian
Technical Exchange and Advisory Meeting on Marine
Structure, pp. 199–207.
Rice J.R. (1968), A path independent integral and the
approximate analysis of strain concentration by
notches and cracks, Journal of Applied Mechanics, 35,
pp. 379–386.
Rybicki E.F., & Kanninen M.F. (1977), A finite element
calculation of stress intensity factors by a modified
crack closure integral, Engineering Fracture Mechanics,
9, pp. 931–938.
Shivakumar K.N., Tan P.W., & Newman Jr. J.C. (1988),
A virtual crack-closure technique for calculating stress
intensity factors for cracked three dimensional bodies,
International Journal of Fracture, 36, pp. R43–R50.
Tanaka S., Okada H., & Okazawa S. (2009), Three
dimensional fracture mechanics analysis for welded
joint structures using virtual crack closure-integral
method, The proceedings of the 19th International off-
shore and polar engineering conference, pp. 415–421.
Tanaka S., Okada H., Ogawa S., & Okazawa S. (2010),
Analysis of Three-dimensional Crack in Welded Joint
Structure using Shell-Solid Zooming Method, The
proceedings of the 19th International offshore and polar
engineering conference, pp. 31–37.
Tanaka S., Okada H., Okazawa S., Xi Y., & Ohtsuki Y.
Analysis of Three-dimensional Surface Cracks in a
Welded Joint Structure using the Shell-Solid Mixed
Method, International Journal of Offshore and Polar
Engineering (IJOPE), submitted.
Tokuda T., Kawai H., Okada H., & Fukui Y. (2010),
Development of Automated Crack Propagation
Analysis System (1st Report, Outlines of the System
and Finite Element Model Generation), Transactions
JSME Ser A, in print. (in Japanese).
Zienkiewicz O.C., & Taylor R.L. (2000). “The finite element
method 5th edition”, Butterworth-Heinemann, Vol. 1.
2/18/2011 5:49:08 PM
 Collision and impact strength

MARSTRUCT.indb 341 2/18/2011 5:49:09 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Ship hull composite plates analysis under blast loads

I. Chirica, D. Boazu, E.F. Beznea & A. Chirica

University “Dunarea de Jos” of Galati, Romania

ABSTRACT: In this paper the results of the study on the protective capacity of ship hull structures
made of composite materials subjected to an explosion of a spherical charge are presented. The main
application of these structures may be followed by design of ship structures of great importance, which
should also be protected against exceptional loads of this kind. In this study, a nonlinear analysis with the
finite-elements was done. The methodology to apply the blast pressure and the mechanism of the blast
wave in free air is given. The space pressure variation is determined by using Friedlander exponential
decay equation. Various parametric calculus to evaluating the behaviour of the ship structure laminated
plate to blast loading: explosive magnitude, distance from source of explosion, plate thickness.

1 INTRODUCTION in (Librescu & Noisier 1990b). In many of these

The explosion produced by accidents or inten-
tionally by terrorist attacks within or immediately
nearby ship hull can cause catastrophic damage
on the ship structure and shutting down of criti-
cal life safety systems. Loss of life and injuries to
crew and passengers can result from many causes,
including direct blast-effects, structural collapse,
debris impact, fire and smoke. The analysis and
design of structure subjected to blast loads require
a detailed understanding of blast phenomena and
the dynamic response of various structural ele-
ments. The most important way to diminish the
injuries due to blast loading is to secure sufficient
stand off distance between the explosion source
and the target structure and reduce the magnitude
of the blast impact wave so that the structure to
be not highly damaged. To achieve these ideas it
is necessary to do various scenarios to evaluating
the behaviour of the ship structure to blast load-
ing (explosive magnitude, distance from source
of explosion, structure scantling etc.). Transient
response of the shells from ship hull structures
to high intensity loads is often investigated in the
context of sonic booms, explosive blasts, and other
shock type pressure loads. Sonic booms and explo-
sions that are initiated sufficiently far from a struc-
ture are often idealized as arriving at all points on
the structure’s surface simultaneously.
The ship hull structures in close proximity to an
explosion experience high intensity-short duration
pressures that vary in time and space.
Most part of the literature available concern-
ing plate response to short-duration, high inten-
sity pressures make this assumption. Experimental
evidence supporting this assumption is provided
works, the time history of the spatially uniform
pressure is described by step-pulse, N-pulse, or
Friedlander equations (Crocker & Hudson 1969).
The response of elastic structures to time-
dependent external excitations, such as sonic boom
and blast loadings, is a subject of much interest in
the design of marine vehicles (Houlston, Slater,
Pegg, & Des Rochers 1985, Gupta, Gregory &
Bitting 1985). For the case of blast loadings, vari-
ous analytical expressions have been proposed and
discussed in (FEMA 2003).
The particular equation used for the pressure-
time history of the load is often chosen to best
match the particular phenomenon; considered.
The time history of overpressures due to explo-
sions is often represented by the modified Fried-
lander exponential decay equation (FEMA 2003).
In the references listed thus far, the blast pres-
sures are of arbitrary magnitude and are applied
uniformly across the structure. Methods presented
in (FEMA 2003) use experimental data from
explosive tests to develop expressions for the blast
overpressure as a function of time and distance
from the blast, as well as charge weight and other
important blast parameters. Very few papers make
use of such a realistic blast load. Most of the lit-
erature available concerning impulsively loaded
plates considers a linear solution for isotropic
plates. There are also many linear solutions avail-
able for impulsively loaded composite plates, and
some of the references listed thus far are of this
type (Crocker & Hudson 1969).
The rapid expansion of the detonation products
creates a shock wave in the surrounding medium,
which for simplicity in this paper is assumed as
being in air. This shock wave in air is known as a

343

MARSTRUCT.indb 343 2/18/2011 5:49:09 PM

 blast wave. Similar to the detonation wave there is
for practical purposes, a discontinuous increase in
pressure, density, temperature and velocity across a
blast wave. The shock-induced compression of the
ambient air also leads to an increase in temperature
behind the shock front. The pre and post shock
states are described by conservation equations for
mass, momentum and energy, collectively known
as Rankine-Hugoniot Jump equations (Kinney &
Graham 1985, Crocker & Hudson 1969).
Predicting the structural response to such an
explosion requires accurate prediction of the
applied pressures and a solution procedure that is
adequate for such transient phenomena.
The work presented here focuses on the struc-
tural response to such close proximity explosions.
In particular, the structures considered include
orthotropic composite and contacting plates sub-
ject to mine blasts. Both linear and a nonlinear
solution are developed for these simulations.
2 CHARACTERISTICS OF THE
BLAST WAVE
A huge or close-to-structure explosion will cause
extreme damage of main structural components
and the entire loss of load-carrying capacity of
structural components. Blast resistant design
of ship hull frames generally provides sufficient
toughness of components and structural system
capable of limiting the possibility of hull collapse.
Therefore, simulation of blast loading and estima-
tion of structure behaviour and damage under blast
loading are very important phase of research to
evaluate the resistance and safety of hull structure
against direct and consequential blast damage.
The explosion dissipates energy forming light,
sound, and very dense and high pressure wave with
initial expansion at very high velocities.
Typical explosive detonations in the free field
create a suddenly rising and rapidly decaying pres-
sure to satisfy equilibrium with surrounding water,
or a shock wave with very short duration.
The range in which the risen pressure decays
back to normal pressure is defined as a positive
phase (see Fig. 1). As the wave front expands, a
negative pressure phase occurs when the pressure
is lower than ambient pressure. The negative phase
has a little effect on the response of structures
(Cooper 1996).
Blast wave is reflected and amplified when the
incident pressure wave is transferred through air or
fluid and contact any structure, causing reflected
pressure. The reflected pressure usually varies
through the weight and type of explosive, stand-
off distance (or distance from the detonation),
and the incident angle of the wave. The reflected
344
MARSTRUCT.indb 344
Figure 1. Typical variation of the blast pressure.
pressure is a blast load on a structure considered in
a design and analysis. In addition, the duration and
amount of amplification of the reflected pressure
can be affected by the shape of target structure or
other objects, such as ground, which can produce
another reflected blast wave.
When a high order explosion is initiated, a
very rapid exothermic chemical reaction occurs.
Since the reaction progresses the solid or liquid
explosive material is converted to very hot, dense,
high-pressure gas. The explosion products initially
expand at very high velocities in an attempt to
reach equilibrium with the surrounding air, caus-
ing a shock wave.
A shock wave consists of highly compressed
air, is traveling radially outward from the source
at supersonic velocities. Only one-third of the
chemical energy available in most high explosives is
released in the detonation process. The remaining
two-thirds is released more slowly as the detona-
tion products mix with air and burn. This after-
burning process has little effect on the initial blast
wave because it occurs much slower than the origi-
nal detonation.
However, later stages of the blast wave can
be affected by the afterburning, particular for
explosions in confined spaces. As the shock wave
expands, pressures decrease rapidly (with the cube
of the distance) because of geometric divergence
and the dissipation of energy in heating the air.
Pressures also decay rapidly over time and have
a very brief span of existence, measured typically
in thousandths of a second, or milliseconds. An
explosion can be visualized as a “bubble” of highly
compressed air that expands until reaching equi-
librium with the surrounding air.
Explosive detonations create an incident blast
wave, characterized by an almost instantaneous rise
from atmospheric pressure to a peak overpressure.
2/18/2011 5:49:10 PM
 As the shock front expands pressure decays back to
ambient pressure, a negative pressure phase occurs
that is usually longer in duration than the positive
phase as it is shown in the figure 1. The negative
phase is usually less important in a design than the
positive phase (Brode 1955).
When the incident pressure wave impinges on
a structure that is not parallel to the direction of
the wave’s travel, it is reflected and reinforced, pro-
ducing what is known as reflected pressure. The
reflected pressure is always greater than the incident
pressure at the same distance from the explosion.
The reflected pressure varies with the angle of inci-
dence of the shock wave. When the shock wave
impinges on a surface that is perpendicular to the
direction it is traveling, the point of impact will
experience the maximum reflected pressure. When
the reflecting surface is parallel to the blast wave,
the minimum reflected pressure or incident pres-
sure will be experienced. In addition to the angle
of incidence, the magnitude of the peak reflected
pressure is dependent on the peak incident pres-
sure, which is a function of the net explosive weight
and distance from the detonation.
Typical reflected pressure coefficients versus
the angle of incidence for different peak incident
pressures are presented in (FEMA 2003). The
reflected pressure coefficient equals to the ratio
of the peak reflected pressure to the peak inci-
dent pressure (Cr = Pr/Pi). The figure 1 shows that
reflected pressures for explosive detonations can
be almost 13 times greater than peak incident pres-
sures and, for all explosions, the reflected pressure
coefficients are significantly greater closer to the
explosion.
Impulse is a measure of the energy from an
explosion imparted to a structure. Both the nega-
tive and positive phases of the pressure-time wave-
form contribute to impulse. Figure 1 shows how
impulse and pressure vary over time from a typical
explosive detonation. The magnitude and distribu-
tion of blast loads on a structure vary greatly with
several factors:
− Explosive properties (type of material, energy
output, and quantity of explosive);
− Location of the detonation relative to the
structure;
− Reinforcement of the pressure pulse through its
interaction with the ground, water or structure
(reflections).
The reflected pressure and the reflected impulse
are the forces to which the ship structure ultimately
responds. These forces vary in time and space over
the exposed surface of the ship hull structure,
depending on the location of the detonation in
relation to the ship hull. Therefore, when analyzing
a structure for a specific blast event, care should be
345
MARSTRUCT.indb 345
taken to identify the worst case explosive detona-
tion location.
In the context of other hazards (e.g., earth-
quakes, winds, waves or tides), an explosive attack
has the following distinguishing features:
− The intensity of the pressures acting on a tar-
geted structure building can be several orders of
magnitude greater than these other hazards. It
is not uncommon for the peak incident pressure
to be in excess of 689475 Pa on a building in an
urban setting for a vehicle weapon parked along
the curb. At these pressure levels, major dam-
ages and failure are expected.
− Explosive pressures decay extremely rapidly with
distance from the source. Therefore, the damages
on the side of the building facing the explosion
may be significantly more severe than on the
opposite side. As a consequence, direct air-blast
damages tend to cause more localized damage.
In an urban setting, however, reflections off sur-
rounding buildings can increase damages to the
opposite side.
− The duration of the event is very short, meas-
ured in thousandths of a second, or milliseconds.
This differs from earthquakes, waves or wind
gusts, which are measured in seconds, or sus-
tained wind or waves situations, which may be
measured in hours. Because of this, the mass
of the structure has a strong mitigating effect
on the response because it takes time to mobi-
lize the mass of the structure. By the time the
mass is mobilized, the loading is gone, thus miti-
gating the response.
The blast wave reaches the peak value in such a
short time that the structure can be assumed to be
loaded instantly. Due to the relative small dimen-
sions of the structure when compared to the blast
and sonic boom wave front, it may also be assumed
that the pressure is uniformly distributed over the
structure.
The first step in blast related research is to pre-
dict blast loads on the structure. For the purpose
of blast resistant design, experiment-based direct
load-calculating methods have been mostly used
to describe the blast pressure on a structure. The
variables of a function to represent a blast pres-
sure time history are peak pressure, impulse,
arrival time, and the duration of the pressure.
These parameters are generally determined by
experimental results. Many documents in this field
provide graphical form displaying the values of
the blast wave parameters as a function of scaled
distance after an explosive weight is converted to
TNT equivalent weight (Brode 1955).
A universal normalized description of the blast
effects can be given by scaling distance relative to
(E/P0)1/3 and scaling pressure relative to P0, where
2/18/2011 5:49:10 PM
 E is the energy release (kJ) and P0 the ambient
pressure (typically 100 kN/m2). For convenience,
however, it is general practice to express the basic
explosive input or charge weight W as an equiva-
lent mass of TNT.
Much of the published data for blast wave
parameters is presented in terms of scaled dis-
tance. At a range of R1 from a reference explosive
mass W1, an over pressure of P will occur, accord-
ing to cube-root scaling, if the explosive mass is
changed to W2, the same overpressure will occur
at a distance R2
R1 R2
Z= = (1)
W1 W2
where R, in meters, is the actual effective distance
from the explosion and W is generally expressed in
kilograms. Scaling laws provide parametric corre-
lations between a particular explosion and a stand-
ard charge of the same substance.
Blast wave parameters for conventional high
explosive materials have been the focus of a
number of studies during the 1950’s and 1960’s.
Estimations of peak overpressure Pso, in kPa, due
to spherical blast based on scaled distance Z were
introduced in (Brode 1955) as:
− for Pso > 1000 kPa
Pso = 670/Z3 + 100
− for 10 kPa < Pso < 1000 kPa
Pso = 97.5/Z + 145.5/Z2 + 585/Z3 – 1.9
The shape of blast wave can be represented
by linear decay using an approximate triangular
equivalents or more realistic exponential decay
shown in figure 1 based on Friedlander equation
which intends to agree with experimental values
of blast pressure (Baker 1973). A modified Fried-
lander’s equation is as follows
⎛ t T0 ⎞
⎛ t Ta ⎞ − A⎜⎝ T0 ⎟⎠
p (t ) = pso ⎜1 − e (2)
⎝ T ⎟⎠
0
where p(t) is blast pressure at time t, pso is peak inci-
dent pressure, T0 is positive phase duration, Ta is
arrival time, and A is a decay coefficient.
Peak reflected pressure is given as a function
of variables such as peak incident pressure, angle
of wave incidence to the surface of an object and
shock front velocity etc. Then, reflected impulse
proportional to the calibrated peak reflected pres-
sure and the corresponding duration of reflected
pressure will be determined (FEMA 2003).
In accordance with above mentioned references
the overpressure associated with the blast pulses
346
MARSTRUCT.indb 346
can be described in terms of the modified Fried-
lander exponential decay equation as (Baker 1973)
a ′t
⎛ t ⎞ −t
py ( s,zz t ) = pm ⎜1 − ⎟ e p (3)
⎝ tp ⎠
where the negative phase of the blast is included. In
Equation (3), pm denotes the peak reflected pressure
in excess of the ambient one; tp denotes the positive
phase duration of the pulse measured from the time
of impact of the structure and a′ denotes a decay
parameter which has to be adjusted to approximate
the overpressure signature from the blast tests.
As concerns the sonic-boom loading, this can be
modeled as an N-shaped pressure pulse arriving at
a normal incidence. Such a pulse corresponds to
an idealized far field overpressure produced by an
aircraft flying supersonically in the earth’s atmos-
phere or by any supersonic projectile rocket or
missile. The overpressure signature of the N-wave
shock pulse can be described by
⎧ ⎛ t⎞
⎪ pm ⎜1 − ⎟ for 0 < t < rtt p
p ( s, z , t ) = ⎨ ⎝ t p ⎠ (4)
⎪0
fo t 0 o t > rtt p
⎩
3 NUMERICAL ANALYSIS OF BLAST
LOAD EFFECTS ON THE COMPOSITE
PLATE
Taking into account the time variation of the pres-
sure load in space, a graphical methodology, pre-
sented in Figure 2, to describe this action on the
plate is used. As it is shown in the figure 2, some of
the assumptions and parameters involved in blast
loading are presented. Also, a finite element model
of the plate structure is presented.
Once the blast distance h is determined, ele-
ments within 45 degrees of the blast normal vector
are divided into groups based upon their average
Figure 2. Characteristics of blast load pressure on
plate mesh.
2/18/2011 5:49:10 PM
 distance to the center node. There is described
the application of a blast load to a finite element
mesh. Only elements within the 45 degrees cone are
loaded. The area within the cone is divided into a
number of rings to determine the pressure acting
on the elements of the mesh.
According to the methods used in this paper an
individual pressure-time history to each element
based on its distance from the blast is assigned.
Each ring from Figure 3 has its own pressure-time
history as it is shown in Figure 4.
The studies are carried out on a square plate
from the side shell placed between two pairs of web
stiffeners, having the dimension axa (a = 500 mm).
So, the plate can be considered as being clamped
on the all sides.
For the studied plate, ten discretization rings
were used to load the panel with the mine blast at
the distance h. Parametric calculus was done for
equivalent TNT mass (W = 0.1 kg, 0.2 kg, 0.5 kg)
placed at the distance h = 0.15 m and h = 0.2 m
from the plate surface.
Figure 3. Distribution of the blast load on the plate
mesh.
Figure 4. Pressure time histories for all rings.
347
MARSTRUCT.indb 347
The spatial distributions and pressure profiles
are illustrated in figure 4 (the ring 1 is the circle
from the middle of the plate). As it is shown in
the figure, the time of arrival is taken the same
for all rings due to the small difference for Ta from
each to other ring, comparing with total time. So,
Ta = 0.01 s and T0 = 0.1 s.
Using the ideal and applied impulse values,
a percent error for the applied impulse can be
determined.
The applied impulse is calculated using the plate
mesh. The area of each of the ring zones is easily
calculated from the formula for the cross sectional
area of a cylinder.
These individual areas are then multiplied by
the corresponding impulse per unit area to obtain
the total applied impulse.
The material is E-glass/polyester having the
symmetric stack. The stack of the shell is accord-
ing to the topologic code [A/B]3 s.
The layers made of material A, have the thick-
ness of 0.25 mm and characteristics:
Ex = 80 GPa, Ey = 80 GPa, Gxy = 10 GPa, μxy = 0.2.
The layers made of material B, have the thick-
ness of 0.1 mm and characteristics:
Ex = 3.4 GPa, Ey = 3.4 GPa, Gxy = 1.3 GPa, μxy = 0.3.
Due to the double symmetry, one quarter of
plate is studied.
In the non-linear calculus, the material is con-
sidered as non-linear one. Tsai-Wu criterion is con-
sidered for the limit state. The dynamic calculus is
done by direct integration with Newmark method
(γ = 0.5; β = 0.25).
The calculus was done for 3 groups of values so
for equivalent TNT mass of 0.215 kg, 0.5 kg and
1 kg, h = 0.2 m and for total thickness of the com-
posite plate (2 mm, 4 mm and 6 mm).
The results of the parametric calculus are pre-
sented in Figures 5–12 (time variation of displace-
ment of the central point of the plate, time variation
of the maximum von Mises stress occurred in the
plate sides).
For the case of plate having the thickness of
2 mm, two cases of plate are considered: plate with
damping and plate without damping. For the first
case, damping was considered as
C = 0.1 K + 0.1 M (5)
Time variation of the maximum von Mises
stress obtained in the plate for the case h = 0.2 m
and W = 0.1 kg, t = 2 mm with damping (figure 5)
and without damping (figure 6) are presented. As
it is seen, the maximum stress in the case of damp-
ing is 2.5 times bigger than the stress in the case of
the plate without damping.
2/18/2011 5:49:12 PM
 In figures 9–12 the results for other parametric
versions are presented.
In figure 9 is presented the variation maximum
von Mises stress on the sides of the plate, with-
out damping in the case: h = 0.15 m; W = 0.5 kg.
Fails in various layers occurred (so by tension and
compression).
Figure 5. Maximum von Mises stress on the sides of the
plate with damping.
Figure 6. Maximum von Mises stress on the sides of the
plate without damping.
Figure 7. Variation of the maximum transversal dis-
placement in the case h = 0.2 m.
348
MARSTRUCT.indb 348
Variation of the transversal displacement of the
plate’s central point, without damping is presented
in figure 10.
In figure 11, variation of the transversal displace-
ment of the plate’s central point, without damping
in the case: h = 0.2 m; W = 1 kg. The calculus is
stopped at time: 0.00012s because in material very
Figure 8. Variation of the maximum transversal
displacement in the case h = 0.15 m.
Figure 9. Maximum von Mises stress on the sides of the
plate, without damping.
Figure 10. Variation of the transversal displacement of
the plate’s central point, without damping.
2/18/2011 5:49:12 PM

Figure 11. Variation of the transversal displacement
of the plate’s central point, without damping (Case:
h = 0.2 m; W = 1 kg).
Figure 12. Variation of the transversal displacement of
the plate’s central point with damping. (Case: h = 0.2 m;
W = 1 kg).
large displacements occur. Tension and compres-
sion fails occur in a lot of elements.
In figure 12, variation of the transversal dis-
placement of the plate’s central point with damp-
ing according to the equation (5), in the case:
h = 0.2 m; W = 1 kg.
4 CONCLUSIONS
The design considerations against extreme events
(bomb blast, high velocity impact) are very
important for high-risks facilities such as navy
and commercial ships. It is recommended that
guidelines on abnormal load cases and provisions
on progressive collapse prevention should be
included in the current ship hull structure design
norms.
Requirements on ductility levels also help
improve the structure performance under severe
load conditions.
349
MARSTRUCT.indb 349
Dynamic calculus was done with COSMOS/M
soft package using specific elements SHELL4L.
The blast wave instantaneously increases to a
value of pressure above the ambient atmospheric
pressure. This is referred to as the side-on overpres-
sure that decays as the shock wave expands out-
ward from the explosion source. After a short time,
the pressure behind the front may drop below the
ambient pressure (figure 2). During such a nega-
tive phase, a partial vacuum is created and air is
sucked in. This is also accompanied by high suc-
tion winds that carry the debris for long distances
away from the explosion source. In the paper this
phase is considered as equal to zero.
The values of the pressure on the plate have
small differences between cases W = 0.1 kg and
W = 0.2 kg. To have a good strength the thickness
of the plate was adapted accordingly.
It is critical for accurate analytical predictions to
obtain pressure maps and time histories for muzzle
blast from a particular blast explosion. This data
should then be fit to the blast pressure model to
obtain duration, delay, and maximum pressure as
a function of location for each exposed element in
a finite element model.
Models should be analyzed using geometric non-
linearity to capture deformation hardening (and
softening) of pressurized skins. Use the minimum
damping necessary to obtain solution convergence.
Use large safety factors (obviously more than
2.0) when sizing structure due to uncertainties
in blast pressure characteristics and structure
dynamic behavior is design requesting.
Structure subjected to blast explosion should
be as compliant as possible while maintaining
structural integrity to minimize transferred loads.
Internal shear joints require special attention to
ensure adequate strength for transferred loads.
According to the analysis, the developed blast
simulation model and optimal design system can
enable the prediction, design and prototyping of
blast-protective composite structures for a wide
range of damage scenarios in various blast events,
ranging from plate damage, localized structural
failure. From the studies, the proposal of a com-
posite structure with special damping system can
help the structure to sustain blast load. The inclu-
sion of a damping material in the composite struc-
ture can absorb energy under blast load and help
to reduce the force transmitted to the main struc-
ture. Also, the damping material helps to reduce
stress concentration in the plate material.
ACKNOWLEDGEMENTS
The work has been performed in the scope of
the Romanian Project PN2—IDEI, Code 512
(2009–2011).
2/18/2011 5:49:15 PM
 REFERENCES
Baker, W., Explosions in air, University of Texas Press,
Austin, 1973.
Brode, H.L., Numerical solution of spherical blast waves,
Journal of Applied Physics, American Institute of
Physics, New York, 1955.
Coggin, J.M., Response of Isotropic and Laminated
Plates to Close Proximity Blast Load, MSc thesis, Vir-
ginia Tech, 2000.
Cooper, P., Explosive engineering, Wiley-VCH, New
York, 1996.
Cordon, J.A., Gniazdowski, N. & Gregory, F.H., The
Design, Testing and Analysis od a Proposed Comn-
posite Hull Technology Mine-Blast-Resistant Vehicle
Floor Panel, ARL-TR-796, 1995.
Crocker, M.J. & Hudson, R.R., Structural Response to
Sonic Booms, J. of Sound and Vibration, 9(3), pp.
454–468, 1969.
FEMA, Risk Management Series, Reference Manual to
Mitigate Potential Terrorist Attacks Against Build-
ings. Providing Protection to People and Buildings,
FEMA, 426, 2003.
Gupta, A.D., Dynamic Analysis of a Flat plate subjected
to an explosive blast,Proc. ASME International Com-
puters in Engineering Conference, vol. 1, pp. 491–496,
1985.
Gupta, A.D., Gregory, F.H. & Bitting, R.L., Dynamic
Response of a Simply Supported Rectangular Plate to
an Explosive Blast, Proc. XIII Southestern Conf. on
Theoretical and Appl. Mech. 1, pp. 385–390, 1985.
350
MARSTRUCT.indb 350
Houlston, R., Slater, J.E., Pegg, N. & Des Rochers, C.G.,
On the analysis of structural response of ship panels
subjected to air blast loading, Computers and Struc-
tures, 21, pp. 273–289, 1985.
Jiang, D., Liu, Y., Qi, C., Ma, Z.D., Raju, B. & Bryzik, W.,
Innovative Composite Structure Design for Blast Pro-
tection, SAE International, 2007.
Kinney, G. & Graham, K., Explosive Shocks in Air, 2-nd
Ed., Springer Verlag, New York, 1985.
Lee, K., Kim, T. & Kim, J., Local response of W-shaped
steel columns under blast loading, Structural Engineer-
ing and Mechanics, Vol. 31, No. 1, 00 25–38, 2009.
Librescu, L. & Noisier, A., Dynamic Response of Aniso-
tropic Composite Panels toTime-Dependent External
Excitations, The 17-th Congress of the International
Council of the AeronauticalSciences, ICAS-90-1.4R,
pp. 2134–2144, 1990.
Librescu, L. & Noisier, A., Response of Shear Deforma-
ble Elastic Laminated Composite Flat Panels to Sonic
Boom and Explosive Blast Loadings, AIAA Journal,
23(2), pp. 345–352, 1990.
Longinow, A. & Mniszewski, K.R., Protecting buildings
against vehicle bomb attacks, Practice Periodical, on
Structural Design and Construction, ASCE, New
York, pp. 51–54, 1996.
Ngo, T., Mendis, P., Gupta, A. & Ramsay, J., BlastLoading
and Blast Effectson Structures—An Overview, EJSE
Special Issue: Loading on Structures, pp. 76–91, 2007.
Yen, C.F. & Jones, M.L., Composite Material Modeling
for Blast Protection, MSC-TFR-3705/CB02, Coun-
termine Division, Fort Belvoir, VA 22060, 1996.
2/18/2011 5:49:16 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

An investigation of a Suezmax tanker grounding accident

S. Ehlers
Aalto University, School of Science and Technology, Marine Technology, Espoo, Finland

D. Polić, A. Klanac & M. Schröder

as2con-alveus ltd., Rijeka, Croatia

ABSTRACT: A grounding accident of a Suezmax tanker, built in compliance with IACS Common
Structural Rules (CSR), is presented and analysed from the structural point of view. The analysis focuses
on the simulation of structural deformations caused by the accidental event. The analysis is performed
using the non-linear finite element method. The validation of the simulation results is performed through
the comparison of calculated deformations with those measured on the actual damaged structure
of the vessel. The measurements were performed in the dry dock prior to the steel replacement. Thereby,
the validity of the finite element simulations is secured and the conclusions on the state of deformation
can be made with confidence. Furthermore, critical structural arrangements can be outlined, in order to
contribute to an improvement of the similar new buildings. The validity of the simulation procedure raises
the belief in the analysis of future grounding events.

1 INTRODUCTION material relation resulted in better convergence and

Quantitative and reliable structural analyses of
grounding accidents have not been done in the past
(see also Pedersen 2010). One reasons for this is the
large number of unknown parameters associated
with the accident, such as missing recordings of
the ships’ motions and the sea bottom consistency.
Therefore, simplified analytical assessments of the
ship bottom damage are carried out (i.e. Hong
and Amdahl 2008, Pedersen and Zhang 2000 and
Zhang 2002). Furthermore simplified numerical
analyses are compared with experimental results by
Simonsen 1997 and Paik and Seo 2007. However,
these analyses are limited in accuracy and are only
carried out for small ship bottom sections, typically
between two frames. Hence, detailed structural
analysis of different ship bottom structures were
carried out by Naar et al. 2002, however, without
experimental validation. Additionally, the insuffi-
ciency of common material and failure models was
presented in a benchmark study with experimental
validation (Ehlers et al. 2008). The primary reason
for this insufficiency was the determination and
implementation of the nonlinear material relation
into the numerical simulation.
However, recent developments of a finite ele-
ment length-dependent material relation until
failure (Ehlers 2010a and b) may be a good choice
to improve the accuracy of grounding simula-
tions. Furthermore, this element length-dependent
compliance of simulation results with the experi-
ments than a conventional power law material rela-
tion and failure criterion proposed by Germanischer
Lloyd (Zhang et al. 2004). Therefore, this paper
employs the non-linear material relation obtained
by Ehlers (2010a) to analyse the bottom damage
of a Suezmax tanker during a grounding accident.
Additionally, the actual bottom damage will be
measured and presented. Furthermore, the simula-
tion will be compared with the measurements and
critical structural elements will be outlined. The
congruence of the simulation procedure raises the
belief in the analysis of future grounding events.
For the reason of confidentiality exact details
of the vessel, the accident cause and the acci-
dent location cannot be presented in this paper.
However, the core parameters relevant for the
simulations and available are presented.
1.1 The Suezmax tanker
The tanker is a Suezmax crude oil tanker delivered
in the year of 2009 and built according to the
IACS Common Structural Rules (CSR). The dis-
placement of the tanker during the accident was
approximately 180.000 ton. The scantling length of
the tanker is 260 m, the moulded breadth is 48 m
and the scantling draught is 17 m. A typical section
of the bilge area found in the affected area of the
tanker is shown as an example in Figure 1.

351

MARSTRUCT.indb 351 2/18/2011 5:49:16 PM

 Figure 1. Typical section of the affected area.

1.2 The grounding accident

The grounding accident of the tanker occurred in
a narrow channel passage with zero visibility due
to dense fog. A sketch of the location is shown in
Figure 2. The width of the channel at the accident
location is approximately 300 m. The accidental
manoeuver initiated with a course in close proximity
to a dredger at the port side channel bank. Hence, Figure 2. Location of the grounding accident in a
the tanker needed to steer to starboard to avoid narrow channel.
a close encounter with the dredger. However, the
strong current against the heading of the tanker
and a significant current increase towards the
starboard channel bank resulted in a high rate of
turn towards the starboard channel bank. Even
with the rudder put hard to port control of the
vessels manoeuvrability could not be gained. As a
result, the tanker hit the starboard side channel Figure 3. Extend of the bottom damage after the
bank. The last reported speed over ground seven grounding accident.
minutes prior to the grounding was 6.5 km. Two
minutes prior to the grounding the engine was
stopped, with no prior speed reduction, and the
grounding resulted in a full stop of the tanker. The
crew described that the tanker hit the channel bank
in a straightforward manor with no noticeable
motions occurring other than the sudden stopping.
The resulting damage of the hull structure concen-
trated at the bilge area in the forepart of the tanker
between frame 89 and 96, see Figures 3 and 4.
Furthermore, the damage indicates that the channel
bank was very hard and therefore able to bring the
tanker to a full stop with rather localized damage. Figure 4. Bilge area damage after the grounding
Minor damage and some scratches initiated from damage.
the bulb bottom shell plating and extended slightly
after frame 96. Overall, only insignificant fracture
of the shell plating occurred at a single location, the deformed bottom shell is obtained in a grid
the remaining outer plating stayed. pattern at each web frame. As a result, the actual
The resulting structural damage, respectively the deformation is known between frame 89 and 96 by
bottom indentation, shown in Figures 3 and 4 is comparing the measured points with the original
measured with a digital laser-based ruler prior to lines of the tanker (see Fig. 5). A Three Dimen-
the steel repair work in the dry dock. Thereby, the sional (3D) presentation of the measured damage
perpendicular distance from the dry dock floor to is shown in Figure 6.

352

MARSTRUCT.indb 352 2/18/2011 5:49:16 PM

 Figure 5. Intact and damaged lines of the tanker.
Figure 6. 3D representation of the measured damage.
1.3 Scope of the analysis
The scope of this study is to simulate the tanker
grounding accident numerically in order to repro-
duce the actual structural damage. Furthermore,
a comparison of the simulated bottom damage with
the deformation measurements will be presented.
Additionally, a visual comparison of the simulated
deformations with a series of photographs taken
during the steel repair will be shown. Therefore,
a quasi-static simulation approach will be presented
utilizing a rigid intender to simulate the bottom
structure. Furthermore, it will be shown that this
assumption of a rigid bottom structure suffices in
this particular case by comparing the theoretically
available kinetic energy with the deformation energy.
Additionally, this study will validate the applicabil-
ity of the nonlinear material relation according
to Ehlers (2010) for this real accident. As a result,
the structural details are investigated and it will be
verified if the replaced steel was sufficient. Further-
more, possible critical details outside the replaced
steel area will be highlighted for future inspections.
2 THE GROUNDING SIMULATION
The solver LS-DYNA version 971 is used for the
grounding simulations. The CAFE modeller (www.
bvbcafe.com) is used to build the finite element
model of the tanker (see Fig. 7). The model is built
between frame 88 and 97, thus it extends for one
web frame at each end over the primarily damaged
zone. Furthermore, the model includes all primary
stiffeners, brackets and openings. The influence
of welding is however not included in this model.
The boundary conditions are applied at frame 88
353
MARSTRUCT.indb 353
Figure 7. Finite element model of the starboard section
of the tanker between frame 88 and 97.
Figure 8. Refined mesh in the grounding area.
and z = 18.2 m and restrict all translational degrees
of freedom. The remaining edges are free. The
structure is modelled using primarily four nodded,
quadrilateral Belytschko-Lin-Tsay shell elements
with 5 integration points through their thickness.
The element size in the grounding area is refined
from 200 × 200 mm to 40 × 40 mm resulting in
a model size of 1.65 million shells (see Fig. 8).
Standard LS-DYNA hourglass control is used for
the simulations (see Hallquist 2007).
The grounding simulations are displacement-
controlled. The rigid indenter, simulating the slope
of the channel bank, is moved at a constant velocity
of 10 m/s following the path of the measured defor-
mation (see Figs. 6 and 9). The choice of a rigid
indenter arises from the fact that the actual inden-
tation is localized and that no information on the
true bottom consistency is available. Furthermore,
the applied velocity is reasonably low so as not to
cause inertia effects resulting from the ships’ masses
(see Konter et al. 2004). The ship motions are not
considered in this analysis, because the crew did not
notice any motions other than the straightforward
hit followed by a full stop. Furthermore, recorded
information about the tankers motions during the
accident is not available. The time-step is control-
led by the bar-wave speed and the following length
measure: the maximum of the shortest element side
or the element area divided by the minimum of the
2/18/2011 5:49:18 PM
 structure, the structure is assumed to follow the
material behaviour presented by Ehlers (2010b).
This assumption is in line with an earlier validation
presented by Ehlers 2010b for stiffened plates built
from mild steel. Additionally, the material relation
obtained by Ehlers is scaled upwards to account for
the increase in yield strength of the high strength
steel found in the tanker, see Figure 10. The failure
strain is assigned according to the element length
as shown in Figure 11. This assignment is consid-
ered valid, because the insignificant fracture did not
occur in the high strength steel.

2.2 Results and discussion

Figure 9. Rigid indenter, respectively channel bank
shape, and location prior to the grounding simulation.
longest side or the longest diagonal of the element.
Instabilities for this time-step measure could not be
found. Additionally, the solution is not mass-scaled.
The automatic single surface contact of
LS-DYNA—see Hallquist 2007—is used to treat
the contact occurring during the simulation with
an assumed static friction coefficient of 0.3. The
reaction force between the rigid indenter and the
bottom structure is obtained with a contact force
transducer penalty card (see Hallquist 2007).
Integrating the resultant contact force over the
deformation length leads to the energy absorbed by
the tanker structure during the grounding accident.
The quasi-static simulation is carried out on a
four core AMD processor with 2.3 GHz at a total
CPU time of 535 hours. The resulting force—and
energy versus damage length curves are shown in
Figure 12. Therein, the higher structural resistance
at the web frames is clearly shown. Furthermore,
the simulated energy absorbed by the structural
deformations reaches 967 MJ, which corre-
sponded well to the available kinetic energy of
the 180.000 ton tanker at 6.5 km. Therefore, the
rigid indenter is a good choice for the presented
grounding simulation, because good agreement

2.1 Material and failure modelling

The material relation until failure according to Ehlers
(2010a) is utilized in this simulation, because it rep-
resents a consistent link between experimental mate-
rial behaviour and the numerical implementation.
In other words, Ehlers describes a procedure to
obtain an element length-dependent material rela-
tion until failure for mild steel, which is suitable for
Figure 10. Nonlinear material relation for mild and
the finite element method. The finite element length-
high strength steel.
dependency of this material relation is achieved
because the strain reference length is clearly identi-
fied on the basis of optical measurements. Hence,
the finite element length has to correspond to this
strain reference, and thereby an element length-
dependent strain and stress relation until failure
is achieved. Furthermore, a constant strain failure
criterion is chosen to delete failing elements and
to simulate rupture even though significant rup-
ture did not occur during the grounding accident.
However, this failure criterion is included in order
to validate the structural deformation and thereby
the failure criterion itself. In other words, failure
should not occur and should therefore not be pre-
dicted at a premature stage of deformation. Because Figure 11. Failure strain versus element length (Ehlers
actual material tests are not available for the tanker 2010b).

354

MARSTRUCT.indb 354 2/18/2011 5:49:19 PM

 1200

Available kinetic energy ~1000 MJ

1000

Energy [MJ]
967 MJ
800
Force [kN]/100

600

400

200

0
0 5 10 15 20 25 30
Damage length [m]

Figure 12. Simulated force versus (--) and energy (−)

Figure 15. Simulation a) versus photograph b) of
versus damage length.
frame 90A.

Figure 13. Simulated damage at the bilge area.

Figure 14. Comparison of the simulated and measured Figure 16. Simulation a) versus photograph b) of
(black) frame deformations. frame 90F.

between the absorbed and available energy is

obtained. Furthermore, the structural defor-
mations (see Figs. 13 and 14) correspond well
to the measured actual damage, and failure of
the material is not predicted by the employed
constant strain failure criterion (see Fig. 11).
The strain range in Figure 13 has been scaled to
the critical value for the 40 mm element length.
Furthermore, the qualitative comparison of the
simulated deformation shape and the photographs
taken during the steel replacement are in very
good correspondence, see Figures 15, 16 and 17.
Therein, it can be seen that the web frame defor-
mations and the buckling of individual stiffeners is
captured accurately. Furthermore, Figure 18 shows
the undamaged and unrepaired structure without
plastic deformations; the strain range is scaled
to 0.05. Additionally, the replaced steel section
presented in Figure 19 includes all deformations. Figure 17. Simulation a) versus photograph b) of
The replaced steel concerned the shell plating and frame 89.

355

MARSTRUCT.indb 355 2/18/2011 5:49:20 PM


Figure 18. Unrepaired structure with no signs of damage.
Figure 19. Replaced structure including the damage.
Figure 20. Internal deformations and critical area at the
hopper tank at frame 95.
side shell longitudinal stiffeners, the longitudinal
bottom side girder, the web frames, other longitu-
dinals and the remaining internal structures shown
in Figure 19. Furthermore, the numerical simula-
tion resulted in one critical area at the inside of
the tank shown in Figure 20 at the hopper plating
around frame 95. This area appears to be prone to
damage due to the transition of the tank top into
the hopper plating and the change in slope in longi-
tudinal direction of the hopper plating. Therefore,
the only improvement to be suggested for future
vessels is to stiffen such areas internally to avoid
rupture and thereby possible leakage of the cargo
into the ballast tanks. In the present case, this area
is repaired with an 800 × 800 mm inset.
3 SUMMARY AND CONCLUSIONS
A grounding simulation procedure on the basis
of bottom damage measurements was shown and
good agreement of the results in terms of absorbed
energy, deformation shape and replaced steel area
are achieved. Furthermore, this good correspond-
ence of the results validates the applicability of
356
MARSTRUCT.indb 356
the material relation until failure for full scale
grounding accidents. Failure itself is not predicted
prematurely, however whether or not rupture
propagation would be predicted accurately with
the current method remains in question due to the
absence of significant rupture. However, the general
applicability of the presented procedure, to assess
the bottom damage during a grounding accident,
can be stated. A general procedure for various and
usually unknown sea bottom consistency remains
for future work. However, for the presented tanker
grounding accident it can be concluded that the
steel replacement was done sufficiently and all criti-
cal and deformed structural members are replaced.
REFERENCES
Ehlers, S., Broekhuijsen, J., Alsos, H.S., Biehl, F., &
Tabri, K. 2008. Simulating the collision response of
ship side structures: A failure criteria benchmark
study. Int Ship Progress; 55; 127–144.
Ehlers, S. 2009a. Strain and stress relation until fracture
for finite element simulations of a thin circular plate.
Thin-Walled Structures; 48 (1); 1–8.
Ehlers, S. 2009b. A procedure to optimize ship side struc-
tures for crashworthiness. Journal of Engineering for
the Maritime Environment; 224; 1–12.
Hallquist, J.O. 2007. LS-DYNA. Keyword User’s Manual,
Version 971. Livermore: Livermore Software Technology
Corporation.
Hong, L., & Amdahl, J. 2008. Crushing resistance of
web girders in ship collision and grounding. Marine
Structures; 21; 374–401.
Konter, A., Broekhuijsen, J., & Vredeveldt, A. 2004.
A quantitative assessment of the factors contribut-
ing to the accuracy of ship collision predictions with
the finite element method. In Proceedings of the Third
International Conference on Collision and Grounding
of Ships; 17–26.
Naar, H., Kujala, P., Simonsen, B.C., & Ludolphy, H.
2002. Comparison of the crashworthiness of various
bottom and side structures. Marine Structures; 15;
443–460.
Paik, J.K., & Seo, J.K. 2007. A method for progressive
structural crashworthiness analysis under collisions
and grounding. Thin-Walled Structures; 45; 15–23.
Pedersen, P.T., & Zhang, S. 2000. Effect of ship structure
and size on grounding and collision damage distribu-
tions. Ocean Engineering; 27; 1161–1179.
Pedersen, P.T. 2010. Review and application of ship col-
lision and grounding analysis procedures. Marine
Structures; 23; 241–262.
Simonsen, B.C. 1997. Ship Grounding on Rock—II.
Validation and Application. Marine Structures; 10;
563–584.
Zhang, L., Egge, E.D., & Bruhms, H. 2004. Approval
procedure concept for alternative arrangements.
In Proceedings of the Third International Conference
on Collision and Grounding of Ships; 87–97.
Zhang, S. 2002. Plate tearing and bottom damage in ship
Grounding. Marine Structures; 15; 101–117.
2/18/2011 5:49:23 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Research on anti-collision capability for double-hull design

for the column of semi-submersible by numerical simulation

Zhiqiang Hu, Gang Chen & Jianmin Yang

State Key Lab of Ocean Engineering, Shanghai Jiao Tong University, Shanghai, China

ABSTRACT: Most of the present semi-submersibles adopt single hull design for their columns.
The anti-collision capability of the single hull column could fail to prevent the water pouring into the
column in case of the collision accident. A double-hull design concept is proposed in the paper. The ring
frame, the inner shell and the outer shell create an integrated system to undertake the collision. Three sets
of finite element models are built. The nonlinear numerical simulation code, MSC.Dytran, is employed
to do the simulation. Three collision scenarios are defined. The structural damages, the curves of
relationship between collision force vs. penetration, and the curves of energy dissipation are obtained, to
evaluate the anti-collision characteristics for the different column design and different collision scenarios.
Consequently, comparisons are made, to evaluate the improvement of the anti-collision characteristic for
the double-hull design, and some suggestions are also provided.

1 INTRODUCTION The new structural design concept is a double

Offshore oil fields are increasing their territory
from the sea area close to the shore, to the sea
area far away from the shore. Many deepwater
floating offshore units are under construction now
or planned in the future, and they are going to
occupy and control more space in the sea surface.
While, at the same time, with the augmentation
of international trading and shipping, ship speed
and the total amount of ships sailing in the sea
are also increasing. Both of the ships and off-
shore units ask for more space in the sea surface.
It seems that collision between the ships and off-
shore units is inevitable, although such kind of
accident seldom happened. However, the accident
of “Deepwater Horizontal” semi-submersible in
Gulf of Mexico informed everyone that the failure
of offshore unit can cause disaster, not only to
lives and properties, but also to the environment,
although the possibility of such kind of accident
has always being regarded as very small. Since the
public is becoming less tolerant towards environ-
mental spoiling, the offshore units are obliged to
ask for a stricter requirement to protect them-
selves in the accidents. An example of such an
accidental event is the collision between the vessel
“Far Symphony” and the drilling platform “West
Venture” in 2004 (Pettersen and Soegaard, 2005).
Therefore, a new kind of structural design is pro-
posed in the paper, on purpose of increasing the
anti-collision capability of the semi-submersible
column in case of serious collision accident.
hull design for semi’s column, for the column has a
possibility of being struck by the passing merchant
vessels in the accident. Some investigations on
the anti-collision capability and the strength
assessment of the semi-submersible have been
executed. Hu (2009) assessed the global strength
of the semi-submersible after the column is struck
by a passing vessel or a drifting object in severe sea
condition. Hu (2010) also made a research on the
collision scenario characteristics for the moored
semi-submersible. The assessment of the anti-
collision capability is of crucial importance, and
numerical simulation technique is always used in
the research to do the assessment.
Numerical simulation technique is the method
that has been widely used. Compared with other
method such as model test, numerical simulation
has the advantage of low-cost, repeatable analysis,
and a relatively short analyzing period. It has been
widely used to analyze the structural perform-
ance in ship collision and grounding. Zhang and
Suzuki (2006) analyzed the selected parameters
of grounding test with LS_DYNA code. Paik
(2007) developed a practical useful technique of
nonlinear finite element method for the simula-
tion of ship structure in collision and grounding.
Zhang (1999) used numerical simulation method
to verify his simplified analytical method for pre-
diction of ship collision force. Simonsen (1997)
verified a simplified grounding damage predic-
tion method with numerical simulation code
LS_DYNA. The numerical simulation method

357

MARSTRUCT.indb 357 2/18/2011 5:49:24 PM

 has also been used in buffer bow design. Kitamura
(2002), Endo (2008), and Yamada (2008) made a
series of buffer bow design and study by numerical
simulation methods. Therefore, the anti-collision
capability is assessed by the numerical simulation
method in the research.
The semi-submersible, the double hull design
for the semi column, and the striking vessel are
introduced in the paper. Three collision scenarios
are designed, which are head on collisions on the
column from the longitudinal direction, from lateral
direction and from the oblique direction. Finite
element models of the whole semi-submersible,
of a single column for both original hull design
and double hull design, and of the striking vessel
are presented. Simulation cases are determined,
and the aim of each simulation is described.
The analysis consists of two steps. The first step
is the analysis for comparing the simulation results
between the whole vessel collision and the single
column collision with original design, on purpose
of proving the single column collision analysis
can be used to do the assessment. Hence, the anti-
collision capability of the new design is assessed by
the simulations with only column model, which can
save simulation cost greatly. Therefore, the second
step of the analysis is the assessment of the anti-
collision capability of the double hull design, with
the comparison between simulation results of the
column with original hull design and double hull
design. Consequently, the anti-collision capability
of the double hull design is summarized.
The purpose of the work presented in this
paper is to propose a new method to raise the anti-
collision capability for semi-submersible’s column.
Although the double hull design may bring
inconvenience for some other aspects, they are not
considered in this paper.
2 DOUBLE HULL DESIGN AND THE
FATAL MOMENT CONCEPTS
The double hull design concept takes advantage of
the free end of the ring frame, which provides a
support of the inner shell. In the original design,
one end of the ring frame is connected with outer
shell, and the other end is free. In the double hull
design, an inner shell is designed in the column and
connected to the other end of ring frame. The inner
shell is designed between the two decks in vertical
direction, and parallel to the outer shell. The inner
shell, the ring frame and the outer shell build an
integrated protection system, which can increase
the anti-collision capability. In the original design,
the free end of the ring frame always deform with-
out constrain. In the double hull design, the ring
frame will deform with restriction. Besides, the
358
MARSTRUCT.indb 358
deck
Outer
shell
Ring
Inner
frame
shell
Original design Double hull design
Figure 1. Double hull design of the semi-submersible’s
column.
integrated protection system also creates many
separated cabins, which can operate as water tight
holds and slow down the water ingress speed in the
column in case of collision accident. The double
hull design of the column is shown in Figure 1.
Among all of the failures of the column structural
components, the failure leads to ingress of sea water
into the column is of crucial importance, for it will
reduce the platform’s buoyance and also do harm to
the balance seriously within a short time. Therefore,
a new concept, named as fatal moment, is proposed.
It means the instantaneous moment when the sea
water ingress freely. In the original design, the fatal
moment is the time when the outer shell ruptures.
In the double hull design, the fatal moment is the
time when the inner shell ruptures. The sea water
can be prevented in the watertight cabins if only
the outer shell ruptures and the inner shell is still
intact. Following assessments will be made for the
structural deformation at the fatal moment.
3 SCANTLING OF THE SEMI-
SUBMERSIBLE AND THE STRIKING
VESSEL
A semi-submersible is chosen as the struck object,
a containership is chosen as the striking vessel and
their scantling data is listed in Table 1 and Table 2,
respectively.
4 FINITE ELEMENT MODELS
Three sets of finite element models are made,
which are:
− Set1, whole semi-submersible and whole striking
vessel
− Set2, a single hull column and a rigid striking bow
− Set3, a double hull column and a rigid strik-
ing bow
2/18/2011 5:49:24 PM
 Table 1. Scantling data of semi-submersible.

Scantling data Unit Prototype data

Length m 114.07
Width m 78.68
Column height m 17.385
Column width m 15.86
Draft m 19.0
Displacement ton 51751.3
Figure 3. Finite element models set2, single hull column
with original design and the striking bow.
Table 2. Scantling data of striking vessel.

Scantling data Unit Prototype data

Length m 201
Displacement ton 36198.9
Draft m 12

Figure 4. Finite element model set3, double hull column

and the striking bow (the outer shell of column is erased
to show the inner shell).

has a continuous vertical bulkhead. The thickness

Figure 2. Finite element models set1, whole semi-
submersible and whole striking vessel.
Structural collision simulations are performed
using the commercial code MSC.DYTRAN 2005.
PATRAN version 2005r3 is used to build the finite
element mode. In set1, both the semi-submersible
and the striking vessel are all built, see Figure 2.
The finite element model consists of fine mesh part
and coarse mesh part. The fine mesh part consists
of a semi column and the bow of the striking vessel.
The rest is defined as coarse mesh part, and the fine
mesh changes to coarse mesh gradually. The struc-
ture is modeled using four nodded, quadrilateral
Belytschko-Tsay (PSHELL1) shell elements for the
finite element models. There are 319248 elements for
the semi-submersible and 261258 elements for the
striking vessel. The surface contact of DYTRAN
is defined to treat the contact occurring during the
simulation with a static friction coefficient of 0.3.
The inertia effects of the semi-submersible and the
striking vessel are defined on their coarse mesh parts
respectively. The structural performance of the
semi and the striking vessel can be exhibited clearly.
The column has different section disposals in two
directions. In transverse direction, it has a convex
shape, and the vertical bulkhead is discontinuous.
Nevertheless, in longitudinal direction, the column
of the shell is 11 mm, the thickness of the frame
web and the deck is 12.5 mm, and the thickness of
the vertical bulkhead and the stiffeners is 10 mm.
The initial striking velocity of the striking vessel
is 6 m/s, and the mass of the striking vessel is
120000 ton.
In set 2, only the semi-submersible column of
the original design and the striking bow are built.
The column is built with fine mesh elements and
Belytschko-Tsay (PSHELL1) shell elements. Other
property of the column model is the same as that in
set1. The column has a fixed boundary condition
at both sides of the upper and lower surfaces. The
striking bow is defined with rigid property and it
strikes the column with velocity of 6 m/s. The finite
element models of set2 are shown in Figure 3. The
finite element models in set3 are similar to those
in set2, only with a double hull design for the semi
column. The finite element models of set3 are
shown in Figure 4.
5 COLLISION SCENARIOS, COMPARISON
METHOD AND SIMULATION CASES
DEFINITION
Three collision scenarios are defined. The striking
vessel strikes the semi-submersible on its column
with three different directions, which are collision
in longitudinal direction, in lateral direction and in

359

MARSTRUCT.indb 359 2/18/2011 5:49:25 PM

 oblique direction. The scenario definition is shown
in Figure 5.
The analysis is executed in two steps:
− The first step is the comparison between the results
of finite element model set1 and those of set2.
− The second step is the comparison between the
results of finite element model set2 and those of
set3.
The purpose of the comparison between the
simulation results of set1 and those of set2 is
to determine whether the simulation of set2 is
conservative. Although the simulation for set1 can
obtain the accurate structural performance, but its
cost on computational time is large quite a while.
If the anti-collision characteristics of the single hull
column are conservative and are able to represent
those of whole vessels collision, the simulation
for only column will be used to assess the anti-
collision characteristics of the double hull design.
The purpose of comparison between the simulation
results of the second step is to assess the anti-collision
characteristics of the double hull design, provided
that the aim of the first step analysis is acquired.
semi
Longitudinal
Nine simulation cases are defined, according to
the two analysis steps. Simulation case 101, 102 and
103 are defined for the collisions between the whole
semi-submersible and the striking vessel. Simulation
201, 202 and 203 are defined for the collisions
between single hull column with original design and
the rigid striking bow. Simulation 301, 302 and 303
are defined for the collisions between double hull
column and the rigid striking bow. The simulation
cases definition are presented in Table 3.
6 SIMULATION RESULTS
AND ASSESSMENT
6.1 Results comparison for the simulations
of whole vessels and original design
Nine numerical simulations are completed and
the results of structural deformations, curves
of collision force vs. penetration, and curves of
distortion energy vs. penetration are obtained. The
structural deformations of the semi-submersible
column and of the striking vessel bow in simulation
case 101 are picked out as example and are shown
in Figure 6. The comparisons of curves of colli-
sion force vs. penetration for the three pairs are
presented in Figure 7, and the comparison of
curves of distortion energy vs. penetration for the
three pairs are presented in Figure 8.
It is indicated in Figure 8 that the curves of
distortion energy in each comparison are close

Oblique

Lateral

Figure 5. Collision scenario definitions.

Table 3. Simulation cases definition.

striking bow shell
Simulation Striking
case number direction Description

101 Longitudinal Collision between

whole vessels
102 Lateral
103 Oblique
201 Longitudinal Collision analysis for
original design
202 Lateral
203 Oblique
301 Longitudinal Collision analysis for
double hull design vertical bulkhead ring frame
302 Lateral
303 Oblique Figure 6. Structural deformations of the semi-column
and striking bow in case 101.

360

MARSTRUCT.indb 360 2/18/2011 5:49:26 PM

 Collision force comparison
very well, which means from the viewpoint of
3.50E+07

3.00E+07
energy dissipation, the numerical simulation for
2.50E+07 a column can represent that for the whole vessels.
Collision force (N)

2.00E+07 101
201
It can also be pointed out from Figure 7 that the
1.50E+07

1.00E+07
differences of curves of collision force vs. penetra-
5.00E+06 tion in each comparison are not large. In the com-
0.00E+00
0 1 2 3 4 5 6 7 parison between 101 and 201, the value of collision
penetration (m)

a. Case 101 and 201 force for single hull is higher than that of whole
Collision force comparison vessels, proving the case 101 brings conservative
3.50E+07
results than that of case 201. In the comparison
3.00E+07

2.50E+07
between 102 and 202, and the comparison between
Collision force (N)

2.00E+07 102
103 and 203, the curves of collision force cross
202
1.50E+07
each other. But the difference is not large.
1.00E+07

5.00E+06
On the basis of these comparisons, it can be
0.00E+00
0 1 2 3 4 5 6
summarized that the simulation of a rigid bow
penetration (m)
striking the column is able to bring a conservative,
b. Case 102 and 202 at least a similar result for that of the simulation
2.50E+07
Collision force comparison
of the whole vessels collision. Thus, in the follow-
ing analysis, the structural performance and anti-
2.00E+07
collision characteristic of the double hull design is
Collision force (N)

1.50E+07
103
203
assessed by the simulations for a single column and
1.00E+07
rigid bow.
5.00E+06

0.00E+00
0 1 2 3
penetration (m)
4 5 6
6.2 Assessment of anti-collision characteristic
c. Case 103 and 203 for the double hull design
Figure 7. Comparison of curves of collision force vs. As introduced in section 2, fatal moment is a
penetration for the first step analysis. factor of crucial importance for the anti-collision
capability of semi-submersible. Values of fatal
moment for simulation cases from 201 to 303
Energy dissipation comparison are picked out from simulation results, and the
9.00E+07

8.00E+07
values of corresponding penetration can also be
7.00E+07
determined. These data is listed in Table 4. In
Distortion energy (J)

6.00E+07

5.00E+07 101
201
Table 4, rupture time means the moment when the
4.00E+07

3.00E+07
outer shell or inner shell ruptures. For original shell
2.00E+07

1.00E+07
design, rupture time of the outer shell is the fatal
0.00E+00
0 1 2 3 4 5 6 7
moment. Nevertheless, for double hull design, the
Penetration (m)
fatal moment is the rupture time of the inner shell.
a. Case 101 and 201 For the comparison between single hull design
8.00E+07
Energy dissipation comparison
and double hull design, the comparisons for col-
7.00E+07 lision force and distortion energy are presented in
6.00E+07
Figure 9 and Figure 10. In Figure 9, the curves of
Distortion energy (J)

5.00E+07

4.00E+07
102
202 collision force vs. penetration for the corresponding
3.00E+07

2.00E+07 simulation cases are presented. In Figure 10, the

1.00E+07

0.00E+00
curves of distortion energy vs. penetration for the
0 1 2 3 4
Penetration (m)
5 6 7
corresponding simulation cases are presented.
b. Case 102 and 202 It has been emphasized in the paper that the fatal
Energy dissipation comparsion
moment is the crucial important time to assess the
6.00E+07

5.00E+07
anti-collision characteristic of the semi’s column,
the comparison is made on the results at the fatal
Distortion energy (J)

4.00E+07

3.00E+07
103
203
time. Although the simulations continue to a large
2.00E+07 penetration state, the assessment only adopts the
1.00E+07
values at the fatal time and the rest of the results
0.00E+00
0 1 2 3
Penetration (m)
4 5 6 7 are used for reference. Because the distortion
c. Case 103 and 203 energy is easy to compare, the energy dissipation
by structural components in each simulation case
Figure 8. Comparison of curves of distortion energy vs. is listed in Table 5, and a summary comparison is
penetration for the first step analysis. presented in Table 6. The structural deformation

361

MARSTRUCT.indb 361 2/18/2011 5:49:27 PM

 Table 4. Values of fatal moment and corresponding penetration.

Simulation case Shell Rupture time (s) Fatal moment (s) Penetration (m)

201 Outer shell 0.229 0.229 1.355

202 0.305 0.305 1.800
203 0.372 0.372 2.198
301 Outer shell 0.191 0.318 1.842
Inner shell 0.318
302 Outer shell 0.254 0.350 1.845
Inner shell 0.350
303 Outer shell 0.320 0.570 3.304
Inner shell 0.570

Collision force comparison Energy dissipation comparison

5.00E+07 1.60E+08
4.50E+07
1.40E+08
4.00E+07
3.50E+07 1.20E+08

Distortion energy (J)

Collision force (N)

3.00E+07 1.00E+08
201
2.50E+07 201
301 8.00E+07
301
2.00E+07
6.00E+07
1.50E+07
1.00E+07 4.00E+07
5.00E+06
2.00E+07
0.00E+00
0 1 2 3 4 5 6 0.00E+00
Penetration (m) 0 1 2 3 4 5 6
Penetration (m)

a. Case 201 and 301

a. Case 101 and 201
Collision force comparison Energy dissipation comparison

3.50E+07 1.20E+08

3.00E+07 1.00E+08

2.50E+07
Distortion energy (J)

8.00E+07
Collision force (N)

2.00E+07 202 202

302 6.00E+07 302
1.50E+07
4.00E+07
1.00E+07

5.00E+06 2.00E+07

0.00E+00 0.00E+00
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Penetration (m) Penetration (m)

b. Case 202 and 302 b. Case 102 and 202

Collision force comparsion Energy dissipation comparsion

3.00E+07 9.00E+07

8.00E+07
2.50E+07
7.00E+07
Distortion energy (J)
Collision force (m)

2.00E+07 6.00E+07

203 5.00E+07 203

1.50E+07 303 303
4.00E+07

1.00E+07 3.00E+07
2.00E+07
5.00E+06
1.00E+07

0.00E+00 0.00E+00
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Penetration (m) Penetration (m)

c. Case 203 and 303 c. Case 103 and 203

Figure 9. Comparison of curves of collision force vs. Figure 10. Comparison of curves of distortion energy
penetration for the first step analysis. vs. penetration for the first step analysis.

modes of the structural components of semi’s semi’s column very much, which means the semi’s
column for double hull design are presented in column can improve its survivability in case of a
Figure 11. serious collision accident. Besides, from the values
in Table 6, it can also be indicated that the scenario
Percentage = (E1−E0)/E0 × 100% (1) when the striking vessel hits the semi’s column in
the longitudinal direction is the most vulnerable
It is indicated from the Figure 9 that the colli- situation, for the energy dissipated by the semi-
sion force of double hull design is bigger than that submersible is the smallest at the fatal moment
of original design from the beginning of the colli- among all of the three scenarios.
sion course. Form Figure 10, it can also be pointed In the original design, the outer shell and the ring
out that the double hull design absorbs more dis- frame of column deform separately, though one
tortion energy than that of the original design. The side of the ring frame is connected with the outer
amount of energy dissipation increase presented in shell. The outer shell has a large space to deform,
Table 6 demonstrates that the double hull design and its major deformation mode is membrane
benefits the energy dissipation capability of the elongation and rupture. The major deformation

362

MARSTRUCT.indb 362 2/18/2011 5:49:27 PM

 Table 5. Energy dissipation by structural components at fatal moment.

Total Energy Energy Energy Energy Energy Energy

Fatal energy dissipated dissipated dissipated dissipated dissipated dissipated
Simulation moment dissipated by outer by inner by frame by vertical by deck by stiffener
case (s) (MJ) shell (MJ) shell (MJ) (MJ) bulkhead (MJ) (MJ) (MJ)

201 0.229 14.485 7.082 / 2.519 3.591 0.444 0.849

202 0.305 18.511 8.456 / 2.104 4.921 2.185 0.844
203 0.372 15.210 8.274 / 4.127 0.066 0.768 1.975
301 0.318 24.612 8.256 3.392 5.146 6.358 0.460 1.000
302 0.350 30.082 12.684 3.741 5.577 5.193 1.772 1.114
303 0.570 33.653 12.438 5.406 10.412 0.048 3.123 2.225

Table 6. Energy comparison at fatal moment. inner shell are membrane elongation and rupture.
The deformation modes of the ring frame and
E0 (MJ) E1 (MJ) Percentage vertical bulkhead are folding and buckling, which
are identical to those of original design. However,
Longitudinal collision
201 vs. 301 14.485 24.612 69.9
the incorporation of the inner shell changes the
Lateral collision
original design’s free deformation mode. The inner
202 vs. 302 18.511 30.082 62.5 shell not only dissipates part of the energy, but
Oblique collision also helps the integrated system to dissipate more
203 vs. 303 15.210 33.653 121.3 energy, for it delays the semi-submersible’s fatal
moment. From the values in Table 5, it is clearly
Note: In Table 6, E0 is total distortion energy by original show that the outer shell, the ring frame and the
design, E1 is total distortion energy by double hull vertical bulkhead dissipate more energy than
design, and the percentage is calculated by equation (1). those in original design, especially the ring frame.
The percentage value demonstrates the improvement The energy dissipation capability of different
effect of the double hull design, compared with that of structural components is described as following:
the original design.
− The energy dissipated by the ring frame increases
quite a while from 2.519MJ to 5.146MJ in case
201 to case 301, from 2.104MJ to 5.577MJ in case
202 to 302, and from 4.127MJ to 10.412MJ in
case 203 to case 303. In all of the three scenarios,
the energy dissipated by the ring frame increase
more than 100 percent.
− The outer shell also enhances its energy
dissipation capability for the double hull design
cases, but the augmentation is not as obvious
as that by the ring frame. For the three pairs of
comparisons, the enhanced energy dissipation
capability of the outer shell is 16.6% in case 201
and 301, 50% in case 202 and 302, and 50.3% in
case 203 and 303. The reason for the outer shell
can dissipate more energy is the delay of the fatal
moment and the increase of penetration.
− The inner shell contributes the energy dissipation
of 3.392MJ in case 301, 3.741MJ in case 302,
Figure 11. Structural deformations of the column struc- and 5.406MJ in case 303. The amount of energy
tural components in case 301 (from left to right: whole
column, inner shell, ring frame, and vertical bulkhead).
dissipated by inner shell is similar to augmented
amount of energy dissipated by outer shell.
− The vertical bulkhead plays an important role in
mode of the ring frame and the vertical bulkhead the energy dissipation for the collision scenario in
is folding and buckling. In double hull design, the longitudinal direction and lateral direction. The
outer shell, the inner shell and the ring frame build double hull design helps the vertical bulkhead
an integrated system to undertake the impact. increases its energy dissipation capability from
The deformation modes of the outer shell and the 3.591MJ to 6.358MJ in case 201 and 301, and it

363

MARSTRUCT.indb 363 2/18/2011 5:49:29 PM

 is an obvious augmentation. In case 202 and 302,
the increase of energy dissipation is only from
4.921MJ to 5.193MJ, and the augmentation is
not large. For oblique collision cases 203 and
303, the vertical bulkhead dissipates little energy.
− The energy dissipated by the deck has an obvious
augmentation in case 303 compared with that in
case 203. It may be the reason that the inner shell
connects to the deck and bring the deck involved
into the folding deformation, which helps the
column dissipates more energy.
− The energy dissipated by the stiffeners are of lit-
tle amount in collision scenarios of longitudinal
and lateral directions. But in oblique collision
scenarios, the stiffeners dissipate more energy,
but the augmentation amount is not large.
7 CONCLUSIONS
With the numerical simulation analyses for semi-
submersible collision scenarios of whole vessels, of
original design column and of the double hull design
column, and also the corresponding comparison
and the assessment, following conclusions can be
drawn.
− A new concept of double hull design for the semi-
submersible’s column is proposed in this paper,
and its effect of increasing the anti-collision
capability is assessed. The double hull design has
an obvious benefit to increase the fatal moment
for the semi-submersible’s column. The double
hull design has an advantage for the anti-collision
capability than that of the original design.
− From viewpoint of assessing anti-collision capa-
bility of semi-submersible’s column, numerical
simulation of the whole vessels can be replaced
by that of single column struck by a rigid bow.
Computational cost can be saved greatly.
− Although the double hull design of the
semi-submersible’s column may have some
disadvantage on the operation, layout and
building cost, it is able to increase the anti-
collision capability obviously. The adoption
of the double hull design concept should be
determined by a compromise of different aspects.
This will be considered in the future work.
364
MARSTRUCT.indb 364
ACKNOWLEDGEMENTS
The work embedded in this paper is supported by
China National Scientific and Technology Major
Project (Grant No. 2008ZX05026–006). The finan-
cial support is acknowledged gratefully.
REFERENCES
Endo, H. & Yamada, Y. 2008. Performance of buffer bow
structure against collision-on the effect in preventing
oil outflow. Marine Technology, 45(1): 54–62.
Hu Zhiqiang, Yang Jianmin, & Xiao Longfei. 2009.
“Global Strength Assessment for Semi-Submersible
Column after Supply Vessel Collision Accident”,
OMAE2009-79082. Proceedings of the 28th Inter-
national Conference on Ocean, Offshore and Arctic
Engineering. May 31–June 5, 2009, Honolulu, USA.
Hu Zhiqiang, Liu Zhenhui, & Jørgen Amdahl. 2010,
“Collision Character Research for Semi-Submersible
through Model Test, Simplified Analytical and
Numerical Simulation Method”, OMAE2010-20253,
Proceedings of the 29th International Conference on
Ocean, Offshore and Arctic Engineering, June 6–11,
2010, Shanghai, China.
Kitamura, O. 2002. FEM approach to the simulation of
collision and grounding damage. Marine Structures,
15: 403–428.
Paik, J.K. 2007, Practical techniques for finite element
modeling to simulate structural crashworthiness in
ship collisions and grounding (Part I: Theory). Ships
and Offshore Structures, 2007, Vol.2, No.1, pp69–80.
Pettersen, E. & Soegaard, L.M. 2005. “A study of
a Collision Incident Evaluated against Ruling”,
OTC2005, Huston, U.S.A, paper 17156.
Simonsen, B.C. 1997. Mechanics of Ship Grounding.
Ph.D. thesis, Department of Naval Architecture
and Offshore Engineering, Technical University of
Denmark.
Yamada, Y. & Endo, H. 2008. Experimental and numeri-
cal study on the collapse strength of the bulbous bow
structures in oblique collision. Marine Technology,
45(1): 42–53.
Zhang, S.M. 1999. The mechanics of ship collisions. Ph.D
Thesis, Department of Naval Architecture and Ocean
Engineering, Technical University of Denmark.
Zhang, A. & Suzuki, K. 2006. Dynamic FE simulations
of the effect of selected parameters on grounding
test results of bottom structures, Ships and Offshore
Structures, 2006, Vol.1, No.2. 117–125.
2/18/2011 5:49:30 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Effect of crushable bow on the overall crashworthiness

in ship collision

Y.T. Huang, K.P. Wu, H.L. Chien, C.M. Chou & K.C. Tseng
Department of Design, CSBC Corporation, Taiwan, Kaohsiung, Taiwan, R.O.C.

C.F. Hung & C.L. Chang

Department of Engineering Science and Ocean Engineering, National Taiwan University, Taipei, Taiwan, R.O.C.

ABSTRACT: Many researches on damage of bow structure and hull structures in ship collision have
been carried out. Considering the failure conditions of striking and struck vessels depends on the overall
longitudinal stiffness of bow structures and the crash resistance of struck hull structures. This study inves-
tigated the effects of damage conditions of different bow structures. The vessels designed and constructed
by CSBC Corporation, Taiwan were selected as simulation scenario, i.e. a 203,000 tons bulk carrier was
struck by an 8,240 TEU container ship. The critical speed is defined as the minimum initial velocity of
striking vessel to cause the rupture of inner hull of struck double hulls. Different bulbous bow structure
with same shape but different longitudinal and transverse stiffness were considered. The nonlinear tran-
sient dynamics analysis of both ships in collision was analyzed. The energy dissipation and reaction force
of different type of bow structures were compared.

1 INTRODUCTION very expensive, and it was performed only in some

Although the velocity of ship is not very fast, the
huge ship mass possesses enormous momentum
and kinetic energy. During the collision the kinetic
energy will redistributed in short time, portion of
kinetic energy of striking ship will be transferred to
struck vessel, and the velocity of striking ship will
be reduced very quickly and it also induces very
high impact loading between two colliding vessels
that may cause serious damage on striking vessel
and struck structures. The catastrophe of oil pollu-
tions on the sea may take place consequently.
In order to prevent the calamity induced by ship
collision and grounding the double hull and mid-
deck design concepts were requested to avoid the oil
escaping from tank while the ship hull was struck
since last 20 years. The impact force of ship struc-
tures subjected to collision and grounding became
an important issue; and numerous researches
about the crashworthiness were carried out in dif-
ferent approach, e.g. theoretical, experimental, and
numerical approaches (Jones & Jouri 1987, Pedersen
et al. 1993, Pedersen & Zhang 2000, Wang 2000,
Kitamura 2002, Lehmann & Peschmann 2002).
Because of the complexity of ship structures
the most of damages of structural components are
multi-damage modes; the small scale experiment
are not easy to predicate the damage condition
of structures in real ship collision. The large scale
experiments for ship collision and grounding are
special projects used as benchmarks for other
researches (Rodd 1996, Wevers & Vredevelt 1999,
Ehlers et al. 2008).
Recently, the analysis of structural damage in
ship grounding and collision analyzed by nonlin-
ear Finite Element (FE) method becomes a prin-
ciple approach in past years. Nevertheless, it needs
large computation resource and the establishment
of FE model of collided vessels and the evaluation
of the calculated results needs huge man power.
The damages of hull structures during collision
can be classified into four fundamental damage
modes, which are the stretching mode, the tearing
mode and the penetration mode of plates, as well
as the denting mode for single girder and crossing
girders. The reasonable simplified approximation
method to predicate the energy dissipation and
impact force for damaged structures were derived
in last three decades. The overall energy dissipation
and impact force of a damaged structures can be
estimated by assembly of these fundamental fail-
ure mechanisms (Pedersen & Zhang 2000, Wang
2000, Hung & Chen 2007, Yamada & Pedersen
2008, Paik & Seo 2007).
In case of a perpendicular collision of two
vessels, the momentum before and after collision is
M2V2 = (M1 + M2 )V (1)
where M1 = mass of struck vessel; M2 = mass of
striking vessel; V2 = initial velocity of striking ship

365

MARSTRUCT.indb 365 2/18/2011 5:49:30 PM

 and V = combined velocity of both vessels after
collision.
The combined velocity of both vessels after col-
lision becomes:
M2
V V2 (2)
M 2 M1
The lost kinetic energy is:
1 1
ΔE M 2V22 (M + M )V 2
2 2
1 M1
= V22M 2 (3)
2 M1 M 2
The lost kinematic may be transferred to the
internal energy absorbed by the damaged structures,
and scattered in to surrounding fluid. If the dam-
age on a struck vessel is based on an assumption of
a rigid bow collision, and the energy absorbed by
striking bow is not considered, the damage extent to
the struck side of a vessel is overestimated. On the
contrary, the damage of striking bow based on the
assumption of rigid struck body the absorbed energy
of damaged structural parts is underestimated. The
failure conditions of damaged structures in ship col-
lision are depending on sharpness and the longitu-
dinal stiffness of striking bow and the resistance of
struck hull structures. The crushable striking bow
and struck hulls are considered simultaneously, the
final result may allow for a more optimally designed
vessel. Gooding (2001) provides a simple closed-
form solution for modeling the collision resistance
of the bulbous bow. Yagi et al. (2009) examined
the buffer bow characteristic (energy absorbing
capability) of sharp entrance angle bow, the results
show its characteristic was superior to that of a
conventional bulbous bow, since much more energy
was dissipated by the plastic deformation of strik-
ing and struck ships until the inner shell of struck
ship ruptured. Yamada & Endo (2008) investigated
the collapse strength and the mechanism of the
bulbous bow structure in case of an oblique colli-
sion by nonlinear FE analysis corresponding to the
experiments. Endo & Yamada (2008) introduced
the critical striking velocity and the critical collision
angle as the measures to represent the performance
level of buffer bow. Yamada et al. (2008) examined
the effects of buffer bow structure in ship-ship col-
lision. Karlsson et al. (2009) proposed a conceptual
deformable inner barrier design with the aim of
improving the collision safety during a collision into
the side of a ship by facilitation of a deep intrusion
before penetration. Those critical events are defined
as the threshold whether the inner hull of struck
tanker is ruptured or not. The effectiveness of buffer
bow is highlighted by focusing on the performance
how much the risk of oil outflow is decreased. This
366
MARSTRUCT.indb 366
paper investigated the strength and crashworthiness
of different longitudinal stiffness bow structures of
striking vessel. The compromise of energy absorbed
by striking bow and struck hull vessel should be
considered for the optimal design for ship-ship
collision.
The effect of stain rate on yield strength is mod-
eled using the Cowper & Symonds (1957) strain
rate model.
1/ n
σ ′0 ⎛ ε ⎞
= 1+ ⎜ ⎟
⎝ D⎠
(4)
σ0
where ε = strain rate; σ ′0 = dynamical yield stress;
σ0 = static yield stress; D and n are the strain rate
parameters.
2 CRASHWORTHINESS OF DIFFERENT
BOW STRUCTURES IN SHIP COLLISION
During last 20 years, in order to prevent the calam-
ity induced by ship collision and grounding the
double hull design concepts were requested to
avoid the oil escaping from tank while the ship
hull was struck. The x-type double hull structure
(Klanac et al. 2005) in this study is assumed as the
ship side of struck vessel.
The model of a striking ship with bulbous bow
and the struck double hull ship in 90-deg collision are
shown in Figure 1. Only part of side shell was taken
as analysis model, and the dimensions are shown in
Figure 2. The span of x-web plate is 800 mm, and
the span of longitudinal web plate is 2,000 mm.
The depth of double hull is 1,200 mm. The analysis
domain of double hull is shown on Figure 3. Four
sides of double hull model are set fixed, i.e. in the
collision process, the relative velocity between strik-
ing bow and struck hull is considered. The material
is assumed bilinear model show on Figure 4, and the
material constants are listed in Table 1.
Figure 1. Analysis model of bow-side collision.
2/18/2011 5:49:30 PM
 (a) FEM model (the outside shell is removed)
Figure 2. X-type web double hull structure and its
dimensions.

(b) Original bow design (c) Bow type A

(d) Bow type B (e) Bow type C

Figure 3. Analysis domain of wall structure.
Figure 5. Four alternative bow structures.

show the structural model. In four alternative bow

structures the shapes are kept the same, and only
the central bulkhead and front transverse structures
are varied to adjust the relative longitudinal and
transverse stiffness of bow structures. The alterna-
tive bow structures shown in Figure 5 are follows.
Bow type A: partial centerline web is removed.
Bow type B: modified from type A. In the front
peak tank, two longitudinal girders and partial
centerline web are removed, the transverse decks is
replaced by a longitudinal girder, to maintain the
Figure 4. Bilinear stress-strain material model. same strength against hydrodynamic load two ver-
tical girders are added. The front transverse bulk-
Table 1. Material constants of analysis models. head is added also.
Bow type C: modified from type B. The front trans-
Elastic modulus (E) 210 GPa verse web is removed.
Density 7860 kg/m3
Poisson’s ratio 0.3 From the view points of longitudinal stiffness, the
Yield stress 300 N/mm2 order is: original design > type A > type B > type C.
Tangent modulus 250 MPa
Rupture strain 0.15
2.2 FE analysis model
Frictional coefficient 0.25
Strain rate parameters D = 40.4 sec−1, n = 5 In the FE analysis model the Belytschko-Tsay shell
element and Hughes-Liu beam element were used.
The element size was set 200 mm, the total number
of element are 39,000 and 58,400 for double hull
2.1 Different bow structure
and ship bow structures, respectively. Although the
An original bow structure of 8,240 TEU container velocity of collision is not high, the strain rate of
ship designed by CSBC was selected as reference damaged structures has same effects on damage
base, shown in Figure 5. The Figure 5(a) show the process, the Cowper & Symonds (1957) strain rate
FE model of inner structures. The other figures rule was taken into consideration.

367

MARSTRUCT.indb 367 2/18/2011 5:49:32 PM

 2.3 Analysis items in this study is the energy absorbed by the ruptured
structures. For comparison purpose the results of
In the simulation of ship collision the density of
rigid bow is shown in the same table.
aft transverse bulkhead of bow model shown in
The original design and bow type A have higher
Figure 1 was adjusted so that its mass is about the
longitudinal stiffness, and have higher indentation
same as the striking ship, 100,200 tons. The initial
depth, reaction force and energy dissipation than
velocity of striking vessel is 5 m/s moving perpen-
bow type B and type C, when the outer plate rup-
dicularly toward side of a struck vessel, i.e. the ini-
tured. Nevertheless, they penetrate the inner plate
tial kinetic energy is about 1,250 MJ. The Impact
of double hull more early, and the total reaction
response analysis of collided structures was per-
force and energy dissipation become smaller. The
formed with LS/DYNA code and the analysis of
bow type B has relative lower reaction force but
collision duration was set for 1.5 seconds.
higher energy dissipation than other three types.
Figure 7 and Figure 8 show the total reac-
2.4 Analysis results and discussions tion force and energy dissipation of double hull
struck by five different bow structures vs. indenta-
Figure 6 shows damage conditions of bow struc-
tion depth, respectively. Compared with the case
tures for original design, bow type B and bow type
of crushable bow, when the outer plate ruptured
C. Sub- Figure 6 (a) to (c) show the deformed shape
in case of rigid bow, the indentation depths and
of outer shells. Bow type B and type C have relative
smaller longitudinal stiffness, the buckling occurred
in low part of outer shell. The front shell of original
design has smaller crushing range. Sub-Figure 6 (d) Table 2. Comparison of the results of impact response
to (f) show the damage conditions of inner structural analysis between different bow structures.
parts. The damage modes of girders are stretching,
twisting, buckling, and tearing. The longitudinal Energy
Structure Indentation Reaction dissipation
bulkheads are buckled and folded. type depth (m) force (kN) (kJ)
Table 2 shows the comparison of the indenta-
tion depth, reaction force of hull boundary and the When outer plate of struck hull rupture
total energy dissipation of double hull structures Original 1.515 5.361 × 104 7.006 × 104
struck by four bow structures when the outer plate Bow type A 1.535 5.491 × 104 7.188 × 104
and the inner plate were ruptured. When striking Bow type B 1.490 4.880 × 104 6.603 × 104
bow impacted on the ship side of struck vessel, Bow type C 1.400 4.596 × 104 5.616 × 104
faceplate of struck ship and striking bow struc-
When inner plate of struck hull ruptured
tures will also be ruptured, the energy dissipation
Original 2.585 5.156 × 104 1.288 × 105
Bow type A 2.700 4.530 × 104 1.406 × 105
Bow type B 2.801 5.578 × 104 1.302 × 105
Bow type C 1.615 5.921 × 104 5.137 × 105

(a) Outer shell of original (d) Inner structure of

bow original bow

(b) Outer shell of bow type B (e) Inner structure of

bow type B

(c) Outer shell of bow type C (f) Inner structure of

bow type C

Figure 6. Deformed shape of different damaged bow Figure 7. Total reaction force of double hull struck by
structures. different bow vs. indentation depth.

368

MARSTRUCT.indb 368 2/18/2011 5:49:34 PM

 Figure 9. Midship section of 203,000 DWT bulk carrier.
Figure 8. Energy dissipation of double hull struck by
different bow vs. indentation depth.

energy dissipations are lower; the reaction forces

are much higher than crushable bow before outer
plate ruptured, and thereafter it is reduced to
about the same level as crushable bow cases. No
internal energy absorbed by rigid bow during col-
lision, the energy dissipation before outer plate
ruptured has small difference from crushable bow
cases; thereafter the energy dissipation becomes
much lower.

3 CRASHWORTHINESS OF REAL SHIP

STRUCTURES IN SHIP COLLISION

The midship section of a 203,000 DWT bulk car-

rier and the fore part of centerline section of an
8,240 TEU container ship are shown in Figure 9
Figure 10. Centerline section of 8,240 TEU container
and Figure 10, respectively. ship.

3.1 FE model
The FE model of bow structure of container is
similar to Figure 5. A cargo hold space was selected
as analysis domain for bulk carrier, and the dimen-
sions of its FE-model are shown in Figure 12.
The model of a striking bow and the struck side
hull is shown in Figure 9 and Figure 10. The mass
of container ship excluding model mass was distrib-
uted on rear bulkhead of model; the mass of bulk
carrier excluding the model was distributed on the
two side transverse bulkheads. The material con-
stants of the FE-model are the same as Table 1.

3.2 Analysis conditions

In the analysis case the displacements of striking
and struck ship are 100,200 tons and 106,500 tons,
respectively. The striking point is assumed on the
center of rotation of struck ship. Figure 11. Model of 203,000 DWT bulk carrier.

369

MARSTRUCT.indb 369 2/18/2011 5:49:36 PM

 Figure 12. Analysis model of ship collision.
Figure 13. Time history of struck bulk carrier (lower
curves) and striking bow (upper curves).
No constrained boundary conditions are speci-
fied for the analysis model. The initial velocity of
striking vessel is 5 m/s moving perpendicularly
toward side of a struck vessel and the initial kinetic
energy of striking ship is about 1,250 MJ. The
Impact response analysis of collided structures was
performed with LS/DYNA code and the analysis
of collision duration was set for 2.0 seconds.
3.3 Analysis results and discussions
The curves in lower part of Figure 13 show the
time history of bulk carrier struck by different
type of bow structures, the curves in upper part are
for the striking bows. Two seconds after collision
the striking and struck ships are still in decelerat-
ing and accelerating. The relative velocity between
striking and struck ships in cases of bow type B
and C is reduced faster than for cases of initial
design and bow type A. i.e. the energy absorbed by
damaged structures for the bow type B and C are
higher than other two.
Analysis collision duration for cases of bow type
A and bow type C were extended to 5.2 seconds.
Figure 14 shows the time history of striking
370
MARSTRUCT.indb 370
Figure 14. Time history of struck and striking ship for
bow type A and C.
Figure 15. Total reaction force of bulk carrier struck by
different types of bow vs. indentation depth.
and struck ships for cases of bow type A and C.
In case of bow type A the velocity of both ship
approached to about the same at about 4 seconds,
it is about 2.4 m/s, from Eq. (3) the final velocity
should be 2.42 m/s. On the contrary, in case of bow
type C the struck ship moved faster than striking
ship after about 2.5 seconds, and both ships are
separated afterward.
The total reaction force and energy dissipation
of different type of bow structures versus time are
shown in Figure 15 and Figure 16. The case for
original bow and both transverse bulk carrier with
fixed boundary condition is shown in the same
figures. The boundary condition on two side bulk-
heads of struck ship with fixed constraints or free
condition but including ship mass has no difference
in first 0.6 seconds, thereafter the case with fixed
boundary condition has little lower reaction force
but slightly higher energy dissipation.
2/18/2011 5:49:37 PM
 (a) Shell of bow structure (c) Inner structure of bow

(b) Struck outside shell (d) Struck inner structure

Figure 18. Damage of structure parts at time

t = 1.99 seconds (Bow type B).
Figure 16. Energy dissipation of double hull struck by
different types of bow vs. indentation depth.
Table 3. Comparison of the results of impact response
analysis between different bow structures.

Energy
Structure Indentation Reaction dissipation
type depth (m) force (kN) (kJ)

When outer plate of struck hull rupture

(a) Shell of bow structure (c) Inner structure of bow
Original bow 0.509 6.718 × 104 1.182 × 105
(fixed BC)
Original 0.513 6.693 × 104 1.155 × 105
Bow type A 0.529 7.144 × 104 1.216 × 105
Bow type B 0.814 7.519 × 104 2.386 × 105
Bow type C 1.032 7.639 × 104 3.119 × 105
When inner plate of struck hull ruptured
Original bow 1.636 7.561 × 104 3.211 × 105
(b) Struck outside shell (d) Struck inner structure (fixed BC)
Original 1.773 7.842 × 104 2.896 × 105
Figure 17. Damage of structure parts at time Bow type A 1.966 8.650 × 104 3.394 × 105
t = 1.99 seconds (Original bow structure). Bow type B – – –
Bow type C – – –
The total reaction force and energy dissipation * Note: the inner hull has no rupture.
for all cases have no difference before the outside
shell ruptured. The original design and bow type
A have higher longitudinal stiffness than type B t = 1.99 seconds, the original bow has higher stiff-
and C. During collision the cases of original bow ness and has only some permanent deformation, the
and type A cause the struck outside shell to dam- inner hull of struck ship is ruptured. In case of the
age more early than cases of bow type B and C. bow type B the forward part of structure has buck-
From eq.(3) the lost kinematic energy should be led; the bow structure becomes a larger flat contact
654 MJ, the energy dissipation of type B and C are surface to impact the struck outside shell; and the
near to 600 MJ at time 2 seconds. struck outside shell has large deformation with small
The forward part of bow type B and bow type ruptured parts, the struck inner hull is not ruptured.
C have smaller stiffness. While the collision occurs, Table 3 shows the comparison of the indentation
the forward part is buckled and more bow struc- depth, total reaction forces of hull boundary and
ture range impact the side shell; the reaction force the total energy dissipation of struck double hull
and energy dissipation are much larger than origi- and striking bow structures when the outside plate
nal bow and type A when the struck outside shell and the inner plate of double hull were ruptured.
is ruptured. For comparison purpose the results in case of both
Figure 17 and Figure 18 show the damage con- transverse bulkhead of bulk carrier setting fixed
ditions of original bow and bow type B at time struck by original design is shown in the same table.

371

MARSTRUCT.indb 371 2/18/2011 5:49:40 PM

 The previously described two different boundary
conditions for original bow structures there were
only small effects on the final damage conditions.
In cases of bow type B and type C have higher reac-
tion force and energy dissipation than the cases of
original bow and type A. The damaged forward
bow part of type B and C absorbed higher quantity
of energy; and the inner hull of struck ship keeps
no rupture after collision in this study case.
4 CONCLUSIONS
1. Compared with the case of crushable bow, when
the outside plate of hull structure is ruptured, the
reaction force for rigid bow case is significantly
higher than crushable bow cases; the energy dis-
sipation is only slightly lower. After rupture of
outside plate, the reaction force is reduced to
about the same level as crushable bow case; and
the energy dissipation becomes much lower.
2. The stiffness of original bow design and bow
type A are larger than types B and C. When
inner plate of double hull is ruptured their
indentation depth and energy dissipation are
lower than bow type B and bow type C; the
bow type B has lowest reaction force but high-
est energy dissipation. The optimum design of
bow structure to improve the overall crashwor-
thiness in ship collision needs further studies.
3. If the damage on a struck hull structures is
based on the assumption of rigid striking bow,
and only the energy absorbed by struck hull is
considered, the energy dissipation during colli-
sion will be underestimated, the reaction force
and damage condition of struck hull will be
overestimated.
4. If the damage on a bow structures is based
on the assumption of rigid struck wall and
only the energy absorbed by bow is considered,
the energy dissipation during collision will be
underestimated.
5. Both side transverse bulkheads of struck ship
are set as fixed boundary condition or free
boundary but carrying ship mass has only
small effect on damage condition in the initial
stage after collision, but may has significant
difference in final damage conditions, because
the struck ship carries no kinetic energy if its
boundaries are fixed.
6. The forward part of bow type B and bow type C
have smaller stiffness. While the collision occurs,
the forward part of bow structure is buckled and
becomes a larger flat contact surface to impact
the side shell; the reaction force and energy dis-
sipation are much larger than original bow and
type A when the struck outside shell is ruptured;
and the inner hull of struck ship keeps no rup-
ture after collision in this study case.
372
MARSTRUCT.indb 372
REFERENCES
Ehlers, S. & Broekhuijsen, J. & Alsos, H.S. & Biehl, F. &
Tabri, K. 2008. Simulating the collision response of
ship side structures: A failure criteria benchmark study.
International Shipbuilding Progress. 55: 127–144.
Endo, H. & Yamada, Y. 2008. Performance of Buffer Bow
Structure against Collision—On the Effect in Prevent-
ing Oil Outflow. Marine Technology. 45(1): 54–62.
Gooding, P.W. 2001. Collision with a Crushable Bow.
Marine Technology. 38(3): 186–192.
Hung, C.F. & Chen, C.P. 2007. The approximate methods
to predicate the crashworthiness of ship double hull
structures. Journal Taiwan SNAME. 26(3): 139–150.
Jones, N. & Jouri, W.S. 1987. A Study of Plate Tearing
for ship Collision and Grounding Damage. Journal of
Ship Research. 31(4): 253–268.
Karlsson, U.B. 2009. Improved collision safety of ships
by an intrusion—tolerant inner side shell. Marine
Technology. 46(3): 165–173.
Kitamura, O. 2002. FEM approach to the simulation of
collision and grounding damage. Marine Structures
15: 403–428.
Klanac, A. & Ehlers, S. & Tabri, K. & Rudan, S. &
Broekhuijsen, J. 2005. Qualitative design assessment of
crashworthy structures. Int. Congress of International
Maritime Association of the Mediterranean (IMAM),
Lisboa, Portugal, 26–30, September 2005.
Lehmann, E. & Peschmann, J. 2002. Energy absorption
by the steel structure of ship in the event of collisions.
Marine Structures. 15: 429–441.
Paik, J.K. & Seo, J.K. 2007. A method for progressive
structural crashworthiness analysis under collisions
and grounding. Thin-Walled Structures. 45(1): 15–23.
Pedersen, P.T. & Valsgaad, S. & Olsen, D. & Spangenberg, S.
1993. Ship impact: bow collision. International Jour-
nal Impact Engineering. 13(2): 163–187.
Pedersen, P.T. & Zhang, S. 2000. Absorbed Energy in Ship
Collisions and Grounding-Revising Minorsky’s Empir-
ical Method. Journal of Ship Research. 44(2): 140–154.
Rodd, J.L. 1996. Observation on Conventional and
Advanced Double Hull Grounding Experiments.
In Int. Conf. On Design and Methodologies for Col-
lision and Grounding Protection of Ship: 13.1–13.13.
San Francisco, USA.
Wang, G. 2000. Behavior of a double hull in a variety
of stranding or collision scenarios. Marine Structures.
13: 147–187.
Wevers, L.J. & Vredevelt. 1999. Full Scale Ship Collision
Experiment. Delft: TNO Report.
Yagi, S. & Kumamoto, H. & Muragishi, O. &
Takaoka, Y. & Shimoda, T. 2009. A study on collision
buffer characteristic of sharp entrance angle bow
structure. Marine Structures. 22(1): 12–23.
Yamada, Y. & Endo, H. 2008. Experimental and
Numerical Study on the Collapse Strength of the
Bulbous Bow Structure in Oblique Collision. Marine
Technology. 45(1): 42–53.
Yamada, Y. & Pedersen, P.T. 2008. A benchmark study
of procedures for analysis of axial crushing of bul-
bous bows. Marine Structures. 21: 257–293.
Yamada, Y. & Endo, H. & Pedersen, P.T. 2008. Effects of
buffer bow structure in ship-ship collision. International
Journal of Offshore and Polar Engineering. 18(2):
133–141.
2/18/2011 5:49:44 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

A study on dynamic grounding of ships

Tan-Hoi Nguyen, J. Amdahl, L. Garrè & Bernt J. Leira

Department of Marine Technology, Norwegian University of Science and Technology, Trondheim, Norway

ABSTRACT: This study introduces a generalized model for the shape of the sea floor which is relevant
for ships during grounding. The sea floor is characterized as a paraboloid that allows a parametric modeling
of the sea floor topology. Nonlinear explicit Finite Element (FE) code LS-DYNA was used to simulate the
ship grounding scenarios. The simplified formulae to estimate the average grounding forces are proposed.
The analytical formulae show reasonable agreement with the results from FE simulations. A simple
computer program was used to simulate the ship response, which is assumed to be two-dimensional rigid
motion, i.e. surge, heave and pitch. The results show that the sea floor topology is indeed a key parameter
in the response analysis of a ship bottom during grounding. The penetration induced by the ship motion
is also a function of ship motion. The envelopes of shear force and bending moment which are induced
during grounding are checked against the DNV’s still water requirements. The ultimate goal of the analysis
is to allow prediction of the risk of penetration into cargo tanks with oil spill as a result, the risk of hull
girder failure and estimation of the required tug force for pulling the vessel off the ground.

1 INTRODUCTION During grounding, the contact forces induced

Although the safety requirements for sea trans-
portation become more and more strict to protect
the human, goods, properties and the environment
against accidents, incidents to ships still occur due
to many reasons, such as harsh weather conditions,
human factors and so on. One of the oil spill
catastrophic in 2009 was caused by the 26,800
dwt bulk carrier Full City, which ran aground at
Langesund off the southern coast of Norway in
heavy weather at the end of July (EMSA 2009).
Together with the application of nonlinear finite
element analysis to the study of ship grounding (to
name a few (Simonsen 1997), (Samuelides et al.
2007), (Alsos and Amdahl 2007)), a comprehensive
set of analytical formulae were also established to
estimate the contact forces and bottom damage
during grounding, see for instance (Pedersen 1997),
(Simonsen 1997), (Paik and Wierzbicki 1997),
(Hong and Amdahl 2008). Grounding problem
is highly dependent on the characteristics of sce-
nario, i.e. the seabed topolology, ship speed, envi-
ronmental condition and so on. Unfortunately,
in the literature few studies have addressed the
influence of the seabed topology. In this respect,
(Nguyen et al. 2010) generated a comprehensive
set of stranding situations by a FE code varying
the size of the seabed topology (modeled by means
of paraboloid surface) and concluded that seabed
topology is a key factor in influencing the response
of bottom structures.
by the seabed will cause surge, heave, pitch and
roll. Together with this motion, the penetration of
the seabed into the bottom hull is also significant,
especially as this penetration in turn interacts with
the contact force, inertial force and hydrodynamic
force. Alsos (2008) developed a simplified in-house
code to simulate the ships response during
grounding as a rigid body motion, in which the
grounding excitation forces are found by means
of the simplified grounding forces approach.
However, the effect of seabed topology was not
fully considered in his work.
In the light of this previous work, the present
study contributes to the understanding of the effect
of the seabed topology, initial level of penetration
and ship speed on the damage extension of the ship
bottom and on the corresponding ship response. The
bottom damage and contact forces are simulated by
FE code LS-DYNA. The seabed is assumed to be
a paraboloid, which characterizes a set of vari-
ous shapes of obstruction. It is remarked that the
present work is relevant only for large shoals which
affect much the rigid motion of ship. The analyti-
cal formulae to estimate the grounding forces are
developed based upon the assumed paraboloid
shape. The envelopes of shear force and bend-
ing moment which are induced during grounding
are also checked against the DNV’s still water
requirements. Additionally, it is also remarked that
the grounding scenario simulated is taking place
on calm water, i.e. no waves actions and flooding,

373

MARSTRUCT.indb 373 2/18/2011 5:49:45 PM

 induced by fracture in the inner bottom plating, a variation of the parameters of the paraboloid of
are accounted for in this study. The present limita- Equation 1. Different pairs of (α, β) characterize
tions will be addressed in future analyses. different seabeds. This is mainly done with reference
to curves describing the resistance of the bottom
versus penetration of the seabed. An example of
2 CASE STUDY these curves is plotted in Figure 1b, showing the
curves obtained for obstruction located at the
The ship is assumed to run aground on seabed of intersection of main girder and bulkhead during
different topologies at velocities 12, 15 and 20 knots. stranding simulations (Nguyen et al. 2010).
The seabed is assumed to penetrate initially into the
ship bottom to the desired level of penetration and x2 y2
it then moves horizontally along the ship bottom. z= 2
+ 2 (1)
a b
The bottom damage and the grounding forces are
first obtained by FE simulations. The simplified
where a2 = αLref ; b2 = βLref ; α and β are scaling
procedure to estimate the average grounding
factors (Fig. 1a) and Lref is a reference length,
forces and ship’s response during grounding is also
corresponding to the interval between adjacent
discussed in the following sections.
floors (4 m). It is also noted that only large seabed
obstructions, i.e. α or β ≥ 7, are considered in
2.1 The FE model of tanker the present study. Particularly, Figure 2 shows
the deformation modes of girders and floors at the
The ship analyzed is a 260 m shuttle tanker
level of penetration of 4.8 m for two cases, namely
consisting of six tanks, the width of the ship
(α = 1, β = 7) and (α = 7, β = 7), computed during
42.5 m, mean draught 15 m. The FE model is
a stranding simulation. It is seen that different
shown in Figure 1a with two tanks: each tank is
seabed topologies will lead to different modes of
32 m long. The LS-DYNA nonlinear explicit FE
deformation. Particularly, more structural members
code is used for simulating the grounding. It is
realized that proper boundaries should simu-
late the actual deformation of a real stranding
scenario, although they are not easily identifiable.
This requirement is particularly relevant when
large contact area exists; in these cases, the defor-
mation field is large and can possibly extend to
the neighboring compartments, whose contribu-
tion must be accounted for. In this study, the end
sections are simply assumed to rotate rigidly about
their neutral axes (i.e. no in-plane deformations and
no rigid translations of the sections are allowed, as
seen in Fig. 1a). Details of the FE model realized
can be found in (Nguyen et al. 2010).

2.2 Seabed topology Figure 1a. FE model of the considered tanker.

Grounding incidents can occur on soft seabed or
hard seabed. The corresponding damage to the
300
hull and interaction between bottom hull and α<β (0.5,0.5)
(1,1)
seabed are therefore strongly dependent on seabed 250 α>β (2,2)
topology. Particularly, if the ship is grounded on (3,3)
(4,4)
the soft seabeds, the hull girder may collapse due 200 (5,5)
Load [MN]

to shear force and bending moment exceeding its (6,6)

(7,7)
capacity. On the other hand, local crushing and 150
(8,8)
tearing of the bottom hull are concerned in hard (9,9)
(10,10)
100
grounding (Simonsen 1997). It is realized that the (2,6)
actual seabed topology is arbitrarily due to its (6,2)
50 (5,3)
complex nature. To the author’s knowledge, there outer fracture
inner fracture
are no reports of the effect of actual seabed on ship 0
0 1 2 3 4 5
grounding in the available literature. On the effort Penetration [m]
to characterize the seabed feature, a set of differ-
ent seabed obstructions is generated by means of Figure 1b. Resistance—penetration curves.

374

MARSTRUCT.indb 374 2/18/2011 5:49:45 PM

 Figure 2. Deformation of girders and floors at penetra-
tion of 4.8 m. a) (α = 1, β = 7); b) (α = 7, β = 7).

Figure 3. Deformation mode for (α = 7, β = 7) at

are involved and the grillage mode of deformation δ = 4.8 m.
activated for the circular case, i.e. (7, 7), while local
deformation mode is exhibited for the elliptical 60
case, i.e. (1, 7). More descriptions of the effect of μ = 0.0
seabed topology on the collapse mode can be found μ = 0.3
50
μ = 0.6
Longi. grounding force [MN]

in Nguyen et al. (2010).

40
2.3 Results of grounding analysis
by FE simulation
30
For each considered obstruction, i.e. pair of (α, β) in
Equation 1, the grounding scenario is simulated as 20
follows: the obstruction is first moved to penetrate
vertically into the bottom to a level of penetration 10
δ = 1.0 m or 2.0 m and it then slides along the
ship bottom with a speed of 10 m/s. The friction
0
between the obstruction and the outer bottom 0 5 10 15 20 25
plate is modeled with different friction coefficients: Grounding length [m]
μ = 0, 0.3, and 0.6. The deformation modes have
been observed in parametric studies of grounding Figure 4a. Longitudinal grounding force (Fx) for
(α = 10, β = 10) at δ = 1.0 m.
performed with LS-DYNA. Particularly, Figure 3
depicts the damage of bottom structures for the
shape (α = 7, β = 7) at the initial level of penetration 75
δ = 2.0 m. μ = 0.0
μ = 0.3
Vertical grounding force [MN]

70
Figures 4 and 5 show the results of the horizontal
μ = 0.6
and vertical components of the grounding forces 65
for the obstruction (α = 10, β = 10) at different
levels of initial penetration. It is seen that the 60
horizontal grounding force is affected very much
by μ. On the other hand, the vertical grounding 55

force is seemingly unaffected by μ for both levels 50

of penetration δ considered. The same observation
was also reported by Alsos (2008). The highest peak 45
forces occur at the bulkheads, where the structure is
40
stiffer. Another interesting observation is that the
average horizontal grounding force for δ = 2.0 m is 35
virtually twice as high as that for δ = 1.0 m. 0 5 10 15
Grounding length [m]
20 25

The above results have been obtained for

symmetric geometries of the seabed. In general, Figure 4b. Vertical grounding force (Fz) for (α = 10,
the seabed topology is arbitrary and asymmetric. β = 10) at δ = 1.0 m.

375

MARSTRUCT.indb 375 2/18/2011 5:49:47 PM

 120 40
μ = 0.0 (7,7)
μ = 0.3 (1,7)
Longi. grounding force [MN]

100 35 (7,1)
μ = 0.6

Longi. grounding force [MN]

30
80

25
60

20
40

15
20
10
0
0 5 10 15 20 25
Grounding length [m] 5
0 5 10 15 20
Displacement [m]
Figure 5a. Longitudinal grounding force (Fx) for
(α = 10, β = 10) at δ = 2.0 m. Figure 6a. Longitudinal grounding force (Fx) for
asymmetric seabeds at δ = 1.0 m and μ = 0.3.
130
μ = 0.0
70
μ = 0.3
Vertical grounding force [MN]

(7,7)
120 μ = 0.6 (1,7)
(7,1)
Vertical grounding force [MN]

60
110

50
100

40
90

30
80

70 20
0 5 10 15 20 25
Grounding length [m]
10
0 5 10 15 20
Figure 5b. Vertical grounding force (Fz) for (α = 10, Displacement [m]
β = 10) at δ = 2.0 m.
Figure 6b. Vertical grounding force (Fz) for asymmetric
In this respect, the present work also considers the seabeds at δ = 1.0 m and μ = 0.3.
effect of asymmetric geometries, e.g. (α = 1, β = 7)
and (α = 7, β = 1). Figure 6 shows the results of seabed surface: this pressure is assumed to be
the grounding forces for the asymmetric seabed equivalent to the pressure under vertical punching
topologies together with those of the symmetric for the same level of penetration. In this respect, the
case (α = 7, β = 7). It is observed that this effect is simplified formulae to estimate the average ground-
considerable. Particularly, the horizontal grounding ing forces (vertical and horizontal components) can
force of (α = 1, β = 7) is highest in correspondence be derived. For the sake of simplicity, only circular
of the bulkhead (Fig. 6a). This result is due to the paraboloid is considered, i.e. a = b in Equation 1.
fact that obstructions with larger size in the trans- The resultant vertical force under vertical punch-
verse direction (y-axis) oppose more resistance in ing which is induced by the uniform pressure p is
the horizontal direction (x-axis). Considering the computed as (refer to (Johnson 1986)):
vertical component, (α = 7, β = 7) features the high-
est forces (Fig. 6b). This is due to the fact that the
Fv
vertical force component depends on the projected Fv π δ a 2 p or p = (2)
contact area which is normal to the z-axis. π δ a2

where p is the average contact pressure during

2.4 Simplified formulae for grounding forces
grounding; δ is the initial level of penetration; a
It is assumed that during grounding the average is the length of the semi-axis of the paraboloid
contact pressure p is uniformly distributed on the (Equation 1). The pressure p is calculated from

376

MARSTRUCT.indb 376 2/18/2011 5:49:48 PM

 vertical force Fv resulted during stranding 120
Vertical punching force FV
simulation for the corresponding obstruction as
sliding FX
plotted in Figure 1b. 100
sliding FZ
The horizontal grounding force can be derived

Grounding force [MN]

simplified FX
by integrating the horizontal pressure on the active 80 simplified FZ
contact surface:
60
4 3
Fx δ aap (3)
3 40

Following Alsos (2008), the vertical grounding 20

force may be assumed as half of the resultant
vertical force at penetration of δ, i.e. Fz = –12 Fv.
0
The simplified formulae are then compared with 0 5 10 15 20
Grounding distance [m]
the FE results for grounding forces, i.e. Fx and
Fz, in Figure 7 for obstruction (α = 10, β = 10). Figure 8a. Resistance—penetration and the grounding
A general agreement between the FE results and forces for (α = 10, β = 10), δ = 1.0 m and μ = 0.3.
the simplified estimates is found for the two levels
of penetration 1.0 m and 2.0 m, although the
120
Vertical punching force FV
120
sliding FX
Vertical punching force FV 100
sliding FZ
sliding FX
Grounding force [MN]

100 simplified FX
sliding FZ
80
Grounding force [MN]

simplified FZ
simplified FX
80 simplified FZ
60

60
40

40
20

20
0
0 5 10 15 20
0 Grounding distance [m]
0 5 10 15 20
Grounding distance [m]
Figure 8b. Resistance—penetration and the grounding
forces for (α = 10, β = 10), δ = 1.0 m and μ = 0.6.
Figure 7a. Resistance—penetration and the grounding
forces for (α = 10, β = 10), μ = 0 and δ = 1.0 m.

horizontal component Fx is a bit lower than the FE

200 results. It is noted that the effect of friction is not
Vertical punching force FV
180
accounted for in Equation 3. The contribution of
sliding FX
friction can be estimated by using Coulomb’s fric-
160 sliding FZ
tion law. This yields the following expression for
Grounding force [MN]

simplified FX
140 the total horizontal grounding force:
simplified FZ
120
Fxtot = Fx + μFz (4)
100

80 where μ is friction coefficient. The effect of friction

60
on the grounding forces are shown in Figure 8 with
two values of friction coefficient, namely 0.3 and
40
0.6 for the obstruction (α = 10, β = 10).
20

0
0 5 10 15 20 3 DYNAMIC GROUNDING
Grounding distance [m]

Figure 7b. Resistance—penetration and the grounding The tanker previously illustrated is used also for the
forces for (α = 10, β = 10), μ = 0 and δ = 2.0 m. analysis of dynamic grounding. At the present state,

377

MARSTRUCT.indb 377 2/18/2011 5:49:51 PM

 only surge, pitch and heave are considered. The
equation of rigid motion is written in Equation 5,
with the assumption that added mass and damping
terms are constant (Faltinsen 1999). It is realized
that this is not entirely correct. However, it can be
acceptable in the view of uncertainties involved
with the shallow water effects and with the seabed
topology in a real grounding event.
( )η + Bη + Cη = g( ,δ)
where A, M, B and C are added mass, mass,
damping and restoring matrices respectively. The
vectors (η η η  ) are displacement, velocity and
acceleration vectors of the ship. Fg(x, δ) is the
excitation force acting at the bottom as a function
of penetration (δ) and location of the obstruction
(x) during grounding. The solutions of Equation 5
can be obtained by Newmark-β integration scheme,
in which the explicit time integration procedure was
implemented in the code based on the lump mass
matrix. The simplified grounding forces provided
by Equations 3–4 with the friction coefficient of
0.3 are used as Fg(x, δ) in Equation 5.
The different variations of grounding forces can
be seen between static and dynamic groundings.
The ship bounces up and down during dynamic
grounding while she moves gradually over the
seabed obstruction in the static case. Moreover,
it is also interesting to see that the ship bounces
up twice for very large shoal, i.e. (α = 20, β = 20),
as shown in Figure 9b. A similar observation goes
also for the penetration as shown in Figure 10a; it
(5) can be seen that the larger the shape, the smaller
the level of penetration. This is due to the fact that
the larger the obstruction, the larger the contact
area. The effect of ship speed on the penetration is
also shown in Figure 10b. As observed, the higher
the speed, the more these curves shift to the right.
It is also realized that the scaling expression of the
contact force is simply assumed to be linear during
the initial stage, although it is not totally correct.
50
Fx - dynamic
45 Fz - dynamic
40 Fx - static
Grounding force [MN]

FZ - static
35
It is realized that the impact force at bow is
difficult to estimate because the formula obtained 30
in Equation 3 is for the area of the cargo tank, and 25
not for bow area. This leads to an unreasonable
20
large impact force at bow due to the large contact
area. In order to circumvent this, a scale factor is 15
essential to scale down the impact force at bow 10
during the first passage of impact. In this respect,
5
it is reasonable to assume that the impact load
is increasing linearly with time. Following Alsos 0
0 50 100 150 200 250
(2008), the excitation force is scaled down when the Grounding distance [m]
width of obstruction (BI) is greater than the dam-
age width (BD) as expressed in Equation 6. Figure 9a. Grounding forces for (α = 7, β = 7) at
δ = 1.0 m and speed of 12 knots.
⎧ BD ( x )
⎪ F ( x, ) BD ( x,
x, ) < BI ( )
Fg ( x,δ ) = ⎨ BI ( ) g (6) 90
⎪F ( x, ) otherwise Fx dynamic
⎩ g 80 Fz dynamic
Fx static
70
Grounding force [MN]

4 RESULTS OF DYNAMIC AND STATIC Fz static

POWERED GROUNDINGS 60

50
4.1 Grounding forces
40
The horizontal and vertical components of the
grounding force are computed for two obstruc- 30

tions, namely (α = 7, β = 7) and (α = 20, β = 20). 20

The obstruction is initially at penetration depth
δ = 1.0 m over the keel line and then moved 10

horizontally with speed of 12 knots along the ship 0

0 50 100 150 200 250
bottom. The static and dynamic groundings are Grounding distance [m]
analyzed and the results are plotted in Figure 9.
Here, static grounding is performed by running Figure 9b. Grounding forces for (α = 20, β = 20) at
dynamic grounding quasi-statically. δ = 1.0 m and speed of 12 knots.

378

MARSTRUCT.indb 378 2/18/2011 5:49:54 PM

 1.4 100
(7,7) dynamic Still water requirement
(10,10) dynamic (7,7)
1.2 80
(20,20) dynamic (10,10)
(7,7) static (20,20)
(10,10) static 60
1

Shear force [MN]

(20,20) static
Penetration [m]

40
0.8
20
0.6
0

0.4
–20

0.2 –40

0 –60
0 50 100 150 200 250 0 50 100 150 200 250
Grounding distance [m] Distance from AP [m]

Figure 10a. Penetration of ship bottom, δ = 1.0 m and Figure 11a. Envelopes of shear force at δ = 1.0 m and
speed of 12 knots. speed of 12 knots.

1.4
12 knots 5000
15 knots Still water requirement
1.2 20 knots 4000 (7,7)
(10,10)
3000 (20,20)
Bending moment [MNm]

1
Penetration [m]

2000
0.8
1000

0.6 0

–1000
0.4
–2000
0.2
–3000

0 –4000
0 50 100 150 200 250 0 50 100 150 200 250
Grounding distance[m] Distance from AP [m]

Figure 10b. Penetration of ship bottom for (α = 20, Figure 11b. Envelopes of bending moment at δ = 1.0 m
β = 20), δ = 1.0 m and speeds of 12, 15 and 20 knots. and speed of 12 knots.

4.2 Shear and bending moment

150
Still water requirement
The envelope curves of shear and bending moments (7,7)
during grounding at two levels of penetration are (10,10)
plotted together with the water requirements from 100 (20,20)

the DNV rules, i.e. 1.0 m and 2.0 m, as shown

Shear force [MN]

in Figures 11–12. Due to the transient dynamic

effect, the large shear force is observed at the initial 50

impact (Figs. 11a, 12a). This leads to a higher value

of shear force than the DNV requirement (DNV
0
2009) at the initial impact of the ship. Moreover,
it can be seen that also the larger the obstruction,
the higher the shear force and bending moment.
–50
This is physical due to the fact that the larger the
obstruction, the higher the contact force. Moreover,
0 50 100 150 200 250
it is more evident that the shear forces and bending Distance from AP [m]
moments induced by large obstructions become
larger than the requirements at high level of Figure 12a. Envelopes of shear force at δ = 2.0 m and
penetration (Fig. 12). This can in turn affect the speed of 12 knots.

379

MARSTRUCT.indb 379 2/18/2011 5:49:57 PM

 5000
Still water requirement
4000 (7,7)
(10,10)
Bending moment [MNm]
3000 (20,20)
2000
1000
0 can fulfill the safety requirement of seaworthiness.
–1000
–2000
–3000
–4000
–5000
0 50 100 150
Distance from AP [m]
Figure 12b. Envelopes of bending moment at δ = 2.0 m
and speed of 12 knots.
overall strength of the ship so that the safety level
for sea-going has to be reassessed. In addition, it is
also observed that shear force and bending moment
are somewhat unaffected by the ship speed.
5 CONCLUSIONS
In this study a generalized seabed topology is used
to simulate the different grounding scenarios. The
results revealed that the seabed geometry is a key
parameter involved in ship grounding. Large shoals
give a high resistance to penetration. The dynamic
grounding was considered by means of nonlinear
finite element analysis with various seabed topolo-
gies, friction coefficients, “initial” levels of penetra-
tion over the keel line and ship speeds. The horizontal
grounding force is much affected by friction while
the vertical component is not. The grounding forces
are increasing virtually twice as the “initial” level of
penetration is doubled. The effect of asymmetric
seabeds is also considerable, especially when the
obstruction has a larger extent in the direction per-
pendicular to the grounding direction of the ship.
Moreover, it is realized that there are uncertainties
related to the actual seabeds which are arbitrary
and unknown. More studies are therefore essential
in order to describe the effect of actual seabed on
ship grounding correctly.
The simplified formulae to estimate the average
grounding forces are proposed for paraboloid
seabeds. The results of the simplified formulae
show reasonable agreement with the results from
FE simulations. It is still compromising to use
these formulae for quick estimate of the average
grounding forces for each seabed topology. The
difference between static and dynamic grounding
is significant. The envelopes of shear force and
bending moment induced by each shape of seabed
MARSTRUCT.indb 380
are computed and checked with the DNV’s still
water requirements. The shear force always exceeds
the rule requirement at the initial impact. Finally,
the residual strength of the damaged hull girder
should be reassessed in case of grounding on a
very large shoal, in order to ensure that the ship
ACKNOWLEDGMENTS
The authors gratefully acknowledge the support
of the Strategic University Program (SUP) “Scena
200 250
RisC&G” at the Norwegian University of Science
and Technology (NTNU). A special gratitude goes
to the Research Council of Norway for providing
the funding for this research. Part of the study
was generated within the European Framework
for Safe, Efficient and Environmentally-Friendly
Ship Operations (FLAGSHIP). The financial sup-
port from the European Commission is gratefully
acknowledged.
REFERENCES
Alsos, H.S. & Amdahl, J. 2007. On the resistance of
tanker bottom structures during stranding. Marine
Structures 20, 218–237.
DNV. 2009. Part 3- Chapter 1- Hull structural design
ships with length 100 metres and above. Det Norske
Veritas.
EMSA. 2009. Maritime accident review. European Mari-
time Safety Agency. Available for download from
www.emsa.europa.eu.
Faltinsen, O.M. 1999. Sea loads on ships and offshore
structures. Cambridge University Press, Cambridge,
United Kingdom.
Hong, L. & Amdahl, J. 2008. Plastic mechanism analysis
of the grounding resistance of ship bottom longitudinal
girders. Ships and Offshore Structures 3, 159–171.
Johnson, D. 1986. Advanced structural mechanics. William
Collins Son & Co. Ltd, London.
Nguyen, T.H., Garrè, L. Amdahl, J. & Leira, B.J. 2010.
Benchmark study on the assessment of ship damage
conditions during stranding. Accepted in Ships and
Offshore Structures.
Paik, J.K. & Wierzbicki, T. 1997. A benchmark study on
crushing and cutting of plated structures. Journal of
Ship Research 41, 147–160.
Pedersen, P.T. 1997. Ship grounding and hull girder
strength. Marine Structures 44, 140–154.
Samuelides, M.S., Amdahl, J. & Dow, R. 2007. Studies on
the behavior of bottom structures during grounding:
Advancements in Marine Structures, C. Guedes Soares
and P. K. Das (Eds), Taylor & Francis Group; London,
U.K: pp. 273–282.
Simonsen, B.C. 1997. Mechanics of ship grounding. Ph. D.
thesis, Department of Naval Architecture and Offshore
Engineering, Technical University of Denmark (DTU).
380
2/18/2011 5:50:00 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Influence of bulbous bow structures on their collision behaviour

M. Schöttelndreyer, I. Tautz, J.M. Kubiczek, W. Fricke & E. Lehmann

Hamburg University of Technology (TUHH), Hamburg, Germany

ABSTRACT: The authors are currently engaged in a collaborative research project which—amongst
others—deals with the comparison of different bulbous bows and side structures. Their corresponding
effects on collision will be investigated. This paper will give a short overview of the background and start-
ing phase of the project. It will focus on developing one determined test model geometry of a bulbous
bow for the next two prospective collision experiments. This geometry remains unchanged for a rigid
bulbous bow as well as for a deformable bulbous bow structure. The requirement of the designed bulbous
bow is to be able to create a significant damage of the ship side structure and itself. This pre-experimental
study shows that minors parametrical changes of bulbous bow geometry will significantly change the
collision behaviour. For this study the finite element method has been used.

1 INTRODUCTION results will provide the basis to validate simplified

simulation approaches needed to realise a reason-
1.1 Background able combination between collision simulation and
damage stability calculation.
This paper reports on actual research work carried
Most of the experimental investigations of the
out in the collaborative project ELKOS (German
past focused on the struck ship like TNO-Test
acronym, for: “Improving collision safety by
(Peschmann, 2001). Only a few researchers consid-
integrating effects of structural arrangements in
ered the inner mechanics of a deformable bulbous
damage stability calculations”). Superior research
bow e.g. Amdahl (1983) and Yamada (2006).
objective is to develop a method, that allows ade-
This collaborative project investigates on the
quate consideration of structural arrangements
one hand the striking ship with its pre-tests and on
that significantly increase collision safety in dam-
the other hand the struck ships in the first collision
age stability calculations.
test with a rigid bulbous bow. In addition to these
The project is funded by the German Federal
approaches experimental tests which will consider
Ministry of Economics and Technology (BMWi).
the whole collision procedure with the striking ship
TUHH is engaged with its institutes “Ship Struc-
as a deformable one and the struck ship too will
tural Design and Analysis” and “Ship Design and
also be carried out in end of 2011.
Ship Safety”. Industrial partner is the German
For these two prospective experiments a geom-
yard Flensburger Schiffbau-Gesellschaft.
etry of a rigid bulbous bow as well as a deform-
The authors focus on experimental investiga-
able bulbous bow structure have to be designed. In
tions of inner mechanics concerning several designs
both cases the exceedance of the maximum load-
of ship side structures and following numerical
ing capacity of the test-plant has to be avoided.
simulations. Furthermore, stiffness of bulbous
Regarding the deformable bulbous bow the defor-
bows will be considered in a realistic manner and
mation of both—bow and side structure—must be
corresponding effects on collision mechanics with
fulfilled.
side structures will be investigated.
This paper will describe the way of finding such
Regarding to the collision experiments with bul-
a bulbous bow with the aim to design suitable test
bous bow and ship side structures, two pre-tests
models.
were carried out on the test facility of the Institute
for Ship Structural Design and Analysis of TUHH
in March and June 2010. Two different deformable
2 EXPERIMENTAL CONDITIONS
bow structures were driven against a rigid plate
(Tautz et al., 2010). After these two pre-tests the
2.1 Pre-tests
test facility was modified and prepared for the first
collision experiment. The start of these experi- Collision tests are carried out on the existing test-
ments is scheduled for summer 2011. Experimental plant of the Institute of Ship Structural Design and

381

MARSTRUCT.indb 381 2/18/2011 5:50:00 PM

 Analysis of TUHH which is adequately supple-
mented and modified for this purpose (Fig. 1).
Collision forces are applied by four hydraulic
cylinders. They are connected with the longitudi-
nal girders of the test-plant and with a cross-beam.
Thereby a closed loop of forces is provided. The
test model of the bulbous bow is located under-
neath the cross-beam. It is driven against a coun-
ter plate which can be assumed as rigid. Collision
forces are measured at the hydraulic cylinders as
well as at pressure load cells underneath the rigid
counter plate. The maximum loading capacity is
4,000 kN.
Hydraulic cylinders are limited at 400 mm
regarding the maximum range of displacement.
Thus larger displacements will be realised by
using appropriate interim pieces between the bul-
bous bow and the cross-beam. This approach is
permissible because the whole test procedure is
quasi-static with maximum speed of 0.5 mm/sec.
Therefore the interruption of the test is assumed
to be permissible at any time.
The two pre-tests confirmed this assumption.
It can be observed that the original load path was
reached after re-loading (Tautz et al., 2010).
2.2 Coming collision-tests
The test-plant offers possibilities to integrate ship
side structures with moderate effort instead of using
the rigid plate. Side structures will be supported
adequately to measure collision forces as well as
membrane forces (Fig. 2). Regarding dimensions,
test models of side structures will be comparable
with those investigated in a very realistic man-
ner in the research project “Life Cycle Design”
(Wevers, 1999) in the Netherlands (scaling ca.
1:3). Regarding the steel design, test model will be
comparable to side structures of RoRo-vessels.
Cross-Beam
Hydraulic Bulbous
Cylinder Bow Rigid
Plate
Longitudinal Girder
Hydraulic
Longitudinal Girder Cylinder
Cross-Beam
Figure 1. Test-plant, configuration bow against rigid
plate.
382
MARSTRUCT.indb 382
Cross-Beam
Bulbous
Hydraulic Support Bow
Cylinder Side Structure
Longitudinal Girder
Figure 2. Test-plant, configuration bow against side
structure.
3 BASIS FOR THE NUMERICAL
SIMULATION
3.1 General
The authors use the latest LS-DYNA version
971/Rev5 (Beta) for their simulations. The inner
and outer shell of the side structure is modelled
with four-noded quadrilateral Belytschko-Lin-Tsay
element type with five integration points through
their thickness. Its size is 33 × 35 mm at the area of
impact. The plate thickness amounts to 5 mm. An
assumed static friction coefficient of 0.10 for steel
on steel without grease, a Poisson ratio of 0.3 and
the material *mat 123 implemented in LS-DYNA
for a modified piecewise linear plasticity is used
for the whole numerical investigation. The contact
treatment is realized by using the contact defini-
tion *CONTACT AUTOMATIC SURFACE TO
SURFACE for the fundamental structure like
outer and inner shell, web of the longitudinal bulb
profiles and the web frame of the side structure.
Even nowadays engineers have difficulties to sim-
ulate collapse of the material in a realistic manner.
The used material has to be investigated and fail-
ure criteria must be defined. In shipbuilding the
following could be described as state of the art and
is also used in this investigation as failure criteria.
Failure is simulated by erasing every finite element
which exceeds a previous defined failure strain in
every integration point through the thickness.
3.2 Proposal for the material model
Due to the fact that material properties of steel
always vary in a certain extent it is necessary to
measure the exact material parameters of the steel
plates which will be used for the test model.
Thus an existing strain-stress curve which was
developed out of an uniaxial and quasi-static
tensile-test for a former investigation is modified.
This curve describes the true stress-strain relation-
ship which is the precise way to simulate the plastic
deformability.
For modifying the existing true strain-stress
curve, a quasi-static tensile-test was carried out
(Fig. 3). The load of this test-plant is restricted.
2/18/2011 5:50:01 PM
 900
800
700

Stress [N/mm2]
600
500
400
300
200 Existing Curve
100
0
Figure 3. Test-plant, tensile-test. 0.00 0.20 0.40 0.60 0.80 1.00
Strain[-]
350
Figure 5. Power law hardening based material relation
300 of existing and scaled curve.
250
Stress [N/mm2]

200 (2008). The presented paper will regard two

150
100 Gauges Top Site
50 Gauges Bottom Side
Average
0
-0.05 0 0.05 0.1 0.15 0.2 0.25
Strain[-]
Figure 4. Measured stress-strain values of specimen.
Hence a material specimen of the test model of
the first pre-test was chosen. The flat bar tension
specimen was prepared after the Norm EN ISO
6892-1. Two strain gauges were applied, one on
the top site and one on the bottom site of the
specimen (Fig. 3). The specimen had a geometry
with a slight bending curvature to the top site.
Therefore the two measured start values of the
strain held opposite signs at the beginning of the
test (Fig. 4).
For modifying the previous true strain-stress
curve the yield stress of the tested material must
be found. For this the linear part is deducted
from the mean strain-stress curve. At the point
of 252 N/mm2 the linear graph disperses from the
average. The remaining part of the strain-stress
curve has been transferred by scaling the previous
existing curve to the new one. The result of the
scaled graph is shown in Figure 5.
For the prospective collision test it was decided
to use the same plate thickness for side structure
and bulbous bow. Therefore the use of the found
power law hardening based material relation is
justified and will be used further on.
3.3 Failure criteria
In the past several different failure criteria were
investigated and compared; see e.g. Ehlers et al.
different criteria. One proposed by Peschmann
(2001) and the second one is a combination of
Rice & Tracey (1969) and Cockcroft & Latham
(1972) named RTCL.
The criterion of Peschmann is a quasi-static
uniaxial tension consideration. It is not able to dif-
ferentiate the applied load between compression
and tension and will delete elements when the criti-
cal rupture strain is reached. It does not matter what
kind of load is applied. Therefore the application
of this criterion must be well deliberated and the
applied load of each single part of construction
must be known. A further disadvantage is that the
critical rupture strain depends on the length of the
elements. By using a coarse mesh the necking will
not fit to the location where the fracture starts.
Thus the ratio of uniform and necking strain must
be scaled by the length of necking:
ε crit = ε g + ε m ⋅ ⎛ te ⎞ ⋅ t
x
(1)
⎝ ⎠ l
εg is the uniform strain, εm is the necking strain,
xe the length of neck, t the plate thickness and l the
element length. Based on tests, Peschmann obtains
the following values for plate thickness smaller
than 12 mm:
ε crit = 0.1 + 0.8 ⋅ t (2)
l
A major advantage of this criterion is that it is
easy to implement in every commercial finite ele-
ment code for non-linear calculations and it needs
less computing time.
The second criterion is the above mentioned
RTCL criterion which considers the influence of
the stress triaxility. This criterion is presented and
verified by Törnqvist (2003). This failure criterion
was recently implemented by Livermore Software

383

MARSTRUCT.indb 383 2/18/2011 5:50:03 PM

 Technology Corporation (2010) in the latest version numerical calculations. The wide difference of this
of LS-DYNA which is used for the calculations. comparison shows the necessity to validate one’s
The criterion defines a damage indicator Di as: calculations by collision experiments in future.
The maximum of the RTCL is more than 100%
1 ⎛σH ⎞ higher than that of Peschmann. For further cal-
ε 0 ∫ f ⎝ σ eq ⎠
Di = d ε eqp (3) culations the RTCL will be taken, even if this
RTCL would lead to a certain overestimation of collision
forces. This is acceptable in order not to exceed the
where maximum loading capacity of the test-plant.

⎧ σH 1
⎪ 0 for ≤−
σ eq 3
⎪
⎪ 2
⎪ σH ⎛σH ⎞
⎪ 1+ σ 12 − 27 ⎜ ⎟
⎛σ ⎞ ⎪ eq ⎝ σ eq
e ⎠ 1 σ 1
f⎜ H⎟ = ⎨ 2 for
o − < H <
⎝ σ eq ⎠ RTCL 3 σ eq 3
⎪
2
σH ⎛σH ⎞
⎪ 3 σ + 12 − 27 ⎜ σ ⎟
⎪ eq ⎝ eq ⎠
⎪
⎪ 1 exp ⎛ 3σ H ⎞ for
σH 1
≥
⎪ 1.65 ⎜ 2σ ⎟ σ eq 3
⎩ ⎝ eq ⎠

ε0 is the uniaxial fracture strain, σH hydrostatic

stress, σeq equivalent stress and dε peq equivalent plas- This criterion takes 10% more time for the
tic strain increment. The critical failure is reached calculation.
when Di reaches the value 1.
For getting conservative results and to be sure
not having forces larger than 4,000 kN, the criteria 3.4 Strain rate
which leads to the highest collision force is sought Hardening of the steel is basically dependent of
for. In order to decide what criterion is to be used the strain rate.
comparative calculations of Peschmann and the A possibility—often mentioned in literature—is
RTCL are made (Fig. 6). For the calculation the to include this phenomenon in numerical calcula-
geometry of the bulbous bow in the pre-test is tions by using the equation of Cowper-Symonds
used, modelled as a ridged body. For this com- which describes the influence of the strain rate on
parison 22,009 elements were used. The specific the yield stress,
description of the FE-Model will be presented in
the next section. 1
ReHdynam ε q
It is easy to see that the stress triaxiality has = 1+ ⎛ ⎞ (4)
a significant influence of the solution of the ReHstat ⎝C ⎠

3500 where ReHdynam is the yield stress under dynamic .

RTCL load, ReHstat the yield stress under static load, ε the
3000 Peschmann
strain rate, C and q are factors obtained from tests.
Reaction Force [kN]

2500 Paik & Thayamballi (2003) proposed for mild steel

C = 40.4 and q = 5.
2000 One can assume that the strain rate in the
1500 planned quasi-static collision tests is rather small
in view of the anticipated velocity of 0.5 mm/sec.
1000 The influence of the strain rate is demonstrated by
500 a simplified geometric model (Fig. 7). Here, the
marked point A in the middle of the side structure
0 beneath the middle of the bulbous bow located
0 250 500 750 1000 1250 1500
Displacement [mm] between two web frames is analysed. aR is the web
frame spacing, set to 800 mm.
Figure 6. Comparison of failure criteria RTCL vs. Assuming that the point A moves with the same
Peschmann. velocity as the bulbous bow (0.5 mm/sec), the

384

MARSTRUCT.indb 384 2/18/2011 5:50:05 PM

 3.5
Formulas
: Cowper & Symonds (1957)
3.0 : Paik et al. (1999)

2.5
/ =1 +
2.0

/
1.5

1.1
1.0
/ =1 +
0.5

0.0

Experiments
Mild steel: High tensile steel:
= 189.6 Mpa = 522.9 MPa
Manjoine (1944) Fujii et al. (1985)
= 217.5 MPa = 373.2 Mpa
Campbell & Cooper (1966) Toyosada et al. (1987)
= 283.0 MPa = 313.8 Mpa
Paik et al. (1999) Toyosada & Goto (1991)

Figure 9. Dynamic yield strength plotted versus strain

rate for mild and high tensile steel; Paik & Thayamballi
Figure 7. Model for estimating the max. strain rate. (2003).

20 strain rate will increase approximately three orders

Investigated Case of magnitude. This still means a small effect of the
strain rate on the results.
15
L+ΔL [m]

4 NUMERICAL SIMULATION
10
Asymptote

5 This investigation concentrates on the worst
0 of the side structure with a collision angle of
0.0 4.0·10-4 . 8.0·10-4
ε [sec-1]
Figure 8. Strain rate of point A moving with a constant
velocity of 0.5 mm/sec.
plate of length L = aR/2 will elongate as shown in
Figure 8.
It is easy to see that the strain rate will not super-
sede the asymptotic value of 1.22 ⋅ 10−3 sec−1.
The outer shell of the side structure fails at a
displacement of 400 mm by using the evaluated
bulbous bow. The simplified model has shown a
strain rate value of 5.20 ⋅ 10−4 sec−1 at a displace-
ment of 400 mm. Considering Figure 9 it is obvi-
ous that both values are located near the yield
stress ratio 1. The result of this simplified geomet-
ric model presents a comparable small influence
of the strain rate at a velocity of 0.5 mm/sec as
shown in the investigation made by Ehlers et al.
(2010). Hence the strain rate will be neglected for
the calculations.
In reality, the collision velocity will be much
larger. Considering additional scale effects, the
4.1 General
case for the struck vessel. This includes that the
impact is located in the middle of two web frames
1.2·10-3 90 degrees.
The examination will distinguish between three
sizes of bulbous bow cross sections. The small-
est bulbous bow will not hit the web frames, the
intermediate bulbous bow will touch the web
frame and the largest bulbous bow will totally
collide with the web frames. The initial designs of
the investigated geometries are basically axisym-
metric. The cross section will have an increasing
elliptical shape along the global vertical axis of
the ship.
4.2 Definition of the local coordinate system
For the simulation of the different geometries of
the bulbous bows the axes of the coordinate system
are defined as shown in the Figures 10 and 11.
As shown in Figure 10 the x-axis is located par-
allel to the longitudinal girder in the middle of the
test-model; the y-axis proceeds from the middle of
the side structure to the longitudinal girder and
the z-axis is the vertical one which points into the
bulbous bow’s penetration direction (Fig. 11). The
web frame spacing aR is 800 mm.

385

MARSTRUCT.indb 385 2/18/2011 5:50:11 PM

 Y X-Y Plane X-Y Plane
(End Section) (End Section)

Longitudinal Girder
Connection Plates

Area of
Impact X-Z Plane Y-Z Plane
Side Structure End Section End Section

1800 [mm]
aR aR aR/2

Figure 10. Top view of a quarter of the side structure;

x-y plane.

Z
Figure 12. Investigated bulbous bows, left: Geometries
Bulbous for α1/βi = 1.00; right: Geometries for β1/aR = 0.80, see
Bow also Table 2.
Connecting
Rod Table 1. Definition of the parameters α and β [mm]
Side Structure describing the end section of the bulbous bows.

Support
aR aR aR/2
X α1/βi = 1.00 α2/βi = 1.50 α3/βi = 2.00

β1/aR = 0.80 α1 = 640 α2 = 960 α3 = 1,280

Figure 11. Side view of a quarter of the side structure β1 = 640 β1 = 640 β1 = 640
and support; x-z plane. β2/aR = 1.02 α1 = 816 α2 = 1,224 α3 = 1,632
β2 = 816 β2 = 816 β2 = 816
β3/aR = 1.25 α1 = 1,000 α2 = 1,500 α3 = 2,000
4.3 Geometry of investigated models
β3 = 1,000 β3 = 1,000 β3 = 1,000
In the beginning of the numerical investigation a β4/aR = 1.50 α1 = 1,200 α2 = 1,800 α3 = 2,400
matrix with the parameters α to β is defined. The β4 = 1,200 β4 = 1,200 β4 = 1,200
parameters describe the end section of the bulbous
bows (Fig. 12). Thereby β describes the breadth where: aR = 800 [mm]
of the bulbous bow. Four different β will be inves-
tigated. The parameter α describes the increas- Table 2. Shortcuts of the definition of the parameters
ing elliptical shape for each β. The parameter β α and β [mm] describing the end section of the bulbous
depends of the web frame spacing and must fulfill bows.
the above mentioned three sizes. The smallest size
is relised with β1/aR = 0.80, the intermediate size α1/βi = 1.00 α2/βi = 1.50 α3/βi = 2.00
with β2 /aR = 1.02 and the largest size where the bul-
bous bow will totally collide with the web frame β1/aR = 0.80 11 12 13
is represented by two geometry variations where β2/aR = 1.02 21 22 23
β3/aR = 1.25 and β4/aR = 1.50. β3/aR = 1.25 31 32 33
The columns of Table 1 show the ratio of the β4/aR = 1.50 41 42 43
breadth β to the frame spacing aR of the side
structure. All values of the first column generate a
rotational symmetric geometry. The rows describe The geometry 21 is the one which was used in
the elliptical distortion of the initial rotational the two pre-tests. All geometries of one row have
symmetric geometries. the same top of forepart. The buttock lines in the
For a well arranged analysis of the found y-z plane are straight and connected to the forepart
geometries following shortcuts will be introduced design at its tangential point of touch (marked in
in Table 2 and will be used from now. Fig. 12).

386

MARSTRUCT.indb 386 2/18/2011 5:50:13 PM

 4.4 Numerical model
For this investigation twelve different geometries
of bulbous bows had to be generated. Therefore
the ANSYS parametric design language was used.
To be able to change the geometry in an acceptable
effort, a parametrical finite element model depend-
ent of α and β has been created. Each geometry
is calculated under the assumption rigid as well as
assumption deformable.
For the investigation with a rigid bulbous bow
it is sufficient to consider a quarter of the test
configuration using symmetrical boundary condi-
tions. This consideration is unusual for this kind of
highly nonlinear calculations with failure. However
it delivers sufficient accuracy of the results needed
for the investigation. Non symmetric effects of the
failure mechanism are not in the focus and have a
neglectable influence on the absolute level of col-
lision forces that are of main interest. This was
approved by some comparative calculations which
considered the whole FE-Model. These will not be
discussed further on in this paper.
With regard to the deformable bulbous bow
the whole test configuration had to be modelled
because of the stringer and the longitudinal girder
located at the section planes of the bulbous bow.
In the experimental set-up it is necessary to rep-
resent the correct stiffness of the supports. Other-
wise wrong membrane forces will be transmitted
to the side structure. The consequence would be a
wrong collapse behaviour of the side structure. The
shell of the supports is modelled with four-noded
elements like the side structure (Fig. 13). Each sup-
port is adapted to both longitudinal girders of the
test-plant with four vertical tension rods. The hori-
zontal connection is designed with four connecting
rods per support. The vertical mounting is realised
by using four load cells. The rods and the load cells
are modelled with truss elements.
Longitudinal
Bulkhead
Stringer
Load Cell
Web
Frame
Connection Rods Bulb Profile
Tension Rods
Figure 13. Finite element model of geometry 43 (1/4-
model).
387
MARSTRUCT.indb 387
The finite element models regarding the rigid
bulbous bow consist of 130,000 degrees of free-
dom in average. For the case of the deform-
able bulbous bows these numbers have to be
quadrupled.
5 RESULTS
5.1 General
According to the technical limitation of the test-
plant it is necessary to find out the feasibility of
the twelve variations with a rigid bulbous bow. All
geometries which exceed the force of 4,000 kN
have to be excluded. The others have to be inves-
tigated again by considering their stiffening sys-
tem. The aim is to find a bulbous bow structure
which deforms both the side structure and itself.
Therefore the case in which either the side struc-
ture or the bulbous bow collapses solely will be
excluded.
5.2 Consideration of rigid bulbous bows
Representatively for all geometries the upper row
and the left column of Table 2 will be taken for
the consideration below. The reaction forces of
the geometries with a constant β and a growing
α are shown in Figure 14. The two load peaks of
the curves represented the penetration of the outer
and inner hull of the side structure.
The growing parameter α has no significant
influence on the point of structural failure. Both
load peaks occur at comparable displacements of
the bow for each of the geometric variations due
to the marginal changing geometry of the forepart
design of the bulbous bow. The reaction force of
the geometries with the same α and a growing β is
pictured in Figure 15.
2500
11
2000 12
Reaction Force [kN]
13
1500
1000
500
0
0 250 500 750 1000 1250 1500
Displacement [mm]
Figure 14. Comparison of the reaction force of the three
geometries of the upper row of Table 2 (β1/aR = 0.80).
2/18/2011 5:50:14 PM
 5000 4500
11 Permissible Collision Force
4500
PermissibleCollisionForce
4000 β1= 0.640 [m]
21

Reaction Force [kN]

4000 3500
31
Reaction Force [kN]

3500 3000
x 41 β2 = 0.816 [m]
3000 2500
2500 2000
2000 1500 β3 = 1.000 [m]
1500 1000
1000 500 β4 = 1.200 [m]
500 0
0 0.4 1.4 2.4
0 250 500 750 1000 1250 1500 Parameter αj [m]
Displacement [mm]
Figure 16. Calculated reaction force of the geometries
Figure 15. Comparison of reaction force of the four
at the first load peak.
geometries of the left column of Table 2 (α1/βi = 1.00).

4500
Permissible Collision Force β1= 0.640 [m] EQ
Table 3. Exclusion of the geometries exceeding the 4000
maximum load capacity of the test-plant. β1= 0.640 [m] DY
Reaction Force [kN] 3500
β2 = 0.816 [m] EQ
α1/βi = 1.00 α2/βi = 1.50 α3/βi = 2.00 3000
β2 = 0.816 [m] DY
2500
β1/aR = 0.80 11 12 13 β3 = 1.000 [m] EQ
2000
β2/aR = 1.02 21 22 23 β3 = 1.000 [m] DY
1500
β3/aR = 1.25 31 32 33
β4 = 1.200 [m] EQ
β4/aR = 1.50 41 42 43 1000
β4 = 1.200 [m] DY
500
0.5 1.5 2.5
Parameter αj [m]
In contrast to Figure 14 the penetration of Figure 17. Comparison of the numerical and approxi-
the outer and inner hull is influenced by the mated reaction force at the first load peak.
growing cross section. This leads to a later fail-
ure of both hulls. For example the geometry with
α1 = β1 = 640 mm destroys the outer hull of the side amounts to 3.8% for the geometries of small
structure at a displacement of about 250 mm, the breadth.
one with α1 = β4 = 1,200 mm by more than twice of A similar relationship of the parameters is found
the above mentioned displacement. Furthermore for the second peak (Fig. 18), approximated by the
the geometry 41 will exceed the maximum load following equation:
capacity of the test-plant as well as geometries 42,
43 and 33 do. These geometries are excluded in FΙΙ = 2417β 2 − 1735β − 822
Table 3. + (2099 − 1044β)(α − β ) (6)
Due to the above mentioned influence of the
bulbous bow geometry one can assume that there Here the maximum deviation is 4.7% between
might be a relationship between the parameters of the numerical value and the value given by equa-
the bulbous bow end section and the load peak. tion 6 for the reaction force of the second load
For this reason the value of the geometries are peak. The graphs show a decreasing gradient with
plotted to Figure 16, connecting the values with an increasing β. An explanation for this phenom-
the same β. enon can be given by the numerical simulation.
Obviously all graphs have an increasing gradi- The damage of the side structure caused by bul-
ent. Therefore an equation has been developed bous bows with large α and β values includes a
which offers the possibility to describe the influ- significant area of the outer shell of the side struc-
ence of the geometry without a simulation: ture by loosing its structural resistance against the
penetrating bulbous bow (Fig. 19).
FΙ = 3387β 2 − 1986β + 1087
+ (328β − 83)(α − β ) (5) 5.3 Consideration of deformable bulbous bows
The regression in equation 5 shows a good with As mentioned above eight remaining geometries
numerical results (Fig. 17). The highest deviation have to be investigated. The aim is to find the case

388

MARSTRUCT.indb 388 2/18/2011 5:50:15 PM

 7000
0.640 [m] EQ
6000
0.640 [m] DY
Reaction Force [kN]

5000 0.816 [m] EQ

4000 0.816 [m] DY
3000 1.000 [m] EQ
2000 1.000 [m] DY
1.200 [m] EQ
1000
1.200 [m] EQ
0
0.4 2.4
Parameter αj [m] Figure 20. Failure of outer hull and deformation of
geometry 23; half of side structure is blanked.
Figure 18. Comparison of the numerical and approxi-
mated reaction force of the geometries at the second 4500
load peak. Geometry 23
4000

Reaction Force [kN]

3500
3000
2500
2000
1500
1000
500
0
0 250 500 750 1000 1250 1500
Displacement [mm]

Figure 21. Reaction force between bulbous bow

geometry 23 and the side structure.
Figure 19. Failure of outer hull and bulb profile of
geometry 43.
6 CONCLUSIONS
Table 4. Exclusion of the geometries with only one This paper describes a way to increase the scien-
damaged collision partner. tific knowledge about the interaction of the bul-
α1/βi = 1.00 α2/βi = 1.50 α3/βi = 2.00
bous bow and side structure design in the case of
collision. The selected bulbous bow configuration
β1/aR = 0.80 SiSt SiSt SiSt allows a direct comparison between the rigid and
β2/aR = 1.02 SiSt SiSt BuBo the deformable bulbous bow.
SiSt The focus of the investigation is on finding a
β3/aR = 1.25 BuBo BuBo – case where both bulbous bow and side structure
are deformed simultaneously. The result shows that
where: SiSt = side structure fails; BuBo = bulbous bow there is only a small range of geometry which leads
fails. to the occurrence of this phenomenon. Minor par-
ametrical changes of bulbous bow geometry will
which lead to a deformation of the bulbous bow totally change the failure mode and therefore the
and a failure of the side structure. The results of collision behaviour.
the numerical calculations are listed in Table 4. The goal is to validate the numerical calculations
Considering the results given in Table 4 it is by future collision tests. This will enable a method
obvious that there is only one geometry bulbous to predict quantitative design parameters.
bow which fulfills the above mentioned criteria, i.e.
geometry 23 (Fig. 20), where α = 1,632 mm and ACKNOWLEDGEMENTS
β = 816 mm.
The impact of geometry 23 on the given side The work was performed within the research
structure generate a collision force of 3,150 kN Project ELKOS, funded by German Federal
which is well below (79%) of the permissible Ministry of Economics and Technology (BMWi)
reaction force (Fig. 21). under project no. 03SX284B.

389

MARSTRUCT.indb 389 2/18/2011 5:50:16 PM

 The authors are responsible for the content of
this paper and wish to thank those, who supported
this project. The authors’ gratitude is particularly
addressed to German yard Flensburger Schiffbau-
Gesellschaft which delivered the cross-beam, the
two supports for the test-plant and the test model.
REFERENCES
Amdahl, J. 1983. Energy Absorption in Ship-platform
Impacts. Report No. UR-83–34, University of
Trondheim: Trondheim.
Cockcroft, M.G. & Latham D.J. 1972. Ductility and the
Workability of Metals. J. Inst. Metals: 33–96.
Ehlers, S. Broekhuijsen, J. Hagbart, S.A. Biehl, F. &
Tabri, K. 2008. Simulating the collision response of
ship side structures: A failure criteria benchmark
study, Int. Ship progress 55: 127–144.
Ehlers, S. Tabri, K. Romanoff, J. & Varsta, P. 2010.
Numerical and Experimental Investigation on the
Collision Resistance of the X-core Structure. Proc.
of 5th Int. Conference on Collision and Grounding of
Ships (ICCGS). Espoo: Aalto University.
Livermore Software Technology Corporation. 2010,
LS-DYNA KEYWORD USER’S MANUAL
VOLUME I: Livermore.
390
MARSTRUCT.indb 390
Paik, J.K. & Thayamballi, A.K. 2003. Ultimate Limit
State Design of Steel-Plated Structures. New York:
John Wiley & Sons.
Peschmann, J. 2001. Berechnung der Energieabsorp-
tion der Stahlstruktur von Schiffen bei Kollision und
Grundberührung. Dissertation, Hamburg University
of Technology: Hamburg.
Riece, J. & Tracey, D. 1969. On the ductile enlargement
of voids in triaxial stress fields. J Mech Phy Solids:
201–217.
Tautz, I., Schöttelndreyer, M., Fricke, W. & Lehmann, E.
2010. Experimental Investigation on Collision Behav-
iour of Bow Structures. Proc. of 5th Int. Conference
on Collision and Grounding of Ships (ICCGS). Espoo:
Aalto University.
Törnqvist, R. 2003. Design of Crashworthy Ship
Structures. Ph.D. Thesis, Danish University of
Technoligy: Lyngby.
Wevers, L.J. & Vredeveldt, A.W. 1999. Full scale ship col-
lision experiments 98. TNO-Report 98-CMC-R0359:
Delft.
Yamada, Y. 2006. Bulbous buffer bows: a measure to
reduce oil spill in tanker collisions. Ph.D. Thesis,
Danish University of Technology: Lyngby.
2/18/2011 5:50:19 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Influence of ship motions in the numerical prediction of ship

collision damage

K. Tabri
Department of Mechanics, Tallinn University of Technology, Estonia

J. Broekhuijsen
Damen Schelde Naval Shipbuilding, The Netherlands

ABSTRACT: The paper studies the influence of coupling between the ship motions and the structural
resistance in the prediction of ship collision damage. Several collision scenarios are simulated numerically
using coupled and decoupled approaches. A coupled approach implies a time-domain simulation, where a
precise description of the whole collision process together with the full time histories of the motions and
forces involved, is achieved. There, the ships’ motions are evaluated in parallel to the structural deforma-
tions i.e. the coupling between external dynamic and inner mechanics is preserved. A decoupled approach
is based on the conservation of momentum and allows faster estimation of the deformation energy with-
out providing exact ship motions. The energy to be absorbed in collision is evaluated based on the ship
masses and velocities, and in predicting this energy there is no coupling from the structural behavior. The
ensuing deformations are evaluated in so-called displacement controlled manner—the struck ship is kept
motionless and the striking ship penetrates it along the direction of its initial velocity. The comparison of
the outcomes of two methods reveals that these two methods result in significantly different predictions
of both the penetration path and the deformation energy required to breach the inner hull.

1 INTRODUCTION The accuracy of the decoupled analysis depends

Numerical simulations are often used to assess the
crashworthiness of ship structures and to evalu-
ate the outcomes of ship collisions. A collision
analysis combines two fields, external dynamics
and internal mechanics (Minorsky, 1959). The
external dynamics evaluates the ship motions dur-
ing the collision, giving as a result the energy to
be absorbed by structural deformations, while the
internal mechanics evaluates the structural defor-
mations the ships structures undergo while absorb-
ing that energy.
Numerical analysis of ship collisions is a
demanding task and significant simplifications
are required to achieve the outcomes in reasonable
time. Thus, it is a common approach to decouple
the analysis of structural deformations from the
analysis of ship motions. Such approach is referred
to as a decoupled approach. In a decoupled analy-
sis the problem is solved in a sequential manner:
first the contact force as a function of a prescribed
penetration path is evaluated in so-called displace-
ment controlled analysis of inner mechanics. The
actual extent of the penetration is obtained by
comparing the area under the force-penetration
curve to the deformation energy evaluated with
some calculation model for external dynamics.
on the level of precision on predicting the penetra-
tion path. This can be done rather precisely for a
symmetric collision, where the striking ship collides
under a right angle at the amidships of the initially
motionless struck ship and only few motion com-
ponents are excited. Statistical studies (Lützen,
2001; Tuovinen, 2005) have, however, indicated
that the majority of collisions are non-symmetric
in one way or another. Often the collision angle
deviates from a right angle, the contact point is not
at the amidships or the struck ship is not initially
motionless. As in non-symmetric ship collisions
more motion components are excited, the penetra-
tion path cannot be predefined at reasonable pre-
cision, but it should be evaluated in parallel with
the ship motions i.e. the coupled analysis has to be
exploited. A coupled approach to numerical ship
collision analysis has been proposed for example
by Le Sourne (2001) and Pill and Tabri (2009).
Brown (2002), Tabri (2010) and Tabri (2010b)
compared the coupled and decoupled approaches
and revealed that while the total deformation energy
is predicted rather precisely by the decoupled
approach, associating this energy with the defor-
mations at certain directions cannot be done at the
same precision for all the collision scenarios. In
some scenarios this results in erroneous prediction

391

MARSTRUCT.indb 391 2/18/2011 5:50:19 PM

 of damage length and depth. However, the above
analyses were based on simplified approaches to
inner mechanics and did not consider the struc-
tural aspects at high precision.
In this paper the influence of coupling is studied
further by considering actual ship structures and by
applying Finite Element (FE) method. The paper
addresses the question whether the ship motions
occurring during the collision have influence in pre-
dicting the collision damage and energy. The paper
gives a brief overview of coupled and decoupled
numerical approaches and discusses their advan-
tages and disadvantages. Several non-symmetric
collision scenarios with different collision veloc-
ity, collision angle and location are simulated with
both approaches using explicit FE code LS-Dyna.
The deformation energy required to breach the
inner hull of the struck ship is compared for both
approaches. In all the simulations the struck ship is
assumed initially motionless and the masses of the
colliding ships are assumed to be equal.
2 COUPLED AND DECOUPLED
ANALYSIS OF SHIP COLLISIONS
2.1 Decoupled analysis
First step in the decoupled analysis of ship colli-
sions is the displacement-controlled evaluation of
structural response. In such an analysis the struck
ship is kept fixed through constrained boundary
nodes and the striking ship collides with it at a con-
stant velocity along a prescribed path. FE model
consists of partially modelled striking and struck
ship as depicted in Figure 1.
The striking ship is often modelled as a rigid
body as the stiffness of the bow area is often supe-
rior to that of the side structure. The striking ship
moves at constant velocity along the prescribed
path and the contact between the ships does neither
influence the speed nor the direction of the strik-
ing ship. Therefore, the rigid body motions of the
Figure 1. Displacement controlled calculations.
392
MARSTRUCT.indb 392
colliding ships are excluded from the analysis and
the emphasis is on the modelling of the structural
response. Displacement controlled simulations are
widely used in collision analyses due to their rather
simple manner and fast simulation runs, see for
example Kitamura (2002) or Ehlers et al. (2008).
Above analysis provides the contact force as a
function of this prescribed penetration path. The
actual extent of the penetration is obtained by
comparing the area under the force-penetration
curve to the deformation energy evaluated with
some calculation model that gives the deformation
energy based on the conservation of momentum
in collision; see for example Pedersen and Zhang
(1998). Such a model uses only ship mass proper-
ties, collision velocity, location and angle for input.
External fluid forces due to the surrounding water
are included only through a constant added mass
and the velocity dependent forces, such as hydro-
dynamic damping and frictional resistance, are
neglected as the actual ship motions cannot be
evaluated, see for example Tabri (2010).
The method has several advantages and disad-
vantages. As the motion dynamics are not part of
the simulation, some numerical manipulations can
be made to shorten the duration of the simulation
run. The main advantage is that the time step size
can be increased using so-called mass-scaling in a
certain range without significant loss of accuracy.
In LS-Dyna the time step size is evaluated on
the basis of the Courant stability criteria, which
uses element dimensions, stiffness and density as
input (Hallquist, 2007). The mass-scaling concept
increases the density of the elements in which the
Courant criterion is not met under the prescribed
time step size minima. With reasonable selection
of the step size the increase in mass and the ensu-
ing dynamic effects remain insignificant and the
mass scaling can be used.
Furthermore, the speed of the striking ship often
exceeds the typical speeds in collision. This, and
the fact that the striking ship is not decelerating
under the action of the contact force, significantly
shortens the duration of the collision process and
the simulation time required. Again, as the dynam-
ics are of secondary importance, the error resulting
from sometimes even unrealistically high collision
speed is usually in acceptable range.
The disadvantages of the displacement con-
trolled method stem from the same aspects as
its advantages—neglected collision dynamics.
As the striking ship is moving at constant speed
and the struck ship remains motionless throughout
the collision, the dynamics of the collision, such
as accelerations and velocities, are not physically
correct. This means that some dynamic effects,
such as strain rate effects in deforming structures,
can not be properly included. Furthermore, under
2/18/2011 5:50:19 PM
 the prescribed displacement the collision process
does not describe any actual collision scenario, but
simply gives an assessment on the crashworthiness
of the side structure. The same force-penetration
curve is used for the scenarios with different veloci-
ties, but, as will be show later in the paper, this can
lead to significant errors.
2.2 Coupled analysis
To properly simulate an actual collision scenario
the analysis of structural response cannot be
decoupled from the ship motions. In a coupled
analysis both ships are allowed to move and the
exact penetration history is defined from the
collision dynamics and mechanics. For coupled
simulations the mass, inertia and hydrodynamic
properties of the ships have to be included in the
FE model as the motion dynamics are part of
the simulation. Therefore, the paper utilizes the
dynamic collision simulation procedure proposed
by Pill and Tabri (2009). In this model, the ships’
motions are limited to the plane of water surface
and thus, the restoring forces are not included.
Furthermore, the forces associated with the hydro-
dynamic damping and frictional resistance are
neglected as their inclusion is not straight forward
and as their share in the energy balance is relatively
low, less than 10% of the total available energy
(Tabri, 2010). Hence, the simulation model concen-
trates on the accurate modeling of the main force
components—the contact and the inertial forces.
In the FE model, the masses and inertias of
the colliding ships are modelled by using a small
number of mass points, see Figure 2. Note that
in the figure and in the subsequent sections the
superscripts A and B denote the striking and the
struck ship, respectively. The striking ship consists
of a modelled bow region and three mass points.
Correspondingly, the struck ship consists also of
three mass points and a part of the side struc-
ture. The mass points are constrained to move
together with the boundary nodes of the modelled
Figure 2. Calculation setup for dynamic collision
simulations.
393
MARSTRUCT.indb 393
structural parts and are thereby acting as bound-
ary conditions. The mass nodes of the striking ship
are given the initial velocity v0.
The hydrodynamic added mass components
associated with translational motions are included
in certain directions only. The surge added masses
of the striking and struck ships, denoted as μ11A and
μ11B in Figure 2, are positioned in the centres of grav-
ity of the ships. The added mass associated with the
sway motion is considered only for the struck ship as
the motions of the striking ship are predominantly
in the surge direction. For the struck ship the sway
added mass is modelled as a single block of addi-
tional mass located on the opposite side of the strik-
ing location, see Figure 2. This added mass block
is constrained to the mass points through a planar
joint, which restricts relative movement in sway
direction and allows the joined entities to move in
the surge direction. Thus, this mass becomes active
only if the struck ship undergoes sway motion.
The advantage of the dynamic simulations is
that they simulate an actual collision event and
no prescriptions other than initial conditions are
required. The drawback is that the mass-scaling
would lead to larger errors compared to displace-
ment controlled simulations and is therefore not
suggested. As both ships are moving, the whole
contact process lasts longer as it takes longer time
for the striking ship to penetrate deep enough into
the struck ship to cause breaching of the inner
hull. Therefore, the dynamic simulations require
significantly longer computation time.
3 NUMERICAL STUDY
Numerical study is conducted to evaluate the dif-
ference between two approaches. Several collision
scenarios between two river tankers are simulated
numerically to study the influence of coupling.
The energy required to breach the inner hull of the
struck ship is evaluated with both approaches and
compared.
In the decoupled displacement controlled analy-
sis the inner mechanics is evaluated without any
influence from the ship motions and thus this
energy depends on the structural properties of
the ships and on the collision angle and location.
Thus, in the decoupled analysis we only look at the
inner mechanics and leave the external dynamics
completely out of the investigation.
In the coupled analysis the inner mechanics and
the external dynamics are coupled and thus, full
description of the collision process is achieved.
Obviously, now the collision process also depends
on the initial velocity of the striking ship and
therefore the collision scenarios are simulated at
different velocities.
2/18/2011 5:50:20 PM
 3.1 Colliding ships
It is assumed that the colliding ships are similar.
Both ships have the same mass of 2771 tons and
the length of 110 m. The striking ship has a rectan-
gular bow shape as seen from Figure 2. The struck
ship is depicted in Figure 3 and it has longitudinally
stiffened double-hull with web-frame spacing of
1.65 m and bulkhead spacing of 8.26 m. The main
dimensions of the ships are presented in Table 1.
The added mass values presented in the table are
calculated with the strip theory (Journée, 1992).
The FE models of the striking bow and the side
structure are the same for the displacement con-
trolled and the dynamic simulations. While the
bow structure is modelled as a rigid body, the side
structure is allowed to undergo large deformations.
Ships are assumed to be made of shipbuilding steel
with yield strength of 284 MPa. Strain-rate sensi-
tivity of the steel material is not considered.
Length of the modelled side structure is 24.75 m,
which is sufficient for the plastic deformations not
to occur at the boundaries. The fracturing of the
structures is described by thru thickness failure
criteria, which background and implementation
Figure 3. Structure of a river tanker.
Table 1. Main dimensions of the ships.
394
MARSTRUCT.indb 394
is explained in Zhang et al. (2004), Scharrer et al.
(2002) and Ehlers et al. (2008). The thru thickness
strain criterion in LS-DYNA erases an element
if the through thickness strain reaches its critical
limit. This approach requires rather fine mesh and
therefore, in the contact area the prevailing element
dimension is 30 × 30 mm. The inner hull is consid-
ered to be breached when at least four elements of
the inner plating are erased.
3.2 Collision scenarios
In total, 13 different collision scenarios are evalu-
ated to study the influence of collision velocity,
angle and location. Collisions to three different lon-
gitudinal locations along the side of the struck ship
as depicted in Figure 4 are studied. In Figure 4 the
extent of the FE model is visualized only for location
1. In other two locations the FE model had the same
overall dimensions and structural configuration.
The eccentricities of different collision locations are
given in Table 2. In all the displacement controlled
simulations the striking ship is travelling at con-
stant velocity of 5 m/s along the direction of its
initial velocity. In dynamic coupled simulations
five different collision velocities and three collision
angles are studied for location 1. In locations 2 and
3 the collision angle is always 45 degrees and the
collision velocity is either 3.5 or 5 m/s. The collision
scenarios are summarized in Table 2.
Figure 4. General arrangement of the struck ship and
the collision locations.
Table 2. Simulated collision scenarios.
2/18/2011 5:50:21 PM
 4 INFLUENCE OF COUPLING
Collision scenarios defined in Section 3.2 are simu-
lated with the coupled and decoupled approaches.
The deformation energy at the time instant when
the inner hull is breached is defined for both
approaches and compared. This deformation
energy consists of energy absorbed by deformation
mechanisms such as bending, stretching and tear-
ing of structural elements and also of the energy
absorbed by the friction between the ships.
The relative difference between the two
approaches is defined as
EC EDC
diff . = (1)
EC
where EC and EDC are deformation energy required
to breach the inner hull in the coupled and the
decoupled analysis, respectively.
The simulation results and the difference between
the coupled and decoupled approach are presented
in Table 3 and discussed in subsequent sections.
The table presents both the total deformation
energy (total) and the energy absorbed by friction
(friction) between the ships. Difference between
the total and friction energy is the energy absorbed
in deformation mechanisms such as crushing, tear-
ing, stretching, bending etc of structural members.
In Table 3, the change in collision angle, Δβ, is
measured at the moment of the breaching of the
Table 3. Simulation results.
395
MARSTRUCT.indb 395
inner hull or at the end of the simulation for the
scenarios, where the breaching did not occur.
4.1 Collision velocity
We first study the influence of the collision velocity
and look at the scenario, where the collision occurs
at location 1 and the collision angle is 45 deg. In
the displacement controlled calculation the inner
hull is breached when the deformation energy
reaches to 3.66 MJ, see Table 3.
The same scenario is simulated with the coupled
approach at five different collision velocities varying
from 2.1 m/s to 5 m/s. The difference in deformation
energy between the two approaches is depicted in
Figure 5, where dashed line presents the scenarios
where the inner hull is not breached and the solid
line presents the scenarios with breaching.
At very low velocity, v0 = 2.13 m/s, the striking
ship does not have enough momentum and the
inner hull is not breached. The total amount of the
energy absorbed in the collision is 2.85 MJ (Table 3).
Increasing the velocity v0 to 2.92 m/s does still not
result in the breaching of the inner hull even though
the deforming ship structures absorb already 70%
more energy than was required to breach the inner
hull in displacement controlled simulation.
Inner hull is breached when the collision veloc-
ity is 3.5 m/s or higher. With v0 = 3.5 m/s the
difference in energy is 70% and the structure is
able to withstand more than predicted with the
decoupled approach. This is a clear indication that
different approaches yield to different collision
dynamics and mechanics. In dynamic coupled
simulations the ships are allowed to move and
thus, the striking ship follows a different path until
it meets the inner hull. Larger portion of the side
structure is included in mitigating the collision
and thus the deformation energy becomes higher.
With the different penetration path and the larger
Figure 5. Difference in deformation energy as a func-
tions collision velocity.
2/18/2011 5:50:22 PM
 portion of the deformed structure also the energy
absorbed in friction is doubled compared to the
displacement controlled simulations.
The difference between the methods decreases
as the velocity v0 increases. This is expected as the
higher velocity leaves less time for the struck ship
to start moving and thus, the penetration path
becomes similar to the straight lines assumed in
displacement controlled simulations.
For the scenario with v0 = 3.5 m/s the difference
in penetration path can be seen when looking at
the damage opening in the outer plating at the
time instant, when the inner hull is breached, see
Figure 6. In dynamic coupled analysis the width
of the damage opening is 3.6 m, while with the
displacement controlled analysis the length of the
opening is only 3.05 m.
The difference between the two approaches
is the highest in the vicinity of so-called critical
velocity, which is the minimum velocity required to
breach the inner hull in a certain collision scenario.
Thus, the critical velocity presents a threshold after
which the consequences of the collision become
more serious. It can be concluded that the evalu-
ation of this critical velocity with the decoupled
analysis presents a conservative approach.
4.2 Collision angle and location
Above analysis showed that the collision veloc-
ity has a strong influence on the precision of the
Figure 6. Damage length in the outer plating at the
moment of breaching of the inner hull.
396
MARSTRUCT.indb 396
decoupled analysis. The influence of the collision
angle and location is studied next. First, the colli-
sions are simulated in location 1 under the angles
30, 45 and 60 deg with velocity v0 either 3.5 m/s or
5 m/s. The influence of the collision angle on the
difference between the approaches is presented in
Figure 7.
Figure 7 indicates that the difference decreases
when the angle β between the ships approaches
to a right angle i.e. β→90 deg. This agrees with
the conclusion of Tabri (2010b) that at a right
angle collisions the coupled and the decoupled
approaches lead to very similar outcome. When
the angle between the ships becomes smaller
(β→0), the motions of the ships become more
severe and the actual penetration path deviates
from the straight line assumed in displacement
controlled simulations and thus, larger differences
occur. With v0 = 3.5 m/s the contact between the
ships is lost before the breaching of the inner hull
occurs.
As a result of the contact, the ships start to
rotate and the angle β changes to some extent
denoted as Δβ in Table 3. The maximum change in
β was approximately -2 degrees corresponding to
almost 2 m of translational displacement between
the ships at the contact region.
Compared to the collision angle, the eccen-
tricity has much lower influence on the differ-
ence between the methods, see Figure 8. With
v0 = 5 m/s the results are almost independent of
the value of the eccentricity, while at v0 = 3.5 m/s
the influence is stronger and the differences
between the methods becomes larger as the eccen-
tricity from the amidships increases. The differ-
ence in energy is already close to 90% i.e. in the
coupled analysis the inner hull of the struck ship
is not yet breached even though the side struc-
ture has already absorbed almost twice the energy
required to breach the inner hull in the decoupled
analysis.
Figure 7. Difference in deformation energy as a func-
tions collision angle.
2/18/2011 5:50:23 PM

Figure 8. Difference in deformation energy as a func-
tions collision location (β = 45 deg).
5 CONCLUSIONS
The paper compared two different approaches
for ship collision simulations with FE method. In
the decoupled displacement controlled method,
the struck ship is kept motionless and the strik-
ing ship penetrates it along the direction of its
initial velocity. The coupled dynamic simulations
consider the actual ship motions in collision and
thus, the collision is described more accurately.
The energy required to breach the inner hull of
the struck ship was evaluated with both methods
and compared.
It was shown that for a certain collision sce-
narios the coupled dynamic simulations reach to
significantly different deformation energy at the
moment of inner hull breaching in comparison to
the decoupled analysis. In the dynamic simulations
the yaw motion of both ships is excited and the
penetration paths became different from those of
displacement controlled simulations. Thus, differ-
ent portion of the side structure of the struck ship
was engaged in energy absorption and the defor-
mation energy becomes different.
The difference in deformation energy between
the two approaches was shown to be up to 90%. The
collision velocity was revealed to have the strongest
influence on the precision of the displacement
controlled approach. Also the collision angle had
a relatively high influence on the precision while
the collision location had only modest effect on the
precision. Large differences between the methods
indicate that the decoupled analysis should be used
with care as in some scenarios the error in energy
predictions can be very high.
397
MARSTRUCT.indb 397
Here, the simulations were only conducted for a
single mass ratio between the ships. The mass ratio
has a strong influence on the collision dynam-
ics and thus also on the differences between the
approaches. The effect of different mass ratios,
structural configurations, forward velocity of the
struck ship etc. on the precision of the coupled
approach is still to be studied.
ACKNOWLEDGEMENTS
This research work has been financially supported
by the European Social Fund (grant agreement no.
MJD110). This help is here kindly appreciated.
REFERENCES
Ehlers, S., Broekhuijsen, J., Alsos, H.S., Biehl, F. &
Tabri, K., 2008. Simulating the collision response
of ship side structures: A failure criteria benchmark
study. Int Ship Progress, 55:127–144.
Hallquist, J.O., 2007. LS-DYNA. Keyword User’s
Manual, Version 971. Livermore Software Technology
Corporation.
Journée, J.M.J., 1992. Strip theory algorithms. Delft Uni-
versity of Technology; Report MEMT 24.
Kitamura, O., 2002. FEM approach to the simulation of
collision and grounding damage, Marine Structures,
15:403–428.
Lützen, M., 2001. Ship Collision Damage. Ph.D. Thesis,
Technical University of Denmark.
Minorsky, V.U., 1959. An analysis of ship collision with
reference to protection of nuclear power plants. J Ship
Research, 3:1–4.
Pedersen, P.T. & Zhang, S., 1998. On Impact Mechanics
in Ship Collisions. Marine Structures, 11: 429–449.
Pill, I. & Tabri, K., 2009. Finite element simulations
of ship collisions: A coupled approach to exter-
nal dynamics and inner mechanics. In: Analysis and
Design of Marine Structures, Guedes Soares, C. and
Das, P.K. (Eds.), Taylor & Francis Group, London,
U.K., 2009, pp. 103–109.
Scharrer, M., Zhang, L. & Egge, E.D., 2002. Final report
MTK0614, Collision calculations in naval design
systems, Report Nr. ESS 2002.183, Version 1/2002-11-22,
Germanischer Lloyd, Hamburg, p. 61.
Tabri, K., 2010. Dynamics of Ship Collisions. Doctoral
dissertation. Aalto University, available at http://lib.
tkk.fi/Diss/2010/isbn9789522482730/
Tabri, K., 2010b. Influence of coupling in the prediction
of ship collision damage. In: Proceedings of 5th Inter-
national Conference on Collision and Grounding of
Ships, Espoo, Finland, 14-16.07.2010. pp. 133–138.
Tuovinen, J., 2005. Statistical Analysis of Ship Collisions.
Master’s Thesis, Helsinki University of Technology.
p. 93.
Zhang, L., Egge, E.D. & Bruhns, H., 2004. Approval
Procedure Concept for Alternative Arrangements,
In: Proceedings of the 3rd International Conference
on Collision and Grounding of Ships, IZU, Japan,
pp. 87–96.
2/18/2011 5:50:24 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Numerical prediction of impact loads in rectangular panels

R. Villavicencio & C. Guedes Soares

Centre for Marine Technology and Engineering (CENTEC), Technical University of Lisbon,
Instituto Superior Técnico, Lisboa, Portugal

ABSTRACT: The paper presents finite element simulations examining the dynamic response of small
rectangular panels struck transversely by a mass. Three panel widths with different stiffener distribution
are considered in the analysis. The influence of the impact velocity and impact location on the final results
is discussed. The material true stress-strain curve, used in the numerical model, is obtained from tensile
tests on the actual material used to fabricate the panels. The mesh size, the true material curve definition
and the critical failure strain estimation are evaluated from the numerical simulation of the tensile test.
A brief study of the definition of the boundary conditions is conducted, in order to determine the
simplest, but accurate assumption that will be used in the finite element model. The force-displacement
curves and the shape of the deformation from the impact simulations are used to compare the plastic
response of the impacted panels.

1 INTRODUCTION the prediction of the extent of damage on stiffened

The structural design of ships concerning collision
requires an accurate prediction of the damage
of stiffened plates under impact loading. Experi-
mental studies on laterally loaded panels have been
conducted in order to derive analytical expressions.
For example, Hagiwara et al. (1983) proposed a
method for predicting low-energy ship collision
damage based on combined experiments, which
determined the initiation of plate fracture, the
effects of structural details and the deformation
of a typical ship panel, concluding that the most
important consideration is the estimation of the
initiation of side shell plate fracture.
Manolakos and Mamalis (1985) used a rigid
plastic analysis for predicting the structural
behaviour of longitudinally framed shell plating
of struck vessel during a minor oblique collision,
finding that a collision at right angles is the most
severe condition.
Experimental and theoretical analyses of later-
ally loaded stiffened plates considering various
boundary conditions were conducted in order to
calculate the upper and lower bound loads needed
to cause plastic collapse (Manolakos and Mamalis
1988) or to investigate the effect of Poisson’s ratio
in the stiffeners (Boot and Moore 1988).
Langseth and Larsen (1990) studied experimen-
tally the plugging capacity of simply supported and
continuous steel plates with stringers in the free
span subjected to a dropped drill-collar making
comparison between dynamic and static tests. Cho
and Lee (2009) developed a simplified method for
plates due to lateral collisions, being the method
validated with experiments.
Finite Element (FE) analysis is a useful tool to
predict the extent of ship collision and its struc-
tural components damage. However, the nonlinear
dynamic analysis should be compared with experi-
mental tests before being used for a structural
design. Unfortunately, experimental tests on full
scale ship collision are rare and very expensive.
One approach is to perform scaled collision test
on typical ship structural members to validate the
numerical methods for impact analysis.
Analysis of individual ship structural compo-
nents under lateral impact loads, such as beams
and plates have been widely analyzed. However,
it is difficult to find results from the comparison
between experimental impact tests and numerical
simulations on stiffened plates. In most of the cases,
the impact response is examined by penetrating the
panels using quasi-static lateral loads or applying
central concentrated loads, which does not con-
sider the extremely high impact forces or vibrations
developed at the first period of contact between
the indenter and the specimen. Thus, the numeri-
cal analyses of stiffened plates have been simplified
to simulate this quasi-static contact or to calculate
the ultimate strength of panels under other type
of loads, like uniaxial compression or laterally dis-
tributed pressures. It is considered that, the effect
of dynamic lateral impact on stiffened plates must
be analyzed considering high initial velocities
in the striking mass, thus representing a realistic
collision.

399

MARSTRUCT.indb 399 2/18/2011 5:50:25 PM

 In spite of that, experimental and numerical
analyses have been performed on stiffened plates
which in general simulates the bottom or side
structures in a ship. For example, Deb and Bootom
(1988) compared experiments with two linear finite
element models for bending of stiffened plates
under transverse loading where the stiffeners were
modeled as beams placed along the plate nodal
lines and within plate elements.
Zhu and Faulkner (1994) reported results from
experimental and numerical studies on the dynamic
response of plates under impact load in minor ship
collision using a simplified model. The combined
work gave a better understanding of the collision
process, especially of the dynamic stress and strain
aspect of the ship plating during the collision.
Caridis et al. (1994) presented a numerical
model for the response of thin plates subjected to
dynamic loading which was compared with test
results on models of ship plating obtaining good
predictions of the permanent deflections.
A methodology for the analysis of multi-stiffened
plates under lateral loading using a sequential
quadratic programming to determine the coeffi-
cient that minimize the total potential of the system
was presented by Bedair (1997) who illustrated the
efficiency of various stiffening configurations.
A comparison between experimental and finite
element results of a scaled double hull structure
representing ship-to-ship collision was presented
by Wu et al. (2004), obtaining good results in terms
of general structural response. However, deviations
were observed being attributed to specific finite
element parameters.
Numerical simulations of the collision response
of ship side structures were performed by Ehlers
et al. (2008). A strong sensitivity of the failure
criteria was found in the comparison between the
experimental and numerical results. Experimental
and numerical results on various configurations
of stiffened plates indented by a spherical indenter
were reported by Alsos et al. (2009a, b). They
investigated the performance of two failure criteria
considering the influence of the element size with
respect to the onset of failure.
An important parameter, predicted by means of
numerical analyses of structures under impact loads,
is the failure due to material fracture. However, it
is difficult to establish a procedure suitable for the
prediction of failure in the engineering practice.
Criteria for predicting the dynamic inelastic failure
of structures has been examined by some authors
(Shen and Jones 1992; Simonsen and Lauridsen
2000; Tabri et al. 2007).
In practical terms, a failure criterion based on
the size of the element, which has been widely
used to evaluate the response under impact in ship
collision, was presented by Zhang et al. (2004) in an
400
MARSTRUCT.indb 400
approval procedure worked out by Germanischer
Lloyd, which provides a standard for evaluation of
alternative solution for design and construction of
cargo vessels. The procedure also includes a material
nonlinearity definition for finite element analysis.
In the present study, numerical simulations of
stiffened plates under impact load are conducted, in
order to predict the energy absorption, maximum
deflection and failure. Complete details of the FE
model are presented, including previous analysis of
the mesh size, the material true stress-strain curve,
and the selection of the boundary conditions. The
definitions in the FE model are mainly based on
previous experimental-numerical analyses con-
ducted by the authors on clamped beams and plates
transversely impacted by a mass (Villavicencio and
Guedes Soares 2009, Villavicencio and Guedes
Soares 2011, Villavicencio et al. in press). The analy-
sis of the sensitivity of different parameters, such as
the impact velocity, type and distribution of stiffen-
ers, width of panel, and impact along the width is
summarized. The force-displacement response of
the impacted panels is used to compare the influence
of different parameters. The shape of the deforma-
tion is also included in this work, thus always taking
into consideration the physics of the impact phe-
nomenon when interpreting the graphical results.
The present work aims at analyzing in small
panels important finite element definitions, such as
material nonlinearities and boundary conditions.
This work is limited to a finite element study, which
is to be validated with experiments on specimens
with the same dimensions as the ones used in the
actual simulations.
2 SCOPE OF STUDY
A group of numerical analysis was defined in order
to study different parameters, such as the impact
velocity, type and distribution of stiffeners, width of
the panel, and impact along the width. The selection
of the simulations is based on the capacity of the
impact machine and the stock of the specimens.
The details of the “further” experimental set up
described in the present Chapter help to under-
stand the definitions adopted in the FE model.
The impact tests will be performed using a
fully instrumented Rosand IFW5 falling weight
machine, which gives the variation with time of
the impact force, velocity, displacement and energy
of the impacted specimens. The experimental set
up can be seen in Figure 1. The two edges of the
panels in the longitudinal direction (length of the
panel) are supported between two thick steel plates
and are compressed by two bolts at each support.
The two edges in the transverse direction (width of
the panel) remain free. The support plates are fixed
2/18/2011 5:50:25 PM
 width; the indexes 1, 2, 3, 4 or 5 denote the type
or distribution of stiffener. In all cases, the panel
length (longitudinal distribution of stiffeners)
is 250 mm, and the stiffened plate thickness is
4.0 mm. Thus, the width to length ratios in Panels
A, B and C are 0.5, 1.0 and 1.5, respectively. The
stiffeners are flat bars 25 × 4 mm and L profiles
50 × 50 × 5 mm. The indenter is a hemi-spherically
ended projectile of diameter 30 mm which uses a
striking mass of 84 kg. The impact velocities and
the impact locations are summarized in Figure 2.
The material of the plate and stiffeners is structural
steel ASTM A36 (ASTM Standards 1989).

3 FINITE ELEMENT MODEL

The computations were carried out using the finite

Figure 1. Experimental set-up. element package LS-DYNA Version 971 (Hallquist
2005) which is appropriate for non-linear explicit
dynamic simulations with large deformations.
To define the FE model, analyses of the material
definition, mesh size, and boundary conditions were
conducted, as described in the following sections. The
selected numerical model is sketched in Figure 3.
The plate and stiffeners were modelled by 4-node
shell elements with 5-integration points through the
thickness, using the default Belytshko-Tsay formu-
lation. The striking mass was modelled in 4-node
shell elements with 2-integration points through
the thickness. The selected mesh sizes of the plate
and the stiffeners were 2.0 mm (see details on
Section 4). The striking mass was modelled using
a mesh size of approximately 1.0 mm, in order to
design sufficiently accurately the semi-spherical
geometry of the indenter.
The supported perimeter of the panel was con-
strained in all degrees of freedom (see details on
Section 5). For the striking mass, only the vertical
translation was free, in which direction the initial
impact velocity was assigned. The contact between
the striking mass and the stiffened plate was
defined as ‘Contact Automatic Single Surface’,
which uses nodal normal projections resulting in
a continuous contact surface. The reaction forces
structure-indenter were obtained by the ‘Contact
Force Transducer Penalty’, which defines the body
(specimen), with respect to the indenter, from which
Figure 2. Design of panels and impact events.

to a strong structural base to prevent their move-

ment. A rubber pad between the impact machine
floor and the structural base allows avoiding vibra-
tions during the impact.
Three widths of panel are considered: 125,
250 and 375 mm (Fig. 2). The panels are denoted
by the letters A, B or C depending on their span Figure 3. Details of finite element model.

401

MARSTRUCT.indb 401 2/18/2011 5:50:25 PM

 the contact forces are measured (Hallquist 2005). 0
600
Connection
A static and a dynamic coefficients of friction GL
500
of 0.3 and 0.1, respectively, were included in the UN+GL
contact definition (Ehlers 2010; Villavicencio et al. 400

Stress [MPa]
in press). Engineering
The selected material from the library of Curve
300
LS-DYNA ‘Mat.024-Piecewice Lineal Plasticity’
defines the plate and the stiffeners, and provides a 200 Steel ASTM A36
Tensile strength 400-550MPa
definition of the true stress-strain curve as an offset Yield point, min 250MPa
100
table (for details, see Section 4). The striking mass Elongation in 200 mm, min 20%
was modelled as a rigid undeformable material.
For this propose, ‘Mat.020-Rigid’ was selected 0.00 0.05 0.10 0.15 0.20 0.25 0.30
Strain [-]
from the material library of LS-DYNA assigning
mild steel mechanical properties. The mechanical
Figure 4. Engineering and true stress-strain curves.
properties are used for determining the sliding
interface parameters in the contact definition
when the rigid body interacts. It is necessary to
give realistic values of these properties, since unre- simulations of the tensile test is summarized as
alistic values may contribute to numerical prob- follows (Villavicencio et al. in press): Only the
lems in the contact definition (Hallquist 2005). length between the clamping edges of the tensile
Since the falling weight assembly was modelled as test specimen was modelled (Fig. 5). For initiating
a simple shell elements semi-sphere, an artificially necking, the width of the specimen at the centre
large density was used to give the same mass as the was gradually reduced by 0.1 mm. Shell elements
one to use in the experiments. with default LS-DYNA formulation were used.
The strain-rate sensitivity behaviour of mate- Default hourglass control was specified. How-
rials in the FE model can be included using the ever, the mesh was designed to avoid hourglassing
coefficients of the Cowper-Simonds constitutive by orientating the elements diagonally (Fig. 5). The
equation (Jones 1989). However, the theoretical translational degrees of freedom were restricted at
behaviour of strain-rate-sensitivity structures one end and at the other end a constant displace-
summarized by Jones (1989) shows that the mate- ment of 100 times the experimental speed was pre-
rial strain rate had a small effect at relatively low scribed (Ehlers and Varsta 2009). The force of the
incident energies when structures such as beams displaced nodes at the free end is obtained from the
and plates are subjected to uniform impulsive calculations. This force, plotted versus the applied
velocities. Consequently, the strain rate sensitivity prescribed displacement, gives the engineering
was omitted from the finite element model. stress-strain behaviour.
Two mesh sizes of 5.0 and 2.0 mm (denoted by
Shell5 and Shell2 respectively) were analyzed, and
4 MATERIAL AND MESH SIZE two true stress-strain curves were used to define
DEFINITION the material. The first true material curve is the one
proposed by Zhang et al. (2004), where the true
The mechanical properties of the material used in stress σt and the true strain εt are expressed by:
the FE models were obtained from the ‘Standard
Specification for Carbon Structural Steel ASTM σt C εt n (1)
A36’ (ASTM Standards 1989). The engineering
where
stress-strain curve is plotted in Figure 4.
Since collision calculations involve extreme n = 1n(1 + Ag) (2)
structural behaviour with both geometrical and
nonlinear effects, the input of material properties and
up to the ultimate tensile stress has a significant
influence on the extent of critical deformation C = Rm (e/n)n (3)
energy. Thus, it is generally recommended to
use true stress-strain relationship, which can be Ag is the maximal uniform strain related to
obtained from a tensile test. the ultimate tensile stress Rm and e is the natural
The experimental tensile test (Fig. 4) was mod- logarithmic constant. If only the ultimate stress
elled in LS-DYNA, in order to verify the plastic Rm (MPa) is available, the following approximation
response of the material and to explore the stress can be used to obtain the proper Ag:
and strain distribution in the neck at failure.
The general procedure to perform the numerical Ag = 1/(0.24 + 0.01395Rm) (4)

402

MARSTRUCT.indb 402 2/18/2011 5:50:26 PM


Stress [MPa]
450 UN+GL
400
350 GL
300 Engineering
250
200
150
100
50
0
0.00 0.05 0.10 0.15
Figure 5. Tensile test simulations.
The second material curve is the true stress-strain
curve until the onset of necking (Dieter 1986),
where the true stress σt and the true strain εt are
expressed in terms of engineering stress σeng and
engineering strain εeng by:
σt = σeng(εeng + 1)
εt = 1n(εeng + 1)
since Eqs. 5 and 6 are valid only until maximum
load, the curve beyond necking was continued
using the Zhang et al. (2004) approximation, as
shown in Figure 4. In the present work, the mate-
rial curve defined by Zhang et al. (2004) is denoted
by “GL” and the true stress-strain curve until the
onset of necking is denoted by “UN + GL”. The
two defined material curves are plotted in Figure 4,
demonstrating deviation between each other until
the point of maximum load.
The critical failure strain adopted in the present
study is the one recommended by Zhang et al.
(2004), which considers the element size, stress
state, and manufacture influence. The failure strain
at the moment of fracture is calculated by:
εf (le) = εg + εe (t/le)
where εg is the uniform strain, εe is the necking
strain, t is the plate thickness, and le is the individual
element length. The recommended values for the
uniform and necking strain for shell elements are
0.056 and 0.540, respectively. It is commonly rec-
ommended that the ratio le/t is not less than 5 for
shell elements. In the actual specimens, the ratio le/t
is 1.25 and 0.8. However, failure benchmark studies
on ship collision (Ehlers et al. 2008) demonstrated
that similar results can be obtained using shell ele-
ments with length similar to the plate thickness.
The tensile test simulations were conducted
using both material curves and both element sizes,
and considering their corresponding failure strain
estimated by Eq. 7. Figure 5 shows the deformed
mesh of the tensile test specimens one instant before
fracture and the resulting numerical engineering
MARSTRUCT.indb 403
stress-strain curves. It is observed that, until
Shell5 maximum load, similar results are obtained using
Shell2
the two material curves and the two mesh sizes.
However, it must be noted that a better approxima-
Tensile Test
tion is given by UN + GL material, which follows
Shell5_GL the experimental curve quite precisely using both
Shell2_GL
Shell5_UN+GL mesh sizes Shell5 and Shell2. Beyond necking, the
Shell2_UN+GL influence of the mesh size is significant, and the
Strain [ - ] results tend to deviate; a better approximation is
0.20 0.25 obtained using the Shell2 model.
Based on the numerical simulation of the tensile
test, the UN + GL material and the mesh size of
2.0 mm are selected to define the FE model of the
stiffened plates. The adopted failure criterion is
considered satisfactory for the mesh size of 2.0 mm,
which gives differences between the numerical and
experimental results of 10 and 5% in the fracture
stress and fracture strain, respectively. It must be
mentioned that the decrease in the mesh size of
the tensile test specimen brings some expected
(5) improvement in the prediction of the point of
fracture. For example, this situation was observed
(6)
when the mesh size of 5.0 mm was decreased to
2.0 mm. Although the numerical simulations tend
to deviate beyond necking, the approximation is
satisfactory for this relatively coarse mesh size.
Panels B1 and B2 were selected to evaluate their
force-displacement response under impact using
the selected UN + GL material. Both mesh sizes
2.0 and 5.0 mm were considered when studying
the influence of the failure criterion for a structure
suffering transverse impact. Figure 6 shows that,
when only plastic deformation occurs, similar
results are obtained using both mesh sizes (e.g.
Panel B2). Similarity between the results before
necking, irrespective of the mesh size, was pre-
dicted by the numerical simulations of the tensile
tests. In Panel B1 (unstiffened plate), the fracture
is reached by Shell5 model, whereas Shell2 model
(7) only predicts the plastic deformation. This situation
was observed in the tensile test simulations, when
the fracture stress-strain in the Shell5 model occurs
earlier, as compared to the Shell2 model, being
100 B2
Shell5_B1
Shell2_B1 B1
80 Shell5_B2
Shell2_B2
Force [kN]
60
40
20
0
0 10 20 30 40
Displacement [mm]
Figure 6. Mesh size analysis: Panels B1 and B2; impact
at mid-span; initial velocity 6.0 m/s.
403
2/18/2011 5:50:27 PM
 clear the mesh size dependence when the critical
failure strain is estimated.
The mesh size of 2.0 mm was selected for two
reasons: a) due to the similitude with the numerical
simulation of the tensile tests b) it is able to define
a better shape of the deformation in the impacted
panels.
5 BOUNDARY CONDITIONS
In the experimental tests it is almost impossible to
satisfy precisely the zero displacement condition
at the supports, because the load capacity of
the structure is strongly dependent on the axial
restraint provided at the supports. Therefore, when
developing numerical models that are to be com-
pared with experimental results, it is necessary that
the real boundary conditions are represented in the
numerical model, instead of the ideal ones. In the
present work was conducted a study of the bound-
ary conditions, in order to select the restraints that
not only represent the experimental supports, but
also simplify the numerical model. The boundary
condition study is mainly based on definitions
adopted in previous experimental-numerical analysis
(Villavicencio and Guedes Soares 2011) using simi-
lar structural supports to clamp the specimens.
Three models were considered in the previous
analysis of the boundary conditions. First, the
support plates were represented using shell
undeformable elements compressing the specimen
as occurred in the experiments. This model is
denoted by “Support Plates” (Fig. 7a). In this
Figure 7. Models of boundary conditions: a) support
plates; b) bolt constrained; c) boundary conditions.
404
MARSTRUCT.indb 404
boundary condition representation, all the support
plate length was modelled and no gap between
the support plates and the supported area of the
specimen was considered. The lower support plate
was constrained in all degrees of freedom. The upper
support plate was constrained in all degrees of free-
dom, except for vertical translation, because a pre-
scribed vertical motion was imposed to compress
the supported surface of the panel simulating the
clamped condition. The value of the prescribed
displacement was equal to εyt/3, where εy is the
yield strain of the material and t is the thickness of
the supported stiffened plate. Using this boundary
condition representation, some small longitudinal
displacements of the supported portion of the
panel between the support plates were observed,
which are due to the high incident energy applied.
The sliding between the support plates gives some
distortion in the force-displacement response,
adding extra transverse deflection at the impact
point and a vibratory response at the end of the
impact event (Fig. 8).
To avoid these small longitudinal displacements,
nodal restrictions were included in the displacements
and rotations of the second model in the zone where
the bolts pass through the plate. This second model
is denoted by “Bolt Constrained” (Fig. 7b). The
restricted perimeter at the bolt holes corresponds
to the contact between holes and pins considered
in the mechanical engineering design. Taking into
account these nodal restrictions, yields a more
realistic simulation of the “expected” experimen-
tal clamped conditions, because the bolts in the
experimental set-up not only press the supported
portion of the panel, but also help avoiding the
longitudinal displacements at the support.
As in FE analysis, the idea is to generate simple
geometries and obtain results similar to those
using complex models. Subsequently, the boundary
restraining displacements and rotations at the
nodes of the supported perimeter was evaluated.
This model is denoted by “Boundary Conditions”
(Fig. 7c). A similar response was observed using
120
Shell2_Support Plates
100 Shell2_Bolt Constrained
Shell2_Boundary Conditions
80
Force [kN]
60
40
20
0
0 5 10 15 20 25 30 35
Displacement [mm]
Figure 8. Boundary condition analysis: Panel A1; impact
at mid-span; initial velocity 6.0 m/s.
2/18/2011 5:50:27 PM
 the models “Bolt Constrained” and “Boundary
Conditions” (Figs. 7b and 7c, respectively). Based on
the good agreement between them, the “Boundary
Conditions” model is selected to evaluate the
plastic response of the panel under impact loads,
i.e. the supported perimeter of the panel was
constrained in all degrees of freedom as summa-
rized in Chapter 3 and Figure 3.
It should be noted that, when the bolts restriction
is considered, some very small longitudinal displace-
ments were noticed between the support plates.
However, as it was mentioned, in experimental
tests it is almost impossible to satisfy precisely the
zero displacement condition at the supports. Thus,
some displacements are also expected in the experi-
mental analysis, although it is considered that they
are very small and will not affect the total response
of the impacted panel.
6 NUMERICAL RESULTS
Only part of the numerical results for the total
number of specimens represented in Figure 2 are
summarized in the present Chapter. The described
results are the most representative of the numeri-
cal simulations and were selected for the run of
“further” experiments to compare their plastic
response. The results are presented in terms of
the force-displacement response and the deformed
shape of the impacted panels.
6.1 Group 1 of tests: impact velocity, width
of panel and stiffener type when the impact
is at the mid-span
Panel types A1-A2-A3 and B1-B2-B3 were numeri-
cally simulated to analyze the influence of the
impact velocity, width of panel and stiffener
type when were impacted at the mid-span. For
example, Figure 9 represents the sensitivity of the
width of panel and the impact velocity for Panels
140
V6
120
Panel A2
100
Force [kN]
80 V4
Panel B2
60
V2
40
V4
20
V2
0
0 10 20
Displacement [mm]
Figure 9. Group 1 of tests: Panels A2 and B2; impact
at mid-span; initial velocities 2.0, 4.0 and 6.0 m/s (V2,
V4 and V6).
MARSTRUCT.indb 405
A2 and B2. It is observed that, when the impact
velocity increases, larger transverse displacements
and impact forces are developed in both panels.
Considering the influence of the width of panel,
it is noticed that, when the width increases, larger
transverse displacements are obtained at the impact
point. However, the impact forces decrease. It
should be noticed that, when the width increases,
the maximum transverse displacement is larger
than the permanent transverse displacement (e.g.
Panel B2). On the other hand, when the panel width
decreases (e.g. Panel A2), the maximum and perma-
nent displacements are more similar in magnitude.
The deformed shape of Panel B2 when impacted
at the mid-span at velocities of 2.0, 4.0 and 6.0 m/s
is shown in Figure 10. It is observed that, the panel
suffers mainly global deformation when impacted
at low incident energies (e.g. impact velocity
2.0 m/s), whereas it suffers both local indentation
and global deformation when the incident energy
increases (e.g. impact velocity 6.0 m/s). The local
indentation plays an important role in the total
deformation of the stiffened plate, being more
than one-quarter of the total deformation. The
plate and the stiffener under the striking mass have
the same spherical shape as the head of the striking
mass, and it is observed that the points in the unde-
formed plate and stiffener move vertically without
any important horizontal displacement.
Panels A1-A3 and B1-B3 showed similar force-
displacement responses to the observed for Panels
A2 and B2 when analyzing the influence of the
impact velocity and the width of panel. The influ-
ence of the stiffener type is similar for panels of
width 125 and 250 mm.
V6
30 40
Figure 10. Shape of deformation: Panel B2; impact at
mid-span; initial velocities 2.0, 4.0 and 6.0 m/s.
405
2/18/2011 5:50:29 PM
 6.2 Group 2 of tests: impact velocity and stiffener
type when the impact is at different locations
Panel types B1, B2 and B3 were numerically
simulated to observe the sensitivity of the stiff-
ener type and the impact velocity when the
impact is at different locations. Figure 11 shows
that, when the impact is at mid-span (Location
1), the stiffener type has a strong influence in the
force-displacement response, resulting in smaller
transverse displacements in the panels with stiff-
eners with high section modulus (e.g. L.50 × 50 ×
5 in Panel B3). However, the impact forces increase
because of the rigid contact. On the other hand,
when the impact is at one-quarter of the support
(Location 2), the stiffener type has a small influence
in the force-displacement response, experiencing in
the three panels similar behaviour under impact loads.
Similar force-displacement response and shape of
the deformation were obtained for the other range
of impact velocities (2.0 and 4.0 m/s). However, at
lower velocities the plates did not suffer fracture
when impacted at one-quarter from the support. Figure 12. Shape of deformation: Panels B1, B2 and
Figure 12 shows the deformed shape of Panels B3; impact at mid-span and at one-quarter from the
B1, B2 and B3 when impacted at both locations support; initial velocity 6.0 m/s.
at the incident velocity of 6.0 m/s. Again, one can
observe the important influence of the stiffener
1.6 B3 B2 B1
type when the impact is at mid-span (Fig. 12a), B1_1
decreasing considerably the local indentation into 1.4 B2_1
B3_1
the plane of the plate that can result in fracture. 1.2 B1_2
Energy [kJ]

Furthermore, the small effect of the stiffener type 1.0 B2_2

is noticed when the panels are impacted at one- B3_2
0.8
quarter of the support, suffering similar shape of 0.6
deformation and fracture propagation (Fig. 12b).
0.4
Figure 13 represents the absorbed energies
with time for the impact event at 6.0 m/s. It is 0.2
noticed that, when the impact is at mid-span the 0.0
time necessary to absorb all the incident energy 0 2 4 6 8 10 12 14
time [ms]
is shorter in stronger panels (stiffeners with high
section modulus, e.g. Panel B3). However, when Figure 13. Absorbed energy with time: Panels B1, B2
the impact is at one-quarter of the support, the and B3; impact at mid-span and at one-quarter from the
time necessary to absorb all the incident energy support; initial velocity 6.0 m/s.
is almost the same for the three panels, thus
corroborating the small influence of the stiffener
120 Impact Point 1
B3_1
type at this impact location. The absorbed energy
100 B1_1 B2_1 during the impact at one-quarter of the support
B2_1 B1_1
80 B3_1
is smaller than the initial kinetic energy, because
Force [kN]

B1_2 of the fracture reached. However, at lower incident

60 B2_2 velocities (2.0 and 4.0 m/s) all the kinetic energy is
B3_2
40 absorbed by the impacted structure.
Similar comments, in terms of force-displacement
20 Impact Point 2
B1-B2-B3
response and shape of the deformation, can be
0 adopted to describe the influence of the stiffener
0 10 20
Displacement [mm]
30 40 type for Panels B1, B2 and B3 when impacted at
4.0 m/s. However, at the lower velocity of 2.0 m/s,
Figure 11. Group 2 of tests: Panels B1, B2 and B3; there is mainly a global deformation for impact
impact at mid-span and at one-quarter from the support; at the mid-span, as well as at one-quarter to the
initial velocity 6.0 m/s. support.

406

MARSTRUCT.indb 406 2/18/2011 5:50:30 PM

 6.3 Group 3 of tests: width of panel, restriction at Important differences were found between
the free edge and impact location Panels B4-B5 and C4-C5 in the shape of the defor-
mation at the free edges of the plate. However,
Panel types B4-B5 and C4-C5 were calculated
at the impact point the global deformation and
under impact loads to analyze the influence of
the local indentation due to the impact is quite
the width of panel, the restriction at the free edge
similar between both stiffened plates, as is shown
and the impact location. The stiffener distribution
in Figure 15 for Panels B4 and B5.
between Panels B4 and B5 is the same in the longi-
The influence of the width of panel when the
tudinal direction. However, in Panel B5 is included
impact point is near the mid-span (Locations
a flat bar 25 × 4 mm in the free edges of the plate
1 and 2) is clearly seen. However, small differences
(Detail A-A, Fig. 2). The same flat bar, 25 × 4 mm,
are found when the impact point is close to the
at the same position, is included in Panel C5.
support. Comparing Panels B4-C4 and B5-C5,
Figure 14 shows that, the reinforcement of the free
larger transverse displacements are observed
edges of the plate in Panels B5 and C5 has a small
when the width increases (Figs. 14a and 14b). On
influence on the force-displacement response, in
the other hand, when the impact point is near
comparison with Panels B4 and C4.
the support (Location 3), the force-displacement
In the three impact locations, a decrease is
response between panels of different width is quite
observed by just few millimetres in the maximum
similar (Fig. 14c).
transverse displacements. It is expected that, the
The influence of the impact location is small
transverse displacements will be significantly
in shorter panels, such as Panels B4 and B5,
affected, if a strong stiffener is considered at the
when impacted above the stiffener (Locations
end connection of the panel’s longitudinal.
1 and 3). For example, Panel B4 experiences
similar transverse displacements at the impact
100
B4_1 point when impacted at Locations 1 and 3 (29.9
B5_1
B5
B4 and 26.8 mm), as shown in Figs. 14a and 14c,
80 C4_1 respectively. However, when the width of panel
C5_1
increases, as in Panels C4 and C5, significant
Force [kN]

60 C5 C4
differences are observed, for example Panel C4
40 reaches 35.9 and 27.8 mm at Locations 1 and
20
3, respectively. Figure 16 represent the time
necessary to absorb the initial impact energy.
0 It is observed that, the impact response is shorter
0 10 20 30 40 when the impact is above the stiffener (Location
a) Displacement [mm]
100
1 and 3). However, when the impact is in the
B4_2
B5 unstiffened zone of the panel (Location 2) the
B5_2
80 C4_2
B4 impact event takes more time.
C5_2
Force [kN]

60

40

20 C5
C4

0
0 10 20 30 40
b) Displacement [mm]
100 B5 B4
B4_3
B5_3 C4
80 C4_3
C5_3 C5
Force [kN]

60

40

20

0
0 10 20 30 40
c) Displacement [mm]

Figure 14. Group 3 of tests: Panels B4-B5 and C4-C5;

impact velocity 6.0 m/s; a) location 1, b) location 2, Figure 15. Shape of deformation: Panels B4 and B5;
c) location 3. impact at locations 1, 2 and 3; initial velocity 6.0 m/s.

407

MARSTRUCT.indb 407 2/18/2011 5:50:33 PM

 1.6
1.4 Impact Point 3
1.2
Impact Point 1
Energy [kJ]
1.0
0.8
Impact Point 2
0.6
0.4
0.2
0.0
0 2 4 6 8 10
time [ms]
Figure 16. Absorbed energy with time: Panels B4
and B5; impact at Locations 1, 2 and 3; initial velocity
6.0 m/s.
7 CONCLUSIONS
Detailed information of the impact response of
rectangular panels has been obtained by non-linear
explicit dynamic simulations. The results give a
good understanding of the force-displacement
response and shape of the deformation in the
impacted panels.
The performed numerical simulations of the ten-
sile test defined well the mesh size, the true stress-
strain curve and the critical failure strain which
were subsequently applied in the panel FE model.
The definition of the true stress-strain curve has
a small influence on the results when only plastic
deformation is observed. However, when fracture
occurs, an accurate definition and its validation by
simulation of tensile tests are necessary. The criti-
cal failure strain proposed by Zhang et al. (2004)
gives good agreement with the tensile test simula-
tions of the steel material.
The representation of the boundary conditions
by nodal restriction at the supported perimeter
gives similar results to the more real representation
with support plates and longitudinal restriction of
the bolts.
The maximum transverse displacement is larger
than the permanent transverse displacement in
wider panels. The difference decreases in shorter
panels.
The local indentation plays an important role in
the total deformation of the stiffened plate, being
more than one-quarter of the total deformation.
For impacts at one-quarter of the support,
the stiffener type has a small influence on the
force-displacement response, showing similar
deformation.
The reinforcement of the free edges of the panel
has a small influence on the plastic response. It is
observed that, the maximum transverse displace-
ments decrease by only few millimetres.
MARSTRUCT.indb 408
The influence of the width of the panel is
evident when the impact point is close to the mid-
span. However, small differences are found when
B4_1 the impact is near the support.
B5_1
B4_2
B5_2
B4_3
ACKNOWLEDGEMENTS
B5_3
The first author has been financed by the Portuguese
Foundation for Science and Technology, under
contract SFRH/BD/46369/2008.
12 14 This work contributes to the activities of
MARSTRUCT VIRTUAL INSTITUTE, (www.
marstruct-vi.com) in particular its Technical
Subcommittee 2.5 on Crashworthiness and Impact
Strength.
REFERENCES
Alsos HS, Amdahl J, Hopperstad O. 2009a. On the
resistance of stiffened plates, Part I: Experiments. Inter-
national Journal of Impact Engineering. 36: 799–807.
Alsos HS, Amdahl J, Hopperstad O. 2009b. On the resist-
ance of stiffened plates, Part II: Numerical analysis.
International Journal of Impact Engineering. 36:
875–887.
ASTM (American Society for Testing and Materials).
1989. Standard Specification for Carbon Structural
Steel.
Bedair OK. 1997. Analysis of stiffened plates under lateral
loading using sequential quadratic programming
(SQP). Computer & Structures. 62 (1): 63–80.
Boot JC, Moore DB. 1988. Stiffened plates subjected to
transverse loading. International Journal of Solid and
Structures. 24 (1): 89–104.
Caridis PA, Samuelides E, Frieze PA. 1994. On the
dynamic response of ship plating under lateral impact.
International Journal of Impact Engineering. 15 (2):
149–164.
Cho S-R, Lee H-S. 2009. Experimental and analytical
investigations on the response of stiffened plates
subjected lo lateral collisions. Marine Structures. 22:
84–95.
Deb A, Bootom M. 1988. Finite element models for stiff-
ened plates under transverse loading. Computer &
Structures. 28 (3): 361–372.
Dieter GE. 1986. Mechanical behavior under tensile and
compressive loads. ASM Handbook. 8: 99–10.
Ehlers S, Broekhuijsen J, Alsos HS, Biehl F, Tabri K.
2008. Simulating the collision response of ship
side structures: A failure criteria benchmark study.
International Shipbuilding Progress. 55: 127–144.
Ehlers S, Varsta P. 2009. Strain and stress relation for
non-linear finite element simulations. Thin-Walled
Structures. 47 (11): 1203–1217.
Ehlers S. 2010. Strain and stress relation relation until
fracture for finite element simulations of a thin circu-
lar plate. Thin-Walled Structures. 48 (1): 1–8.
Hagiwara K, Takanabe H, Kawano H. 1983. A proposed
method of predicting ship collision damage. Interna-
tional Journal of Impact Engineering. 1 (3): 257–279.
408
2/18/2011 5:50:35 PM
 Hallquist JO. 2005. LS-DYNA Theory Manual. Livermore
Software Technology Corporation.
Jones N. 1989. Structural Impact. Cambridge University
Press.
Langseth M, Larsen PK. 1990. Dropped objects’ plug-
ging capacity of steel plates: An experimental investi-
gation. International Journal of Impact Engineering.
9 (3): 289–316.
Manolakos DE, Mamalis AG. 1985. On ship collisions:
The plastic collapse of longitudinally framed shell
plating subjected to oblique loading. International
Journal of Impact Engineering. 3 (1): 41–55.
Manolakos DE, Mamalis AG. 1988. Limit analysis for
laterally loaded stiffened plates. International Journal
of Mechanical Sciences. 30 (6): 441–447.
Shen WQ, Jones N. 1992. A failure criterion for beams
under impulsive loading. International Journal of
Impact Engineering. 12 (1): 101–121.
Simonsen BC, Lauridsen LP. 2000. Energy absorption
and ductile failure in metal sheets under lateral inden-
tation by a sphere. International Journal of Impact
Engineering. 24 (10): 1017–1039.
Tabri K, Alsos H, Broekhuijsen J, Ehlers S. 2007.
A benchmark study on ductile failure criteria for shell
elements in multiaxial stress state. Advancements
in Marine Structures, Guedes Soares & Das (Eds.).
Taylor & Francis Group: London. 401–409.
409
MARSTRUCT.indb 409
Villavicencio R, Guedes Soares C. 2009. Response of
clamped beams to impact loading along their length.
Proceedings 13th Congress of International Maritime
Association of Mediterranean, Istanbul, Turkey; Vol. 2,
pp. 437–445.
Villavicencio R, Guedes Soares C. 2011. Numerical mod-
elling of the boundary conditions on beams stuck
transversely by a mass. International Journal of Impact
Engineering. doi: 10.1016/j.ijimpeng.2010.12.006.
Villavicencio R, Sutherland LS, Guedes Soares C. In
press. Numerical simulation of transversely impacted,
clamped circular aluminium plates. Ships and Offshore
Structures.
Wu F, Spong R, Wang G. 2004. Using Numerical Simula-
tion to Analyze Ship Collision. Proceedings of the 3rd
International Conference on Collision and Grounding
of Ships, Izu, Japan; pp. 27–33.
Zhang L, Egge ED, Bruhns H. 2004. Approval procedure
concept for alternative arrangements. Proceedings of
the 3rd International Conference on Collision and
Grounding of Ships, Izu, Japan; pp. 87–96.
Zhu L, Faulkner D. 1994. Dynamic inelastic behaviour of
plates in minor ship collisions. International Journal
of Impact Engineering. 15 (2): 165–178.
2/18/2011 5:50:35 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Influence of the neutral axis displacement on the residual strength

of a damaged tanker double bottom structure

R. Villavicencio & C. Guedes Soares

Centre for Marine Technology and Engineering (CENTEC), Technical University of Lisbon, Instituto Superior
Técnico, Lisboa, Portugal

Z. Liu & J. Amdahl

Marine Technology Department, Norwegian University of Technology and Science, Trondheim, Norway

ABSTRACT: The paper presents a method to estimate the displacement of the neutral axis of a damaged
ship subjected to bending moment. The method uses the load-end shortening relationship obtained from
previous numerical double bottom residual strength calculations establishing force equilibrium over the
whole transverse section of the ship. The set of estimated neutral axes allows the definition of a new load
condition to induce a higher bending moment on the double bottom structure. The new residual strength
is compared with previous numerical calculations for the intact and minor damaged double bottom in
which the bending moment was induced considering the neutral axis fixed during the entire simulation.
The comparison showed a reduction in the capacity of the double bottom structure when the bending
moment is induced by the new load condition.

1 INTRODUCTION Guedes Soares et al. (2008) benchmarked the

In the design of the midship section it is necessary
to know the ultimate longitudinal hull girder
strength. The ship structure is exposed to different
types of loads related to extreme environmental
conditions or accidental events such as collision
and grounding. In case of structural damage,
due to collision or grounding, the hull girder
strength may be reduced compared with the intact
condition. In order to estimate the damage due to
tanker collision and grounding, it is necessary to
investigate the global dynamic behaviour and the
local plastic response of the individual ship struc-
tural elements. If a fracture occurs, the oil spill will
be the most severe consequence. Thus, improve-
ments in the design of tanker double bottoms are
essential to obtain a high level of safety at sea and
protection of marine ecosystems.
The ultimate hull girder capacity of a ship
can be estimated by simplified analytical expres-
sions or by fully nonlinear finite element analysis.
Simplified structural analysis methods are widely
accepted as reliable and fast tools to obtain the
longitudinal strength of an intact ship (e.g. Smith
1977; Yao and Nikolov 1991, 1992; Gordo et al.
1996). In minor ship collision or grounding sce-
narios the damaged panel could contribute to the
hull girder residual strength. Thus, the neglect of
the damaged structure in the estimation of the
residual strength is not a straightforward decision.
ability of simplified structural methods based
on the Smith’s formulation (Smith 1977) to pre-
dict the ultimate strength of a damaged ship and
compared the results with finite elements analysis,
finding good agreement for intact and damaged
conditions. However, due to the complexity of the
problem, it is recommended to study the collapse
response of ship hulls by numerical procedures. On
the other hand, the nonlinear finite element analy-
sis of the hull girder requires huge computational
sources. Thus, simplified finite element models of
double bottom structures can be used to analyze its
resistance capacity in the first stage of design.
It must be noted that the modelling of the dou-
ble bottom structure requires adopting some sim-
plifications. For example, the ship structure above
the tank top can be represented by boundary
conditions, or the induced bending moment can
be locally applied without considering the global
effect induced by all the ship structure. The main
consequence of omitting the ship structure above
the tank top is that the displacement of the neutral
axis due to the nonlinear response of the material
cannot be reproduced during the numerical simu-
lations. This shifting in the position of the neutral
axis is especially important in damaged double
bottoms, because the damaged structural elements
could not contribute to the hull girder capacity.
Thus, they should be removed provoking a higher
displacement of the neutral axis.

411

MARSTRUCT.indb 411 2/18/2011 5:50:35 PM

 Ultimate strength analyses of stiffened panels
typical of intact ship structures have been per-
formed by several authors (Guedes Soares and
Gordo 1997; Paik and Kim 2002; etc.). Moreover,
several analyses of the intact hull girder capac-
ity subjected to bending moment have been con-
ducted. For example, Bai et al. (1993) presented
a numerical procedure for ship hull under com-
plicated loads where a set of finite elements were
defined accounting geometrical and material
nonlinearities. Gordo et al. (1996) presented a
simplified method to represent the behaviour of
stiffened plate columns allowing the prediction
of the strength degradation due to corrosion and
residual stresses. Qi et al. (2005) studied theoreti-
cal methods for large tankers which were based
on experimental works comparing the results with
various finite element and theoretical methods and
also with the common structural rules for double
hull tankers. Luis et al. (2007) studied the effect of
bottom and side damage on the ultimate strength
of double hull tankers. Amlashi and Moan (2008)
summarized the definitions for the nonlinear finite
element modelling to estimate the ultimate hull
girder capacity of a bulk carrier under alternate
hold loading condition.
Some studies have been conducted on stiffened
panels and hull girder capacity after denting dam-
age such as collision or grounding. For example,
Smith and Dow (1981) reviewed various forms of
structural damage on ships and offshore structures,
suggesting analytical methods for evaluation of the
residual stiffness and strength. A recent study of
similar type of stiffened panels under localized
damage was presented by Guedes Soares et al.
(2008a). Pedersen (1994) presented a mathemati-
cal model for analysis of ship hull loading due to
grounding where the grounding forces, the sectional
shear forces and the bending moments were deter-
mined and related to the ultimate capacity of the
hull girder. Wang and Chen (2002) derived analyti-
cal equations for the residual strength of ships with
damages due to grounding or collision which were
verified with direct calculations of sample com-
mercial ships for a broad spectrum of accidents.
The finite element analysis of damaged double
bottoms has been focused on grounding scenarios
where the structure is penetrated until fracture in
the inner plate (Naar et al. 2002; Alsos and Amdahl
2007; etc.). After such a major damage, the residual
strength of the double bottom loses importance.
However, in minor grounding scenarios, the ulti-
mate capacity of the double bottom structure to
sustain compressive loads plays an important role
to ensure the safety at sea and avoid environmental
catastrophes.
The present study is a continuation of a previous
numerical investigation (Liu and Amdahl unpubl.)
412
MARSTRUCT.indb 412
estimating the associated residual strength in intact
and minor damaged double bottom structures.
In the previous study, the bending moment was
induced considering the neutral axis fixed during
the simulation. In this paper, a method to estimate
the displacement of the neutral axis due to the
nonlinear response of a ship subjected to bending
moment is presented. The method uses the force
equilibrium over the whole transverse section of
the ship and the load-end shortening relationship
of the double bottom obtained from the previ-
ous simulations. A new load condition to induce
a higher bending moment on the double bottom
structure is derived. The new residual strength is
compared with the previous numerical calculations
considering the intact condition and one case with
damage.
2 DOUBLE BOTTOM IN STUDY
The double bottom used in the present work was
previously analyzed (Liu and Amdahl unpubl.)
performing numerical simulations yielding the
residual strength in the intact and damaged con-
ditions. In the present Chapter, the finite element
model and some of the results are summarized to
understand the actual calculations.
The structure of the double bottom corresponds
to a “Shuttle Tanker 140.000 m3”. The main dimen-
sions of the ship are: length between perpendicu-
lars 256.5 m, moulded breadth 42.5 m, depth to
upper deck 22.0 m and design draft 15.0 m. The
double bottom was modelled between four floors
located near the midship section of the ship. Only
the starboard side was considered in the analysis.
The main dimensions of the double bottom model
are: length 12.0 m, width 21.25 m and height
2.65 m. The space between frames is 4.0 m. Figure 1
represents the scantling used in the finite element
model, which was simplified with respect to the
original design and Figure 2 shows the isometric
view of the finite element model.
The computations were carried out using the finite
element package LS-DYNA Version 971 (Hallquist
2005). The plating and longitudinals were modelled
with 4-node shell elements with 5-integration points
using the Hughes-Liu formulation. The mesh size
was approximately 100 mm. The magnitude of the
Figure 1. Scantling of the double bottom structure (FE
model).
2/18/2011 5:50:36 PM

Figure 2. Finite element model (fore floor is hidden).
defined initial imperfections corresponded to fabri-
cation tolerances and the shape was defined using
sinusoidal expressions.
The true stress-strain relationship of the mate-
rial was defined by an isotropic plasticity model
using a power law hardening rule (‘Mat.18 Power
Law Plasticity’ in LS-DYNA). The inputs of the
material were: yield stress 275 MPa, strength coef-
ficient 740 and hardening exponent 0.24; strain
rate effects were ignored. Since only large plastic
deformations were studied, failure was not required
in the material definition.
Symmetry boundary conditions were defined at
the center line and the fore and aft floors were fully
clamped. Both boundary conditions were prop-
erly modified to induce a bending moment. As
the tanker was considered in full load condition,
water and oil pressure were applied on the bottom
and tank top, respectively. In the intact case the
sequence of the loads was pressure and a bending
moment whereas in the damaged case the sequence
was pressure, indentation damage and a bending
moment. In both intact and damaged cases the
bending moment was induced by defining pre-
scribed displacements and rotations on the fore
and aft floors.
The hogging bending moment applied on the
double bottom structure induced axial strains and
stresses which were represented by “mean strains”
and “resistance forces”. The “mean strain” was
obtained by the ratio between the longitudinal
displacement of a node located in the central fore
floor and the span of the stiffeners. The “resistance
forces” were obtained from elements located for-
ward the central aft floor. It must be noted that the
resistance forces were given for only one side (star-
board) of the double bottom, which represented all
the thickness of the centre vertical keel.
Figure 3 shows the “resistance force-mean strain”
relationships of the intact and damaged cases. The
selected indenter to induce damage has triangular
shape, the depth is 150 mm and causes damage
to five stiffeners located between the second and
third side girder. The “total” resistance force is a
summation of the resistance force of the structural
elements located below and above the local neutral
413
MARSTRUCT.indb 413
300 Outer Buckling
Inner Buckling
250
Intact
Resistance Force [MN]
"total"
200
Damage
150
100
50 "inner"
"outer"
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
mean strain [%]
Figure 3. Ultimate strength of double bottom. Previous
intact and damaged cases.
axis of the double bottom (1340 at BL; BL is the
base line) denoted in this study by “outer” and
“inner” elements, respectively. Due to the rotation
applied on the fore and aft floor to induce a bend-
ing moment, the inner and outer elements buckled
at different induced strains, see Figure 3.
The damage on the bottom panel reduced its
capacity to sustain a compressive load. This reduc-
tion is more evident in the outer than in the inner
elements. The post-buckling response of the inner
elements is less influenced by the indentation of the
central bottom panel. The intact ultimate strength
of the double bottom occurred when the outer ele-
ments buckled, whereas in the damaged case it was
observed at the buckling of the inner elements.
3 NEUTRAL AXIS DISPLACEMENT
The method used to estimate the displacement of
the neutral axis, due to the nonlinear response of
a ship subjected to bending moment, is based on
the criterion provided by IACS (2006). According
to this criterion, the neutral axis is determined
by establishing force equilibrium over the whole
transverse section of the ship.
The double bottom was idealized for its repre-
sentation in the “previous” finite element model.
The structure of the ship above the tank top was
used to estimate the neutral axis and the stress-
strain distribution of the structural elements.
Some simplifications of the structure above the
tank top were adopted. However, the scantlings
of the plates and the longitudinal were the same
as the ones used in the original design. Figure 4-a
shows this simplified structure where the phantom
lines represent the camber of the upper deck and
the diagonal inner plating of the bilge tank used in
the original design. In the simplification, they were
considered parallel to the base and the centre line,
respectively. The midship section above the tank
top was divided into ordinary stiffener elements
2/18/2011 5:50:36 PM
 (IACS 2006), which were further simplified to
longer plates with equivalent thickness for panels
having the same type of stiffener, see Figure 4-b.
The transverse section of the double bottom was
the same as the one used in the finite element
model. The neutral axis was estimated in 9840 mm
above the base line (9840 at BL).
The “resistance force-mean strain” relation-
ship of the previous study was transformed to a
“stress-strain” curve. The nonlinear response of
the structural elements between the tank top and
the neutral axis was described by the exact load-
end shortening curve obtained from the previous
numerical simulation, which was simplified by
trend lines. As the elements above the neutral axis
are lengthened, they were considered to fail by elas-
to-plastic collapse (IACS 2006). The structural ele-
ments responses above and below the neutral axis
are plotted in Figure 5. It must be noted that the
elements above the neutral axis are under tension
(lengthening), whereas elements bellow the neutral
axis are under compression (shortening). Points 1
to 6 denote different slopes of the end shorten-
ing curve, which are used to analyze the results.
Once the structural elements, geometrical neutral
axis and material response are defined, the change
Figure 4. Simplified structure and stiffener elements.
Figure 5. Stress-strain curves (or so called load-end
shortening curves).
414
MARSTRUCT.indb 414
in the position of the neutral axis was estimated
according to the following procedure.
The calculated “mean strain” on the double
bottom was used to estimate the linear strain dis-
tribution above the tank top, see Figure 6-a. The
magnitude of any strain is positioned at half of the
double bottom height (1340 at BL), then a linear
strain distribution is drawn passing through the
geometrical neutral axis (9840 at BL) and ending
at the upper deck (22000 at BL). Based on this lin-
ear strain distribution, the stress in each structural
element induced by the strain was obtained from
the derived stress-strain curves (Fig. 5). Using this
initial stress distribution (Fig. 6-a), the axial forces
in each structural element were obtained (Fi = Aiσi;
i denotes each structural element). These forces
were summed to obtain the total axial force on the
midship section. It must be noted that the stresses
are positive for elements under tension and nega-
tive for elements under compression. In the lin-
ear response (Point 1, Fig. 5) the summation of
the axial forces is zero. However, in the nonlinear
response the total force is negative or positive,
because the neutral axis has been moved up or
down, respectively.
The “new” position of the neutral axis was deter-
mined by iterative calculations implemented in a
simple numerical subroutine, which checked the
longitudinal force equilibrium over the whole trans-
verse section until reaching zero magnitude of the
summation of the axial forces (ΣAiσi = ΣAjσj; i, j
denote tension and compression, respectively).
Using the “new” neutral axis, the “correct” stress-
strain distribution in the structural elements of the
midship section can be estimated, see Figure 6-b.
The described procedure was used to determine
the neutral axis for different strains of the load-
end shortening curve, e.g. Figure 7 represents
the stress-strain distribution over the transversal
Figure 6. Longitudinal force equilibrium.
2/18/2011 5:50:37 PM
 22000 4 REPRESENTATION AND RESULTS
zi [mm]
Point 1 The different positions of the neutral axis were
Point 2 used to define a “new load condition” to induce
Point 3 bending moment in the finite element model by
Point 4
Point 5
increasing the magnitude of the prescribed longitu-
dinal displacement of the fore floor, thus inducing
compression. Figure 9 shows the various positions
of the neutral axis for each induced strain. The set
Base Line Strain [n.u.]
of neutral axes (zi) is normalized by the depth up
0 to the upper deck (D). It is observed that the neu-
-0.006 -0.004 -0.002 0.000 0.002 0.004 0.006 tral axis is constant in the linear material response
and coincides with the geometrically estimated.
However, the neutral axis increases gradually in the
22000
nonlinear response. Points 1 to 6 of the load-end
zi [mm]
shortening curve are represented as “small breaks”
in the tendency of the derived curve. The break
between Points 3 and 4 represents approximately
Point 1 the starting point of yielding at the upper deck.
Point 2 In the previous study the bending moment in the
Point 3 double bottom was induced by prescribing certain
Point 4
Point 5
longitudinal displacements and rotations on the
fore and aft floors (Fig. 10-a). These displacements
Base Line Stress [N/mm ]
2
and rotations were defined by a constant linear
0
and angular velocity, because the neutral axis was
-300 -200 -100 0 100 200 300
fixed during the simulation, see Figure 10-b. In the
Figure 7. Strain and stress distribution at midship present work, the set of “new” neutral axes was
section.
1.0
z/D
6.0 2 0.9
0.8
Bending Moment [MN-m]

5.0
3 0.7
1 4
4.0 5 0.6 2
6 0.5
6
3.0 0.4 5
0.3 1 3 4
2.0
0.2
1.0 0.1 Mean Strain [%]
0.0
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Mean Strain [%]
Figure 9. Estimation of the neutral axis displacement.
Figure 8. Bending moment-mean strain response for
the set of neutral axes.

section at Points 1 to 5. The vertical displacement

of the neutral axis for the various magnitudes of
mean strain is shown. As previously mentioned,
the strain distribution is linear for all cases. When
the stress distribution starts to be nonlinear, the
displacement of the neutral axis increases rapidly.
As a “cross check” of the procedure, the bending
moment at different mean strains, considering the
change in the position of the neutral axis, was
estimated by summing up the moment contribu-
tion of each axial force in each structural element
(M = ΣAiσizi, i denotes each structural element;
σi and zi are positive values), see Figure 8. Figure 10. Definition of loads.

415

MARSTRUCT.indb 415 2/18/2011 5:50:39 PM

 used to define a variable linear velocity in terms of words, in the new simulation the double bottom
displacement with time (Fig. 10-c). structure is subjected to a higher compressive load
The comparison between the “new” displace- with a smaller contribution of the bending effect.
ment-time curve used in the finite element model However, the difference between previous and new
and the “previous” one is shown in Figure 11. maximum resistance forces is insignificant (less
It must be noted that the new curve represent a than 2.5%).
simplification of the “exact” curve (dashed line). As mentioned, in the previous model, due to
The later was obtained by considering all the esti- the rotation applied on the fore and aft floor to
mated positions of the neutral axis. The simplifica- induce a bending moment, the inner and outer ele-
tion of the curve allows comparing the predictions ments buckle at different induced strains. This dif-
of the elastic response. The breakpoint in the new ference is also observed herein, showing a smaller
curve (time 0.1 s) corresponds to the starting point distance between the buckling peaks, i.e. buckling
of the nonlinear response in both models. at the inner elements is reached at a lower imposed
The total intact load-end shortening curves of curvature. After buckling at the inner elements, the
both models, with and without displacement of post-buckling response reduces rapidly the resist-
the neutral axis, are compared in Figure 12, where ance force of the double bottom structure.
the resistance force is represented in terms of the It must be noted that the most important dif-
imposed curvature (ki = εizi) describing the dis- ference between the previous and the new model
placement of the neutral axis in the new model. It is the shortening in the distance between the buck-
is observed that, due to the simplification in the ling points at the outer and inner elements. This
new loading condition, illustrated in Figure 11, the difference is just due to the increment in the rate
linear response of both models is the same. of the compressive load which determines the
Subsequently, the buckling at the outer ele- critical response of the double bottom structure.
ments is slightly overestimated by the new calcula- It is convenient to remember that, in the case of
tion due to the smaller contribution of the rotation pure compression only one peak can result from
with respect to the increment in the prescribed the simulation if the outer and inner structures
displacement (Figs. 10-b and 10-c). In other are similar and the maximum resistance force is
higher. In the new model, a similar response is
observed where the distance between the buck-
80
ling peaks was decreased and the peak force was
70 slightly increased.
Displacement [mm]

60 The resistance force of the structural elements

50 bellow and above the local neutral axis of the dou-
40 ble bottom is also shown in Figure 12. For the outer
30 elements, the response of both models is coinci-
Previous dent until the maximum resistance force, and after
20
New
10 Exact
this peak the capacity of the new model is slightly
reduced. For the inner elements, the response of
0
0.0 0.1 0.2 0.3 0.4 0.5
both models is coincident until the point where
time [s] previous and new models used the same rate of
prescribed displacement, but after this point the
Figure 11. Displacement-time curve. “break” in the load-end response is represented by
the new model. This break is due to the increment
in the prescribed displacement of the new model
300 Outer Buckling
(Fig. 11; time 0.1 s). After this break, the response
Inner Buckling
250 of the new model is shortened, i.e. buckling and
Resistance Force [MN]

"total" post-buckling responses occur at lower imposed

200 "Previous"
"New"
curvature. In both models the magnitude of the
150 maximum resistance force is similar.
In the previous analysis, the bending moment
100
"break"
after damage was imposed using the same load
50 Inner Elements curve of the intact case. Thus, the “new damaged”
Outer Elements model uses the same load curves defined for the
0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
“new intact” model. The new damaged load curve
Curvature*10
-3
was adjusted to obtain the same linear response
of the previous damaged model. The results from
Figure 12. Load-end shortening curves. Intact case. the comparison between “the previous and the

416

MARSTRUCT.indb 416 2/18/2011 5:50:41 PM

 300
250
Resistance Force [MN]
200
150
100
50
0 tanker bottom structures during stranding. Marine
0.0 0.2 0.4 0.6
Figure 13. Load-end shortening curves. Damaged case.
new damaged simulations” (Fig. 13) are similar to
those observed for the intact case. In particular, the
buckling at the outer and inner elements is slightly
overestimated by the new models and the distance
between the buckling peaks is shortening.
5 CONCLUSIONS
A method to estimate the displacement of the neu-
tral axis of a damaged ship subjected to bending
moment is described. The method uses the load-
end shortening relationship obtained from previ-
ous numerical residual strength calculations of
a double bottom establishing force equilibrium
over the whole transverse section of the ship. The
method allows the definition of a new load con-
dition to induce a higher bending moment on the
double bottom structure.
The ultimate strength of the structural elements
below and above the local neutral axis of the dou-
ble bottom structure are slightly overestimated
in the new calculations, demonstrating that the
double bottom structure is subjected to a higher
compressive load at smaller induced strains, and,
consequently experiences smaller contribution of
induced bending moment.
The most important difference between the pre-
vious and the new finite element calculations is the
shortening in the distance between the buckling of
the elements below and above the local neutral axis
of the double bottom (outer and inner elements).
This difference is due to the increment in the rate
of the compressive load. The conclusions are valid
for intact and damaged cases.
ACKNOWLEDGEMENTS
This work has been performed within the project
FLAGSHIP (European Framework for Safe,
MARSTRUCT.indb 417
Efficient and Environmentally-friendly Ship
Operations) that is partially financed by the Euro-
pean Union through contract TIP5-CT-2006-
031406.
REFERENCES
Alsos HS, & Amdahl J. 2007. On the resistance of
0.8 1.0 1.2 1.4 Structures; 20: 218–237.
-3
Curvature*10 Amlashi H, & Moan T. 2008. Ultimate strength analysis
of a bulk carrier hull girder under alternate hold load-
ing condition—A case study Part 1: Nonlinear finite
element modelling and ultimate hull girder capacity.
Marine Structures; 21: 327–352.
Bai Y, Bendiksen E, & Pedersen PT. 1993. Collapse anal-
ysis of ship hulls. Marine Structures; 6: 485–507.
Gordo JM, Guedes Soares C, & Faulkner D. 1996.
Approximate assessment of the ultimate longitudinal
strength of the hull girder. Journal of Ship Research;
4 (1): 60–69.
Guedes Soares C, & Gordo JM. 1997. Design methods
for stiffened plates under predominantly uniaxial
compression. Marine Structures; 10 (6): 465–497.
Guedes Soares C, Luís RM, Teixeira AP, Quesnel T,
Nikolov PI, Steen E, Khan IA, Toderan C, Olaru VD,
Bollero A, & Taczala M. 2008a. Parametric Study
on the Collapse Strength of Rectangular Plates with
Localized Imperfections under Inplane Compression.
International Shipbuilding Progress; 55: 63–85.
Guedes Soares C, Luís RM, Nikolov P, Downes J,
Taczala M, Modiga M, Quesnel T, Toderan C, &
Samuelides M. 2008b. Benchmark study on the use
of simplified structural codes to predict the ulti-
mate strength of a damaged ship hull. International
Shipbuilding Progress; 55: 87–107.
Hallquist JO. 2005. LS-DYNA Keyword user’s man-
ual, Version 971. Livermore Software Technology
Corporation, Livermore, California.
IACS (International Association of Classification
Societies). 2006. Common Structural Rules for Double
Hull Oil Tankers.
Liu Z, & Amdahl J, unpub. Numerical and simplified
methods for analysis of the residual strength of ship
double bottoms.
Luís RM, Hussein AW, & Guedes Soares C. 2007. On
the Effect of Damage to the Ultimate Longitudi-
nal Strength of Double Hull Tankers. Proc. 10th
Int. Symp. on Practical Design of Ships and Other
Floating Structures (PRADS’07); ABS, Houston,
Texas, USA.
Naar H, Kujala P, Simonsen BC, & Ludolphy H. 2002.
Comparison of the crashworthiness of various bot-
tom and side structures. Marine Structures; 15:
443–460.
Paik JK, & Kim BJ. 2002. Ultimate strength formula-
tions for stiffened panels under combined axial load,
in-plane bending and lateral pressure: a benchmark
study. Thin-Walled Structures; 40 (1): 45–83.
Pedersen P. 1994. Ship grounding and hull-girder
strength. Marine Structures; 7: 1–29.
417
2/18/2011 5:50:42 PM
 Qi E, Cui W, & Wan Z. 2005. Comparative study of
ultimate hull girder strength of large double hull
tankers. Marine Structures; 18: 227–249.
Smith CS. 1977. Influence of local compressive failure
on ultimate longitudinal strength of a ship’s hull. In:
Proceedings of the PRADS: International Symposium
on Practical Design in Shipbuilding; N. Zosengakkai
ed., Society of Naval Architects of Japan, Tokyo,
Japan, 18–20 October 1977, p. 73–79.
Smith CS, & Dow RS. 1981. Residual strength of
damaged steel ship structures and offshore platforms.
Journal of Constructional Steel Research; 1 (4): 2–15.
418
MARSTRUCT.indb 418
Wang G, & Chen Y. 2002. Longitudinal strength of ships
with accidental damages. Marine Structures; 15 (2):
119–138.
Yao T, & Nikolov PI. 1991. Progressive collapse analysis
of a ship’s hull under longitudinal bending. J. Soc.
Naval Arch. of Japan; 170: 449–461.
Yao T, & Nikolov PI. 1992. Progressive collapse analy-
sis of a ship’s hull under longitudinal bending (2nd
report). J. Soc. Naval Arch. of Japan; 172: 437–446.
2/18/2011 5:50:43 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Numerical simulation of laterally impacted clamped circular

steel plates

R. Villavicencio, L. Sutherland & C. Guedes Soares

Centre for Marine Technology and Engineering (CENTEC), Technical University of Lisbon, Instituto Superior
Técnico, Lisboa, Portugal

ABSTRACT: The paper presents numerical simulations of previously reported drop weight impact tests
examining the dynamic response of fully clamped steel circular plates struck transversely at the centre
by a mass with a spherical indenter. The impact velocity varied from 0.5 to 6.0 m/s. The plates showed
no visible damage at the very low incident energies, but suffered both permanent indentation and global
deformation as incident energy was increased. The numerical modelling was performed using the LS-
DYNA nonlinear, dynamic finite element software and is based on a previous finite element analysis
of aluminium plates under lateral impact. The numerical calculations used can accurately predict the
response of deflections, forces and absorbed energies, using both shell and solid element types. However,
solid elements were required to obtain a satisfactorily accurate prediction of the deformed shape includ-
ing the indentation into the thickness. The steel plastic response is compared and discussed with that of
similar aluminium plates.

1 INTRODUCTION or vibrations developed at the instant of contact

The structural design of ships concerning collision
requires an accurate prediction of the damage
of stiffened plates under impact loading. Finite
Element (FE) analysis is a useful tool to predict
the extent of ship collision and consequent dam-
age to structural components. However, the non-
linear dynamic analysis should be compared with
experimental tests before being used for structural
design. Unfortunately, experimental tests on full
scale ship collision are rare and very expensive.
One approach is to perform scaled collision test
on typical ship structural members to validate the
numerical methods for impact analysis.
Theoretical and experimental analyses of indi-
vidual ship structural components under lateral
impact loads, such as plates, have been widely
analyzed. For example, Shen (1995) examined
the dynamic plastic response of thin circular
plates transversely and centrally struck by a mass
with a conical head and a spherical nose. Wang
et al. (1998) derived approximate formulas for
the load-deflection relationship of a rigid-plastic
circular plate deflected by a rigid sphere. However,
it is difficult to find comparative results between
experimental tests and numerical simulations on
circular plates struck laterally by a mass. In most
cases, the impact response is examined by penetrat-
ing the plates using quasi-static lateral loads, which
does not consider the extremely high impact forces
between the indenter and the specimen. For
example, Simonsen and Lauridsen (2000) studied
the mechanics of lateral indentation of a rigid
sphere into a thin, ductile metal plate, including
experiments, analytical theories and numerical
calculations. Tabri et al. (2007) performed FE sim-
ulations on thick steel plates where the specimens
were statically deformed by a spherical punch
using different failure criteria. Ehlers (2010) pro-
posed a true stress and strain relation to simulate
plate punching experiments until fracture with the
FE method.
The present study is a continuation of the
experimental investigation reported by Sutherland
and Guedes Soares (2009) on fully clamped
composite, aluminium and steel circular plates
subjected to lateral impact. In this paper, numeri-
cal simulations of tests on steel plates are presented
in order to predict their maximum deflection,
force and absorbed energy. The FE model is
based on previous numerical simulations of lateral
impacts on aluminium 5083/H111 circular plates
(Villavicencio et al. 2010).
A brief summary of the experimental set up and
the definitions adopted in the numerical model are
described. Then comparisons were made between
the numerical and experimental force-displacement
responses for selected (low, medium and high)
velocities in order both to validate and optimise
the numerical simulations. Using the optimised

419

MARSTRUCT.indb 419 2/18/2011 5:50:43 PM

 models from the previous step, the experimental
and numerical maximum deflection and maximum
force behaviours are then compared for all tested
impact velocities. The deformation shape is
analyzed using both shell and solid elements types
and different mesh sizes, and additional com-
ments on the stress and strain distributions, and
on the energy partitioning, are then given. Finally,
the behaviour of the current steel plates is com-
pared to those of the aluminium plates studied in
Villavicencio et al. (2010).
2 EXPERIMENTAL DETAILS
The experimental tests represent an impact event
in which a fully clamped circular plate is struck
by a mass G traveling with an initial velocity V0 at
the centre of the plate (Fig. 1). After impact, the
striker G is assumed to remain in contact with the
plate. Therefore, the striker and the struck point of
the plate have an initial velocity V0 at the instant
of contact and a common velocity throughout
the entire response. The maximum total deforma-
tion Wt is divided into two parts: maximum local
indentation Wi and maximum global deflection W
(Shen 1995).
The impact tests were performed using a
fully instrumented Rosand IFW5 falling weight
machine, which gives the variation with time
of the impact force, velocity, displacement and
energy of the impacted specimens. The experimen-
tal setup can be seen in Figure 2. Specimen plates
Figure 1. Clamped circular plate struck laterally at the
centre by a mass.
Figure 2. Experimental clamped condition.
420
MARSTRUCT.indb 420
600
Thick (4.0 mm)
500
465MPa Thin (1.4 mm)
Stress [MPa]
400
350MPa
365MPa
300
230MPa
200
100
0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
Strain [-]
Figure 3. Engineering curves of material.
were 200 mm square and were fully clamped by
four bolts between two thick 200 mm square
steel plates with internal diameter 100 mm. The
indenter was a hemi-spherically ended projectile of
radius 5.0 mm.
In order to investigate the effects of both global
deformation and local indentation, tests were car-
ried out on steel plates for two thicknesses 1.4 and
4.0 mm (thin and thick plates respectively), using
an impact mass of 3.1 and 4.9 kg respectively.
Tests were carried out on virgin specimens for
a range of impact velocities, from 0.5 to 6.0 m/s
approximately.
The material for the thick plates is ‘Structural
hot rolled steel grade S235JR’ whereas for the thin
plates is ‘Structural cold rolled steel grade ST12’.
The mechanical properties of the material were
obtained from tensile tests carried out on the same
plates from which the impact specimens were cut.
The engineering curves of both thicknesses are
plotted in Figure 3. Full experimental details and
discussions of the experimental results may be
found in Sutherland and Guedes Soares (2009).
3 NUMERICAL MODEL
The computations were carried out using the
finite element package LS-DYNA Version 971
(Hallquist 2005) which is appropriate for non-
linear explicit dynamic simulations with large
deformations. The FE model is based on the
numerical analysis on circular aluminium plates
conducted by Villavicencio et al. 2010, where a
sensitivity analysis of the mesh size and the model-
ling of support plates to represent the experimen-
tal boundary condition were completed. In the
present section the main characteristics of the FE
model and improvements with respect to the previ-
ous model are summarized (Fig. 4).
The specimen plates were modelled with 4-node
shell elements with 4-integration points through the
2/18/2011 5:50:43 PM

Figure 4. Details of finite element model (Villavicencio
et al. 2010).
thickness (Belytshko-Lin-Tsay formulation) and
with 8-node solid elements with 1-integration point
(constant stress solid element formulation), both
element formulations are the default in LS-DYNA.
The mesh sizes of the shell elements were 2 × 2 and
1 × 1 mm for both plate thicknesses (denoted by
Shell2 and Shell1 respectively). The mesh size of
the solid elements was 1 × 1 × 1 mm for the thick
plates and 1 × 1 × 0.35 mm for the thin plates
(denoted by Solid1 in both cases), in order to give
the same number of integration points through the
thickness as in the corresponding shell models.
The support plates simulate the boundary con-
ditions of the specimen plate, compressing the
specimen as occurred in the experiments. For the
striking mass only the vertical translation was
free, in which direction the initial impact velocity
was assigned. The indenter-specimen and support
plates-specimen contact was defined as ‘Automatic
Surface to Surface’ (Hallquist 2005).
The support plates and the striking mass were
modelled as a rigid un-deformable material.
‘Mat.020-Rigid’ was selected from the material
library of LS-DYNA and steel mechanical prop-
erties assigned. Since the falling weight was mod-
elled as a simple sphere, an artificially large density
was used to give the same mass as the one used in
the experiments. To represent the specimen plate
the material ‘Mat.024-Piecewice linear plasticity’
was selected and defined according to the exact
true stress-strain curve until the onset of necking
(Dieter 1986), where the true stress σt and the true
strain εt are expressed in terms of the engineering
stress σe and engineering strain εe (obtained from
the tensile tests, and shown in Fig. 3) by:
σt = σe (εe + 1)
εt = 1n (εe + 1)
Since for the impact tests considered here only
plastic deformation was observed, failure strain
was not required for the material definition.
MARSTRUCT.indb 421
Strain-rate sensitivity was considered by includ-
ing nominal strain-rate coefficients (Jones 1989).
However, this resulted in predicting lower displace-
ments than the experimental results and hence
these strain-rate coefficients were not included in
the model.
In order to try to investigate the source of vibra-
tions seen in the experimental force-deflection
results, the time interval between outputs was
decreased ten times in comparison with the
previous analysis on aluminium plates.
4 RESULTS AND DISCUSSION
4.1 Force-displacement response
Initially, in order to optimise the numerical model in
terms of mesh size and element type (shell or solid),
four impact velocities from the full set of experi-
mental tests were considered. Force-displacement
test results of a high, low and two intermediate
velocities were compared to the numerically simu-
lated impact responses.
For the thick plates (4.0 mm), using any of the
three numerical models (Shell2, Shell1 or Solid1)
gave very similar results for all four impact veloci-
ties, and in general the numerical simulations
compare well with the experimental results. As
an example, Figure 5 compares the experimental
and numerical results for the thick plate impacted
at the highest velocity of 5.9 m/s. From this figure
it can be seen that, in terms of force-displacement
response, the Solid1 model does not give better pre-
dictions than the computationally less demanding
Shell2 or Shell1 models, although the solid models
give slightly higher maximum deflection and lower
maximum impact force. This was also found to be
the case for the thin plates (1.4 mm).
Figures 6 and 7 compare the experimental
(dashed line) and numerical (continuous line)
force-displacement results for all four incident
25
Experimental
Shell2
20 Shell1
Solid1
Force [kN]
15
10
5
(1)
0
(2)
0 1 2 3 4 5 6
Displacement [mm]
Figure 5. Force-displacement response. Thick plate
(4.0 mm) impacted at 5.9 m/s.
421
2/18/2011 5:50:45 PM
 25
20
Force [kN]
15
10
5
0.88m/s 2.65m/s
0 0.00
0 1 2 3
Displacement [mm]
Figure 6. Force-displacement response thick steel plates.
Experimental (dashed line); Numerical (continuous line).
10
8 accurate. The source of this vibratory response
4.39m/s
Force [kN]
6
2.55m/s
4
0.60m/s
2 to move.
0 a sphere, but a hemispherical-ended bar con-
0 2 4 6
Displacement [mm]
Figure 7. Force-displacement response thin steel plates.
Experimental (dashed line); Numerical (continuous line).
velocities considered for the thick and thin plates
respectively. Good agreement between numerical
and experimental results is seen for the whole range
of impact velocities, and the overlapping parts of
the response for different velocities are coincident.
However, for the thin plates deflection and force
are slightly overestimated by the numerical model
at the higher impact velocities (4.39 and 5.90 m/s).
For both thin and thick plates discrepancies were
seen between the experimental and numerical
force-displacement responses for impacts at the
lowest velocities (0.88 and 0.60 m/s respectively)
as illustrated in Figure 8. These are thought to be
due to experimental measurement errors which
become significant when compared to the very low
forces and displacements seen at these low incident
energies.
All of the experimental and numerical responses
in Figures 5 to 8 show oscillations during the initial
stages of the impact, which are more pronounced
in the experimental results, becoming more so as
the severity of the impact test was increased. These
vibrations are also more apparent for the thick than
for the thin plates. However, since the oscillations
MARSTRUCT.indb 422
1.00
Experimental
Shell2
Shell1
0.75 Solid1
Force [kN]
0.50
0.25
4.80m/s 5.90m/s
4 5 6 0.00 0.25 0.50 0.75 1.00 1.25 1.50
Displacement [mm]
Figure 8. Force-displacement response. Thin plate
(1.4 mm) impacted at 0.6 m/s.
5.90m/s are harmonic and sufficient cycles occur prior to
the end of the test the energy values are sufficiently
could be due to two possible phenomena:
i. Due to the inertia of the plate, at the initiation
of contact the impact mass and plate vibrate
around the contact stiffness as the plate begins
ii. Since in the experiments the impact mass is not
8 10 12 nected via a load cell to a large mass, vibrations
may be set up in this (and/or other) part(s) of
the impact machine.
Phenomena (i) would be apparent in both
experimental and numerical responses. However,
phenomena (ii) would only be apparent in the
experimental response.
The fact that vibrations of the same frequency
were noted in the experimental force measure-
ments after separation of the impact mass and
plate at the end of the impact event suggest that
vibrations in the test machine (ii) were present in
the experimental results. Also, the presence of (less
severe) vibrations in the numerical results suggest
that oscillations due to the contact stiffness-plate
inertia phenomena (i) are also present in the exper-
imental results, hence the larger vibrations seen
experimentally.
Since only oscillations due to (i) are valid mate-
rial impact responses, it would be pertinent to filter
out any experimental oscillations due to machine
vibrations. However, in practice this filtering may
well affect the impact response of interest, and it is
extremely difficult to ensure that only the machine
vibrations are filtered out. These results show
that the filtering and interpretation of impact test
results is by no means a straightforward, clear-cut
process.
Since no significant increase in accuracy was
seen in the results described above through using
422
2/18/2011 5:50:46 PM
 the finer meshed Shell1, or the Solid1 models, the
less computationally expensive Shell2 model was
then used to calculate the maximum transversal
deflection and impact force for the full range of
impact velocities carried out in the experiments.
Figures 9 and 10 show these experimental-
numerical comparisons. Excellent agreement for
all incident velocities is shown in Figure 9 between
the numerical and experimental maximum deflec-
tion results for thick plates. Figure 9 shows that the
experimental maximum deflection is also very well
predicted by the numerical analyses for thin plates
at lower impact velocities. However, the experi-
mental thin plate values of maximum deflection
and maximum force in Figures 9 and 10 respec-
tively become slightly lower than their numerical
counterparts when the impact velocities exceed
3.0 m/s. This is seen to be due to the slightly dif-
ferent shapes of the force-deflection responses of
Figure 7, where the experimental curves at higher
velocities exhibit slightly higher stiffness, but the
drop of force to zero at the end of the impact event
occurs at a lower deflection than for the numerical
response. In fact, when the experimental and numer-
ical energy values (obtained by integration of the
14
Maximum Deflection [mm]
relevant force-displacement curves) are compared
the numerical results underestimate the amount of
impact energy irreversibly absorbed by the plates
by an average of 12% of the incident energy.
It is not clear exactly what causes these discrep-
ancies, but it appears that mechanism(s) of energy
absorption seen in reality are not modelled cor-
rectly in the numerical analyses, for example the
dissipation of heat generated during plastic defor-
mation. It is also possible that this slightly higher
stiffer response is due to strain-rate effects that
were not included in the FEA.
Since the numerical model is very sensitive to
the true stress-strain curve, any slight errors in this
input could explain discrepancies between numeri-
cal and experimental results. However, since this
behaviour was obtained through in-house tensile
tests cut from the same sheets from which the
impacted plates were cut, confidence in this curve
is high.
Despite generally good agreement, the thick
plate experimental maximum forces are over-
predicted by the numerical model, especially at
velocities lower than 2.5 m/s, at which point an
inflection is seen in both the experimental and
numerical curves. This inflection could be due to a
change from the mainly contact-indentation domi-
nated damage seen at lower velocities to global plate

12 1.4 mm
plastic deformations at higher velocities. It is sug-
Numerical
gested that this over-prediction of the experimen-
10
tal impact force could be due to the way in which
8 Experimental the local contact is modelled numerically, since this
6 4.0 mm
would only become significant in impacts where
contact damage is dominant, as is the case for the
4 thick plates at lower incident velocities. This could
2 be because the striking mass was modelled as a
rigid body, when in fact deformation at the impac-
0
tor nose could have occurred experimentally.
0 1 2 3 4 5 6 7
Impact Velocity [m/s]

Figure 9. Maximum deflection-velocity. Thick and thin

4.2 Plastic deformation, maximum stresses
plates. and energy partition
Once the force-displacement response was validated
against the experimental results, the numerical
25 simulations were then used to give information
4.0 mm concerning the specimen plastic deformation, the
Maximum Force [kN]

20 Numerical maximum stresses, and the partition of the impact

energy.
15 Experimental It is beneficial to be able to predict the shape
of the deformation due to both global deflection
10 1.4 mm
and local indentation. Local indentation is divided
in two parts: local out-of-plane plate deformation
5 (where the plate wraps around the indenter), and
the indentation of the mass into the plate thickness.
0 For the thick plates the Solid1 model gives a bet-
0 1 2 3 4 5 6 7
Impact Velocity [m/s] ter definition of the shape of the deformation than
does the Shell2 or Shell1 model, because only solid
Figure 10. Maximum force-velocity. Thick and thin plates. elements are able to model the change in plate

423

MARSTRUCT.indb 423 2/18/2011 5:50:47 PM

 thickness associated with the indentation of the
mass into the plate at the impact point, which is
significant for the thick plates (Fig. 11a). For the
thin plates indentation is more significant in terms
of out-of-plane plate deformation, (Fig. 11b) and
hence the only requirement is a fine shell mesh for
accurate modelling of the local indentation of the
thin plates. A fine mesh is also the only require-
ment for the accurate modelling of the global plate
deformation of both thin and thick plates.
In Villavicencio et al. (2010) the maximum stress
for impacted aluminium plates was found to occur Figure 12. Maximum von Mises stress distribution.
on the surface opposite to the impact point for Impact velocity 5.9 m/s.
both shell and solid models, but that near the sup-
port the stresses did not differ between upper and
100 Steel 4.0 mm
lower faces. The numerical models of the current
steel plates also show that the maximum plastic
80
strain occurs on the lower surface opposite to the
impact point, and indicate the presence of plas-

Energy [J]
60
tic strain throughout the plate thickness at high Kinetic
Internal
impact velocities. However, at lower impact ener- 40 Sliding
gies, elements near the neutral axis did not undergo Internal+Sliding
Total
plastic strain. 20
Effective stresses were also observed to decrease
gradually from the impact point to the supports. 0
Figure 12 shows the maximum von Mises stress dis- 0 1 2 3 4
tribution on the plate lower surface when impacted time [ms]
at the highest velocity. For the thick plates only that 100 Steel 1.4 mm
part of the specimen within the annular cut-out in
the support plates is affected by the impact. How- 80
ever, for the thin plates the stresses in the clamped
Energy [J]

area of the specimen are also seen to be affected 60

by the impact. This is thought to be mainly due to
the slippage of the specimen plate between the sup- 40 Kinetic
port plates due to the higher deflections and hence Internal
Sliding
membrane forces experienced by the thin plates. 20 Internal+Sliding
Total
Temporal variations in the energy during
the impact system are shown in Figure 13. The 0
kinetic energy of the falling mass is dissipated 0 1 2 3 4
time [ms]

Figure 13. Energy partitioning. Impact velocity 5.9 m/s.

as a combination of internal and sliding energies

Figure 11. Shape of the deformation. Impact velocity
5.9 m/s.
(the hourglass energy was zero in the numerical
simulations). The internal energy consists of elas-
tic strain energy and plastically dissipated energy.
The sliding energy (or contact energy) is due to the
friction included in the support plates-specimen
contact definition. During the initial instants of
the impact event the internal energy is due to elas-
tic deformation only, and is almost entirely respon-
sible for the dissipation of kinetic energy as the
sliding energy is extremely small. However, as the
impact progresses both plastic deformation energy
becomes significant and sliding energy increases.
Although the latter remains a very small contri-
bution to the total energy conversion in all cases,

424

MARSTRUCT.indb 424 2/18/2011 5:50:48 PM

 the proportion of sliding energy is slightly higher
for the thin plates due to the small radial displace-
ments between the support plates (slippage) expe-
rienced by these specimens (Fig. 12).
4.3 Comparisons with aluminium impact response
The impact tests on the present steel plates were
part of a wider experimental study by Sutherland
and Guedes Soares (2009), which also considered
the aluminium plates numerically simulated in
Villavicencio et al. (2010). Since impact response
is a structural and not a material property, the
plate thicknesses of the different materials were
chosen to give plate bending equivalence so that
valid comparisons between the different materials
were possible. Hence, the thicknesses of the alu-
minium plates were 6.0 and 2.0 mm (thick and thin
respectively). Here, the experimental and numeri-
cal impact responses of the steel and aluminium
plates are compared.
The improvements included in the FE model of
the steel plates have been retrospectively considered
in the aluminium FE model, and the numerical
results of the latter updated accordingly. In the
new aluminium models, the shell elements have 6
and 4 integration points for the 6.0 and 2.0 mm
thick plates, respectively. The thin plate solid model
used 4 elements through the thickness instead of
the 2 defined in the previous model. However, for
the thick plates the through-thickness number
of elements of the solid model was maintained
at 6. These modifications were included in order
to define the same number of integration points
through the thickness for both shell and solid
models. The time interval between outputs was
also decreased ten-fold.
The modified aluminium numerical models were
again initially evaluated and optimised in terms of
element type, using the maximum and minimum
experimental impact velocities for each plate thick-
ness in this case. However, it was assumed that,
as found previously for the aluminium plates in
Villavicencio et al. (2010), and for the steel plate
impacts of this paper, mesh size was not signifi-
cant and hence only the Solid1 and Shell2 elements
were considered. As for the steel plates, the alumin-
ium shell and solid models give very similar force-
displacement results, and good agreement with
the experimental results is seen. As an example,
Figure 14 gives a comparison of the experimental
and numerical results for a thin aluminium plate
impacted at 5.9 m/s.
For both thicknesses, the improvements to
the aluminium numerical model gave better
agreement between the shell and solid models,
which has been illustrated by the inclusion of
the ‘old’ Shell2 model results in Figure 14. Very
425
MARSTRUCT.indb 425
12
Experimental
Shell2
10 Solid1
Shell2_Old
8
Force [kN]
6
4
2
0
0 2 4 6 8 10 12 14
Displacement [mm]
Figure 14. Force-displacement response thin alumin-
ium plate. Impact velocity 5.9 m/s.
similar discrepancies between the numerical and
experimental force-deflection responses were seen
for the aluminium plates at the very lowest inci-
dent velocities as were seen for the steel specimens
(Fig. 8). Again, these are thought to be due to the
relatively high significance of experimental errors
when compared to the very small force and dis-
placement measurements that must be made for
these very low-energy impact events.
As for the steel plates, since there was seen to
be no advantage in using the more computation-
ally expensive solid elements, the Shell2 model
was selected to calculate the maximum deflections
and forces for all experimental impact velocities,
as shown in Figures 15 and 16. These plots show
very good, and similar, agreement between experi-
mental and numerical results as seen for the steel
plates in Figures 9 and 10. However, the agree-
ment between experimental and numerical maxi-
mum deflection for the thin aluminium plates seen
in Figure 15 continues up to the highest impact
velocities (as opposed to the numerical overestima-
tion of maximum deflection at high impact ener-
gies for the steel plates seen in Fig. 9), although the
experimental curve does also dip at higher veloci-
ties. The slight overestimation of maximum force
for thin steel plates at higher velocities (Fig. 10) is
also seen in Figure 16 for the aluminium tests, and
again more impact energy is irreversibly absorbed
experimentally than numerically predicted.
The overestimation of maximum impact force
by the numerical model seen for thick steel plates
(Fig. 10) is seen to be even more significant in
Figure 16 for the thick aluminium plates. Again,
it is thought that this is due to the way in which the
local impactor-plate contact/indentation is modelled.
The inflection observed in the force-velocity curve
of the thick steel plates at around 2.5 m/s, thought
to be due to the initiation of significant global
plastic deformation, is also seen in that of the thick
aluminium plates.
2/18/2011 5:50:51 PM
 14 of plastic strain throughout the plate thickness at
high impact velocities was seen. However, at lower
Maximum Deflection [mm]

12 2.0 mm
impact energies elements near the neutral axis were
10 Experimental not predicted to undergo plastic strain.
8 Numerical Although generally very good agreement was
6.0 mm obtained, two discrepancies between numerical
6
and experimental were observed, that were very
4 similar for both steel and aluminium impacts:
2 i. For thin plates at higher velocities, numerical
0 predictions overestimated the experimental
0 1 2 3 4 5 6 7 maximum forces, and underestimated the energy
Impact Velocity [m/s] irreversibly absorbed. The numerical simulations
also overestimated the thin steel plate maximum
Figure 15. Maximum deflection-velocity. experimental deflection for higher velocity
impacts. A slightly stiffer response was also seen
25 experimentally. It is postulated that this could be
due to more slippage, more plastic deformation
6.0 mm
20 generated heat dissipation, or strain-rate depend-
Maximum Force [kN]

Numerical
15
Experimental
10
5 plate-support contact has been modelled.
0
0 1 2 3 4 5
Impact Velocity [m/s]
Figure 16. Maximum force-velocity.
5 CONCLUSIONS
Detailed information of the impact response of
clamped steel and aluminium circular plates has
been obtained through non-linear explicit dynamic
simulations using the LS-DYNA software package.
The results obtained were in good agreement with
those of previous experimental tests, indicating
that relatively coarse meshes using shell elements
are sufficient to predict the maximum deflections
and forces for the entire plastic response of both
steel and aluminium.
The support plates representing the experimen-
tal boundary conditions were capable of represent-
ing some small longitudinal displacement of the
specimen plate between the supports, especially for
the thin plates at higher velocities.
The numerical simulations gave a good under-
standing of the shape of the deformation in plates
subjected to impact loads. In terms of indentation,
a fine meshed solid model gave a better definition
of the shape of the deformation since this was able
to model both the deformation around the indenter
and also the change in plate thickness at the impact
point. The maximum plastic strain occurred on
the lower surface of the plate, and the presence
ant effects, or a combination of these effects.
ii. For thick plates the maximum forces seen exper-
imentally were overestimated by the numerical
2.0 mm model. It is thought that this is due to the way
in which the impactor-plate local contact, and
Further work would be beneficial to refine the
model in terms of the static coefficients of fric-
6 7
tion used, the modelling of the striking mass,
possible material property strain-rate effects, and
additional energy absorption mechanisms. Since
it is highly probable that more than one of these
phenomena could be influential, and more impor-
tantly that there may well be interactions between
these effects, any further work must be carefully
designed to give a numerical model optimised in
terms of all such effects.
The investigation of the impact of thicker plates
would also be worthwhile. These could be geomet-
rically scaled-up versions of the present plates, or
thicker plates of similar size to the present ones. In
the former case, a dimensional analysis approach
to develop scaling laws, such as used with suc-
cess in Sutherland and Guedes Soares (2007) for
laminated glass-polyester plates, could be used.
This would also identify any unexpected scaling
model shortcomings, or ‘size effects’ that could
lead to inaccurate prediction of full-scale in-service
behaviour from laboratory scale tests.
For the latter case of plates of lower length
to thickness ratio, a far more contact/indenta-
tion dominated behaviour, with little global plate
deformation or influence of boundary conditions,
would be expected. Hence, the use of refined solid
element models would be necessary to represent
the indentation into the plate thickness. The study
of lower length to thickness ratio plates could also
be used to approximate the behaviour expected for
thinner plates at, or near to, stiffeners.

426

MARSTRUCT.indb 426 2/18/2011 5:50:52 PM

 ACKNOWLEDGMENTS
The work has been performed in the scope of the
project MARSTRUCT, Network of Excellence
on Marine Structures (http://www.mar.ist.utl.pt/
marstruct/), which has been financed by the EU
through the GROWTH Programme under con-
tract TNE3-CT-2003-506141.
The first author has been financed by the
Portuguese Foundation for Science and Technol-
ogy, under contract SFRH/BD/46369/2008.
REFERENCES
Dieter GE. 1986. Mechanical behavior under tensile and
compressive loads. ASM Handbook. 8: 99–10.
Ehlers S. 2010. Strain and stress relation relation until
fracture for finite element simulations of a thin circu-
lar plate. Thin-Walled Structures. 48 (1): 1–8.
Hallquist JO. 2005. LS-DYNA Theory Manual. Liver-
more Software Technology Corporation.
Jones N. 1989. Structural Impact. Cambridge University
Press.
Shen WQ. 1995. Dynamic plastic response of thin circular
plates struck transversely by nonblunt masses. Int. J.
Solids Structures. 32 (14): 2009–2021.
427
MARSTRUCT.indb 427
Simonsen BC, Lauridsen LP. 2000. Energy absorption
and ductile failure in metal sheets under lateral inden-
tation by a sphere. International Journal of Impact
Engineering. 24 (10): 1017–1039.
Sutherland LS, Guedes Soares C. 2007. Scaling of impact
on low fibre-volume glass–polyester laminates. Com-
posites: Part A. 38: 307–317.
Sutherland LS, Guedes Soares C. 2009. Impact behav-
iour of GRP, aluminium and steel plates. Analysis and
Design of Marine Structures; Guedes Soares & Das
(eds). Taylor & Francis Group: London. 293–300.
Tabri K, Alsos H, Broekhuijsen J, Ehlers S. 2007.
A benchmark study on ductile failure criteria for shell
elements in multiaxial stress state. Advancements
in Marine Structures, Guedes Soares & Das (eds).
Taylor & Francis Group: London. 401–409.
Villavicencio R, Sutherland LS, Guedes Soares C. 2010.
Numerical simulation of transversely impacted,
clamped circular aluminium plates. 5th International
Conference on Collision and Grounding of Ships;
Espoo, Finland. 104–112.
Wang G, Ohtsubo H, Arita K. 1998. Large deflection of
a rigid-plastic circular plate pressed by a sphere. Jour-
nal of Applied Mechanics. 65 (2): 533–535.
2/18/2011 5:50:53 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Study on the residual ultimate longitudinal strength of hull girder

of a bulk carrier against a sagging moment after ship collision

Yasuhira Yamada & Yoshitaka Ogawa

National Maritime Research Institute, Tokyo, Japan

ABSTRACT: The purpose of this study is to investigate residual Ultimate Longitudinal Strength (ULS)
of a panamax bulk carrier being subjected to sagging moment. ULS of panamax bulk carrier in intact
condition is estimated using large-scale nonlinear FEM analysis code LS-DYNA as well as simplified
analysis method based on the Smith method. ULS in intact condition estimated by FEA is compared
with that estimated by simplified method. It can be found that ULS estimated by FEA and that by
simplified analysis show a fairly good agreement although small discrepancy can be seen. Moreover ULS
in damaged condition is estimated assuming collision at midship region of the side shell, where damaged
area is artificially created as a first phase of the research project.

1 INTRODUCTION
Tier I: Goals
1.1 Goal Based ship construction
Standard (GBS)
A Goal Based ship construction Standards (GBS) Tier II: Functional Requirements
for oil tankers and bulk carriers was adopted as
a new SOLAS convention at the 87th session of
Tier III: Verification of conformity
Maritime Safety Committee (MSC87) in IMO
(2010a). Figure 1 shows the illustration of an
overall framework of GBS, where 5 tiers can be Tier IV:Rules and regulations for ship design
seen. Functional requirement of tier II among and construction
5 tiers are intended to allow the performance
based criteria other than the conventional pre- Tier V: Industry practices and standards
scriptive criteria.

Figure 1. Illustration of GBS framework.

1.2 Residual strength
“Residual Strength” is included in Tier II as one of
functional requirements as follows (IMO, 2010b).
“II.5 Residual strength
Ships shall be designed to have sufficient strength
to withstand the wave and internal loads in speci-
fied damaged conditions such as collision, ground-
ing or flooding. Residual strength calculations shall
take into account the ultimate reserve capacity of
the hull girder, including permanent deformation
and post-buckling behaviour. Actual foreseeable
scenarios shall be investigated in this regard as far
as is reasonably practicable.”
Consequently “residual strength” in case of
damage becomes mandatory to newly oil tankers
and bulk carriers constructed after SOLAS con-
vention comes into force. That is, those ships shall
have to possess sufficient residual strength after
damaged such as collision, grounding and/or
flooding. However, it is not necessarily clear the
way to comply with the functional requirements.
In the conventional rules and regulations, there
was no detailed description of them because they
are newly developed functional requirements in
the GBS. Therefore, administrator or classifica-
tion societies have to prepare for ways to fulfill
these functional requirements. It is important to
assess the safety level ensured by the functional
residual strength from the technical point of
view.
With regards to residual strength of ships, several
studies have been carried out (Wang et al, 2002a;
Wang et al, 2002b; Das & Chuang, 2007; Hussein &

429

MARSTRUCT.indb 429 2/18/2011 5:50:53 PM

 Guedes Soares, 2009; Nikolopoulos et al, 2010).
Hussein & Guedes Soares (2009) investigated
residual strength of three double hull tankers by
using simplified method (Smith type method) and
carried out reliability assessment. Nikolopoulos
et al (2010) has investigated the compliance of
existing prescriptive regulations with GBS regard-
ing the residual strength of ship. It is concluded
by Nikolopoulos et al (2010) that “Having seen all
the existing regulation on the topic we conclude
that none of them can be described as fully GBS
compliant.” It might be true that none of these reg-
ulations explicitly fully cover the residual strength
requirement in GBS. However it might be also
promising that existing rules implicitly satisfy the
residual strength requirement in GBS depending
on various factors such as damage scenarios, wave
induced load and so on. Therefore it is important
to quantitatively investigate residual strength of
ships after damage.
Various residual strengths after damage can be
defined depending on loads (global/local vertical/
horizontal bending moment, local pressure and so
on). In the present study main focus is laid on resid-
ual Ultimate Longitudinal Strength (ULS) of glo-
bal hull girder under sagging moment after collision
damage. Loads that are applicable to the damaged
ship are also very important since collapse (failure)
of ships take place if loads are larger than strength
capacity of damaged ships (see Eq. (1)).
P(f ): L − S > 0 (1)
where, P( f ): failure probability, L: Loads,
S: Strength capacity. Since loads and strength
deviate as random variables, collapse of damaged
ships can be evaluated by using reliability analysis
as shown in several previous studies (eg. Fang &
Das, 2005).
Moreover collisions and grounding often take
place in ports or near shoreline. In evaluating
collapse of damaged ships it can be pointed out
that average wave induced bending moment in
these area might be smaller than that in the middle
of the Atlantic/Pacific Ocean. In fact such effect
has already taken into account in IACS Common
Structural Rules (CSR). In this study, however,
main focus is laid only on strength capacity of
damaged ships. Loads applicable to the damaged
ships as well as evaluation of collapse of damaged
ships (Eq. (1)) will be investigated in the future
study.
Ultimate Longitudinal Strength (ULS) of global
hull girder under vertical bending moment might
be decreased after damages such as collisions
and groundings. Therefore it is important to keep
Residual Ultimate Longitudinal Strength (RULS)
of ships so that ships do not break into two parts
430
MARSTRUCT.indb 430
until damaged ships could arrive at ship yards for
repair. Therefore ULS can be one of important
strength which ships should possess even after
damages. In order to investigate “RULS” of ships
after “collision”, a project has started in National
Maritime Research Institute since April of 2010.
In order to accurately investigate collapse
mechanism of “damaged ships” under vertical
bending moment non-linear FE analysis using
commercial code LS-DYNA is adopted where two-
steps-approach is adopted. First step is collision
analysis. Second step is ultimate strength analysis
with keeping residual stress and deformation
induced by the first step of collision analysis,
which are usually disregarded in the previous
RULS researches.
As a first phase of the project, ultimate strength
with intact condition and its numerical method
are investigated using nonlinear FEM code LS-
DYNA. Although implicit methods are widely
used to assess ULS of ships, applicability of
explicit method is investigated considering consist-
ency with first step of collision analysis as well as
effectiveness of two-step-approach. In this study,
ULS of hull girder of a bulk carrier under sag-
ging bending moment is investigated using large
scale finite element model. Collapse mechanism of
the bulk carrier is investigated using time domain
simulation and discussed in detail. In addition
to the intact ULS, damaged ULS is calculated
with assuming damaged area simply by removing
elements. The effect of damaged condition on the
reduction of ULS is discussed in detail. Finally
some of numerical methodologies are summarized
in order to assess RULS of hull girder after colli-
sion as a next step of the project.
2 FINITE ELEMENT MODEL
In order to investigate RULS of ships, a cape
size bulk carrier is adopted as a target ship.
Figure 2 shows a finite element model of the bulk
carrier, where main particulars of the ship are also
shown in Table 1.
Figure 2. Finite element model.
2/18/2011 5:50:54 PM
 Table 1. Main particulars.
Length Lpp 278 m Generally it can be more probable that a ship is
Breadth Bm 45 m not a quite new ship when being struck by another
Depth Dm 24 m ship. Rather it can be expected that a ship is a more
Dead weight DW 170281 ton
Table 2. Number of nodes and elements.
Nodes 234,779
Elements 322,487
Table 2 shows a number of nodes and elem-
ents used in the finite element model. In order to
carry out ship-ship collision analysis as well as
ULS analysis, three cargo holds (No. 4, No. 5 and
No. 6 holds) around amidships are modeled with
elastic-plastic elements, and remaining fore and
aft parts are modeled with rigid element. No. 5
cargo hold is set to target cargo hold and mod-
eled with fine mesh and adjacent cargo hold of
No. 4 and No. 6 are modeled with slightly coarse
mesh than No. 5 hold. Usually in ship-ship colli-
sion, only the struck side of target hold is mod-
eled with fine mesh and the other side might be
modeled with rather coarse mesh or rigid ele-
ments in order to reduce a computation time. In
this study, however, in order to carry out ULS
analysis after collision it is important both sides
are symmetrically modeled, therefore both sides
of the target hold (No. 5) are modeled with
fine mesh, consequently a number of elements
increases considerably.
It is noted that since both sides are modeled with
elastic-plastic elements effect of global hull girder
horizontal bending moment can also be considered
in this model especially being subjected to collision.
Plates and web stiffeners are modeled with shell
elements and flange of stiffeners are modeled with
beam elements in fine mesh region. Default element
types of LS-DYNA are used for both beam and
shell elements respectively. That is Belytscko-Tsay
element for shell elements, Hughes-Liu element
for beam elements. The ship is designed to com-
ply with a class rule, and built of typical mild steel
and high tensile steel. Material model of MAT24
is used where strain hardening effect and fracture
of elements are taken into account. The strain rate
effect can also be taken into account by MAT24.
However in this study in order to investigate quasi-
static ULS and also to compare ULS estimated by
FEA with that by smith method, strain rate effect
is not taken into account. Strain rate effect can be
taken into account in order to investigate dynamic
ULS of ships considering time duration of wave
431
MARSTRUCT.indb 431
bending moment in the next phase of the present
research project.
or less aged ship with plate diminution due to corro-
sion and so on. Therefore it is assumed that a ship
is in aging degradation condition. In order to con-
sider such effect, diminution of plate thickness is
considered by adopting “net scantling” condition as
prescribed by CSR (IACS, 2009) is adopted in both
FEA and simplified analysis. That is analysis are per-
formed based on t = tnet + 50% corrosion addition.
3 ULTIMATE LONGITUDINAL
STRENGTH (ULS) IN INTACT
CONDITION
Before carrying out RULS analysis, ULS in intact
condition is investigated using LS-DYNA, where
net scantling condition is considered (IACS, 2009).
No. 5 hold at midship is set to target hold since this
hold is supposed to be subjected to a most severe
bending moment and that damage caused by ship
collision at mid of a struck ship is supposed to
cause one of the most severe damage on the struck
side shell.
3.1 Analysis method
It can be assumed that the target hold is supposed
to be subject to a uniform bending moment.
Therefore as a loading condition, a same sagging
bending moment is applied to fore and aft T.BHD
of the target Hold (No. 5 Hold) respectively. It is
assumed that a plain T.BHD section remains plain
before and after loading, which is usually assumed
in Smith type simplified method. Moreover in order
to evenly distribute a bending moment applied to
T.BHD into many nodes consisting of a T.BHD
section, each T.BHD is constrained as a rigid
plane using a keyword of “CONSTRAINED_
RIGID” respectively. The effect of this modeling
on the result is supposed to be small since T.BHD is
located far from the center of the target hold where
focus is laid on. In explicit FEA a finite time needs
to be specified to apply load, and a load increasing
time (Δt [s]) needs to be gradually increased in order
to reduce dynamic effect. Therefore in the present
study an applied load is linearly increased from
0 [GNm] up to 20 [GNm] during t = 0 [s] to t = Δt [s].
A maximum applied moment is kept constant after
t = Δt [s] as shown in Figure 7 (black solid line).
Moreover in order to investigate the effect of load
increasing time on results sensitivity analyses are
carried out with varying Δt. In the present study
Δt = 0.3 [s] and 2.0 s [s] are shown.
2/18/2011 5:50:55 PM
 3.2 Analysis results
Figures 3–6 show deformation of the ship, where
it can be seen that upper deck of the target hold
collapse at the almost center of the target hold
although the accurate position is slightly at the aft
side from the center.
It is confirmed by the observation of simula-
tion results that after buckling of upper deck
at almost the longitudinal center of the target
hold, large out-of-plane deformation starts tak-
ing place at the same part of upper deck as well
as surrounding part of the upper deck. In the later
stage of collapse of upper deck, several out-of-plane
deformation become to merge one by one and to
form a global folding pattern of the target hold
as shown in Figures 4–7 and Figure 9 show time
histories of applied moment and reaction moment
of the section at the center of the target cargo
Figure 3. Collapse of the ship due to vertical bending
moment (Intact Condition).
Figure 4. Collapse of the ship due to vertical bending
moment (Intact Condition).
Figure 5. Collapse of the ship due to vertical bending
moment (Intact Condition).
432
MARSTRUCT.indb 432
Figure 6. Collapse of the ship due to vertical bending
moment (Intact Condition).
M02-C35a
(Mmax=16 GNm, Δt=2.0s, DT2MS=10ms)
30
Applied Moment
Moment [GNm]
20 Reaction Moment
10
0
0.0 0.5 1.0 1.5 2.0 2.5
Time[s]
Figure 7. Time histories of bending moments
(Δt = 2.0 s).
hold for Δt = 2.0 [s] and Δt = 0.3 [s] respectively.
It is confirmed by carrying out sensitivity analysis
that the effect of load increasing time on ULS is
significant, and that Mu decreases gradually as Δt
increases mainly due to reduction of dynamic effect.
Therefore at least around Δt > = 2.0 [s] might be
required to obtain quasi-static Mu as far as present
FE model and analysis conditions. Therefore in the
present study Δt = 2.0 is used hereafter. It can be
seen in Figure 7 that reaction moment increases as
the applied moment increase, and that after reach-
ing to ULS at around t = 2.0 [s] reaction moment
decrease considerably with some fluctuations.
Figure 8 and Figure 10 show relation between
reaction moment and curvature. Assuming the
beam theory for bending of global hull girder
(See Fig. 11), curvature (1/ρ) is calculated by rotation
angles of two T.BHDs by the following formula
1 θ1 θ 2
ρ= L
(2)
where θ1 and θ2 represents rotation angle of
two T.BHD respectively. L represents distance
between two T.BHDs which is equal to 25.38 [m].
θ1 and θ2 are usually almost same value, but slightly
different due to that structure of fore and aft T.BHD
of the target hold are not completely symmetry
2/18/2011 5:50:55 PM
 M02-C35a (ex. horizontal girders are attached only fore side of
(Mmax=16GNm, Δt=2.0s, DT2MS=10ms)
30
T.BHD and so on). Figure 8 shows relation between
reaction moment and curvature. It can be seen in
Figure 8 that ULS decreases gradually after reach-
Moment [GNm]

20
ing ULS. It is obtained from the present analysis that
ULS in intact condition (Mu,intact) is 15.22 [GNm].
10
Mu,intact = 15.22 [GNm] (3)
0
0.0E+00 2.0E-04 4.0E-04 6.0E-04 8.0E-04 Moreover fully plastic bending moment (Mp)
1/ρ[1/m] at the center of the target cargo hold is manually
calculated with using excel sheet where axial
Figure 8. Relation between reaction bending moment yielding force of all members on the transverse
and curvature (Δt = 2.0 s). section are calculated. Mp can be obtained as:

M02-C05a Mp = 19.48 [GNm] (4)

(Mmax=20GNm, Δt=0.3s, DT2MS=10ms)
30
Applied Moment
Therefore non-dimensional ULS of the present
ship at the target hold can be calculated as:
Moment [GNm]

Reaction Moment
20
Mu
M u* = (5)
10 Mp

Mu* can be calculated as 15.22/19.48 = 0.85.

0 According to definition of Mu*, Mu* should
0.0 0.2 0.4 0.6 0.8 1.0 satisfy with
Time [s]
* (6)
Figure 9. Time histories of bending moments u
(Δt = 0.3 s).
M02-C05a Eq. (6) is confirmed in the present analysis.
(Mmax=20GNm, Δt=0.3s, DT2MS=10ms) This value can also be useful especially when
30
considering Mu* in different size and type of ships
and comparing thsese Mu* each other.
Moment [GNm]

20
For verification and comparison purpose,
ULS of the ship in intact condition is estimated
10 using Smith type simplified method (Smith,
1977) using the software MARS2000 (BV, 2000).
0 Figure 12 shows an example figure of the half
0.0E+00 2.0E-04 4.0E-04 6.0E-04 8.0E-04 transverse section of the ship in intact condition
1/ρ[1/m] obtained by MARS2000. ULS in net scantling
is calculated in MARS2000. Figure 13 shows
Figure 10. Relation between reaction bending moment comparison of ULS between Simplified analysis
and curvature (Δt = 0.3 s).

curvature center

θ1 θ2
UP.DK.

Neutral Axis

BTM. ρ

T.BHD. L T.BHD.

Figure 11. Relation between curvature radius and Figure 12. Transverse section of the ship (Intact condi-
rotation angle of T.BHD. tion, MARS2000).

433

MARSTRUCT.indb 433 2/18/2011 5:51:00 PM

 25
20
14.44 15.22
Mu [GNm]
15
10
0
MARS2000 FEA (C35a)
Figure 13. Comparison of ULS in intact condition
between simplified method and FEA.
and FEA. It can be seen in Figure 13 that ULS
estimated by FEA gives slightly larger results than
that by simplified analysis. This is presumably due
to that strain hardening effect is not taken into
account in simplified method; consequently results
by simplified method might give underestimation.
Another reason might be due to that initial
deflection due to welding is not taken into account
in FEA. Considering these situations actual value
of ULS of the ship might lie between these two val-
ues. It can be said that fairly good correlation can
be achieved between simplified method and FEA.
4 ULTIMATE LONGITUDINAL
STRENGTH (ULS) IN DAMAGED
CONDITION
In this section, ULS in damaged condition is esti-
mated using FEA as well as simplified analysis,
where a collision at midship region of the ship
is assumed as an accident scenario. Hussein &
Guedes Soares (2009) assumed a vertical collision
damage of ship from upper deck to downward
in estimating RULS by using MARS2000. Since
two-dimensional transverse section is considered
in the smith type simplified method, extent of
longitudinal damage is not explicitly considered.
However rapid estimation of RULS by the Smith
type method is very useful and effective from the
view point of initial step of ship design. However
it is important and useful to compare RULS esti-
mated by simplified analysis and that by FEA. On
the other hand a 3-D ship-ship collision simulation
is going to be carried out in the present research
project in order to set up actual damage of the
struck ship. However it is not practical to carry
out a lot ofship-ship collision due to restriction by
time-consuming large scale simulation.
Therefore considering these above as well as a
first phase of the research project, artificial vertical
damages, which are similar to those assumed by
Hussein & Guedes Soares (2009) on the side shell,
434
MARSTRUCT.indb 434
Figure 14. Transverse section of the ship (Damaged
condition, MARS2000).
are assumed as shown in Figure 14. Dm denotes
molded depth of the ship, and h denotes height
of damage hole. In order to investigate effect of
damage height h on RULS, several analysis are
carried out with varying h values. The effect of lon-
gitudinal damage extent is not taken into account
and is assumed to be one element size. Therefore one
column of elements are removed along the damaged
transverse section on the side shell, consequently
rectangle damge area is assumed. Since one of the
main objectives of this section is to understand the
difference of RULS estimated by simplified analysis
and that by FEA, same assumption as Hussein &
Guedes Soares (2009) is adopted. That is, vertical
damage hole lies between upper deck and some-
where on the side shell of the struck ship. Further
pattern of h definition can be studied in the future.
Hussein & Guedes Soares (2009) defined Residual
Strength Index (RIF), which represents a ratio of
ultimate longitudinal strength between damaged
condition and intact condition. In this study, fol-
lowing Hussein & Guedes Soares (2009), Residual
Strength Index (RSI) is defined as follows;
M u,damage
S =
RSI (7)
M u,int act
where, Mu,damage denotes a ULS in damaged
condition, Mu,intact denotes a ULS in intact
condition. From its definition RSI should satisfy
following equation.
0 ≤ RSI ≤ 1 (8)
Figure 14 shows a illustration of the transverse
section in MARS2000, where both sides of the
ship is modeled in order to estimate ULS in dam-
age condition at one side shell.
5 ANALYSIS RESULTS
Figure 15 shows relation between Mu and h/D.
Figure 16 shows relation between RSI and h/D,
where RSI denotes residual strength index defined
by Eq. (7). h/D denotes non-dimensional damage
height, where h denotes height of damage hole and
2/18/2011 5:51:06 PM
 16.0
15.22 FEA (M16GNm_T2.0s)
15.0
Simplified
Mu [GNm]
14.44
14.0
13.0
12.0
0 0.2 0.4 0.6
h/D
Figure 15. Relation between Mu and h/D.
1.0 Simplified
FEA (M16GNm_T2.0s)
0.9
RSI
0.8
0 0.2 0.4 0.6
h/D
Figure 16. Relation between RSI and h/D.
D denotes moulded depth of the ship. It can be seen
in Figure 16 that RSI decreases as the h/D increases.
It is interesting to note that relation between
RSI and h/D is not linear although linear regres-
sion formula is obtained in Hussein & Guedes
Soares (2009). One of reasons can be due to
that ship types are different (a bulk carrier in
the present study; tankers in Hussein & Guedes
Soares, 2009). Bulk carrier have large opening on
upper deck while tankers do not have it. Another
reason might be presumably due to that damage
close to neutral axis does not affect much on
decrease of ULS. In this study damage length h is
assumed to always includes upper deck as shown in
Figure 14. Lower limit of damage area goes down
as h increases. Therefore outer shell structures
between upper deck and vertical center of depth
are removed if h = 0.5. Structures close to neutral
axis do not contribute much to bending resistance
although structures close to upper deck contrib-
ute much to bending resistance. Therefore it can
be presumed that reduction of Mu is large if h/D
is relatively small, and reduction of Mu is small if
h/D is relatively large. It can be seen in Figure 15
that reduction rate of Mu (tangent line of curve)
decreases as h/D increases.
It can also be seen in Figure 16 that curve
obtained by FEA and that obtained by simplified
analysis have good correlation if h/D is smaller
than 0.1. On the other hand, it can also be seen
in Figure 16 that RSI obtained by FEA is smaller
MARSTRUCT.indb 435
than RSI obtained by simplified analysis if h/D is
larger than 0.1.
It should be pointed out that RSI is largely
dependent on Mu,intact (denominator in Eq. (7)),
and Mu,intact estimated by FEA is larger than
that estimated by simplified analysis. Therefore
RSI estimated by FEA becomes lower than that
estimated by simplified analysis if absolute value
0.8 1 of RULS is close or same. In order to overcome
this issue another residual strength index RSI2 can
be proposed in the following.
M u,damage
S 2=
RSI (9)
Mp
where Mp denotes fully plastic bending moment
of the hull girder section as calculated in Eq. (3).
Figure 17 shows relation between RSI2 and dimen-
sionless damage height h/D.
It can be seen in Figure 17 that RSI2 decrease
as h/D increases in both FEA and simplified
analysis, which is the same tendency observed in
Figure 16.
0.8 1
It can be seen in Figure 15 and Figure 17 that
Mu and RSI2 estimated by FEA is larger than
that estimated by simplified analysis if h/D < 0.3
especially due to that ULS in intact condition is
different between FEA and simplified analysis. It can
also be seen in Figure 15 and Figure 17 that Mu and
RSI2 estimated by FEA becomes close to that esti-
mated by simplified analysis if h/D > 0.3. It is inter-
esting to point out that difference of absolute value
of Mu decreases as h/D increases. This is presum-
ably due to that causes which make difference of
Mu in intact condition gradually disappear as h/D
increases although causes at only one side disappear
and causes at the other side still remain.
In another word, it can be said that the index
RSI is different for the two approaches because
their predictions for the reduction of Mu of the
damaged ship are different as shown in Figure 15
and Figure 17. Therefore any residual strength
index with the same denominator would provide
similar results to those obtained using RSI2.
1.0
FEA (M16GNm_T2.0s)
0.9 Simplified
0.8
RSI2
0.7
0.6
0 0.2 0.4 0.6 0.8 1
h/D
Figure 17. Relation between RSI2 and h/D.
435
2/18/2011 5:51:08 PM
 By comparing Figure 16 and Figure 17, it can be
found that RSI estimated by simplified analysis is
larger than that estimated by FEA while opposite
tendency can be seen in RSI2. This is presumably
due to that RSI estimated by FEA becomes smaller
since denominator of Mu,int act estimated by FEA is
larger than Mu, int act estimated by simplified analysis.
However it can be presumed that unsymmetrical
collapse mechanism in case of damaged condition
might also affect. More detailed investigation can
be carried out in the future work. Considering these
above, it can be said that RSI is useful in order to
observe strength reduction in damaged condition
as compared with that in intact condition.
However it can be found that accuracy of RSI is
largely dependent on ULS in intact condition since
ULS in intact condition is used as denominator.
Residual Strength Index 2 (RSI2) proposed by the
present study gives stable index, and might be use-
ful in comparing actual RULS.
6 CONCLUSIONS
Ultimate Longitudinal Strength (ULS) in both
intact and damaged condition of hull girder of a
bulk carrier are investigated with using nonlinear
finite element simulation as well as with using Smith
type simplified analysis. LS-DYNA is used for FEA
and MARS2000 is used for simplified analysis.
Following concluding remarks can be obtained.
1. ULS in intact condition estimated by FEA and
that by simplified analysis shows fairly good
agreement although small discrepancy can be
seen. As far as a limited number of analysis, it is
found that ULS estimated by simplified analy-
sis gives slightly more conservative results than
that estimated by FEA.
2. It is found from FEA and simplified analysis that
RULS after side damage can be reduced about
7%–15% if dimensionless damage height is up
to around 50%–65% of the depth of a ship.
3. RULS estimated by FEA and that by simplified
analysis shows fairly good agreement especially
when dimensionless damage height is large pre-
sumably due to that initial imperfection effect
might be decreased in severe damaged condition.
4. RSI is useful in order to observe strength reduction
in damaged condition as compared with that in
intact condition. However it can be found that
accuracy of RSI is largely dependent on ULS in
intact condition since ULS in intact condition is
used as denominator of the index RSI.
5. Residual Strength Index 2 (RSI2) is newly pro-
posed in the present study, where fully plastic
bending moment of hull girder is used for the
denominator of the new index RSI2. RSI2 can
be useful index in comparing absolute value
436
MARSTRUCT.indb 436
of different RULS for different ships although
reduction of ULS from that in intact condition
can not be observed directly.
6. Combination of RSI as well as RSI2 can be
promising index to investigate and evaluate
RULS of ships.
It is noted that present results can be derived
by carrying out FEA and simplified analysis of
the specific bulk carrier in limited number of
calculation. More calculation results and investiga-
tion for another ship (size, type) as well as another
damage condition are necessary in order to derive
more general conclusion.
ACKNOWLEDGEMENTS
A part of the present research work was sup-
ported by Grants-in-Aid for Scientific Research
(B) (20360400). Authors would gratefully thanks to
Japan Society for the Promotion of Science (JSPS)
as well as the Ministry of Education, Culture, Sports,
Science and Technology (MEXT) for their support.
REFERENCES
BV (2000). “MARS2000 User’s guide”, BureauVeritas.
Marine Division Development Department.
Das, P.K. & Chuang, F. (2007). “Residual Strength
and Survivability of Bulk Carriers after Grounding
and Collision”, “Journal of Ship Research”, Vol. 51,
No. 2, pp. 137–149.
Hussein, A.W. & Guedes Soares, C. (2009). “Reliability
and residual strength of both double hull tankers
designed according to the new IACS common struc-
tural rules”, Ocean Engineering 36, pp. 1446–1459.
IACS (2009). “Common Structural Rule (CSR) for oil
tankers.”
IMO (2010a). “Amendments to the 1974 SOLAS
Convention, and to the FSS and LSA Codes”,
Maritime Safety Committee, MSC 87/3.
IMO (2010b). “Adoption of the International Goal-based
Ship Construction Standards for Bulk Carriers and Oil
Tankers”, Maritime Safety Committee, MSC 87/3/2.
Nikolopoulos, L., Samuelides, M. & Zaraphonitis, G.
(2010). “Investigating the compliance of existing
regulation with Goal Based Standards regarding the
residual strength of a damaged ship”, Proceedings
of SNAME Annual Meeting and Ship Production
Symposium, Vol. 2, pp. 162–169.
Smith, C.S. (1977). “Influence of Local Compression
Failure on Ultimate Longitudinal Strength of a Ship’s
Hull. In: Proceedings of the International Sympo-
sium on Practical Design in Shipbuilding. (PRADS),
Tokyo, Japan, pp. 73–79.
Wang, G., Spencer, J. & Chen, Y. (2002a). “Assessment of
a ship’s performance in accidents”, Marine Structures
Vol. 15.
Wang, G., Chen, Y., Zhang, H. & Peng, H. (2002b).
“Longitudinal Strength of ships with accidental
damages”, Marine Structures Vol. 15.
2/18/2011 5:51:13 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Comparison of resistance to penetration of stiffened panels

with T- and Y-stiffeners

M.R. Zareei
Chabahar Maritime University, Chabahar, Iran

ABSTRACT: Non-linear finite element method is a powerful tool for analyzing ship collision problem
and has been seen more and more applications in the recent years. Ship collision simulations still face
some difficulties in terms of providing reliable results. The accuracy of numerical modeling results sig-
nificantly depends on the proper definition of the phenomenon and careful control of the some critical
parameters such as rupture criteria. A series of FEM indentation analysis have been carried out quasi-
statically on various configurations of stiffened steel panels with T- and Y- stiffeners. The analysis method
presented has been implemented into explicit finite element code LS-DYNA, being linked with ANSYS
for pre- and post-processing of the related computations.

1 INTRODUCTION driven by the momentum of the ship. Depending

Stiffened plates are used as main supporting mem-
bers in many civil as well as marine structural appli-
cations. They typically consist of a plate with equally
spaced stiffeners welded on one side, often with
intermediate transverse stiffeners or bulkheads.
The most common stiffener cross-sections are
bulb, flat bar or T- and L-sections. Such structural
arrangements are common for both steel and alu-
minium structures.
Recently, some new steel double hull structures
were invented to achieve better energy absorption
capacity. Y-, rectangular box section, trapezoidal
box section, X- and corrugated frames are some of
new innovation stiffened structures. The intention
of these new designs was to prevent early crack
occurrence during a collision or grounding. Some
examples of this type illustrates in Figure 1.
Ship grounding scenarios can be divided into two
sub groups: ‘‘stranding’’ and ‘‘powered grounding’’.
During powered grounding, structural damage is
Figure 1. Non-conventional rectangular box, hat, n,
and M stiffeners and its attached effective plating (Badran
et al. (2007)).
on the topology of the sea floor, the ship may
either tear open or slide over ground. In the case of
stranding, it is assumed that the ship settles on the
sea floor without being subjected to sway or surge
motions. Damage of the hull is a consequence of
receding tides and wave loads. Collision analyses
of ship structures are increasingly being performed
both experimentally and numerically.
The objective of this paper is to study struc-
tural failure mechanisms in ships subjected to
stranding. The resistance to penetration of the
hull plating is of special interest. This is because
onset of fracture in the outer hull and in cargo
tanks is directly linked to oil spill and compart-
ment flooding. Moreover, onset of fracture results
in degraded structural capacity. This is especially
the case for structural members such as hull panels
which carry loads primarily by membrane action.
A ship is assumed to settle vertically on a rock. It is
further assumed that contact actions are local and
restricted to one plate section. The scenario is ana-
lyzed by conducting a series of panel indentation
numerical analysis. Various configurations of stiff-
ened panels are loaded laterally by a cone shaped
indenter until fracture occurs.
Many researchers study collision and stranding
of stiffened panels and ship hull structures. Most
recently, Quinton (2007) study collision scenario
on side shell of oil tanker ship with suggestion
of new stiffened panel arrangement. Alsos et al.
(2009a) perform a series of indentation tests quasi-
statically to study stranding on various configura-
tions of stiffened panels. The results of the scaled
down tests are reported in the first part of two-part

437

MARSTRUCT.indb 437 2/18/2011 5:51:13 PM

 companion paper. Also Alsos et al. (2009b) in part
II presents results from numerical analyses with
focus on fracture prediction. Ehlers (2009) numeri-
cally study a procedure to optimize ship side struc-
tures for crashworthiness.
Analysis of accidental response of marine struc-
tures is more and more performed by means of the
finite element method. A series of FEM indentation
Analysis have been carried out quasi-statically on
various configurations of stiffened steel panels with
T- and Y- stiffeners. The analysis method presented
has been implemented into explicit finite element
code LS-DYNA, being linked with ANSYS for pre-
and post-processing of the related computations.
Finally, Comparison of resistance to penetration
in stranding of stiffened steel panels with T- and
Y-stiffeners in ship structures is studied.
Although the study is aimed towards accidental
scenarios considering collision and grounding of
ships, the investigation also apply to other types of
problems. This may for instance be the response
of deck structures which are hit by dropped objects,
or the response of stiffened panels subjected to
explosions or ice actions.
2 NUMERICAL SIMULATION
Recent advances in computers and calculation algo-
rithms have made non-linear finite element analysis
a viable tool for assessing the internal mechanics
of collisions. Two types of FE methodologies are
relevant, namely implicit and explicit techniques.
Implicit methodologies obtain solution by simul-
taneous solving of systems of equations needing
frequent updating of the stiffness matrix for nonlin-
ear FE analysis. This places demands on the equa-
tion solver and the computer capacity especially in
terms of memory resources and CPU cycles. Explicit
codes obtain system solutions based on mass matrix
that remain constant not needing frequent updating
as in the case of stiffness approach. Explicit method
however needs smaller time steps to comply with
stability requirement for equation solving.
To analyze a collision and grounding accident
involving high non-linearity and phenomena such
as colliding surfaces contact, friction and rupture,
the explicit methodology is suitable. The required
calculation efforts are fewer than the commonly
used implicit methods. Converge of calcula-
tions is much easier to realize. Explicit method-
ologies based computer codes include ABAQUS/
Explicit, DYTRAN, LS-DYNA, PAM-CRASH
and ANSYS/LS-DYNA, and implicit method-
ologies based codes include ABAQUS/Standard,
ANSYS, MARC and NASTRAN.
Non-linear finite element (FEM) is a powerful
tool for analyzing ship collision problem and has
438
MARSTRUCT.indb 438
been seen more and more applications in the recent
years. Expectedly, many more FEM simulation
application will be seen in the coming years. Rapid
advances in computer technology and software
capacity have made FEM simulation a viable choice.
Ship collision simulations still face some difficulties
in terms of providing reliable results. The reliability
of the numerical simulation results largely depends
on the proper modeling of the phenomenon and
realistic consideration of aspects, such as rupture
criteria, non-linearity, element type, friction coeffi-
cient, and mesh fineness (element length).
In this paper investigation of parameters on the
stranding of stiffened panels with Y-stiffeners was
done. It will thus be possible to compare the crash-
worthiness of various structural arrangements
under similar collision scenarios.
3 FINITE ELEMENT RESULTS
AND DISCUSSION
As mentioned above, non-linear finite element
method modeling is the basis for analysis of
collision and stranding. ANSYS/LS-DYNA was
used to solve these numerical models.
SOLID164 element is used for the non-
deformable cone structure. Stiffened panels are
loaded laterally by a cone shaped indenter until frac-
ture occurs. SOLID164 is an 8-node brick element,
but for these analyses, the degenerate 4-node tetra-
hedral version is used for ease of meshing. This is
justified because all SOLID164 elements are rigid
elements. Only for simple modeling and meshing
3D element for indenter was used.
SHELL163 elements are used for stiffened panels
(deformable structure). SHELL163 is a 4-node
planar quadrilateral element. The default element
formulation (Belytschko-Tsay with reduced integra-
tion) is used for computational efficiency. A shear
correction factor of SHR = 5/6 is used along with 5
through-thickness integration points.
The automatic general contact algorithm is used
exclusively in these models. Automatic implies that
the outside normal for each contact surface is auto-
matically determined for all element contact. This
allows random contact between all elements to be
accounted for. The element contact coefficient of
friction is based on coulomb friction model and is
depend on the relative velocity of the element in
contact.
μc = FD + (FS – FD)e−DC(Vrel) (1)
where: FS: static friction coefficient.
FD: dynamic friction coefficient.
Vrel: contact relative velocity.
DC: exponential delay coefficient.
2/18/2011 5:51:14 PM
 FS and FD values used are for wet mild steel to
mild steel contact as reported in some papers are:
FS = 0.3, FD = 0.3 and DC = 7.0. In this analy-
sis only FS was used, because these analyses are
quasi-static.
A plastic kinematic material model is used in
this paper. This model simulates plastic deforma-
tion through linear interpolation between yield
stress and failure strain. The slope of this line is
called tangent modulus (Etan) or plastic modulus,
and is input with the other material properties.
Material strain rate dependency is incorporated
using the Cowper-Symonds model which calcu-
lates a dynamic yield stress, σD, by scaling the static
yield stress, σy, with a strain rate dependent factor
that given in Equation 2.
1 stress state is tri-axial, which complicates the devel-
⎛∈ ⎞P
1+ ⎜ ⎟
⎝ C⎠
. based on evaluation of the thickness measure-
where: ∈ = strain rate
C, P = Cowper-Symonds parameters
The material model properties that used in this
paper are given in Table 1. Yield stresses of used
material are given in Table 2.
A Cowper-symonds material property was
selected in ANSYS/LS-DYNA material list. In
this material properties C and P parameters (strain
rate parameters) must be entered. In these analyses
“dynamic relaxation” option in ANSYS was
applied and then only static parameters were used
in analysis.
Two of effective parameters on collision and
stranding are failure strain and element size. The
rupture failure may be explained using metal
forming theory. The maximum strain that the
material can sustain is limited by the local plas-
tic instability. At the failure, plastic deformations
concentrate on local areas. These areas have typi-
cal dimension of plate thickness. The failure proc-
ess can be divided into diffuse necking and local
necking phases. Diffuse necking develops slowly
as a result of strain rate hardening, and occurs
when the load reaches the maximum value. The
final failure occurs by local necking during which
the deformation is concentrated on a small area
away from which the structure remains almost un-
deformed. This theory of localized necking has
been developed for thin metal sheets, in which the
assumption of plane stress is valid and the failure
criteria can be based on the bi-axial principal strain
state formulations. For typical ship structures the
(2) opment of simple rupture criteria.
In this study, the definition of failure strain
ments as recommended by Alsos et al. (2009a) is
employed due to its simplicity. The values of fail-
ure strain that used in this paper are taken from
Equation 3 and Table 2.
t
ε cr + (εn n) (3)
le
where: εcr = failure stain.
n = power law coefficient.
t = thickness of element.
le = element length.
εn = failure strain at t = le.
4 ALSOS TESTS (VERIFICATION)

Table 1. Material model properties. A recent years, some research was done on colli-
sion and grounding of ship structure and stiffened
Density [kg/m3] ρ 7850 panels. One of them is penetration of stiffened
Young modulus [GPa] E 210 panel both experimentally and numerically by
Poisson ratio [−] ν 0.303 Alsos et al. (2009a, b).
Tangent modulus [GPa] Etan 1 Verification studies were conducted in order to
Cowper-Symonds strain C 40.4 check present analysis for using in panels ground-
rate parameter [Hz] ing damage studies. To achieve this verification,
Cowper-Symonds strain P 5 stiffened panels of Alsos et al. (2009a) was selected
rate parameter and analyzed numerically to compare present anal-
ysis with experimental analysis.
Alsos et al. papers deal with hull damage in
ships which were subjected to grounding actions.
Table 2. Material failure properties. A ship was assumed to settle vertically on a rock.
It was further assumed that contact actions were
Plate Stiffener
local and restricted to one plate section. The sce-
σy 285 340 nario was analyzed by conducting a series of panel
εn 0.71 0.71 indentation experiments. Various configurations
n 0.24 0.225 of stiffened panels were loaded laterally by a cone
shaped indenter until fracture occurred.

439

MARSTRUCT.indb 439 2/18/2011 5:51:14 PM

 That study published in two part companion
papers. Part one reported observations from the
experiments and Part two Numerical Analysis
deals with the numerical simulations of the same
tests.
In this paper, part I of these papers selected
for verification. For more details, please see Alsos
et al. (2009 a, b).

4.1 Penetration tests

Five indentation experiments were described in
Part 1. These had been conducted on 5 mm thick
stiffened steel plates, each 1200 mm long and
720 mm wide. A short description of stiffened
panels is given in Table 3 and Figure 2. The objec-
tive behind the tests was to study the deformation
response and the resistance to indentation of the
panels. The tests were carried out by forcing a
model of the seabed, a so called ‘‘indenter’’, later-
ally into the plate. This was performed up to and
beyond the point of fracture.
Material properties of panels were illustrated
in Alsos et al. (2009a). For verification we use
material properties that shown previously. Main
material properties such as yield stress and
Cowper-Symonds parameters are base on Alsos
et al. (2009a).
In this verification only US and 1-FB speci-
mens are modeled and analyzed. Each specimen
is modeled using two different element size includ-
ing 25 and 40 mm. The finite element models are
illustrated in Figure 3. The results of the finite
element analysis are compared with the experi-
mental results. Focus is on the force-deformation
relationship.

4.2 Unstiffened panel (US panel)

Figure 4 illustrates the force-indentation curve
Figure 2. Component configurations of Alsos et al.
for panel US experimentally and also numerically tests: a) illustrates the typical component, while b) and
with 2 element length. In experimental curve a ver- c) shows the transverse and longitudinal cross section
tical line is seen. Because hydraulic jack in Alsos with and without stiffeners, respectively, d) indenter and
experiment rig had a maximum stroke of 150 mm, loaded specimen (Alsos et al. (2009a)).
unloading had to be conducted after about 130 mm
deformation before the experiment was continued.
It was interesting to observe that after reloading, the force did not immediately enter the level prior
to unloading. In experiment, the panel reached a
maximum force of 1500 kN after about 200 mm
Table 3. Plate–stiffener configuration. indentation when the plate suddenly fractures.
In numerical analysis, the panel reaches about 1395
Specimen Num. of stiff. Stiffener type
and 1585 kN at 178 and 194 mm indentation in
US None – – 40 mm and 25 mm element length respectively.
1-FB one Flat bar (FB) 120 × 6 mm In Figure 5 distribution of plastic strain in plate
1-HP one Bulb (HP) 120 × 6 mm at failure initiation is shown. From this figure, we
2-FB two Flat bar (FB) 120 × 6 mm can see that largest strain occur in the plating adja-
2-HP two Bulb (HP) 120 × 6 mm cent to the indenter contact area and failure of US
panel initiate from this zone. For understanding

440

MARSTRUCT.indb 440 2/18/2011 5:51:15 PM


Figure 3. Finite element model of 1-FB model, a) upper
side with 25 mm element, b) lower side with 40 mm
element.
Figure 4. Force-indentation behavior of panel US.
failure initiation we can compare Figure 5 and 6. In
Figure 6 pictures of US panel experimentally test
with Alsos before and after of fracture illustrated.
4.3 Panel with 1-FB
Figure 7 illustrates the force-indentation curve for
1-FB panel experimentally and also numerically
with 25 and 40 mm element length.
In experiment, the panel reaches a maximum force
of about 1410 kN after about 160 mm indentation.
441
MARSTRUCT.indb 441
Figure 5. Deformation and plastic strain distribution of
US panel at failure initiation (element solution).
Figure 6. Indentation of US panel, a) prior to fracture,
and b) after fracture (Alsos et al. (2009a)).
With increase the indentation, FB stiffener starts
tripping about the intersection of plate and stiffener
and eventually folds over to the side. This happens
after 140 mm indentation.
In numerical analysis, the panel reaches about
1403 and 1566 kN at 168 and 186 mm indentation
in 40 mm and 25 mm element length respectively.
Same tripping occurs at numerical simulation
in about 130 mm indentation. This tripping was
shown in middle length of stiffener in Figure 8(a).
2/18/2011 5:51:17 PM

Figure 7. Force-indentation behavior of panel 1-FB.
Figure 8. Deformation and plastic strain distribution
of 1-FB panel at fracture initiation (element solution).
In Alsos tests fracture of 1-FB panel initiated
from base of stiffener. In this paper, simulations
show that fracture initiate in the base plate next to
the stiffener intersection. Distribution of plastic
strain and location of fracture initiation illustrates
in Figure 8.
In these verification analyses, comparison shows
a good agreement between the experimental results
of Alsos et al. and present numerical analysis.
5 FINITE ELEMENT ANALYSIS
5.1 Stiffened panels for analysis
The geometrical characteristics of the analyzed
stiffened panels are given in Tables 4 and 5. Two
groups of panels have been considered. Group 1
and 2 are panels with Y- and T-stiffeners respec-
tively. Each group has 3 sets of panels. Each set
of panels in two groups has same flexural rigidity.
For example, this means that set 1 of group 1 and
2 has same β and λ. Definition of β and λ are given
442
MARSTRUCT.indb 442
in Equation 4 and 5. Dimensional parameters of
Y-stiffener panels (Cross sectional geometry) are
shown in Figure 9. Also Sample Y-stiffener finite
element model was shown in Figure 10. Boundary
condition of these panels assumed to be simply
support added with fixed rotation around bound-
ary lines like as previous panels. With respect to
the accuracy of verification results in present FEM
analysis only 40 mm element length selected.
b σy
β= (4)
t E
a σy (5)
λ=
πt E
where: b = BT; t = t1; a = L; E = Young modulus;
σy = Yield stress and r = Panel radius of gyration.
5.2 FEM results and discussion
The simulated indentation YB and TB panels are
depicted in Figures 11 and 13. Also deformation
shape and plastic strain distribution of YB-1 and
YB-3 at failure initiation are shown in Figures 12
and for TB-3 is shown in Figure 14.
Deformation and plastic strain distribution in
all sets of two groups are almost similar together.
In YB panels, YB-1 to YB-3 reaches 2142, 2801
and 3247 kN at 208, 212 and 206 mm indentation.
In TB panels, TB-1 to TB-3 reaches 1327, 1258
and 1527 kN at 164, 142 and 146 mm indentation.
From the above results, this can be seen that
failure initiation in YB panels start at larger inden-
tation than TB panels. Also in general YB panels
have very greater contact force than TB panels.
Comparison of YB and TB panels based on con-
tact force illustrate in Figure 15.
Simulations of YB panels are approximately
similar to TB panels. Failure behavior of these 2
types of stiffeners is almost similar together. In
these panels failure initiates at the base plate next
to the stiffeners. These types of failure initiation
occur in all sets of these stiffeners. These failure
initiations illustrate in Figure 12 and 14 for YB
and TB panels respectively.
In TB panels, as the indentation is continued,
the stiffener start tripping about the weld toe.
This tripping occurs very large in TB-1 panel. In
other panels tripping is very low. In TB-1 panel,
large tripping causes to failure initiation in plate
away from connecting line of stiffener and plate
see Figure 16. In other TB panels, low tripping
occurs at strain concentration in plate adjacent to
the stiffener, see Figure 14. This situation in TB-1
toward to others cause that TB-1 panel has larger
contact force than TB-2 panel.
2/18/2011 5:51:20 PM
 Table 4. Geometrical characteristics stiffened panels with Y-stiffener (YB panels)-all dimensions in [mm].

Y stiffener

Plate Hat section Tee section

Model ID L BT t1 B1 B2 H1 t2 H2 B3 t3

YB-1 1200 800 5 120 80 80 5 80 80 5

YB-2 6 160 120 120 6 120 80 6
YB-3 7 200 160 120 7 120 120 7

Table 5. Geometrical characteristics stiffened panels

with T-stiffener (TB panels)-all dimensions in [mm].

Plate Tee section

Model ID L BT t1 Hw Bf t2

TB-1 1200 800 5 137 80 5

TB-2 6 185 120 6
TB-3 7 191 120 7

Figure 11. Force-indentation behavior of YB panel.

Figure 9. Cross-sectional geometry of panels with

Y-stiffener.

Figure 12. Deformation and plastic strain distribution

of YB panels at failure initiation (element solution),
Figure 10. Finite element model of Y-stiffener. a) YB-1 panel upper side, b) YB-3 panel lower side.

443

MARSTRUCT.indb 443 2/18/2011 5:51:23 PM


Figure 13. Force-indentation behavior of TB panel.
Figure 14. Deformation and plastic strain distribution
of TB-3 panel at failure initiation (element solution).
Figure 15. Comparison of YB and TB panels.
In YB panels only strain concentration occurs
at intersection of panel and stiffener at ends of the
panels. In general contact force of panel increased
with increase of bending rigidity of panels except
particular strain concentration.
444
MARSTRUCT.indb 444
Figure 16. Deformation and plastic strain distribution
of TB-1 panel at failure initiation (nodal solution).
Each set of 2 groups has same bending rigid-
ity. Weight ratio of YB to TB panels in set 1 is
1.297, while contact force ratio is 1.614. Same
as this, weight ratio of set 2 and 3 are 1.349 and
1.426 respectively and contact force ration is 2.226
and 2.126. This comparison shows that YB panels
compared to TB panels have greater contact force
to weight.
6 CONCLUSIONS
The purpose of this analysis was to investigate and
understand the behavior and failure mechanisms
of the stiffened steel panels with T- and Y-stiffeners
during grounding. The resistance to penetration of
the hull plating is of special interest. This is a very
complex process and computational calculations
are dependent on mesh size, failure criteria types
of loading and the software that is being used in
the analysis.
To achieve this study, a series of FEM indenta-
tion Analysis have been carried out quasi-statically
on various configurations of stiffened steel panels
with these stiffeners. The analysis method presented
has been implemented into explicit finite element
code LS-DYNA, being linked with ANSYS for pre-
and post-processing of the related computations.
For these analyses plastic kinematic material
property was selected. This material property
is validated with a comparison of experimental
results for stiffened panels. The element-length
dependent constant strain failure criteria and
selected material property proved to be sufficiently
accurate.
Overall the results obtained from the FEA
simulations of penetration are acceptable when
compared to the actual experiments. The grounding
simulation also showed good correlation with pre-
vious published results where the comparisons of
the penetration force gave close correlation.
2/18/2011 5:51:27 PM
 The following related aspects are considered
with analysis of TB and YB panels:
• In general YB toward to TB panels withstand
against larger indentation contact force.
• In general, increase in panel bending rigidity will
result increase contact force.
• YB panels have higher contact force to weight
than TB panels.
• Contact forces to weight of YB to TB panels in
set 1 to 3 are 1.244, 1.65 and 1.49 respectively.
• In YB and TB panels failure initiation and fail-
ure extension are similar.
In the future, this type of stiffener can be used to
find a minimum weight of ship structure like side
structure for a certain collision energy level if the
real scenario and ship types are known.
445
MARSTRUCT.indb 445
REFERENCES
Alsos, H.S. & Amdahl, J. 2009a. On the resistance to pen-
etration of stiffened plates, Part I: Experiments. Int. J.
Impact Engineering 36:799–807.
Alsos, H.S., Amdahl, J. & Hopperstad, O.S. 2009b. On
the resistance to penetration of stiffened plates, Part
II: numerical analysis. Int. J. Impact Engineering
36:875–887.
Badran, S.F., Nassef, A.O. & Metwalli, S. 2007. Stability
of Y-stiffeners in ship plating under uniaxial compres-
sive loads. Ships and Offshore Structures 2(1):87–94.
Ehlers, S. 2009. A procedure to optimize ship side struc-
tures for crashworthiness, J. of engineering for mari-
time environment 224:1–11.
Quinton, B.W.T. 2007. Ship collision mitigation: redesign
of an oil tanker side shell. 8th Canadian marine hydro-
dynamics and structures conference, St. John’s, 16–17
October. NL: Canada.
2/18/2011 5:51:30 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Ultimate load calculation during stranding

Bernadette Zipfel & Eike Lehmann

Hamburg University of Technology, Hamburg

ABSTRACT: If a ship strands it is important to estimate the acting forces and moments in respect to its
ultimate hull girder strength. The influence of the shear force in combination with the bending moment
needs to be taken into account. Therefore a procedure to calculate the moment-shear force interaction
curve of a ship’s cross section is introduced. Three specimens of an experiment (carried out by Reckling
et al.) are recalculated with the FE method to validate the procedure. In the experiment longitudinal
stiffened girders are loaded with pure bending. After the FE results are satisfactory equal to those of the
experiment and the computing time is optimised, the shear force is additionally introduced by different
load cases. Enough nodes for the moment-shear force interaction curve are found. Three normalised
curves are presented- one for each of the three specimens. Various cross sections and lengths of the
specimen result in different curves.

1 INTRODUCTION method is verified and a realistic moment-shear

Grounding accidents can lead to the loss of human
lives, severe environmental consequences and eco-
nomical loss. Therefore it is important to estimate
the additional forces and moments on the ship
structure resulting from a grounding incident. This
paper has a focus on soft grounding. During soft
grounding, the energy dissipation of the structure
is low. The puncture of the outer hull structure,
which is a characteristic of hard grounding, does
not occur. The problem is that stranded ships can
be subjected to the tide. If the water level is reduced,
the hydrostatic forces significantly change. An
additional shear force is introduced into the hull
structures. The consequence can be global dam-
age and the reduction of the ultimate hull girder
strength, see (ISSC 2009). Especially the additional
shear force at the stranding area has a huge influ-
ence on the ultimate hull girder strength. There
are no simple methods to estimate its influence in
combination with the acting bending moments.
The aim of this paper is to find curves that give a
moment-shear force interaction for a cross section
similar to a ship. Further the influence of different
cross sections shall be discussed. The used method
shall be applicable for any cross section, so that
for a given ship a moment-shear force interaction
curve can be found quickly.
A pure bending moment experiment is recalcu-
lated with the finite-element method. The results
of the experiment are known. When the results
of the FE method are adequately similar to
those of the experiment the shear force is intro-
duced to the model. So the procedure in the FE
force interaction curve can be found.
2 THE EXPERIMENT
An experiment carried out and published by
Reckling et al. (1984) is chosen. Reckling tested
seven different box girders with longitudinal stiff-
eners under pure bending load. The results were
not compared to computing methods, but to exist-
ing formulas. The experiment is chosen because
the cross section of the box girder is similar to a
ship (tanker). Furthermore all relevant results were
published and better experiments of longitudinal
stiffened structures are not known.
In the following two subsections a short sum-
mary of the paper is given for knowing the relevant
details for the upcoming FE calculations.
2.1 Specimens and test facilities
Seven specimens with a length between 1210 and
1800 mm and different cross sections were used.
Three of them are later computed. The relevant
geometrical data are given in Table 1 and Figure 1.
Beside the longitudinal stiffeners each specimen
has two transverse stiffeners, which are located
at x = ±250 mm (see Fig. 2). The used coordinate
system is given is Figures 1 and 2. All specimens
collapse at the top plate between the two transverse
stiffeners.
Most specimens are connected via box-shaped
sub-carriers with HEB-400 beams. The HEB beams
(European wide flange beams) are produced in

447

MARSTRUCT.indb 447 2/18/2011 5:51:30 PM

 Table 1. Dimensions of cross section. Table 2. Results of tension test and calculated reference
value.
Nr lk [mm] t [mm] b [mm] h [mm]
Nr σF [N/mm2] E [N/mm2] Me [kNm] fe [mm]
31 1410 2.5 120 133
22 1210 2.5 85.7 100 31 255 2.09 ⋅ 105 231.7 15
23 1410 2.5 85.7 100 22 255 2.09 ⋅ 105 256.6 11.1
23 239 2.09 ⋅ 105 240.5 14.1
Stiffeners [mm]

Top/bottom Side shell

gained by the Moiré-procedure (Hugenschmidt
31 4 L-profiles 30 × 20 × 3 2 I-profiles 30 × 2.5 2007) after the treatment.
22 6 L-profiles 30 × 20 × 3 3 I-profiles 30 × 2.5 To determine the yielding point of each speci-
23 6 L-profiles 30 × 20 × 3 3 I-profiles 30 × 2.5 men, tension tests with sheet metal samples
after DIN standard were carried out. The sheet
metal samples were also treated with stress relief
annealing. The relevant results, which are needed
for the FE calculation, are given in Table 2.

2.2 Analysis of tests

Figure 1. Cross section.
Figure 2. Experimental setup.
accordance with DIN-1025. The subcarriers have
the same cross section as its specific specimen.
Three specimens (among others number 31) are
directly connected with the HEB beam. For the
connection transverse plates are welded on each
end of the specimens, of the subcarriers and of
the HEB beams. The plates are screwed together.
The length between the bearings as well as the
distance between the bearing and the load intro-
duction are always the same. The load is applied
via two hydraulic pillars, which have an increasing
velocity of 0.3 mm/min during the experiment. The
experimental setup is shown in Figure 2.
Except of one pilot test specimen all specimens
are subjected to stress relief heat treatment to
reduce the pre-deformation. The top plate’s resid-
ual pre-deformation of specimen 31, 22 and 23 were
Normalised moment-displacement curves and
the deformation of the top plate are given as test
results. The measured moments and displace-
ments are divided by the elastic reference value.
The elastic limit moment Me and the appendant
displacement fe are calculated for a hypothetical
girder, which has the same cross section as the cor-
responding test specimen and can be loaded by the
plastic limit moment Mp without buckling until
full plasticity is reached. Both calculated reference
value are given in Table 2.
Recalculations by Ansys (2008) and MARC
MENTAT (MSC software 2008) revealed that the
displacement fe is given with a wrong unit in the
paper. In Table 2 the units are correct.
The normalised moment-displacement curves
of all tests have a lesser gradient than the curves
of the equivalent hypothetical girders. The authors
state that the different gradients are due to pre-
deformations and manufacturing-related initial
stresses, that are just partly reduced by the stress
relief heat treatment.
All specimens collapse due to buckling in the
top plate between the transverse structures. The
top plate’s deformation of specimen 31, 22 and 23
are given after the ultimate strength is reached and
again gained by the Moiré-procedure.
Another interesting result of the test series is
that one cannot directly conclude from the pre-
deformation to the appearance of buckles after
reaching the ultimate bending moment.
3 FINITE ELEMENT CALCULATION
Three specimens of the experiment (number 31,
22 and 23) are recalculated with the finite element
method to validate the computational procedure.

448

MARSTRUCT.indb 448 2/18/2011 5:51:30 PM

 The geometry of the specimens and the setup are 3.2 Pure bending
rebuilded. The three specimens of the experiment
All models collapse because of buckling in the top
and their results are called in the following ‘speci-
plate. The resulting deformation and the ultimate
men 31, 22 or 23’ whereas the FE results are called
strength depend on the material model, the ele-
‘model 31, 22 or 23’.
ment size, the timestep size, the pre-deformation
First the finite element models and its mate-
and the initial stress.
rial model that are used for all FE calculations are
The material model is given by the experiment.
introduced. Then the results of the FE recalcula-
The element size is optimised in respect of the com-
tion of the above described experiment are given.
puting time and the results. The element size in the
The chapter is called pure bending, since this is the
middle part of the model (−250 mm < x < 250 mm)
used load case. At last the models are loaded with
is 10 mm. 20 mm are chosen for the elements from
different load cases in which both—moment and
x = ±250 mm until x = ±750 mm. The remaining
shear force—are acting, to gain the interaction of
elements have a size of 30 mm. The element size is
moment and shear force.
small enough to capture the pre-deformation and
the buckling shape in the top plate.
3.1 Generals The pre-deformation and the timestep size are
varied for model 31 to find the sensitivity of the
For all FE calculations the program LS-DYNA
results and to save modelling and computational
(Livermore Software Technology Corporation
time. These calculations reveal that it is also needed
2009) is used.
to introduce initial stress to get satisfactory agree-
ment with the experimental results.
3.1.1 Models
Three different FE models are calculated. The 3.2.1 Model 31
geometry of the models is given in chapter 2.1. The top plate is divided into five areas to reach
All models are build with shell-elements (Liver- more clearness in the upcoming discussion, see
more Software Technology Corporation 2009). Figure 4.
The boundary conditions are applied at the
nodes of the lower flange of the HEB structure. 3.2.2 Pre-deformation
In each calculation time of 0.1 s a displacement of Model 31 is calculated with three different pre-
0.3 mm is brought into the model. The total dis- deformation conditions. First no deformation
placement is 30 mm at each side. is applied. Then a circular pre-deformation is
brought into the model. At last the original
pre-deformation given in the paper (Reckling et al.
3.1.2 Material model
1984) is computed, see Figure 5.
The material model is nonlinear. Strain harden-
In all three cases the buckling picture after
ing is taken into account. A stress-stain curve
reaching the ultimate strength as well as the ulti-
that is known from experiments performed by the
mate strength itself differ. The following Figures 6,
Institute of Ship Structural Design and Analysis
7 and 8 show the displacement in z-direction of
located at Hamburg University of Technology is
the top plate exactly at the timestep where the
scaled for the given yield stress (see Table 2). The
ultimate strength is reached (see first pictures
following graph 3 shows the stress-strain curve for
of Figs. 6, 7 and 8) and shortly after (see second
model 31 and 22.
picture). The colour scale is always fitted to the

800

Model 31, 22
700

600
Stress [N/mm2]

500

400

300

200

100

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Strain

Figure 3. Stress-stain curve of model 31 and 22. Figure 4. Areas on top plate.

449

MARSTRUCT.indb 449 2/18/2011 5:51:31 PM

 Figure 5. Pre-deformation of specimen 31 (Reckling
et al. 1984).

Figure 7. Model 31 with circular pre-deformation:

Z-displacement of top plate at and after ultimate bending
moment.

top plate. The shown maximum/minimum dis-

Figure 6. Model 31 with no Pre-deformation:
Z-displacement of top plate at and after ultimate bending
moment.
placement includes the deflection of the complete
system in negative z-direction.
If no pre-deformation is applied, one buckle
occurs in negative z-direction in the middle of
the top plate (area 3). Its deepest point with
20 mm of displacement is right in the middle
of the plate between the longitudinal stiffeners.
After reaching the ultimate bending moment
additional buckles establish between the remain-
ing stiffeners. Whereas area 2 and 4 buckle in
positive, the outer areas (1 and 5) deform in nega-
tive z-direction. Between the two pictures the
maximal displacement differ 13 mm measured at
the same node in area 3.

450

MARSTRUCT.indb 450 2/18/2011 5:51:32 PM

 buckles distinguish a little bit cause by the different
geometry of the pre-deformation. The maximal
buckle still occurs in area 3 (24 mm) and only area
1 buckles in positive z-direction.
Figure 9 plots the ultimate bending moment M
at the position x = 0 over the displacement f that is
measured at x = 0; y = 0. All data of the upcoming
moment-displacement curves are taken form the
same reference cross-section respectively point.
The gradient of all curves is the same until the
ultimate bending moment is reached. The gradient
of the moment-displacement curve of specimen
31 is less (see also Fig. 11). The highest moment
(2.46 E + 08 Nmm) out of the three variations
occurs when no pre-deformation is applied onto
the model.
The calculation with the circular pre-deformation
reaches a maximal moment of 2.30 E + 08 Nmm.
The difference between the moments of the cir-
cular and the original (2.35 E + 08 Nmm) pre-
deformation is 2.1%.
All curves have a jump, because the buckles
deepen suddenly. The moment-displacement curve
of the model with no pre-deformation reaches
the highest ultimate bending moment and mostly

2.5E+08
No Pre-Deformation
2.4E+08
Cicular Pre-Deformation

2.3E+08 Original Pre-Deformation

2.2E+08

2.1E+08
M [Nmm]

2.0E+08

1.9E+08

1.8E+08

1.7E+08

1.6E+08

1.5E+08
9.0 11.0 13.0 15.0 17.0 19.0 21.0 23.0
f [mm]
Figure 8. Model 31 with Original Pre-deformation:
Z-displacement of top plate at and after ultimate bend-
ing moment. Figure 9. Ultimate bending moment: No, circular and
original pre-deformation.

2.4E+08
For a circular pre-deformation buckling is Normal Timestep

clearly seen in area 3 and 4 when the ultimate bend- 2.3E+08

Fine Timestep

ing moment is reached. The maximal displacement

occurs in area 3 with 17 mm. Shortly after, one 2.2E+08

buckle between each stiffener develops. They are at

M [Nmm]

the edge of the circular pre-deformation. Now the 2.1E+08

deepest buckle is in area 4 with a displacement of

2.0E+08
24 mm. The outer areas (1 and 5) deform in positive
z-direction. The buckling picture of the original 1.9E+08
model shows buckles in area 2 until 5 on both sides
of the pre-deformation. With 16 mm the maximal 1.8E+08

displacement can be found in the middle (area 3). 11.0 13.0 15.0 17.0
f [mm]
19.0 21.0 23.0

Some timesteps after the ultimate bending moment

is reached the deformation is similar to this with Figure 10. Ultimate bending moment: Influence of
the circular pre-deformation. The position of the timestep size.

451

MARSTRUCT.indb 451 2/18/2011 5:51:34 PM

 2.5E+08
2.0E+08
1.5E+08
M [Nmm]
1.0E+08
5.0E+07
0.0E+00
0.0 5.0 10.0
f [mm]
Figure 11. Ultimate bending moments of model 31.
decreases. Its largest buckle deepens 13 mm from
timestep 0.404 s to 0.424 s, which is also the
largest change in the displacement of the three
calculations.
The maximum moment of the model with the
circular pre-deformation is closer to the moment
of specimen 31. But the collapse mode of the
model with the original pre-deformation is more
similar to the experiment, although the buck-
les between the stiffeners develop on the left
side of the pre-deformation in the experiment
(see Reckling et al. 1984).
For the following calculations the original pre-
deformation is taken.
3.2.3 Timestep size
In the moment-displacement curve a sudden decline
is seen and buckling in the top plate occurs from
one timestep to the next. To reduce this phenom-
enon the timestep is refined around the ultimate
bending moment capacity of the model. Between
calculation time 0.3 s and 0.45 s the timestep size is
set from 5.0 E − 03 to 5.0 E − 05. The calculation
needs eight hours instead of one.
The maximum moment, that are reached in
Figure 10, are almost the same. The decline of
the curve is smoother for finer timestepping, but
the sudden decline cannot be reduced. The buck-
ling picture are for both cases the same. So the
results are similar, but the computational time is
not acceptable. Thus all following simulations are
calculated with the ‘normal’ timestep size.
3.2.4 Initial stress
The gradient of the moment-displacement curves
in Figure 9 is identical to the gradient of the ideal
case. In the ideal case the material model is ideal
elastic and no pre-deformation is applied. The
moment-displacement curve in the ideal case is
similar to the curve of the hypothetical girder until
the girder starts to plasticise. In the FE method
a hypothetical girder which can be loaded by the
MARSTRUCT.indb 452
plastic limit moment without buckling until full
plasticity is reached, cannot be realized.
As stated by Reckling et al. the manufacturing-
related initial stresses are just partly reduced by the
stress relief heat treatment so that a smaller gradi-
Ideal 31
Specimen 31
ent is seen in the moment-displacement curve. To
FEM 31
gain satisfactory results in the FE simulation it is
Initial Stress 31 necessary to include the initial stress.
To demonstrate the effect of initial stress, the
model with the original pre-deformation is taken.
The stiffeners of the top plate are loaded with a
15.0 20.0 25.0
temperature of 600°C. The model is solved until the
temperature is reached and not cooled down. The
procedure introduces initial stress in x-direction
close to the yield stress (see also Lehmann 2004).
After the temperature solution the displacement is
applied onto the model as described above.
In Figure 11 four moment-displacement curves
are plotted. The curve ‘Ideal 31’ is the ideal case,
‘Specimen 31’ shows the result of the experiment,
‘FEM 31’ is the same as the curve with the title
‘Original Pre-Deformation’ given in Figure 9 and
the curve ‘Inital Stress 31’ belongs to the model
which is loaded with initial stress.
The gradient of the model ‘Inital Stress 31’
is very close to the gradient of ‘Specimen 31’.
The total course of the moments is not identical
but the maximum moment is exactly the same
(231.7 kNm), whereas the moment of the calcu-
lation without initial stress has as a deviation of
1.38%. The sudden decrease of the curve from the
FE calculation is still existent.
The buckling picture shows a good conformance
with this of the experiment. The particular buckles
are as well on the left side of the pre-deformation
(see Fig. 12) and the deepest deformation occurs in
area 3. The pre-deformation is still seen after the
ultimate strength is reached.
The results of the FE simulations are satisfac-
tory equal to those of the experiment. So the pro-
cedure in the FE method is validated by reality and
further calculations with the model will give rea-
sonable results. The influence of the initial stress
on the moment-shear force interaction can be
neglected. There is no difference for the relation of
moment and shear force if the simulation is with
or without initial stress. To save computational
time the initial stress is not applied in the following
calculations.
3.2.5 Model 22 and 23
Model 22 and 23 are calculated with the original
pre-deformation, the ‘normal’ timestep size and
without initial stress. The moment-displacement
curves of all three models are given in Figure 13.
Model 31 is shown in Figure 13 to see the influ-
ence of the different cross sections and lengths of
the specimen. The difference in the three graphs is
452
2/18/2011 5:51:36 PM
 the same if the models are computed with initial
stresses. The Figure is given because this is one of
the upcoming load cases for the moment-shear
force interaction curves. As said above the initial
stress has no influence on the interaction.
The decrease of the moment curves are much
smoother for model 22 and 23 as for model 31.
The collapse of the top plate does not occur so
suddenly which can also be seen in the buckling
pictures. Model 22 and 23 are stiffer than model
31 and therefore the total collapse takes more time.
But in contrast to the experiments both curves
decrease. The maximum bending moments differ
2–3% from the experiments.
In total, the results of model 22 and 32 are
as good as for model 31 (without initial stress)
compared to the experiments. When initial stress
is introduced to model 22 and 23 the results will be
as similar to the experiment as it is for model 31.
Consequently model 22 and 23 give just as good
results as model 31 for the moment-shear force
interaction.

3.3 Bending moment and shear force

The models are computed with the original
pre-deformation, a timestep of 5.0 E − 3 and with-
out initial stress, since the maximal moment/shear
force is the result needed for the moment-shear
force interaction.
To introduce the shear force to the models two
additional load cases are used. The pure bending
load case of the experiment is called ‘case 1’. In
‘case 2’ only one force is acting, whereas in ‘case 3’
the second force operates in the opposite direction.
Figure 14 clarifies the relation between the bend-
ing moment and the shear force.
The second force (on the right hand side) is var-
ied in case 1 and 3 to gain sampling points for the
Figure 12. Model 31 with initial stress: Z-displacement. moment-shear force interaction curve. Table 3 gives
an overview of all calculated cases.
3.0E+08
3.3.1 Model 31
Figure 15 gives the courses of the moments for each
load case over the displacement f ′ which is applied
2.5E+08

to the model. The associated shear forces are plotted

2.0E+08
in Figure 16. For more clearness the mechanical
M [Nmm]

1.5E+08 FEM 31
Specimen 31
FEM 22
1.0E+08 Specimen 22
FEM 23
Specimen 23

5.0E+07

0.0E+00
0.0 5.0 10.0 15.0 20.0 25.0 30.0
f [mm]

Figure 13. Ultimate bending moments of model 31, 22

and 23. Figure 14. Load cases 1–3.

453

MARSTRUCT.indb 453 2/18/2011 5:51:36 PM

 Table 3. Load cases. The maximum bending moments respectively
shear forces are nominated by their plastic limit
Case Force P value. The plastic limit moment is given by:
1 −1 ⋅ P
Mpl = σF ⋅ Wpl (1)
1aa −5/6 ⋅ P
1a −1/2 ⋅ P
where as σF is the yield stress and Wpl is the plastic
1b −1/6 ⋅ P
section modulus in Equation 1. The plastic shear
2 0⋅P stress can be calculated after Reckling (1967) or
3 1⋅P Pedersen (1990):
3a 1/2 ⋅ P
3b 1/30 ⋅ P σF
Tpl = (2)
3
2.5E+08
Case 1
Case 1aa The plastic shear stress Tpl is multiplied by the
Case 1a
2.0E+08 Case 1b area of the side shell to get the plastic limit shear
Case 2
Case 3b
force Qpl. The side shell carries the shear force.
1.5E+08
Case 3a
Case 3
The following Figure 17 gives the nominated
moment-shear force interaction curve for the dif-
M [Nmm]

ferent load cases.

1.0E+08

3.3.2 Model 22 and 23

5.0E+07
For model 22 and 23 only the moment-shear force
interaction curves are given. In Figure 18 the
0.0E+00
0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 curves for all three models are plotted.
f' [mm]

Figure 15. Model 31: Moments of load cases 1–3. 1.0

Case 1

0.9 Case 1aa

Case 1a
0.8
Case 1b
3.0E+05
Case 1 0.7 Case 2
Case 1aa Case 3b
Case 1a 0.6
Case 3a
2.5E+05 Case 1b
M/Mpl

0.5 Case 3
Case 2
Case 3b
0.4
2.0E+05 Case 3a
Case 3 0.3
Q [N]

1.5E+05 0.2

0.1

1.0E+05 0.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
Q/Qpl
5.0E+04

Figure 17. Model 31: Moment-shear force interaction

0.0E+00
0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 curve.
f' [mm]

Figure 16. Model 31: Shear forces of load cases 1–3. 1.0
Model 31

0.9 Model 22

Model 23
0.8
systems are additionally given in Figure 15. The
0.7
cases reverses from the moment-displacement
curves to the shear force- displacement curves, e.g. 0.6
M/Mpl

the case with the highest bending moment (case 1) 0.5

has a minimal shear force. Of course, the shear 0.4

force is not zero in case 1 and the bending moment 0.3

is not zero in case 3. 0.2

At the same displacement where the cases 1, 0.1

1aa, 1a, 1b and 2 reach the maximum moment the 0.0

equivalent shear force is read out for the moment- 0 0.1 0.2 0.3 0.4 0.5 0.6
Q/Qpl
0.7 0.8 0.9 1 1.1 1.2

shear force interaction curve. In the cases 3, 3a, 3b

and 3c the maximum shear force and its associated Figure 18. Model 31: Moment-shear force interaction
bending moment is taken. curve.

454

MARSTRUCT.indb 454 2/18/2011 5:51:39 PM

 Model 22 has the highest moment-shear force
interaction curve, whereas model 31 produces the
lowest curve. The curve of model 23 lies again in
between the two other models as expected (com-
pare Fig. 13). The course of curve 22 and 23 are
similar. The different length also influences the
curves. Model 31 with its different cross section
produces a different graph. Non of the models
reaches the plastic limit moment for load case 1.
The ultimate capacity of the girder is not equal to
the full plastic capacity due to the pre-deformation
and the post buckling behaviour. In addition the
shear force, which is small but not zero, reduces the
ultimate capacity of the girder.
For all models Q/Qpl is greater than one for
load case 3. Apparently the shear force is not only
carried by the side shell.
4 CONCLUSIONS
The usual influence of mesh size and material
model on the results are existent in the described
FE-calculations. The material model was given by
the experiment and the mesh size is optimized.
If the pre-deformation is modelled as exactly as
possible the resulting buckling mode is not auto-
matically the same as in the experiment. Neither
for the FE-calculation nor for the experiment (see
conclusion in Reckling et al. 1984) it is possible to
predict the collapse mode of the girder due to the
known pre-deformation.
Varying the timestep size does not eliminate the
jump in the moment-displacement curve which is
not seen in the experiment. The introduction of
initial stress results in an identical maximum bend-
ing moment and a similar gradient of the moment-
displacement curve. The results of the calculation
with the original pre-deformation and initial stress
are satisfactory equal to those of the experiment.
Therefore the used models are validated and the
further calculation with the additional shear force
give reasonable results.
Various load cases give the supporting points
for the moment-shear force interaction curve. For
each model a curve is gained.
Not only different cross sections lead to a dif-
ferent moment-shear force interaction but also
different length of the models. If more stiffeners
in the side, bottom and top plate are used, higher
nominated forces and moments can be carried by
the structure.
The described procedure to gain the moment-
shear force interaction curve for a cross section is
455
MARSTRUCT.indb 455
simple and was verified by reality. For any cross
section of a ship a curve can be calculated by the
presented procedure. So the influence of the shear
force on the ultimate hull girder strength during
stranding can be evaluated.
5 RECOMMENDATION FOR FURTHER
WORK
The scaling of the moment-shear force interaction
curve should be reviewed. Numerical results are
normalised by analytical formulas.
A moment-shear force interaction curve for a
section of a vessel needs to be gained by the above
described procedure.
Than the sections will be completely mod-
elled in the finite-element method and a strand-
ing scenario will be simulated. The results of the
stranding simulation will be compared to the
moment-shear force interaction curve to verify
the simulation.
ACKNOWLEDGEMENTS
The work was performed within the Research
Training Group “Ports for Container Ships
of Future Generations” located at Hamburg
University of Technology. The project is financed
by the German Research Foundation (Deutsche
Forschungsgemein-schaft, DFG).
REFERENCES
Ansys (2008). Release 11.0 Documentation for ANSYS.
Hugenschmidt, M. (2007). Lasermesstechnik, Di-agnostik
der Kurzzeitphysik. Springer Verlag.
ISSC (2009). Committee V.1: Damage Assessment After
Accidental Events.
Lehmann, E. (2004). Grundzüge des Schiffbaus.
Elbewerkstätten.
Livermore Software Technology Corporation (2009).
LS-DYNA Keyword User’s Manual Volume I (Version
971/Release 4 Beta).
MSC software (2008). Marc 2008 r1: Volume A—E.
Petersen, C. (1990). Stahlbau. Vieweg.
Reckling, K.-A. (1967). Plastizitätstheorie und ihre
Anwendung auf Festigkeitsprobleme. Springer-Verlag.
Reckling, K.-A., Dannenberg, L. & Immenkötter, K.
(1984). Die experimentelle Ermittlung der Tragfähig-
keit längsversteifter Kastenträger unter reiner
Biegebeanspruchung. Schiff und Hafen; Heft 3 und 4.
2/18/2011 5:51:41 PM
 Dynamic analysis

MARSTRUCT.indb 457 2/18/2011 5:51:41 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Methods for hull structure strength analysis and ships service

life evaluation, under extreme hydroelastic wave loads,
for a large oil-tanker

L. Domnisoru & A. Chirica

University “Dunarea de Jos” of Galati, Naval Architecture Faculty, Galati, Romania

ABSTRACT: This paper focuses on the numerical methods for the ship hull structure strengths and
fatigue analyses, in order to estimate the initial design ship service life period. The applied methods have
been improved in the Marstruct Project. The numerical analyses are divided in three-interlinked parts.
The first part includes the hull strength analysis method, based on 3D/1D-FEM models, under equivalent
quasi-static head wave loads. The second part presents the ship hull dynamic response analysis method,
based on non-linear hydroelasticity theory with second order wave spectrum. The third part includes the
fatigue analysis method for the initial ship hull structure, based on the long-term prediction ship dynamic
response, the cumulative damage ratio and the design S-N material curves. The numerical analyses are
carried out for a large double hull oil-tanker with 275 m length between perpendiculars. The full and bal-
last loading cases are analysed. The numerical results outline the extreme hydroelastic wave loads and the
ships initial service life evaluation.

1 INTRODUCTION The standard seakeeping analyses of the wave

The present day ship design rules require the
assessment of the ship hull strength and the evalu-
ation of the ship service life starting from the early
design steps, based on the initial ship hull structure
concept.
This study is focused on the link between: the
ship strength analysis with the finite element
method, the waves induced ship hydroelastic
response analysis and the ship structure fatigue
analysis, based on the methods developed and
improved in the frame of the Marstruct Project.
In this study, the numerical analyses are carried
out on a double hull large oil-tanker (IACS 2008,
GL 2008), with elastic girder, considering the full
and ballast conditions, under extreme head irregu-
lar wave loads. The large oil-tanker initial design
characteristics are presented in Section 2.
The nowadays ship hull structure strength rules
require to develop three-dimensional (3D) models,
based on the FEM Finite Element Method (IACS
2008, Jang & Hong 2009). Instead of develop-
ing structural models extended only on several
cargo-holds (Hughes 1988, Mansour & Liu, 2008),
the 3D-FEM full ship length models (Rozbicki,
Purnendu & Crow 2001, Domnisoru 2006,
2008) are to be developed, in order to increase
the ship structure strength assessment accuracy.
In Section 3 is presented the oil-tanker hull struc-
ture 3D/1D-FEM strength analysis.
induced ship dynamic response include only the
ship rigid hull oscillations (Bhattacharyya 1978,
Bertram 2000). For large elastic ships, the wave
induced ship dynamic response includes oscil-
lations (low frequency) and vibration (high
frequency) components (Bishop & Price 1979).
In Section 4 the large oil-tanker wave induced ship
hull dynamic response is obtained, in the hypoth-
eses of the hydroelasticity theory, including the
following components: the linear and non-linear
oscillation response, the springing phenomenon,
linear and non-linear steady state vibration
response, due to the ship structure-wave reso-
nance, and the whipping phenomenon, bottom
and side slamming induced transitory vibration
response (Guedes Soares 1999, Hirdaris, Price &
Temarel 2003, Hirdaris & Ge 2005, Domnisoru
1998, 2008, Perunovic & Jensen 2005, Park &
Temarel 2007).
In order to evaluate the large oil-tanker initial
design service life, this study includes in Section 5
the ship hull structure fatigue strength analysis,
based on the Palmgren-Miner cumulative damage
ratio method (GL 2004, 2008, IACS 2008, Fricke &
Kahl 2005, Garbatov et al. 2005, Mansour & Liu
2008), with steel standard S-N fatigue design curves
and the World Wide Trade wave significant height
histogram. The large oil-tanker structural fatigue
analysis is based on the ship hull stresses induced by
the extreme wave loads, computed in Sections 3–4,

459

MARSTRUCT.indb 459 2/18/2011 5:51:41 PM

 under full cargo and ballast conditions with the Iy Syn− n
same probability in the travel scenario. WB ,D = ; K τn nS
S = ⋅ αS (1)
eB ,D I y tS
The numerical analyses are based on eigen pro-
gram codes and Cosmos/M (SWCM 2007) FEM where: eD, eB are the distances of the neutral axis to
program. the deck and bottom shells, Syn−n is the transversal
section static moment at the neutral axis; tS is the
thickness of the side shell, αS is the shearing stress
2 THE SHIP 1D-BEAM AND 3D-FEM coefficient of the side shell at neutral axis, based
MODELS FOR THE NUMERICAL on the multicell tanker amidships section shearing
ANALYSES stresses distribution.
Based on the Germanischer Lloyd’s Rules (GL
In this study is considered a large oil-tanker TK 2008), Table 3 presents the material isotropic steel
167000 tdw, with the main dimensions in Table 1 characteristics and the admissible stresses. The
and the ship offset section lines in Figure 1. structural damping coefficient β for welded ship
The double hull oil-tanker structure is designed structures is selected according to Johnson &
according to the Germanischer Lloyd Hull Struc- Tamita (Bishop & Price 1979).
ture Rules (GL 2008) and the Common Structural Table 4 presents the full cargo and ballast main
Rules for Double Hull Oil Tankers (IACS 2008). loading cases, according to the Ship Classification
Table 2 presents the amidships structure char- Society Rules for the double hull oil-tankers (IACS
acteristics: the moment of inertia Iy [m4], the total 2008). Figures 2a, b present the large oil-tanker
transversal section area A [m2], the equivalent mass diagrams for the two main loading cases.
shearing area Afz [m2], the neutral axis coefficient Table 5 presents the natural hull oscillation and
of maximum shearing tangential stress in the side vibration mode frequencies, dry hull or with hydro-
shell Kτ n−nS [m−2], the deck and bottom bending dynamic masses (wet hull), based on 1D-FEM
modules WD, WB [m3]. model of the large oil-tanker TK 167000 tdw.
Figure 3 presents the modal form functions (rigid
and elastic) for the full cargo loading case (dry hull
Table 1. The large oil-tanker TK 167000 tdw main at vibration modes).
dimensions. The 1D-FEM equivalent beam ship girder
model (Domnisoru 1998, 2006) of the large oil-
LOA [m] 285.00 dMLD [m] 17.622
tanker hull TK 167000 tdw is developed for simpli-
LBP [m] 275.00 cB 0.834
BMLD [m] 46.00 Vcargo [m3] 188500
fied strength analysis under equivalent quasi-static
DMLD [m] 23.80 Dw [t] 167000
wave loads (Section 3) and ship dynamic response
analysis under irregular wave loads (Section 4),

Table 3. The material isotropic steel characteristics.

Table 2. TK 167000 tdw amidships section characteristics.
2
E [N/m ] 2.1 1011 ρm[t/m3] 7.7
Iy [m4] 838.283 eB [m] 11.912 G [N/m2] 8.1 1010 β 0.001
A [m2] 9.432 eD [m] 13.088
Afz [m2] 4.822 WB [m3] 70.375 ReH-A [N/mm2] 235 ReH-AH32 [N/mm2] 315
Kτn−n S [m−2] 0.429 WD [m3] 64.048 Rm-A [N/mm2] 400 Rm-AH32 [N/mm2] 440
σadm-A [N/mm2] 175 σadm-AH32 [N/mm2] 224
τadm-A [N/mm2] 110 τadm-AH32 [N/mm2] 141

Table 4. TK 167000 tdw large oil-tanker loading

cases*.

Load Δ dm daft dfore vs

No case [t] [m] [m] [m] [Knots]

1 Full 191680 17.622 17.742 17.492 15

cargo
2 Ballast 146570 13.711 15.677 11.889 16

*where Δ is the displacement; dm,daft,dfore are the ship

Figure 1. The large oil-tanker TK 167000 tdw offset draughts: medium, aft peak, fore peak; vs is the ship
section lines. design speed.

460

MARSTRUCT.indb 460 2/18/2011 5:51:41 PM

 1400
mx [t/m] TK Oil-Tanker Ship Full Cargo : Mass Diagram
based on the hydroelasticity theory. The equiva-
1260 lent ship girder 1D-FEM has 40 equal length beam
1120 elements along the ship, so that from the original
980 ship offset lines (Figure 1) are selected only the
840 transversal sections disposed at the middle of each
700 element. The diagrams of the transversal section
560 characteristics (Table 2) are idealised with trap-
420 ezoidal distributions along the ship.
280 The 3D-CAD/FEM model (Domnisoru 2006,
140 x[m] 2008) of the large oil-tanker TK 167000 tdw
0
3 12 18 29 37 46 55 66 73 94 104 122 143 160 170 192 214 223 238 248 263 273 285
is developed using the SolidWorks Cosmos/M
(SWCM 2007) FEM program, for the enhanced
Figure 2a. The oil-tanker mass diagram—full cargo ship hull strength analysis (Section 3). Table 6
load case 1. presents the main characteristics of the 3D-CAD/
FEM model. The 3D-FEM model (including stiff-
eners) is based on auto-mesh procedure with thick
1400
mx[t/m] TK Oil-Tanker Ship Ballast : Mass Diagram
triangular shell elements (SHELL3T), ensuring
1260
the mesh convergence also on surfaces with high
1120
curvature. Figures 4a, b, c, d present the 3D-FEM
980
model, full length extended. In Section 3 the
840
3D-FEM model boundary conditions and loading
700
cases are presented.
560
Sections 3–5 include the numerical analyses
420 results.
280
x[m]

140

0 Table 6. The large oil-tanker 3D-CAD/FEM model

1 8 19 31 39 51 71 93 110 123 143 149 171 192 214 224 239 248 268 277 285
characteristics.
Figure 2b. The oil-tanker mass diagram—ballast load Number of points PTmax 16310
case 2. Number of curves CRmax 40802
Number of surfaces SFmax 18598
Number of nodes NDmax 31977
Table 5. TK 167000 tdw oil-tanker natural modes fre-
Number of SHELL3T ELmax 104275
quencies f [Hz].
elements
Modes Oscillations Vibrations Number of element groups EG 275
Average/maximum average/max 0.9/1.8
No. Load case 0 1 2 3 4 EL size [m]
Support condition NDaft 5990
1 Full dry – – 0.866 1.904 3.434 aft-peak node
cargo wet 0.088 0.097 0.629 1.478 2.536 Support condition NDfore 4214
2 Ballast dry – – 0.897 2.085 3.882 fore-peak node
wet 0.093 0.100 0.645 1.505 2.750

TK Oil-Tanker Ship Full Cargo : Natural Modes

1.00

0.80

0.60

0.40

0.20

0.00

-0.20

-0.40

-0.60

-0.80

-1.00
0.00 28.50 57.00 85.50 114.00 142.50 171.00 199.50 228.00 256.50 285.00

Mode 0 / 0.088Hz Mode 1 / 0.097 Hz Mode 2 / 0.629 Hz Mode 3 / 1.478 Hz Mode 4 / 2.536 Hz x [m]

Figure 3. The oil-tanker natural modes—full cargo

load case. Figure 4a. The large oil-tanker 3D-FEM hull model.

461

MARSTRUCT.indb 461 2/18/2011 5:51:43 PM


Figure 4b. The large oil-tanker 3D-FEM model, aft-peak
part.
Figure 4c. The large oil-tanker 3D-FEM model, cargo-
holds (view with the centre line longitudinal bulkhead
removed).
Figure 4d. The large oil-tanker 3D-FEM model, fore-
peak part.
3 THE SHIP STRENGTH ANALYSIS
UNDER EQUIVALENT QUASI-STATIC
WAVE LOADS
In this section, the large oil-tanker TK 167000 tdw
initial design hull structure strength is analysed,
mainly based on the 3D-FEM model extended on
462
MARSTRUCT.indb 462
the whole ship length (Figs. 4a, b, c, d), under the
equivalent quasi-static head wave loads.
The main steps of the ship strength analysis,
based on 3D-FEM model extended on the whole
ship length (Domnisoru 2006), are the following:
− the 3D-CAD ship hull offset lines generation;
− the 3D-CAD ship hull structure model;
− the 3D-FEM ship hull structure mesh model;
− the boundary conditions on the 3D-FEM model;
Two types of boundary conditions are used: the
symmetry condition at the nodes placed in the cen-
tre plane of the ship (one side model); the vertical
support condition at two nodes placed at the ship
hull ends (in the centre plane), noted NDaft, NDfore at
aft and fore peak. For the vertical equilibrium con-
ditions, in still water or equivalent quasi-static head
wave cases, the two vertical supports reaction forces
RFZ(NDaft), RFZ(NDfore) have to become zero.
− the loading conditions on the 3D-FEM model;
Three types of loads acting on the ship hull
structure are used: the gravity load from the hull
structure weight and other mass components of
the displacement; the oil-cargo or water ballast
modelled as pressure load over the cargo-tanks or
ballast-tanks inner shell; the equivalent quasi-static
head wave pressure load, acting on external ship
hull shell, for the following cases: hw = 0 (still water)
and hw ≠ 0 (wave), according to the statistical val-
ues from Ship Classification Society Rules (IACS
2008). An iterative procedure for the free floating
and trim condition equilibrium is applied, imple-
mented with eigen GEO macro-commands file
(Domnisoru 2006), in the SolidWorks Cosmos/M
(SWCM 2007) FEM program, used as linear-
elastic structural solver for each iteration.
The iterative procedure includes two main
parts:
a. the free floating condition, having as objec-
tive numerical function the sum of vertical
reaction forces at the two nodes from the ship
extremities:
RFZeq = RFZ(NDaft) + RFZ(NDfore) → 0 (2a)
b. the free trim and floating condition, having as
objective numerical functions the vertical reac-
tion forces at each two nodes from the ship
extremities:
RFZ(NDaft) → 0 RFZ(NDfore) → 0 (2b)
− the numerical results evaluation.
The following numerical results are obtained from
the 3D-FEM ship hull strength analysis: the free
floating and trim equilibrium parameters, the ship
2/18/2011 5:51:45 PM
 hull deformations, the ship hull stress distributions The 3D-FEM ship hull structure strength anal-
and also the prediction of the higher risk structural ysis is compared to the classical ship equivalent
domains. 1D-girder strength analysis (Table 8), based on
For the maximum 3D-FEM equivalent von Mises eigen program code (Domnisoru 2006).
stress values in deck and bottom shells, taking into Section 2 presents the numerical data of the
account the hot spot domains at the intersection with 3D-FEM and 1D-girder oil-tanker models.
transversal bulkheads, frames and also side, double- Table 4 and Figures 2a, b present the two
side and central line bulkhead panels (Figs. 9a, b), loading cases, full cargo and ballast, for the large
the influence coefficients are calculated: oil-tanker.
The external quasi-static head wave pressure,
K3DvonM|D,B = σvonM3DFEMmax|D,B/M3DFEM (3) with height hw = 0–12 m, step δhw = 1 m is applied
on the 3D-FEM hull model, using an iterative
Table 7. The large oil-tanker TK 167000 tdw buckling procedure for the vertical in plane equilibrium
factor and the maximum vertical deflection |wz| [m], refer- condition. The sagging and hogging head
ence wave hw = 10.692 m. wave conditions are considered. Based on the
Germanischer Lloyd’s Rules (GL 2008), the Ship
Full cargo Ballast Load case Classification Society equivalent quasi-static
statistical wave height for the TK 167000 tdw
Sagging Hogging Sagging Hogging Criterion
oil-tanker is hw = 10.692 m.
1.348 1.371 1.758 1.763 BBuckling > 1 For the two loading cases (Table 4), the
0.372 0.170 0.212 0.243 max ⋅ wz[m] numerical results of the ship strength analyses
0.64 0.29 0.37 0.42 wzmax/wadm < 1 are synthesized in Tables 7, 8 and Figures 5–9
(see Conclusions 1–3).

Table 8. The maximum TK 167000 tdw oil-tanker stresses, for a) the full cargo load case and b) the ballast load
case, under sagging and hogging quasi-static equivalent head wave conditions, with the reference wave height
hw = 10.692 m.

Sagging Hogging

Load case 1D-girder 3D-FEM 3D/1D 1D-girder 3D-FEM 3D/1D

(1) Deck normal stresses |σxD-max| [N/mm2]: σadm-AH232 = 224

a) full: 167.75 194.25 1.16 68.85 79.71 1.16
σmax/σadm 0.748 0.866 0.307 0.355
b) ballast: 119.02 134.97 1.13 111.10 138.38 1.25
σmax/σadm 0.530 0.602 0.495 0.617
(2) Bottom normal stresses |σxB-max| [N/mm2]: σadm-A = 175
a) full: 151.46 160.77 1.06 62.16 81.34 1.31
σmax/σadm 0.865 0.919 0.355 0.465
b) ballast: 107.45 155.80 1.45 100.31 167.20 1.67
σmax/σadm 0.614 0.890 0.573 0.955
(3) Side shell tangential stresses, in the neutral axis of the ship transversal sections |τxzSnn-max| [N/mm2]: τadm-A = 110
a) full: 56.31 64.68 1.15 27.88 32.97 1.18
τmax/τadm 0.512 0.588 0.253 0.300
b) ballast: 71.67 78.85 1.10 43.75 46.99 1.07
τmax/τadm 0.652 0.717 0.398 0.427
(4) Deck von Mises stresses |σvonD-max| [N/mm2]: σadm-AH32 = 224
a) full: 167.75 184.91 1.10 68.85 77.30 1.12
σmax/σadm 0.748 0.824 0.307 0.345
b) ballast: 119.02 124.74 1.05 111.10 132.00 1.19
σmax/σadm 0.530 0.556 0.495 0.588
(5) Bottom von Mises stresses |σvonB-max| [N/mm2]: σadm-A = 175
a) full: 151.46 156.32 1.03 62.16 73.48 1.18
σmax/σadm 0.865 0.893 0.355 0.420
b) ballast: 107.45 148.90 1.39 100.31 155.10 1.55
σmax/σadm 0.614 0.851 0.573 0.886

463

MARSTRUCT.indb 463 2/18/2011 5:51:46 PM

 2 2
σx [N/mm ] DECK 1D-beam Model Sagging / Quasi-static Wave / TK Oil-tanker Full Cargo σVON max [N/mm ] DECK 3D-FEM Model Sagging / Quasi-static Wave / TK Oil-tanker Ballast
25.00 225.00

0.00 200.00

-25.00
175.00

-50.00
150.00
-75.00
125.00
-100.00
100.00
-125.00
75.00
-150.00

50.00
-175.00

-200.00 25.00

-225.00 0.00
0.00 28.50 57.00 85.50 114.00 142.50 171.00 199.50 228.00 256.50 285.00 0.00 28.50 57.00 85.50 114.00 142.50 171.00 199.50 228.00 256.50 285.00

hw=0m hw=1m hw=2m hw=3m hw=4m hw=5m hw=6m hw=7m x [m] hw=0m hw=1m hw=2m hw=3m hw=4m hw=5m hw=6m hw=7m x [m]
hw=8m hw=9m hw=10m hw=10.69 hw=11m hw=12m adm hw=8m hw=9m hw=10m hw=10.69 hw=11m hw=12m adm

Figure 5a. 1D-girder model, σxD deck normal stress Figure 7a. 3D-FEM model, maximum von Mises deck
[N/mm2], head wave sagging condition (hw = 0–12 m), full stress [N/mm2], head wave sagging condition, ballast.
cargo.
2
σVON max [N/mm ] DECK 3D-FEM Model Hogging / Quasi-static Wave / TK Oil-tanker Ballast
225.00

2
σx max [N/mm ] DECK 3D-FEM Model Sagging / Quasi-static Wave / TK Oil-tanker Full Cargo 200.00
25.00

175.00
0.00

-25.00 150.00

-50.00 125.00

-75.00 100.00

-100.00
75.00

-125.00
50.00
-150.00
25.00
-175.00
0.00
-200.00 0.00 28.50 57.00 85.50 114.00 142.50 171.00 199.50 228.00 256.50 285.00

-225.00 hw=0m hw=1m hw=2m hw=3m hw=4m hw=5m hw=6m hw=7m x [m]
0.00 28.50 57.00 85.50 114.00 142.50 171.00 199.50 228.00 256.50 285.00 hw=8m hw=9m hw=10m hw=10.69 hw=11m hw=12m adm

hw=0m hw=1m hw=2m hw=3m hw=4m hw=5m hw=6m hw=7m x [m]

hw=8m hw=9m hw=10m hw=10.69 hw=11m hw=12m adm

Figure 7b. 3D-FEM model, maximum von Mises deck

Figure 5b. 3D-FEM model, maximum σxD deck stress [N/mm2], head wave hogging condition, ballast.
normal stress [N/mm2], head wave sagging condition
(hw = 0–12 m), full cargo.

2
σVON max [N/mm ] DECK 3D-FEM Model Sagging / Quasi-static Wave / TK Oil-tanker Full Cargo
225.00

200.00

175.00

150.00

125.00

100.00

75.00

50.00

25.00

0.00
0.00 28.50 57.00 85.50 114.00 142.50 171.00 199.50 228.00 256.50 285.00

hw=0m hw=1m hw=2m hw=3m hw=4m hw=5m hw=6m hw=7m x [m]

hw=8m hw=9m hw=10m hw=10.69 hw=11m hw=12m adm

Figure 8a. Wave pressure at hw = 10.692 m, sagging, full

Figure 6a. 3D-FEM model, maximum von Mises deck cargo.
stress [N/mm2], head wave sagging condition, full cargo.

2
σVON max [N/mm ] DECK 3D-FEM Model Hogging / Quasi-static Wave / TK Oil-tanker Full Cargo
225.00

200.00

175.00

150.00

125.00

100.00

75.00

50.00

25.00

0.00
0.00 28.50 57.00 85.50 114.00 142.50 171.00 199.50 228.00 256.50 285.00

hw=0m hw=1m hw=2m hw=3m hw=4m hw=5m hw=6m hw=7m x [m]

hw=8m hw=9m hw=10m hw=10.69 hw=11m hw=12m adm

Figure 6b. 3D-FEM model, maximum von Mises deck Figure 8b. Wave pressure at hW = 10.692 m, hogging, full
stress [N/mm2], head wave hogging condition, full cargo. cargo.

464

MARSTRUCT.indb 464 2/18/2011 5:51:47 PM


Figure 9a. σvonM [KN/m2], hw = 10.692 m, x = 66–216 m,
sagging, full.
Figure 9b. σvonM [KN/m2], hw = 10.692 m, x = 66–216 m,
hogging, full.
In Tables 7 and 8 are included the following
results:
− the ship hull girder buckling factor BBuckling > 1
and the maximum ship girder deflection in verti-
cal direction wz[m], based on 3D-FEM model,
with the admissible hull deflection value
wadm = L/500 = 0.570 m, for quasi-static wave ref-
erence height hw = 10.692 m (Table 7).
− the maximum deck and bottom shell normal
stress σxD,Bmax [N/mm2], based on 3D-FEM and
1D-girder models, inclusive stress ratio 3D/1D,
for quasi-static wave height hw = 10.692 m
(Table 8 (1), (2));
− the maximum side shell tangential stresses at
the neutral axis τxzSn−nmax [N/mm2], based on
3D-FEM and 1D-girder models, inclusive
stress ratio 3D/1D, for quasi-static wave height
hw = 10.692 m (Table 8(3));
− the maximum deck and bottom shell equiva-
lent von Mises stress σvonD,Bmax [N/mm2], based
on 3D-FEM and 1D-girder models, and ratio
3D/1D, for reference wave height hw = 10.692 m
(Table 8 (4),(5)).
In Figures 5–9 are included the following results:
− the deck shell normal stress σxD diagrams, based
on (a) 1D-girder and (b) 3D-FEM models, for
465
MARSTRUCT.indb 465
head wave height hw = 0–12 m, full cargo load
case, sagging quasi-static equivalent wave con-
dition (Figs. 5a, b), pointing out the deck shell
hot-spot stresses from the intersections with
transversal and also vertical panels;
− the maximum deck shell equivalent von Mises
stress σvonDmax diagrams, based on 3D-FEM
model, for head wave height hw = 0–12 m, full
cargo and ballast load cases, sagging and hog-
ging quasi-static equivalent wave conditions
(Figs. 6–7a, b);
− the external wave pressure distribution, for head
wave height hw = 10.692 m, acting on the bottom
and side shell, based on 3D-FEM model, applied
with eigen iterative GEO macro-commands file,
full cargo load case, sagging and hogging quasi-
static equivalent wave conditions (Figs. 8a, b);
− the equivalent von Mises stress distribution, for
head wave height hw = 10.692 m, in the ship cen-
tral part, x = 66–216 m, based on the 3D-FEM
model, full cargo load case, sagging and hogging
(Figs. 9a, b).
4 THE HYDROELASTIC RESPONSE
ANALYSIS
In this section, the large oil-tanker TK 167000 tdw
linear and non-linear dynamic response in irregular
head waves is analysed, based on the hydroelastic-
ity theory, including: ship oscillation and vibra-
tion, the bottom and side slamming phenomena,
the springing and whipping phenomena, under the
following hypotheses:
− the ship hull is modelled with 1D-FEM finite
element method, using Timoshenko elastic beam
finite elements (Hughes 1988);
− the ship offset lines are modelled with the con-
formal multi-parametric transformation method
(Bishop & Price 1979);
− the hydrodynamic excitation forces are modelled
according to hydroelasticity and strip theory
(2D-Flow), with non-linear and slamming terms,
based on a generalized Gerristma & Beukelman
model (Domnisoru 1998);
− the hydrodynamic coefficients are calculated
based on the Porter & Vugts 2D potential fluid
flow method (Bishop & Price 1979, Domnisoru
2006);
− the ship dynamic response is decomposed,
according to the modal analysis technique,
on ship oscillation (low frequency, rigid hull)
and vibration (high frequency, dry elastic hull)
modes (Bishop & Price 1979);
− the restoring stiffness includes the heave and
pitch and the elastic modes terms (Domnisoru
1998, 2006);
2/18/2011 5:51:53 PM
 − the excitation is the external head wave, model − the ship non-linear dynamic response;
Longuet-Higgins, with second order interference − the spectral analysis of the total ship dynamic
components (Price & Bishop 1974, Perunovic & response with the FFT Fast Fourier Transfor-
Jensen 2005, Domnisoru 1998, 2006). mation, short-term statistical parameters. From
the heave and pitch rigid hull modes response
Based on the theoretical model, in detail pre-
results oscillation statistics and from elastic hull
sented in (Domnisoru 1998, 2006), the numerical
modes response results vibration statistics.
analyses are carried out with eigen program DYN,
including the linear STABY and the non-linear The dynamic analyses are carried out for the
TRANZY solvers. head waves first order spectra ITTC (Price &
The STABY module, the steady state ship Bishop 1974) with the significant wave height
dynamic response, includes linear oscillation and h1/3 = 0–12 m, step δh1/3 = 0.5 m, according to the
springing components. This code program module Beaufort scale Blevel = 0–11.
includes the following main steps: Section 2 presents the large oil-tanker TK 167000
tdw 1D-FEM girder model. Table 4 includes the
− input ship data, natural modes, first order wave
full cargo and ballast loading cases.
spectra ITTC (Price & Bishop 1974);
The numerical results for the hydroelas-
− the calculation of Longuet-Higgins time domain
tic response are synthesized in Tables 9–11
wave elevation, with random components
and Figures 10–13, for both loading cases (see
phases;
Conclusions 4–7).
ηw ( t ) ∑ ⎡⎣ ηwc (x, ωe ) ω et − ηws ( ω e ) i ω et ⎤⎦ Table 9a. The ratios between the significant displace-
(ω e ) ments and deformations %w1/3vib/w1/3osc, full cargo case
(4) (reference h1/3 = 12 m).
where ωe is the encountering ship-wave circular fre- %vib/osc %vib/osc Bottom Side
quency of a wave component; ηwc,s(x,ωe) are the ωe x/L linear non-linear slamming slamming
wave frequency domain components;
0.00 5.86 5.90 no yes
− the calculation of motion equations system 0.50 4.91 5.08 – –
terms: structural, hydrodynamic and wave exci- 1.00 5.00 5.22 no yes
tation forces; Average 5.26 5.40 Slamming occurrence
− the solution in frequency domain of the ship linear
dynamic response on each wave component ωe;
− the calculation of linear ship dynamic response, Table 9b. The ratios between the significant displace-
based on spectral composition in time domain. ments and deformations %w1/3vib/w1/3osc, ballast case
(reference h1/3 = 12 m).
The TRANZY module, the non-linear and
transitory ship dynamic response, includes non- %vib/osc %vib/osc Bottom Side
linear oscillations and springing, bottom and side x/L linear non-linear slamming slamming
slamming and whipping components.
This code program module includes the follow- 0.00 7.21 7.83 no yes
ing main steps: 0.50 5.10 5.64 – –
1.00 6.11 6.49 yes yes
− the ship data input and solution of STABY Average 6.14 6.65 Slamming occurrence
module;
− the calculation of the non-linear motion equa-
tions system terms; Table 10a. The maximum ratios for the significant
Because the excitation force includes the bending moments and shearing forces max(%M1/3vib/
unknown non-linear dynamic response, it is nec- M1/3osc, %T1/3vib/T1/3osc), full cargo load case (reference
h1/3 = 12 m).
essary to use an iterative algorithm for the time
domain solution of non-linear motion equations %vib/osc %vib/osc
(Domnisoru 1998). x/L linear non-linear Springing Whipping
− the solution of the differential non-linear motion
0.25 12.52 33.66 Linear:
equations system, using a time domain integra-
0.50 14.00 42.59 Small High
tion procedure, β-Newmark, at each iteration, 0.75 13.50 43.49 Non-linear:
with simulation time Ts = 80s and time step Average 13.34 39.91 Medium
δt = 0.01s (ftriggering = 100 Hz);

466

MARSTRUCT.indb 466 2/18/2011 5:51:55 PM

 Table 10b. The maximum ratios for the significant Table 11b. The maximum significant stresses (dynamic
bending moments and shearing forces max(%M1/3vib/ response), added to still water values, ballast (bottom
M1/3osc, %T1/3vib/T1/3osc), ballast case (reference h1/3 = 12 m). and side σadm-A = 175 N/mm2 and τadm-A = 110 N/mm2/deck
σadm-AH32 = 224 N/mm2).
%vib/osc %vib/osc
x/L linear non-linear Springing Whipping Maximum stress [N/mm2]
(taking as reference Strength h1/3 [m] Beaufort
0.25 11.81 63.55 Linear: h1/3 = 12 m) criterion limit limit
0.50 12.32 75.15 Small Higher
0.75 12.65 63.43 Non-linear: σ1/3max_LIN + |sw| 89.07 0.397 < 1 12 11
Average 12.26 67.37 Medium deck
σ1/3max_LIN + |sw| 75.65 0.432 < 1 12 11
bottom
Table 11a. The maximum significant stresses (dynamic σ1/3max_NL + |sw| 125.54 0.560 < 1 12 11
response), added to still water values, full cargo (bottom deck
and side σadm-A = 175 N/mm2 and τadm-A = 110 N/mm2/deck σ1/3max_NL + |sw| 106.31 0.608 < 1 12 11
σadm-AH32 = 224 N/mm2). bottom
Maximum stress [N/mm2] τ1/3max_LIN + |sw| 56.92 0.517 < 1 12 11
(taking as reference Strength h1/3 [m] Beaufort side n-n
h1/3 = 12 m) criterion limit limit τ1/3max_NL + |sw| 67.21 0.611 < 1 12 11
side n-n
σ1/3max_LIN + |sw| 153.67 0.685 < 1 12 11
deck
σ1/3max_LIN + |sw| 131.49 0.751 < 1 12 11
bottom η wave [m]
7.50
Section x/L=0.5 Wave LH & ITTC h1/3=12m
Time Record : Wave

σ1/3max_NL + |sw| 175.54 0.782 < 1 12 11 6.00

deck 4.50

σ1/3max_NL + |sw| 149.77 0.856 < 1 12 11 3.00

bottom 1.50

τ1/3max_LIN + |sw| 0.461 < 1

0.00
50.71 12 11 -1.50

side n-n -3.00

τ1/3max_NL + |sw| 56.55 0.514 < 1 12 11 -4.50

side n-n -6.00

-7.50
0 10 20 30 40 50 60 70 t [s] 80

In Tables 9–11, for simulation time 80s and Figure 10a. The wave L-H time record (h1/3 = 12 m,
8000 records, the following statistical results are x/L = 0.5).
included:
− the ratio for the significant deformation on the η wave [m] Amplitude Spectrum FFT Analysis : Wave

fundamental natural vibration mode and the sig- 3.00

Section x/L=0.5 Wave LH & ITTC h1/3=12m

2.75

nificant vertical displacement of the ship rigid 2.50

hull oscillations %w1/3vib/w1/3osc (Tables 9a, b); 2.25

− the maximum ratios for the significant bending

2.00

1.75

moments and shearing forces, on fundamental 1.50

1.25

natural vibration mode and the ship rigid hull 1.00

oscillations, max(%M1/3vib/M1/3osc , %T1/3vib/T1/3osc) 0.75

0.50

(Tables 10a, b); 0.25

− the maximum significant (σ1/3, τ1/3) normal deck 0.00

0.00 0.50 1.00 1.50 2.00 2.50 f [Hz] 3.00

and bottom shells, side shell tangential neutral

axis stresses (according to Equation 5), added to Figure 10b. The wave L-H amplitude spectrum
still water (σsw, τsw) stresses (Tables 11a, b). (h1/3 = 12 m, x/L = 0.5).

σ13
1/ , M13
1/ W ,B τ1/13
1/3 = T/ max
a Kτ n − nS
σ13
1/ ,B + σ swD,B τ1///3maxSn − n + τ swSn
S −n
swSn (5)
≤1 ≤1
σ adm τ adm

467

MARSTRUCT.indb 467 2/18/2011 5:51:56 PM

 2
Mosc_nl [KNm] Oil-tanker Full Cargo Section x/L=0.5 vs=15Knots Wave LH & ITTC h1/3=12m σx1/3nl+|ac| [N/mm ] DECK Oil-tanker Full Cargo vs=15 Knots & Wave LH & ITTC
Non-Linear Analysis Time Record : Bending Moment - Oscillation Response 225.00
8.0E+06

200.00
6.0E+06
175.00

4.0E+06
150.00

2.0E+06 125.00

0.0E+00 100.00

75.00
-2.0E+06
50.00
-4.0E+06
25.00

-6.0E+06
0.00
0.00 28.50 57.00 85.50 114.00 142.50 171.00 199.50 228.00 256.50 285.00
-8.0E+06 h1/3=0m h1/3=1m h1/3=2m h1/3=3m h1/3=4m h1/3=5m h1/3=6m x[m]
0 10 20 30 40 50 60 70 t [s] 80
h1/3=7m h1/3=8m h1/3=9m h1/3=10m h1/3=11m h1/3=12m Sig_adm

Figure 11a. Oscillation bending moment time record, Figure 13a. Maximum significant normal deck stress
non-linear analysis, wave h1/3 = 12 m, vs = 15 Knots, [N/mm2], non-linear analysis + still water, h1/3 = 0–12 m,
x/L = 0.5, full cargo case. vs = 15 Knots, full cargo case.

2
Mosc_nl [KNm] Oil-tanker Full Cargo Section x/L=0.5 vs=15Knots Wave LH & ITTC h1/3=12m σx1/3nl+|ac| [N/mm ] DECK Oil-tanker Ballast vs=16 Knots & Wave LH & ITTC
Non-Linear Amplitude Spectrum FFT Analysis : Bending Moment - Oscillation Response 225.00
2.5E+06
200.00

175.00
2.0E+06
150.00

125.00
1.5E+06

100.00

75.00
1.0E+06

50.00

25.00
5.0E+05

0.00
0.00 28.50 57.00 85.50 114.00 142.50 171.00 199.50 228.00 256.50 285.00

0.0E+00 h1/3=0m h1/3=1m h1/3=2m h1/3=3m h1/3=4m h1/3=5m h1/3=6m x[m]

0.00 0.50 1.00 1.50 2.00 2.50 f [Hz] 3.00 h1/3=7m h1/3=8m h1/3=9m h1/3=10m h1/3=11m h1/3=12m Sig_adm

Figure 11b. Oscillation bending moment amplitude Figure 13b. Maximum significant normal deck stress
spectrum FFT, non-linear analysis, wave h1/3 = 12 m, [N/mm2], non-linear analysis + still water, h1/3 = 0–12 m,
vs = 15 Knots, x/L = 0.5, full cargo. vs = 16 Knots, ballast case.

M_nl [KNm]
In Figures 10–13 are included the following
Oil-tanker Full Cargo Section x/L=0.5 vs=15Knots Wave LH & ITTC h1/3=12m

8.0E+06
Non-Linear Analysis Time Record : Bending Moment - Hydroelastic Response
results:
− Figures 10a, b present the time record and the
6.0E+06

4.0E+06
amplitude spectrum for the Longuet-Higgins
2.0E+06

(L-H) wave, with first order wave spectrum ITTC

h1/3 = 12 m, at amidships section x/L = 0.5;
0.0E+00

− Figures 11a, b present the time record and the

-2.0E+06

-4.0E+06

-6.0E+06
amplitude spectrum for the non-linear oscilla-
-8.0E+06
tion bending moment (non-linear seakeeping
0 10 20 30 40 50 60 70 t [s] 80
response on heave and pitch modes), head wave
Figure 12a. Hydroelastic bending moment time record, h1/3 = 12 m, at amidships section x/L = 0.5, full
non-linear analysis, wave h1/3 = 12 m, vs = 15 Knots, cargo load case;
x/L = 0.5, full cargo. − Figures 12a, b present the time record and the
amplitude spectrum FFT, for the non-linear
hydro-elastic bending moment, head wave
M_nl [KNm]

2.5E+06
Oil-tanker Full Cargo Section x/L=0.5 vs=15Knots Wave LH & ITTC h1/3=12m
Non-Linear Amplitude Spectrum FFT Analysis : Bending Moment - Hydroelastic Response h1/3 = 12 m, at amidships section x/L = 0.5, full
cargo load case;
2.0E+06
− Figures 13a, b present the maximum hydroelas-
1.5E+06
tic significant deck normal stresses, both loading
cases.
1.0E+06

5.0E+05

5 THE SHIP FATIGUE ANALYSIS

0.0E+00
0.00 0.50 1.00 1.50 2.00 2.50 f [Hz] 3.00
AND THE INITIAL SHIP SERVICE
LIFE EVALUATION
Figure 12b. Hydroelastic bending moment ampli-
tude spectrum FFT, non-linear analysis, h1/3 = 12 m, In this section, the large oil-tanker TK 167000
vs = 15 Knots, x/L = 0.5, full cargo. tdw initial service life evaluation is analysed,

468

MARSTRUCT.indb 468 2/18/2011 5:51:57 PM

 based on the fatigue strength assessment of the
ship hull structure, using the maximum stresses
obtained in the deck shell for extreme wave loads
(Sections 3–4).
The ship fatigue strength criterion is analysed
applying the cumulative damage ratio D method
(Palmgren-Miner method), with the steel standard
design S-N curves (IACS 2008, GL 2004, 2008).
From the short-term prediction analysis of the
ship dynamic response, the oscillation (low fre-
quency, rigid hull modes) and vibration (high fre-
quency, elastic hull modes) significant stress σ1/3
components, function to the head waves signifi-
cant height h1/3, are obtained.
In the hypotheses of a hydroelastic ship dynamic
response, for a reference time of R = 20 years, the
cumulative damage ratio D has the expression:
D = Dosc + Dvib; ni_osc,vib = pi ⋅ nmax_osc,vib
M
n
Dosc,vib = ∑ Ni _ osc,vibi
i=1 i _ osc
osc ,vib
i
nmax_osc,vib = 3.1536107 R ⋅ fosc,vib;
where: fosc,vib the natural ship frequencies for oscil-
lation and vibration modes (Table 5); nmax_osc,vib the
maximum number of cycles; pi(h1/3i), i = 1, M the
probabilities of World Wide Trade (WWT) wave
significant height h1/3 histogram (Fig. 14) (Price &
Bishop 1974); ni_osc,vib the number of stress cycles
for h1/3i; Ni_osc,vib the number of endured stress cycles
from the steel standard design S-N curves for a
stress range Δσi_osc,vib; fc the correction for non-sym-
metric cycles.
For the oil-tanker the full cargo and ballast load
conditions are considered with the same occur-
rence probability scenario. The cumulative damage
ratio D has the following expression:
D = 0.5 ⋅ Dfull + 0.5 ⋅ Dballast ≤ 1;
where L [years] is the estimated ship service life.
pi World Wide Trade h1/3 Histogram LTP
0.40
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0.00
0.5 1 1.5 2 2.5 3 3.5 4 4.5
Figure 14. The World Wide Trade wave height h1/3
histogram.
MARSTRUCT.indb 469
Because the significant bending moments are
obtained with a 1D-FEM/2D-Flow ship dynamic
model (Section 4) and the deck shell hot spot stresses
at intersections with transversal and vertical panels
are obtained on 3D-FEM model (Section 3), in
order to obtain an equivalent 3D/1D combined
statistical model, the significant deck equivalent
von Mises stresses σ3D/1DvonM1/3 are evaluated based
on the ship significant bending moments M1Dbeam1/3
(Section 4) and the stress influence coefficients
k3DvonM (Equation 3), with next expression:
σ3D/1DvonM1/3_Deck ≅ k3DvonM_Deck ⋅ M1Dbeam1/3 (8)
This study includes the oil-tanker initial design
structure, having not all of the structural details
defined. In this case, the numerical initial fatigue cri-
terion check is carried out only for the longitudinal
butt weld joints of the ship hull deck shell, which
according the Germanischer Lloyds Rules (GL 2008,
; Ni_osc,vib = fSN (Δσi_osc,vib) (6) Part 1, Chapter 1, Section 20C, Table 20.3) is type B1
detail category, with 125 N/mm2 the fatigue strength
reference value from steel S-N curve for 2 ⋅ 106 cycles.
Δσi_osc,vib = 2σ1/3i_osc,vib ⋅ fc Tables 12a, b present a synthesize of the fatigue
analyses for the large oil-tanker deck shell (type
B1 detail), using the Palmgren-Miner cumulative
damage ratio D and the design S-N curves, as
following:
− the fatigue criterion check, based on stresses
from seakeeping analysis (IACS 2008), including
linear and non-linear oscillations components
(Table 12a);
− the fatigue criterion check, based on stresses
from hydroelastic analysis, including linear and
non-linear oscillations and vibrations compo-
nents (Table 12b).
In Tables 12a, b, the influence of the butt weld
joints welding quality (GL 2008), standard or very
L = 20/D (7)
Table 12a. Fatigue criterion, based on D ratio and
design S-N curves, with stresses from seakeeping analysis
(only oscillations).
WWT histogram Full Ballast Combined
Analyses Welding DSN_ full DSN_ball DSN L [years]
1D-beam model, σx1/3 deck normal stress
Standard 0.430 0.365 0.397 >20
Linear
Nonlinear Standard 0.574 0.403 0.488 >20
Nonlinear Very good 0.231 0.157 0.194 >20
3D/1D model, σvonM1/3 deck maximum equivalent
h1/3 [m] von Mises stress
5 5.5 6 6.5 7 7.5 8 8.5 9 9.5 10 10.5 11 11.5 12
Linear Standard 0.623 0.835 0.729 >20
Nonlinear Standard 0.821 0.916 0.869 >20
Nonlinear Very good 0.343 0.390 0.367 >20
469
2/18/2011 5:52:01 PM
 Table 12b. Fatigue criterion, based on D ratio and
design S-N curves, with stresses from hydroelastic analysis
(oscillations and vibrations).
WWT histogram Full Ballast Combined
Analyses Welding DSN_ full DSN_ball DSN L [years]
1D-beam model, σx1/3 deck normal stress
Linear Standard 0.439 0.365 0.402 >20
Nonlinear Standard 0.725 0.757 0.741 >20
Nonlinear Very good 0.281 0.273 0.277 >20
3D/1D model, σvonM1/3 deck maximum equivalent von
Mises stress
Linear Standard 0.638 0.836 0.737 >20
Nonlinear Standard 1.064 1.915 1.490 13.4
Nonlinear Very good 0.423 0.731 0.577 >20
good welding, is taken into account for the initial
ship service life evaluation. Also, in Tables 12a, b
are considered the D factor differences between
1D and equivalent 3D/1D (with hot spot stresses)
models (see Conclusions 8–11).
6 CONCLUSIONS
Based on the numerical results from Sections 3–5,
for the large oil-tanker TK 167000 tdw hull struc-
ture (Section 2), the following conclusions are
resulting:
1. The buckling factor and the maximum vertical
deflection satisfy the admissible limits, based on
the 3D-FEM oil-tanker hull model (Table 7).
2. The average stress differences between the
3D-FEM and 1D-beam models (Table 8), for
the ship strength analyses with quasi-static head
wave loads (reference height hw = 10.692 m),
are: normal stress deck +17.50% and bottom
+37.25%, side tangential stress in neutral axis
+12.5%, von Mises stress deck +11.50% and
bottom +28.75%. Figures 5–7, 9a, b are pointing
out that the stress differences between 3D-FEM
and 1D-girder models are significant around
the transversal bulkheads structure that can not
be included in the 1D-girder model.
3. The maximum stress values obtained at ship
strength analyses with 3D-FEM model, under
quasi-static head wave loads, for both loading
cases, are smaller than the admissible values, with
ratio σ,τmax/σ,τadm = 0.24 ÷ 0.96 ≤ 1 (Table 8). The
maximum stress values are encountered in the
deck shell (Figs. 5–7a, b, Figs. 9a, b and Table 8).
4. The hydroelastic ship girder vertical deforma-
tions are small comparing to the ship oscillation
displacements 5.26 ÷ 6.65% (Tables 9a, b).
470
MARSTRUCT.indb 470
5. Side slamming for both loading cases and
bottom slamming for ballast load case have
high probability to occur at ship extremities
(Tables 9a, b), inducing whipping with high
intensity (Tables 10a, b). The linear springing
has small intensity. The non-linear springing
has medium intensity, due to the ship high
motion amplitudes and low first natural vibra-
tion frequency (0.629 ÷ 0.645 Hz, Table 5),
inducing hydrodynamic non-linearity sources.
6. From the statistical short-term prediction of
the hydroelastic ship hull response, taking as
reference the oscillation significant bending
moments and shear forces, the first order
vibration component (Figs. 11, 12a, b) rep-
resent 11.81 ÷ 14.00% at linear analyses, at
non-linear analyses 33.66 ÷ 43.49% full cargo
case and 63.43 ÷ 75.15% ballast load case
(Tables 10a, b). The extreme wave loads are
induced mainly by the slamming and high
intensity whipping phenomena.
7. The hydroelastic ship response analyses
numerical results (Tables 11a, b), indicate that
the maximum stresses are obtained in the deck
shell (Figs. 13a, b).
8. The fatigue criterion based on seakeeping
(oscillations) stresses (Table 12a) is satisfied,
for 1D-beam model DSN = 0.194/0.488 < 1
and also for equivalent 3D/1D model
DSN = 0.367 ÷ 0.869 < 1, resulting a initial serv-
ice life over L > 20 years.
9. The fatigue criterion based on hydroe-
lastic (oscillations and vibration) stresses
(Table 12b) is satisfied for 1D-beam model
DSN = 0.277 ÷ 0.741 < 1. For equivalent 3D/1D
model, non-linear analysis, standard welding
quality results DSN = 1.490 > 1 and L = 13.4 < 20
years. Considering a very good welding case,
results DSN = 0.577 < 1 and L > 20 years.
10. Based on the S-N curves formulation with
slope exponent m = 3 ÷ 5 > 1 (GL 2008), small
Δσi3D/Δσi1D stress ratios (Table 8) induce larger
Ni1D/Ni3D cycle ratios, resulting larger D3D/D1D
fatigue ratios (Table 12a, b).
11. The numerical results from fatigue analyses
(Tables 12a, b) point out that the hydroelastic
stresses have to be used instead of standard
seakeeping stresses, with non-linear equivalent
3D/1D model, in order to obtain a more realis-
tic initial ship service life evaluation.
12. The numerical results of this study, carried out for
the initial design structure of a large oil-tanker,
based on the link between the 3D/1D-FEM
hull model and the hydroelasticity analysis with
1D-beam/2D-Flow models, are pointing out on
short-term the extreme wave loads induced in
the ship hull structure and on long term are pre-
dicting the ships initial service life.
2/18/2011 5:52:02 PM
 ACKNOWLEDGMENTS
The work has been performed in the scope of
the project MARSTRUCT, Network of Excel-
lence on Marine Structures, (www.mar.ist.utl.pt/
marstruct/), which has been financed by the EU
through the GROWTH Programme under con-
tract TNE3-CT-2003-506141.
REFERENCES
Bertram, V. 2000. Practical ship hydrodynamics. Oxford:
Butterworth Heinemann.
Bhattacharyya, R. 1978. Dynamics of marine vehicles.
New York: John Wiley & Sons Publication.
Bishop, R.E.D. & Price, W.G. 1979. Hydroelasticity of
ships. Cambridge: University Press Cambridge.
Domnisoru, L. 2006. Structural analysis and hydroelas-
ticity of ships. Galati: University “Lower Danube”
Press.
Domnisoru, L. & Domnisoru, D. 1998. The unified anal-
ysis of springing and whipping phenomena. Trans-
actions of the Royal Institution of Naval Architects
London 140(A): 19–36.
Domnisoru, L., Dumitru, D. & Ioan, A. 2008. Numeri-
cal methods for hull structure strengths analysis and
ships service life evaluation, for a LPG carrier. OMAE
15–20 June 2008, Estoril: 509–518.
Fricke, W. & Kahl, A. 2005. Comparison of different
structural stress approaches for fatigue assessment
of welded ship structures. Marine Structures 18:
473–488.
GL 2004. Guidelines for fatigue strength analyses of ship
structures. Hamburg: Germanischer Lloyd.
GL 2008. Germanischer Lloyd’s Rules. Hamburg.
Garbatov, Y., Tomasevic, S. & Guedes Soares, C. 2005.
Fatigue damage assessment of a newly built FPSO
hull. Guedes Soares. C., Garbatov Y. & Fonseca N.,
(editors). Maritime Transportation and Exploata-
tion of Ocean and Costal Resources, Taylor & Francis
Group, London: 423–428.
471
MARSTRUCT.indb 471
Guedes Soares, C. 1999. Special issue on loads on marine
structures. Marine Structures 12(3): 129–209.
Hirdaris, S.E., Price, W.G. & Temarel, P. 2003. Two
and three-dimensional hydroelastic modelling of a
bulk carrier in regular waves. Marine Structures 16:
627–658.
Hirdaris, S.E. & Ge Chunhua. 2005. Review and intro-
duction to hydroelasticity of ships. Report 8. London:
Lloyd’s Register.
Hughes, O.F. 1988. Ship structural design. A rationally-
based, computer-aided optimization approach. New
Jersey: The Society of Naval Architects and Marine
Engineering.
IACS 2008. Common structural rules for double hull oil
tankers. London: International Association of Clas-
sification Societies.
Jang, C.D. & Hong, S.Y. (editors) 2009. Proceeding of the
17th international ship and offshore structures congress
(ISSC). Volumes 1 & 2. Seoul National University.
Mansour, A. & Liu, D. 2008. Strength of ships and ocean
structures. New Jersey: The Society of Naval Archi-
tects and Marine Engineering.
Park, J.H. & Temarel, P. 2007. The influence of nonlin-
earities on wave-induced motions and loads predicted
by two-dimensional hydroelasticity analysis. ABS-
PRADS 1–5 Oct. 2007, Houston (1): 27–34.
Perunovic, J.V. & Jensen, J.J. 2005. Non-linear springing
excitation due to a bidirectional wave field. Marine
Structures 18: 332–358.
Price, W.G. & Bishop, R.E.D. 1974. Probabilistic theory
of ship dynamics. London: Chapman and Hall.
Rozbicki, M., Das Purnendu, K. & Crow, A. 2001.
The preliminary finite element modelling of a full
ship. International Shipbuilding Progress Delft
48(2): 213–225.
SWCM. 2007. SolidWorks Cosmos/M FEM program user
guide. Dassault Systems SolidWorks Corporation.
2/18/2011 5:52:03 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Dynamic analysis in the marine environment considering

FSI—Ship-like structure case

A. Martínez Cimadevilla
Technical Division, Navantia S.A., Madrid, Spain

ABSTRACT: The vibratory behaviour of marine structures like the typical case of a vessel is a complex
phenomenon that involves the evaluation of different aspects as the assessment of the external added
mass water effect and its influence on the vibration characteristics. This added mass water effect can be
properly established by considering the fluid-structure interaction due to the surrounding water.
From that evaluation, reliable dynamic parameters were derived. Two analyses types were applied: a
forced response linear dynamic analysis (harmonic acoustic finite element method) and an unsymmetric
modal analysis, both considering the Fluid-Structure Interaction (FSI).
A Lagrangian Eulerian finite element formulation was adopted for the FSI simulation.
The model employed for such analysis, was a simplified one: a model corresponding to a ship-like
structure (rectangular section hull girder). This model was considered representative enough to highlight
the main features of a marine structure vibration analysis.
In addition, a comparison between analytical values and Finite Element results related to frequencies
data was carried out. The compared values showed a good agreement.

1 INTRODUCTION (Harmonic Acoustic Finite Element Method), and

The study of the global vibratory behaviour of a
marine structure such as a ship, involves the evalu-
ation of different effects, being perhaps the most
remarkable, the surrounding water effect, due to
its influence on the system natural frequencies and
response values.
Different techniques exist to incorporate the above
mentioned added mass water effect in a dynamic
analysis. In the present paper, the technique employed
is based on a Finite Element (FE) modelling of the
surrounding water by using acoustic fluid elements,
and the modelling of the vessel by the corresponding
structural elements. Therefore, the Fluid Structure
Interaction (FSI) that occurs due to the system
vibration can be properly represented.
A simplified model was devised for the structure.
That model corresponds to a rectangular section hull
girder, i.e. a ship-like structure (structural arrangement
consisting of a stiffened shell, transverse bulkheads
and decks). However, that simplified feature does
not represent an inconvenient to highlight the main
facts involving the analysis to be carried out.
Structural model scantlings do not fulfil a specific
standard, like rules of a classification society, but
are adequate for a suitable dynamic behaviour, i.e.
avoidance of vibration local problems due to lack
of the required stiffness.
The analyses types applied, corresponded to
a Forced Response Linear Dynamic Analysis
Unsymmetric Modal Analysis, both considering
the Fluid-Structure Interaction (FSI).
All Finite Element (FE) calculations, have been
performed by using ANSYS Code.
Analysis results for the prediction of system
global vibratory behaviour, consisted of a data
set containing global vibration modal parameters,
harmonic pressure levels, and response curves
(displacement vs. frequency) at selected locations,
considered of interest from a vibratory standpoint.
In addition, the response curves could be used
as analytical data, in an experimental-analytical
correlation case, if necessary.
Traditionally, the Lewis method was the usual
approach to tackle with the added mass effect issue.
That method was regularly employed until the
introduction of boundary and acoustic elements in
FE codes, allowing a more adequate representation
of the mentioned effect.
Authors, like Cabos et al. (2003), have contribu-
ted to the development of analyses, contemplating
such elements.
2 MODEL DESCRIPTION
2.1 Hull girder model
Figure 1 depicts the rectangular section hull
girder global view. Basically, the main components

473

MARSTRUCT.indb 473 2/18/2011 5:52:03 PM


Figure 1. Hull girder global view.
Figure 2. Fluid mesh model.
Figure 3. Fluid mesh model & hull girder model.
are: a transversely web frame ring system together
with stiffened panels, and transverse bulkheads.
Hull Girder scantlings were developed to fulfil
an adequate dynamic behaviour, i.e. avoidance of
vibration local problems due to lack of the required
stiffness, and no specific strength standard was
applied. Considered material was steel.
Total model length is L = 252.3 m, Breadth
B = 32.2 m, Depth D = 32.4 m, and Draught T = 6.8 m
resulting in a displacement of Δ = 56620 t.
Transverse section moment of inertia about
global y axis is 241.90 m4 and about global z axis
is 229.1 m4.
Element density and concentrated masses were
applied in order to achieve the target total weight.
2.2 Fluid model
The fluid model has two zones, a contact zone sur-
rounding the pressure hull, having lagrangian finite
element formulation at the fluid-structure inter-
face, and a non-contact zone, having eulerian finite
474
MARSTRUCT.indb 474
element formulation, outside of the fluid-structure
interface.
Fluid mesh and extent, i.e. the amount of sur-
rounding fluid to include in the model, were studied
in order to obtain acceptable analysis results.
Figure 2 provides an exploded view of fluid
mesh model, and Figure 3 includes hull girder
model in the previous model.
Boundary conditions applied on the FE fluid
model, correspond to a non-reflecting surface, sur-
face limiting the outer limits of the fluid model.
3 DYNAMIC ANALYSIS CONSIDERING FSI
Two types of analyses were evaluated, the first one
corresponded to a forced response analysis tak-
ing into account the vertical direction only, and
the second one to an unsymmetric modal analysis
contemplating all directions but focused to obtain
axial, horizontal and coupled modes. In this way,
the typical analyses (forced and modal) are pre-
sented together with the corresponding analysis
results, providing a complete results set.
3.1 Forced response dynamic analysis method
(including FSI) description
A forced response dynamic analysis considering
harmonic excitations is the type of analysis that
usually is carried out to assess the vibratory
behaviour of a ship under the exciting sources.
In the present case, a simplification regarding the
exciting sources was done, and the only exciting
source considered was a vertical harmonic force
applied at the aft end bottom of the hull girder
model. Therefore, the response values will corre-
spond to such vertical direction (global z axis).
By performing the analysis using a frequency
sweep between 0.1 and 3 Hz, range of interest for
the hull girder global modes, the ship-like structure
global vibratory behaviour can be established.
In addition, a 3% damping ratio was assumed
for the structural system (ξ = 0.03).
3.2 Unsymmetric modal analysis method
(including FSI) description
In the case of a model considering FSI, the system
matrices became unsymmetric at the interface,
and therefore that type of modal analysis is the
adequate to obtain the modes of interest (axial,
horizontal and coupled).
3.3 Governing equations of Fluid Structure
Interaction (FSI)
The theoretical basis to take into account the effect
of the surrounded fluid on the vibrating structure,
2/18/2011 5:52:03 PM
 is based on the coupling of the fluid and structure {Np}: the element shape function for pressure
equations at the interface. {Nu}: the element shape function for displacements
The discretized structural dynamics equation {n}: the unit normal to the interface S
can be formulated as follows: By considering the fluid pressure acting at the
interface, the Eq. (1) for structural dynamics can
[Ms ]{u} [Cs
C ]{u} + [ Ks ]{u} {Fs
F} (1) be written as:

where: [Ms ]{u} [Cs

C ]{u} + [ Ks ]{u} {Fs
F } {Ffs
F } (6)
[Ms] → structural mass matrix, [Cs] → structural
damping matrix, [Ks] → structural stiffness matrix, {Ffs} = The fluid pressure load vector at the inter-
⎨Fs⎬ → applied load vector, ⎨u⎬ → nodal displace- face, which can be calculated by integrating the
ment vector. pressure over the area of the surface.
In ANSYS Code, the behaviour of the water After integration, the following equation is
pressure is described with the acoustic wave obtained:
equation (Helmholtz’s equation). This equation
is derived from the Navier-Stokes equation of ⇒{ } = [ ]{ } (7)
motion and the continuity equation, using the fol-
lowing assumptions: The complete finite element discretized equations
− The fluid is compressible for the fluid interaction case are derived by
− The fluid is inviscid (no viscous dissipation) substituting Eqs. (4) and (7) into Eqs. (3) and (6).
− The flow is irrotational As a result of the previous substitution, the
− There is no mean flow of the fluid complete finite element discretized equations for
− The mean density and pressure are uniform the fluid interaction case are written in combined
throughout the fluid (changes of mean den- form:
sity and pressure in different areas of the fluid
domain are small) ⎡ Ms 0 ⎤ ⎧ u ⎫ ⎡Cs 0 ⎤ ⎧ u ⎫
⎢ ⎥ ⎨  ⎬ + ⎢ ⎥⎨ ⎬
⎣ ρ0 R
T
The acoustic wave equation is given by: Mf ⎦ ⎩P
M ⎭ ⎣ 0 Cf ⎦ ⎩P ⎭
⎡ Ks − R ⎤ ⎧ u ⎫ ⎧Fs ⎫
1 ∂2 P +⎢ ⎥⎨ ⎬ = ⎨ ⎬ (8)
ΔP = (2) ⎣ 0 Kf ⎦ ⎩P ⎭ ⎩ 0 ⎭
c 2 ∂t 2

where: Therefore, in a FSI calculation the fluid element

P: fluid pressure, c: sonic speed in the fluid medium, at the interface generates all the matrices with
t: time, Δ: laplacian operator. subscript f in addition to the coupling matrices
Written in matrix notation, the discretized [R] and ρo [R]T. The matrices with subscript s are
acoustic wave equation (Helmholtz’s equation) at generated by the compatible structural element
the interface is: used in the model.
On a fluid-structure interface the system matrix
equations are unsymmetric due to the coupling
[Mf ]{P} [Cf
C ]{P} + [ Kf ]{P} {Fsf
F } (3) matrix [R], and outside the fluid-structure interface
the system matrix equations became symmetric,
where: the coupling matrix is null.
Mf: Fluid equivalent “mass” matrix; Cf: Fluid
equivalent “damping” matrix; Kf: Fluid equivalent
“Stiffness” matrix. The fluid “load” produced by 4 GLOBAL VIBRATION ANALYSIS
structure displacement at the interface: {Fsf}. RESULTS
That fluid “load” at the interface can be repre-
sented by: 4.1 Forced response dynamic analysis results
Analysis results correspond to the vertical direc-
{Fsf
s } ρo [R ] {u}
T
(4) tion only (vertical excitation force). Different
excitation directions could be applied to the sys-
where: tem and obtaining the corresponding response, if
−ρ0: mean fluid density. required.
Coupling matrix [R] given by: In the following, results data set is presented.
The set contains information considered relevant
[ ] ∫{ }{ p}T {n}d(( ) (5) to predict the system vibratory behaviour: mode
S shapes, displacements and harmonic pressures at

475

MARSTRUCT.indb 475 2/18/2011 5:52:04 PM

 resonant frequencies, together with response curves
at locations of interest (displacement amplitude vs.
frequency).
Output units are: Displacements (m), Pressures
(kPa).
Table 1, summarizes system global vertical fre-
quencies. Global modes correspond to participat-
ing modes ranging from two node (2N) through
five node (5N) mode.
Participating modes displacements are illus-
trated in Figure 4 through 7.
Figure 6. 4N vertical mode—displacements.
Displacements levels at each plot are in line with
expected values. The same applies to the mode
shape.
In addition, the displacements plots together
with the response curves represent a good basis to
assess analysis results consistency.
Harmonic Pressures plots apart from provid-
ing the pressure distribution at the considered fre-
quency, contain the information to be used in an
acoustic analysis, by transforming the pressures
into sound pressure levels.

Table 1. Global vertical frequencies summary. Figure 7. 5N vertical mode—displacements.

Mode Frequency (Hz)

2N 0.448
3N 1.057
4N 1.724
5N 2.42

Figure 8. 2N vertical mode—harmonic pressures.

Figure 4. 2N vertical mode—displacements.

Figure 9. 4N vertical mode—harmonic pressures.

Harmonic Pressures distribution at resonant

frequencies 0.448 Hz and 1.724 Hz, are illustrated
in Figures 8 and 9.

4.2 Unsymmetric modal analysis results

Analysis results correspond to the axial, horizon-
Figure 5. 3N vertical mode—displacements. tal direction and coupled modes. The obtained

476

MARSTRUCT.indb 476 2/18/2011 5:52:07 PM

 modes complement the previous results data
(vertical direction) and together provide a com-
plete dynamic results set.
Table 2, summarizes system global axial, hori-
zontal and coupled frequencies. Global horizontal
modes correspond to participating modes ranging
from two node (2N) through five node (5N) mode.
Participating Mode Shape Displacements are
illustrated in Figure 10 through 15.
Global Hull Girder horizontal frequencies
obtained values are in line with expected typi-
cal values, i.e. horizontal frequencies are in the
1.40–1.50% range of the vertical frequencies values.
Figure 12. 4N horizontal mode shape displacements.
Table 2. Global Ax., horizontal & coupled frequencies
summary.

Mode Frequency (Hz)

2N horizontal 0.645
3N horizontal 1.476
4N horizontal 2.423
5N horizontal 3.416
Axial 0.665
Torsional 2.429

Figure 13. 5N horizontal mode shape displacements.

Figure 10. 2N horizontal mode shape displacements.

Figure 14. Axial mode shape displacements.

Figure 11. 3N horizontal mode shape displacements. Figure 15. Torsional mode shape displacements.

477

MARSTRUCT.indb 477 2/18/2011 5:52:10 PM

 Displacements levels at each plot are in line with
expected values. The same applies to the mode
shape.

5 HULL GIRDER—RESPONSE CURVES

Response curves provide specific information

at locations of interest. That information could
be used in different situations, e.g. as a predicted
analytical data against a required standard, or in
an analytical—experimental correlation study. At
the same time response curves can be used as an
indication of system response data consistency. Figure 17. Response curve “A”.
Location information corresponding to the
selected nodes is depicted in Table 3 and Figure 16
below.
Resulting response curves of vertical displace-
ments (amplitudes) vs. frequencies are illustrated
in Figure 17 through 22.
A typical commentary at each curve plot related
to the curves values fulfilment (against a required
standard) should appear, but this has been omitted
in our case due to the data base nature: simplified
model and excitation.
Nevertheless, the displacements levels are in
line with expected values. The same applies to the
curves shape.
Figure 18. Response curve “B”.

Table 3. Hull girder—response curves location

description.

Curve Location Description

A Section @ 0.011L & CL, z = 32.4 m

B Section @ 0.023L & CL, z = 32.4 m
C Section @ 0.034L & CL, z = 32.4 m
D Section @ 0.540L & CL, z = 32.4 m
E Section @ 0.770L & CL, z = 32.4 m
F Section @ 0.480L & CL, z = 0 m

Notes: L → Hull girder total length (252.3 m); CL →

Hull girder centre line; D → Hull girder depth (32.4 m).
Figure 19. Response curve “C”.

Figure 16. Response curves location. Figure 20. Response curve “D”.

478

MARSTRUCT.indb 478 2/18/2011 5:52:14 PM

 Table 4. Vertical parameter α1—rectangular
cross section.

Ratio a/b α1

0.1 2.23
0.2 1.98
0.5 1.70
1.0 1.51
2.0 1.36
5.0 1.21

Notes: a → a = B/2 (B = 32.2 m); b → b = T

(T = 6.8 m).
Figure 21. Response curve “E”.

In our case a/b = 2.368, and by linear

interpolation, the obtained α1 value is 1.342.
The vertical added mass per unit length is
given by:
2
α1 ⎛ B ⎞
Amv = ρπ ⎜ ⎟
ρ (9)
2 ⎝ 2⎠
ρ = 1.025/9.81(t ⋅ s2/m4)
⇒ Amv = 57.076(t ⋅ s2/m2)
Hull girder mass per unit length: m = Δ/L
Figure 22. Response curve “F”. ⇒ m = 22.878 (t ⋅ s2/m2)
Hull girder vertical bending frequency in air,
calculated as a free-free 2N beam frequency in air:
6 FREQUENCIES COMPARISON—FE
RESULTS VS. ANALYTICAL RESULTS fa = 0.834 Hz
Corresponding frequency in a fluid (sea water):
In order to have an indication of the validity of the
Finite Element (FE) results, a comparison against 1
analytical results was carried out. The employed fw = 0.5
fa (10)
parameter for comparison was the hull girder ⎛ Amv ⎞
⎜⎝1 + ⎟
mode frequency value. m ⎠
Two analytical formulations were applied, the
first one was based on information related to the fw = 0.446 Hz
added mass of floating rectangular cross sections,
Higher modes vertical bending natural frequen-
described by Sedov (1965), and the estimation of
cies, were obtained from Johannessen & Skaar
the two-noded vertical (and horizontal) bending
(1980) expression:
frequency corresponded to a free-free beam case.
The second one, was based on Kumai (1968) equa- Nnv = N2v (n − 1)μ (11)
tions for the added mass and natural frequency of
the two-noded vertical bending case. where N2v is the hull girder 2N vertical natural fre-
For the higher hull girder modes vertical (and quency, n is the number of nodes.
horizontal) frequencies estimation, the formula It was assumed a typical block coefficient for
from Johannessen & Skaar (1980) was used. the ship types, and Table 5 summarizes the result-
ing distribution.
6.1 Frequency evalution based on added mass By linear interpolation of Table 5 values, con-
Sedov formulation sidering a block coefficient of 1.0 (box girder case),
the derived μ value is 1.20.
6.1.1 Vertical case Therefore, the Equation (11) for the present box
Table 4, contains parameters for the two-noded girder case is:
vertical added mass calculation of floating rectan-
gular cross sections from Sedov (1965). Nnv = 0.446 (n − 1)1.20 (12)

479

MARSTRUCT.indb 479 2/18/2011 5:52:16 PM

 Table 5. μ parameter distribution. Hence, from the obtained information, a good
agreement between the Finite Element (FE) and
Ship type Block coefficient (Cb) μ the analytical frequency data is noted.
Cargo ship 0.68 0.845
6.1.2 Horizontal case
Bulk carrier 0.80 1.00
Table 7, contains parameters for the two-noded
Tanker 0.82 1.02
horizontal added mass calculation of floating rec-
tangular cross sections from Sedov (1965).
In our case a/b = 2.368, and by linear interpola-
Table 6. Global vertical frequencies (Eq. 12 results). tion, the obtained α 2 value is 1.734.
The horizontal added mass per unit length is
Mode Frequency (Hz) given by:
2N 0.446
α2
3N 1.025 Amh ρπ (T )2 (13)
4N 1.667 2
5N 2.354
ρ = 1.025/9.81(t ⋅ s2/m4)
⇒ Amh = 13.162(t ⋅ s2/m2)
2.5
Frequencies Comparison Hull girder mass per unit length: m = Δ/L
⇒ m = 22.878 (t ⋅ s2/m2)
2

Hull Girder horizontal bending frequency in air,

f ANSYS 1.5 calculated as a free-free 2N beam frequency in air:
(Hz)

fa = 0.811 Hz
fS

Corresponding frequency in a fluid (sea water):

0.5

1
0 fw = 0.5
fa (14)
2 2.5 3 3.5 4 4.5 5
⎛ Amh ⎞
⎜⎝1 + ⎟
Nd

m ⎠
Node Number

Figure 23. Frequency results—FE values (black solid

line) & Eq. 12 values (blue dash line). fw = 0.646 HZ
Higher modes horizontal bending natural fre-
Frequencies Comparison (Ratio)
quencies, were obtained from Johannessen & Skaar
1.04
(1980) expression:

1.03 Nnh = N2h(n − 1)μ (15)

(Frequencies Ratio)

R ANSYSS
1.02
where:
N2h is the hull girder 2N horizontal natural fre-
quency, n is the number of nodes.
1.01

Table 7. Horizontal parameter α2—rectangular

1
2 2.5 3 3.5 4 4.5 5
cross section.
Nd
Node Number
Ratio a/b α2
Figure 24. Frequency results ratio.
0.1 1.14
0.2 1.21
0.5 1.36
The first four vertical hull girder frequencies cal- 1.0 1.51
culated using Eq. (12) are shown in Table 6. 2.0 1.70
Frequency results from FE and Eq. 12 evaluations 5.0 1.98
(Table 1 and 6 data) are presented in Figure 23.
And the ratio FE480 / Eq. 12 frequencies are shown Notes: a → a = B/2 (B = 32.2 m); b → b = T
in Figure 24. (T = 6.8 m).

480

MARSTRUCT.indb 480 2/18/2011 5:52:18 PM

 It was assumed a typical block coefficient for the 1.004
Frequencies Comparison (Ratio)

ship types, and Table 8 summarizes the resulting

distribution. 1.002
By linear interpolation of Table 8 values,
considering a block coefficient of 1.0 (box girder

(Frequencies Ratio)
case), the derived μ value is 1.20. R ANSYSS
1

Therefore, the Equation (15) for the present box

girder case is: 0.998

Nnh = 0.646(n − 1)1.20 (16) 0.996

The first four horizontal hull girder frequencies

0.994
calculated using Eq. (16) are shown in Table 9. 2 2.5 3 3.5 4 4.5 5
Nd
Frequency results from FE and Eq. 16 evalu- Node Number

ations (Table 2 and 9 data) are presented in

Figure 25. Figure 26. Frequency results ratio.
And the ratio FE / Eq. 16 frequencies is shown
in Figure 26.
Therefore, in the horizontal case a very good 6.2 Frequency evalution based on Kumai
agreement between the Finite Element (FE) and formulation
the analytical frequency results is obtained. Kumai (1968) expressions for the estimation of the
two-node vertical bending hull girder frequencies
of ships, are as follows:
Table 8. μ parameter distribution.
⎛ 1 B⎞
Δv 1.2 Δ (17)
Ship type Block coefficient (Cb) μ ⎝ 3T⎠
Cargo ship 0.68 0.845 where:
Bulk carrier 0.80 1.00 Δv is the virtual displacement, including added
Tanker 0.82 1.02
mass of water, in kg.
Δ = ship displacement, in kg.
B = 32.2 m, T = 6.8 m.
Table 9. Global horizontal frequencies (Eq. 16 results).
0.5
Mode Frequency (Hz) ⎛ I ⎞
N2 v = φ ⎜ 3⎟
(18)
2N 0.646 ⎝ΔvL ⎠
3N 1.484
4N 2.414 where:
5N 3.41 φ = 1.61 × 106, I = 241.90 m4, L = 252.3 m.
N2v is the hull girder 2N vertical natural frequency,
in Hz.
Obtained N2v value from Eq. (18) is:
Frequencies Comparison
4

N2v = 0.498 Hz
3 And the Eq. (12), for the Kumai values becomes:
f ANSYS
Nnv = 0.498 (n − 1)1.20 (19)
(Hz)

fS 2

Calculating the first four vertical hull girder fre-

1
quencies using Eq. (19), the derived results are pre-
sented in Table 10.
Frequency results from FE and Eq. 19 eval-
0
2 2.5 3 3.5 4 4.5 5
uations (Table 1 and 10 data) are shown in
Nd Figure 27.
Node Number
A greater difference exists against the FE values,
Figure 25. Frequency results—FE values (black solid compared to the previous frequencies results
line) & Eq. 16 values (blue dash line). (in 6.1), in particular the N2v case.

481

MARSTRUCT.indb 481 2/18/2011 5:52:20 PM

 Table 10. Global vertical frequencies Figure 28 presents Figure 27 data in the form of
(Eq. 19 results). a ratio (ratio of FE / Eq. 19 frequencies).
In general terms, the frequencies obtained from
Mode Frequency (Hz) Kumai show an acceptable agreement in relation
2N 0.498 to the FE values.
3N 1.145 The above commentary is backed by the exist-
4N 1.862 ing data of FE calculations performed on ships
5N 2.629 and their comparison with the estimation from
Kumai´s formulas (N2v case). The obtained devia-
tions in the present evaluation are in line with such
values.
Frequencies Comparison
3

2.5 7 CONCLUSIONS

2 An evaluation to predict the global vibratory

f ANSYS
behaviour of a marine structure such as a ship
(Hz)

fK 1.5 was carried out. The marine structure considered

in the analysis was a simplified one, i.e. a rec-
1 tangular section hull girder (ship-like structure).
However, that element allowed us to highlight the
0.5
analysis main features involving the analysis to
be performed, namely: added mass water effect,
0
2 2.5 3 3.5 4 4.5 5 modal parameters and forced response analysis
Nd
Node Number
results.
The technique employed for analysis was based
Figure 27. Frequency results—FE values (black solid on the creation of a Finite Element (FE) model of
line) & Eq. 19 values (blue dash line). the whole system, i.e. surrounding water and struc-
ture, allowing a proper representation of the Fluid
Structure Interaction (FSI), that occurs due to the
0.93
Frequencies Comparison (Ratio) structure vibration. Calculations were performed
by using ANSYS code.
0.925 In addition, in order to have an indication of the
validity of the obtained FE results, a comparison
0.92 related to the hull girder natural frequencies was
(Frequencies Ratio)

carried out. The mentioned comparison, involved

0.915
R ANSYSK the calculation of hull girder natural frequencies
0.91
from analytical expressions. Two sources for such
expressions were used, Sedov (1965) and Kumai
0.905 (1968). Compared results showed a good agree-
ment, in particular the results employing Sedov
0.9
(1965) formulation (Box Hull Girder).
The introduction of acoustic fluid elements in
0.895
2 2.5 3 3.5 4 4.5 5 commercial codes like ANSYS, date from the nine-
Nd
Node Number ties, however code developers have carried out
periodic updates on the calculation algorithms,
Figure 28. Frequency results ratio (FE/Kumai freq.). and as a consequence, the presented method could
be considered a state of the art approach.
Finally, it should be noted that a usual task in
Such difference could be related to the back- a vibration study apart from the analytical side
ground formulation, Sedov expressions are based is the experimental side too. Consequently, the
on a box girder case (Cb = 1.0), however Kumai experimental data collected during sea trials shall
formulas are based on ships cases (Cb < 1.0), influ- allow to verify the obtained analytical response
encing the mode added mass value. data set validity or identify potential deficiencies if
The consequence is an “artificially” higher N2v exist, by both data sources correlation. Therefore,
frequency estimated value, employing Kumai´s confidence in the analytical results could be estab-
formula. lished. That regular correlation activity should be

482

MARSTRUCT.indb 482 2/18/2011 5:52:22 PM

 the appropriate route in achieving the method and
tool calibration goal.
REFERENCES
Aker, J.C., 2005, An Interpretation of the Physics of
Fluid-Structure Interaction in the Frequency Domain,
23rd CADFEM User’s Meeting, 2005.
Baker, A.J., 1983, Finite Element Computational Fluid
Mechanics, Mc Graw-Hill Book, New York.
Bathe, K.J., 1996, Finite Element Procedures, Prentice
Hall, Englewood Cliffs.
Benson, D.J., 1992, Computational Methods in
Lagrangian and Eulerian Hydrocodes. Comp. Meth.
Appl. Mech. Eng. 1992b, 99 (2–3): 235–394.
Cabos, C., & Ihlenburg, F., 2003, Vibrational Analysis of
Ships with coupled Finite and Boundary Elements,
Journal of Computational Acoustics, 11(1): 91–114.
Cook, R.D., 1981, Concepts and Applications of Finite
Element Analysis, John Wiley and Sons, New York.
483
MARSTRUCT.indb 483
Imgrund, M., 2008, ANSYS Manual, ANSYS Release 11.0.
Johannessen, H., & Skaar, K.T., 1980, “Guidelines for
Prevention of Excessive Ship Vibration”, SNAME
Transactions, Vol. 88.
Kumai, T., 1968, “On the Estimation of Natural
Frequencies of Vertical Vibration of Ships”, Report
of Research Institute for Applied Mechanics, Vol. 16,
No. 54.
Lewis, F.M., 1929, “The Inertia of the Water
Surrounding a Vibrating Ship”, SNAME Transactions,
Vol. 37.
Sedov, L.I., 1965, Two Dimensional Problems in Hydro-
dynamics and Aeronautics, John Wiley, Interscience,
New York.
Suoli, M., et al., 2004, ALE and Fluid Structure Interac-
tion, Materials Science Forum, Vols. 465–466.
Thomson, W.T., 1998, Theory of Vibration, Prentice
Hall, Upper Saddle River.
Zienkiewicz, O.C., Taylor, R.L. & Nithiarasu, P., 2006,
The Finite Element Method for Fluid Dynamics,
Elsevier.
2/18/2011 5:52:23 PM
 Experimental analysis of structures

MARSTRUCT.indb 485 2/18/2011 5:52:23 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

An experimental study on fatigue crack propagation life of T-joint

fillet specimen considering residual stress under storm loading

S.H. Kim, K.S. Kim, J.H. Lee, C.H. Yu & W.H. You
Department of Naval Architecture and Ocean Engineering, Inha University, Incheon, Korea

ABSTRACT: Fatigue crack propagation life assessment of a cargo hold is an overriding consideration
in the design of marine vessel such as crude oil, LNG, and LPG carriers. This paper studies the crack
propagation life of T-joint welded structure on which typical fatigue cracks are frequently initiated when
the marine vessels experience the storm load. Welding residual stresses are calculated to investigate its
effects on the fatigue life. Thereafter the residual stress distribution is applied to the AFGROW life
prediction program which incorporates the loading, the welding residual stress, and the geometric shape
of the structure. The fatigue tests of the T-joint welded specimen under storm loading show the beach
mark clearly generated on the cut section of the specimen. The crack propagation life estimated based
on the beach mark are compared with that of AFGROW to validate the life prediction. Evaluation of
the remaining service is discussed in order to allow the remaining fatigue lifetime to be estimated for the
marine vessel’s cargo hold if the random load or the storm load is introduced to the vessel.

1 INTRODUCTION to the most frequent experience the fatigue damage

Large welded structures such as ships and offshore
structure have many fatigue crack occurrence at
weld joints because of the stress concentrations
or weld defects. When these cracks are detected at
weld joints, it is necessary to assess the remaining
service life until the crack grows and leads to a
leakage of cargo oil or sea water on inner hull or
outer hull. It is also related with damage tolerance
design to guarantee a certain life to its complete
failure after the crack detection. In this respect,
the crack propagation approach plays a key role to
assess the remaining life and structural redundancy
and to establish the inspection planning. Also,
the reliability of fatigue life under harsh environ-
ments like storm condition comprising overload-
ing should be guaranteed, because the offshore
plants such as FPSO have to remain in designated
location to extract crude oil or natural gas for about
20 years. It has been studied extensively about the
occurrence of fatigue crack to estimate remain-
ing life at weld joints. Fatigue tests and fatigue
crack propagation analysis about welded parts
are carried out by Bartera et al. (2005). Kim et al.
(2004) analyzed the effects when overloads applied
in the middle of constant fatigue load to welded
structures, and proposed an improved formula for
crack propagation life.
In this paper, fatigue tests are carried out under
the storm loading in T-joint fillet which is reported
during the operation of ship and ocean structures.
Storm loading is generated based on the IACS34
wave scatter data with a storm model proposed by
Tomita et al (1995). Also, the experimental results and
analytical results according to comparison between
beach mark and analytical results are evaluated.
Finally, assessment of remaining crack propagation
life is suggested when the fatigue crack of T-joint fil-
let welding is initiated due to the storm loading.
2 FATIGUE TESTS
2.1 Specimen and material properties
The shape of specimen is T-joint and dimensions
of the specimen are shown in Figure 1. And the
material of the specimen was AH32 steel plate and
its mechanical properties are listed in Table 1.
2.2 Fatigue test condition
All test specimens were conducted in a servo-
hydraulic, closed-loop mechanical test machine
with 500 kN load capacity, which was interfaced
to a computer for machine control and data
acquisition. Also all experiments were performed
based on storm loading proposed by Tomita
et al. (1995), in air and at room temperature.
Figure 2 shows the time history of storm loading
applied to the specimen.

487

MARSTRUCT.indb 487 2/18/2011 5:52:23 PM

 Table 1. Mechanical properties of AH32. Figure 3 shows the load case of experiments to
assess the influence of storm loading, Experimental
Young’s Yield Ultimate conditions are summarized as follows:
modulus strength strength
– Step 1: The fatigue tests are divided into three
Type GPa MPa MPa cases under constant amplitude loading, test
AH32 211 315 459
results are compared with an existing method
and S-N curve is obtained.
– Step 2: Fatigue tests are performed on several
stress ranges and mean stresses.
– Step 3: Fatigue tests are performed on constant
storm blocks and varying mean stress.
– Step 4: Fatigue tests are performed on varying
mean stress and loading procedure considering
randomly actual sea state.

3 DEFINITION OF CRACK INITIATION

3.1 Measurement of strain and 5% strain

range drop
Figure 1. Dimension of specimen. Fatigue initiation life is generally estimated based
on S-N curve based on the nominal stress, hot spot
stress and notch stress. The approach based on hot
spot stress including the geometric stress concen-
tration is the most practical method for predicting
fatigue strength of welded joint. This study
measures structural stress based on linear extrapo-
lation based on the strain measured at 0.5 t and
1.5 t from the weld toe proposed by DNV-RP-203
(2008) (Figure 4). 5% strain range drop method
was used to determine the initial life. The crack ini-
tiation life is defined as the number of cycles where
strain range is reduced by 95% of initial strain
range due to decreased load transfer (Figure 5).
Figure 2. Time history of storm loading. 5% strain range drop method can give an accurate
initial life in fatigue test under constant amplitude
loading, while it draws an inaccurate conclusion in
fatigue test under variable amplitude loading.

3.2 Beach mark measurement method

Though crack length on the surface can be detected
by macroscopic observation during the fatigue test,
crack length through the specimen depth is invisible.

Figure 3. Experimental loading condition.

*Mean stress changes from the first magnitude after three
storm loading blocks are given to the specimen. Figure 4. Hot spot stress measurement.

488

MARSTRUCT.indb 488 2/18/2011 5:52:23 PM

 In order to calculate the three dimensional welding
residual stress, MSC.Marc was used. Chemical
compositions and mechanical properties of the
specimen are presented in Table 2. Also the elec-
trode of DW-55LSR which has diameter of 1.2 mm
is used for the experiments. Chemical compositions
and mechanical properties of the electrodes are
also presented in Table 3 and Table 4. Welding
conditions are presented Table 5.
Yoo et al. (2010) has performed the finite ele-
ment analysis of the welding residual stress for
the specimen tested by present study. Figure 7a
shows the temperature distribution and the
Figure 5. 5% strain range drop. residual stresses obtained by finite element analy-
sis. The residual stresses distribution at the hot
spot which generates the maximum stress con-
centration reaches the level of the yield stress
(Figure 7b).
Residual stress distribution shifts under repeated
loading. This phenomenon is called “Shake down
effect” and its schematic process is explained in
Figure 8. A is the location of the initial residual
stress that is similar to tensile yield stress. If Smax
applies to specimen, point A moves to point B.
If applied stress is removed, point B moves to
point C, not point A. Here again, if Smax1 applies
to specimen, point C moves, through B, to point D.
If applied stress is removed, point D finally moves
Figure 6a. Measurement of initial crack length. to point E. Residual stress of yield stress level is
released by several cyclic tensile loadings over the

Table 2. Chemical component of AH32.

C Si Mn P S Ni Cu Mo

0.15 0.14 0.91 0.19 0.44 0.01 0.02 0.02

Table 3. Chemical component of DW-55LSR.

Figure 6b. Measurement of initial crack length.
C Si Mn P S Ni
The markings are formed due to the retardation
0.04 0.36 1.37 0.008 0.008 1.4
of the crack growth subsequent to the overload.
Crack initiation is defined through visual obser-
vation when the crack length on the surface Table 4. Mechanical component of DW-55LSR.
reaches 15 mm. One beach mark interval is one
storm cycles block. Therefore crack growth rate Offset yield Ultimate
is obtained from the measurement of beach mark strength [MPa] strength [MPa]
space. Figure 6a–b show the beach mark inspected
on the specimen. DW-55LSR 550 585

Table 5. Welding condition of FCAW*.

4 RESIDUAL STRESS
Current Voltage Speed Arc efficiency
4.1 Residual stress analysis
A V cm/min %
Residual stress occurs at weld joints of ships
and offshore structure due to heating, melt- 230 27 50 80
ing and cooling. It has a significant impact on
crack initiation and propagation of structures. *Flux core arc welding.

489

MARSTRUCT.indb 489 2/18/2011 5:52:24 PM

 where Smax = maximum nominal stress; Smin =
minimum nominal stress; SR = weld residual stress;
K = stress concentration factor; and Sy = yield
strength.
Released residual stress which is calculated by
“shake down effect theory” applies to crack propa-
gation analysis with AFGROW software.

5 COMPARISON OF TEST RESULTS WITH

ANALYTICAL RESULTS OF AFGROW
Figure 7a. Residual stress analysis.
5.1 Experimental results
The total life is divided into crack initiation life
(Ni) and propagation life (Np) based on beach
mark measurement results. The beach mark meas-
urement method based on actual crack lengths is
more accurate than 5% strain range drop method.
When a crack initiates far away from the extrapo-
lated straight line of the strain gauge location,
crack initiation cycles from 5% strain range drop
are generally overestimated.
In step 2 experiments with the same mean
stress and different stress amplitude, the effect of
stress amplitude on crack initiation is dominant.
Figure 7b. Residual stress distribution. Comparison of specimens S2C1 and S2C2 shown in
Figure 9a explains that the portion of crack propa-
gation life is low when the whole fatigue life is large.
On the order hand, the portion of crack propaga-
tion life is high when the applied loading is large.
In step 3 experiments with a different mean
stress and the same stress amplitude, the effect
of mean stress is significant but the mean stress
shift is not dominant. In step 3 case 1, case 2 and
case 3, fatigue tests are carried out to evaluate the
mean stress effect (35 MPa, 0 MPa and −35 MPa)
under single storm loading at different mean stress.
The whole fatigue life increases according to the
reduced mean stress. But in the case of crack prop-
agation life based on beach mark, crack propaga-
Figure 8. Shake down effect. tion life doesn’t increase according to increasing
total fatigue life as shown in Figure 9b. Conse-
quently, if mean stress changes into compressive
course of this series. The residual stress relaxation
stress, initial life increases with the rate of total life.
can be explained as follows:
But crack propagation life is almost not affected by
1. In a loading cycle, if the applied stress, Smax, that mean stress changing into compressive stress.
is larger than formerly experienced maximum Meanwhile, in the case of S3C4 (70∼ −70 MPa),
stress, the residual stress, SR, is replaced by the S3C5 (35∼ −35 MPa) and S3C6 (70∼0 MPa) to
following value. confirm mean stress shift, crack initiation and
propagation life is almost not affected by mean
SR = Sy − K ⋅ Smax (1) stress shift. The mean stress shift means a mean
stress change after three storm loadings.
2. In a loading cycle, if SR + K ⋅ Smin is less than −SY, Tests of step 4 are performed under realistic
i.e. if the concerning location yields under com- storm loading. Storm order is randomly arranged
pression, the residual stress is given as to reflect real sea conditions. And different mean
stress is also tested as step 3. In S4C6, storm
SR = −Sy − K ⋅ Smin (2) order is arranged by lower storm to higher storm

490

MARSTRUCT.indb 490 2/18/2011 5:52:26 PM


Figure 9a. Step 2 experimental results.
Figure 9b. Step 3 experimental results.
Figure 9c. Step 4 experimental results.
(Random 2). All experiments of step 4 under
realistic storm loading show reasonable results as
shown in Figure 9c.
In the case of S4C1, 2, 3 and 4 to confirm mean
stress shift, if mean stress changes into compres-
sive stress, initial life increases.
Results of the Case 1 and Case 5 which are
70 MPa and 0 MPa show that crack initiation and
propagation life increase.
In the case of S4C1 and S4C6 to confirm the
applied loading order effect, it is difficult to
make a conclusion due to lack of the number of
experiments. But in case of the comparison with
the only previous two results, crack propagation
life of S4C6 which is arranged by lower storm
to higher storm shows more increase than S4C1.
If overloading is randomly applied, crack closure
phenomenon is expected to have an effect on crack
propagation life. This phenomenon appears due to
the relationship between crack closure and crack
491
MARSTRUCT.indb 491
opening stress intensity factor. In this regard,
Kikukawa et al. (1983) proves that Kop remains at
a high level without a decrease when overload is
applied continually by experiments. In S4C1, Kop
remains at a high level due to maximum stress of
randomly distributed storm blocks. Thus, Kop of
S4C1 is higher than Kop of S4C6 by crack closure
and applied loading of S4C1 decreases accord-
ing to decreasing effective stress intensity factor.
Finally, the portion of crack propagation life of
S4C1 is lower than that of S4C6.
5.2 Comparison of analysis results with beach
mark test
A crack growth analysis was performed by using
AFGROW (2006). AFGROW considers the
Forman equation, NASGROW, Harter T-method
and Walker equation which are crack propagation
theory. Material properties are selected from
database provided by AFGROW. In this study,
fatigue crack growth analysis is performed based
upon the NASGROW Equation of the fatigue
crack growth theories. NASGROW equation is
expressed as follow:
p
⎛ ΔKth ⎞
1−
⎤ ⎜⎝ ⎟
n
da ⎡( − f ) ΔK ⎠
=C⎢ ΔK ⎥ (3)
⎣ ( − R)
q
dN ⎦ ⎛ ΔK max ⎞
⎜⎝ 1 −
ΔK crit ⎟⎠
where a = crack length; N = fatigue life; C = Paris
crack growth rate constant; n = Paris exponent in
NASGROW; p & q = exponents in NASGROW;
R = stress ratio; f = Kop /Kmax; ΔK = stress inten-
sity range; ΔKth = threshold stress intensity range;
ΔKmax = maximum stress intensity range; and
ΔKcrit = critical stress intensity.
Figure 10–11 compare the crack lengths both
observed by beach mark test and analytical results
estimated by AFGROW.
AFGROW results are compared with the beach
mark measurement in order to confirm the resid-
ual stress effect as shown in Figure 11. Residual
stress is completely released by shake down effect
when applied stress is larger than 80% of yield
strength and load ratio is high (Table 6). In case of
step 3 case 2, if the residual stress doesn’t exist, the
life estimated by AFGROW is larger than that of
the experimental results. However, if the residual
stress obtained by FEA is considered, the life of
the AFGROW analysis results is similar to the life
of the experimental results. Results based on beach
mark are calculated such that inaccurate crack
lengths near initial crack are linearly extrapolated
by using the visible crack length near initial crack.
2/18/2011 5:52:27 PM
 As validity of AFGROW program was proved
by Kim et al. (2008), the residual stress can be cal-
culated by AFGROW at the propagated crack of
the structures.

6 PREDICTION OF FATIGUE LIFE USING

THE MODIFIED PARIS LAW

This study proposes that a method predicting

Figure 10. Crack growth rate both beach mark and
AFGROW without residual stress.
fatigue life is proposed based on the modified Paris
(1963) equation, when the crack length on the sur-
face reaches 15 mm. Paris law equation using the
stress intensity factor is modified by following
equations:

daX
( )
m
= C ( K )m = C α S π × aX (4)
dN
NX

ΔK α ΔS π a (5)

⎡ 1 ⎤ 1
2 ⋅ ⎢ (m − 2) / 2 ⎥ − (m − 2) / 2
⎣ ai ⎦ ain
ΔN = (6)
( m 2 )C ( ΔS )m ( )m / 2 ( )m

where aX = length to the X region; daX = length

of the X region; dNX = fatigue crack propagation
life of the X region; C = Paris crack growth rate
Figure 11. Crack growth rate both beach mark and constant; m = Paris exponent; C = redefined Paris
AFGROW with residual stress. crack growth rate constant; m = redefined Paris
exponent; ΔS = nominal stress range; α = geom-
Table 6. Residual stress relaxation. etry factor; ai = initial crack length (15 mm); and
ain = invisible crack length.
Residual stress Figure 12 shows the sectional view of the
specimen and Figure 13 explains the relationship
Case R* Ry** % between da/dN and ΔK in the case 1 of step 2.
X, Y and Z denote summation of the macro-
Step 2 1 −0.33 0.68 – scopic beach mark spaces, invisible crack length
2 −0.52 0.94 – and the defined initial crack length (15 mm),
3 −0.44 0.81 – respectively.
4 −0.52 0.94 – The equivalent stress (αΔS) considering geom-
5 −1 0.58 20 etry effects is calculated by daX, dNX, C and m
Step 3 1 −0.73 0.82 – (Equation 4). daX means the length of X. dNX
2 −1 0.71 15 denotes the crack propagation life of the X region.
3 −1.38 0.6 40 C and m are material properties.
4 −0.52/−1.93 0.94 – ΔK is defined by the Equation (5) based on the
5 −0.73/−1.38 0.82 – equivalent stress (αΔS) and crack length (a) of
6 −0.52/−1 0.94 – each beach mark along the X.
Step 4 1 −0.52 0.94 –
2 −0.46/−1.93 0.84 –
3 −0.69/−1.38 0.73 20
4 −0.46/−1 0.84 –
5 −1 0.71 20
6 −0.52 0.94 –

*Ratio of minimum stress/maximum stress. **Ratio of

maximum stress/yield stress. Figure 12. Section of the specimen in the case 1 of step 2.

492

MARSTRUCT.indb 492 2/18/2011 5:52:29 PM


Figure 13. Relationship between da/dN and ΔK in the
case 1 of step 2.
da/dN is also calculated by the cycle of storm
block and the length of one beach mark along the
X region.
It is possible to obtain the relationship between
da/dN and ΔK and extrapolate the Paris equation
which can be defined by new C and m. It is also
possible to calculate the fatigue propagation life of
the Y region by fatigue life equation (Equation 6).
Fatigue life of the Z region is total fatigue life minus
the crack propagation life of the X and Y region.
7 CONCLUSIONS
Fatigue tests are carried out under the variable
storm loading to estimate the remaining service life
of the T-joint welded structure. The remaining serv-
ice life obtained by fatigue tests is compared with
that of analytic method suggested by AFGROW.
The followings are found from this study:
– When a crack initiates far away from the extrap-
olated straight line of the strain gauge location,
crack initiation life from 5% strain range drop is
generally overestimated.
– The portion of crack propagation life increases
if the applied storm loading is high.
– If mean stress changes into compressive stress,
initiated crack life increases with the rate of
total life. But crack propagation life is almost
not affected by mean stress changing into
compressive stress.
– If storm blocks are randomly applied, the
portion of crack propagation life decreases
493
MARSTRUCT.indb 493
according to increasing crack opening stress
intensity factor.
– Welding residual stress should be considered
to estimate the remaining service life of
structures.
– A method predicting fatigue life of the defined
crack length under variable amplitude loading
is proposed based on the modified Paris
equation.
ACKNOWLEDGEMENTS
This work was supported by the research grant of
Samsung Heavy Industries and National Research
Foundation of Korea.
REFERENCES
Bartera, S, Molenta, L, Goldsmitha, N. & Jones, R.
2005. An Experimental Evaluation of Fatigue Crack
Growth. Engineering Failure Analysis 12(1): 99–128.
Det Norske Veritas. 2008. Fatigue Design of Offshore
Steel Structures. Recommended Practice RP-C203.
Det Norske Veritas, Norway.
Harter, J.A. 2006. AFGROW User Guide and Techni-
cal Manual. AFRL-VA-TR-2006-XXXX, Air Force
Research Laboratory in United States.
Kikukawa, M, Jono, M. & Kondo, Y. 1983. Fatigue
Crack Closure and Estimation Method of Crack
Propagation Rate under Stationary Varying Loading
Conditions including Random Loading (2nd Report,
Extension of the Method to High Propagation Rate
Regime). Japan Society of Mechanical Engineers
49–439A: 278–285.
Kim, K.S, Shim, C.S. & Cho, H.M. 2004. A Study on Eval-
uation of Overload Effects on Fatigue Crack Growth.
Key Engineering Materials 261–263: 1169–1178.
Kim, K.S, Shim, C.S, Kwon, Y.B, Ko, H.S, Ki, H.G. &
Viswanathan, K.K. 2008. A study of Crack Propaga-
tion and Fatigue Life Prediction on Welded Joints of
Ship Structure (II). Journal of the Society of Naval
Architects of Korea 45–6: 679–687.
Paris, P.C. & Erdogan, F. 1963. A Critical Analysis of
Crack Propagation Laws. Trans. ASME. J. Basic Eng.
D85: 528–534.
Tomita, Y, Matoba, M. & Kawabe, H. 1995. Fatigue
crack growth behavior under random loading model
simulating real encountered wave condition. Marine
Structures 8–4: 407–422.
Yoo, M.J, Lee, J.H, Hwang, S.Y, Kim, K.S. & Kim, S.C.
2010. A Study of Predicting 3-dimensional Weld-
ing Residual Stress Distribution for T-joint Fillet
Specimen. Journal of KWJS 28–2: 68–74.
2/18/2011 5:52:32 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Experimental study on slamming load of the trimaran

Sheng Peng, Huangxiang Sun & Jingxia Yue

Wuhan University of Technology, Wuhan, Hubei, China

Weiguo Wu
Wuhan University of Technology, Wuhan, Hubei, China
Key Laboratory of High-speed Ship Engineering, Ministry of Education, Wuhan, Hubei, China

ABSTRACT: In order to investigate slamming load distribution and its relation to the impact velocity,
a small scale trimaran model test was performed. During the water entry from a series of heights, the
peak pressure of the main hull, side hull and wet-deck were obtained. Based on the model test and the
similarity theory, this paper proposed a reasonable and effective prediction method of the trimaran
slamming loads.

1 INTRODUCTION numerical methods of fully nonlinear theory of

Trimaran, with its large deck area and excellent naviga-
tion performance, more and more world-widespread
attention is arisen by the shipping industry. Mean-
while, trimaran suffers wave slamming load during
the high-speed sailing caused by frequent water entry
and exit, especially in the cross structure where seri-
ously impact by the sputtered wave. Its unique struc-
ture makes the slamming study different from the
traditional wedge and plate water entry. Trimaran
slamming contains both wedged body (main-hull
and side hulls) and flat body (wet-deck) water entry.
The impact of structure rush into the water is a com-
plex fluid-structure interaction process. In theoretical
calculation and numerical simulation, there are many
approximations, such as the provision constraints of
the end of the structure, the assumption of ideal fluid
which is inviscid, incompressible and irrotational
flow, ignoring air cushion of the bottom and so on.
Thus, a more realistic test method can reveal law of
the trimaran water impact.
Predecessors have done similar experiments
in water entry. For example, Chuang (1966) first
found that the air cushion in the flat structure into
the water plays an important role, and completed a
series of tests and shot contact moments between
structure and the water surface by a high-speed cam-
era. To explore the relationship between slamming
pressure and velocity, Ochi (1970, 1973) did three
types of impact pressure measuring test, and draw
the slamming pressure of all the three tests are simi-
lar to be proportional to the square of the impact
speed. Norway MARINTEK Zhao and Faltinsen
(1996) did body of typical profile of the bow-flare
falling into the water experiments, the accuracy of
hydrodynamic was verified by comparing the calcu-
lated and experimental results of the impact pres-
sure and its distribution. Michael (2007) adopted
the method of combining the additional mass theo-
retical simulation and tests to study cross structure
slamming problems of a SWATH. Anders (2007)
obtained the bow slamming pressure of VISBY
stealth trimaran in a prototype testing, and returns
to slamming pressure formula of VISBY stealth tri-
maran according to test results. Cao (2008) meas-
ured the slamming pressure of the trimaran model
and its distribution at a free-fall water entry experi-
ment, and indicated that the pressure of the cross
structure was higher than the main hull.
In order to investigate the slamming load dis-
tribution and the relationship between the impact
velocity and the slamming load, a small scale tri-
maran model test campaign of water entry from
a series of heights is performed. Based on the
similarity theory, the peak pressure of the main
hull, side hull and wet-deck are obtained. And the
peak slamming pressure prediction formula of the
trimaran is elicited by regression analysis.
2 EXPERIMENTAL MODEL DESIGN
2.1 Similarity analysis of the water entry model
2.1.1 Decision of the basic quantities
The slamming loads of trimaran are mainly fluid
force, inertia force and moment of inertia, accord-
ing to the principle of the physical parameters
selecting, pressure P and bending moment M are
selected to represent the basic quantities describing
the external physical loads.

495

MARSTRUCT.indb 495 2/18/2011 5:52:34 PM

 In the stress test, to measure the normal stress test conditions should be followed. So get the
and shear stress, σ is selected to be representa- following π relationship:
tive basic quantity. Elastic modulus E, Poisson's
ratio μ and stiffness D, describing the basic physi- ML3 P ⎛ Et 2 I m V
cal properties of structural materials, only two of σ , σ = f ⎜⎝ ρL2 , μ , L3δ , ρL3 , δ 0 , gL ,
them are independent. As mentioned above, tri-
maran model is divided into thin-shelled beams γ P Li ⎞ (4)
model, so describing the physical characteristics , , ⎟
L ρ E ρV 2 L ⎠
are liner geometric, inertia moment I and thickness
geometric δ.
Parameters of low-speed water entry test are: Similarity index:
geometric parameters l1, l2, l3, which represent
2
geometry features of body impacting the water CM CP C EC t 2
= 1, 1, = 1, C P C ρCV
and characteristic length L; the center of grav- 3
C L Cσ Cσ C ρC L
2

ity coordinates of the falling body xc, yc, zc, the Cm

3
mass m; The initial kinematic parameters, such CI C L Cδ , 3
= 1, CV = C gC L C Li C L (5)
as entry angle θ0, initial attack angle δ0, ini- C ρC L
tial velocity v0, acceleration of gravity g, water Cμ 1, Cδ 0 = 1, C Fr 1 Cγ = C L C EC ρ
density ρ and kinematic viscosity γ. Impact the
water, unsteady movement; it must also include
the time t. Because of the nature of momentum, 2.1.3 Conversion between model and the real ship
inertial force must be considered, and guarantee If meet the formula (5) criteria, in order to ensure
an equal Froude number while Reynolds number hydrodynamic similarity, Froude number must be
can be ignored. equal, there is:

2.1.2 Similarity criterion VS = Vm CL (6)

As mentioned above, the general functional
describing the slamming process of trimaran as
VS: speed of the prototype trimaran
follow (Detailed in Appendix 1):
Vm: speed of the model trimaran
CL: geometry scaling ratio
⎛M P ⎞
f ⎜ , ,E, μ I v0 m δ 0 ,gg ρ γ t, Fr ⎟ = 0 (1) Meet the dynamic similarity, there is:
⎝σ σ i
⎠

Equation (1) can be written as the π exponential CP C ρCV 2 CL means PS /Pm = CL (7)
term as follow:
PS: the slamming pressure of prototype trimaran
α1 α2 Pm: the slamming pressure of model trimaran
⎛M⎞ ⎛ P⎞
π =⎜σ ⎟ ⎜⎝ σ ⎟⎠ E α 3 μα 4 α5
i
α6 α7
0
α8
δ 0α 9
⎝ ⎠ That is guaranteed to derive the slamming pres-
sure P of the prototype trimaran using this scale
×g α 10 α 11 α 12 α 13
ρ t γ Fr α 114
(2) based on meeting the similarity criteria:

Similarity criteria can be directly written by π PS = Pm * CL

matrix as following:
2.2 The design of similar model
ML3 P Et 2
π1 = ; π2 = σ ; π3 = 2 ; In this experiment, guided by the similarity theory,
σ ρL the model design is based on the structure type
I m provided by the actual prototype structure and
π 4 μ π5 = 3 ; π6 = 3 ; laboratory tests experience. The length scale ratio
Lδ ρL
(3) between prototype and the model is 6:1, while the
V γ
π 7 = Fr = π8 = ; thickness ratio is 2:1. Model total length is 1.0 m,
gL
L L ρE breadth is 2.4 m and depth is 0.9 m.
Li P After determining the geometric scale ratio, a
π9 ; π 10 δ ; π 11 =
L ρV 2 typical cross section and two longitudinal profiles
were designed to meet similarity criteria. The axis
In the Equation (3), π1, π 2 are non-qualitative ratio of the profile was set by 1:6, and the inertia
criteria, and π3, π9 are qualitative criteria, which ratio of the section moment is 1:432, meetting with

496

MARSTRUCT.indb 496 2/18/2011 5:52:34 PM

 can ensure the self-correction of free-fall posture
when the model fall with the car along the track
into the water (see the Fig. 4).
Test equipments include piezoelectric pressure
sensor, photoelectric switch, B & K3560D & PULS
data acquisition and analysis system, high speed
digital cameras and related equipment supply and
cord. Pressure sensors arrange in the main hull,
side hull and cross structure of the trimaran, and
test the slamming pressure. Photoelectric switches
can measured the actual speed at the moment of
model attaching to the water surface.
Figure 1. Typical cross-section structure and sensor
arrangement of the model.
3.2 Experiment principle and procedues
In the experiment, to simulate slamming phenom-
enon, the trimaran model falls into the still water
surface to get the relative impact speed. It is in a
freefall at a height of 0.75 m, 1 m, 1.25 m, 1.5 m,
1.75 m, 2.0 m, 2.25 m, 2.5 m, 2.75 m, 3.00 m above
the water surface. Photoelectric switch is used to
test the impact speed (see Table 1) of the trimaradn
model and pressure transmitter is used to test the
slamming pressure of main hull and cross struc-
ture of trimaran model.
Figure 2. Trimaran model. It would be tested 5 times in every height, and
debug the instrument to Test Status before every
test. Trimaran model is lifted to specified alti-
tude assignment, and then the model falls down
impacting the water surface. In the process of
falling, the model first passes through the pho-
toelectric switch which generates photoelectric
signal and transfer the signal to the collection

Figure 3. Sensors arrangement.

the slimlarity index: CI = CL3Cδ . Figure 1 respec-

tively model cross section diagram, where B stands
for the width of the main hull and b stands for
the width of the cross-structure. Figures 2 and 3,
respectively, the internal structure as a model built
drawings and models.

3 SLAMMING EXPERIMENT
Figure 4. Experimental set-up arrangements.
3.1 Experimental set-up
Test setup includes the water tank, test tower, carry-
Table 1. Experimental and calculated impact speed.
ing car and a winch with manual clutch. The external
dimension of the pool is 8.90 m × 4.90 m × 3.15 m, Drop height (m) 0.75 1.00 1.25 1.50 1.75
while the inner size is 8.00 m × 4.00 m × 3.15 m. Test V (m/s) 2.264 2.452 2.668 2.836 3.123
tower, 6 meters heights with four parallel tracks, is Drop height (m) 2.00 2.25 2.50 2.75 3.00
fixed on the top of the test tank. The carrying car V (m/s) 3.344 3.619 3.968 4.236 4.432
which connected with the model is fixed through
eight guide wheel mounted within the track, which Where V stands for the relative impact speed.

497

MARSTRUCT.indb 497 2/18/2011 5:52:37 PM

 Figure 6. The slamming pressure time history of one
Figure 5. The moment of the model impacting the test point of the main hull (drop height 3.00 m).
water surface.

instrument, and the collection instrument will

record the signal automatically. The model keeps
on falling, part of main hull impact the water
surface. Waves, splashed by the water entry of
the main hull, sputtering the cross structure and
side hulls, bursting out the sound of a loud slam-
ming at the same time. Meanwhile, the collection
instrument record the I-V converter time-history
from the pressure transmitter and acceleration
Figure 7. The slamming peak pressure distribution
transducer. After the first experiment, lift the along the horizontal position of the main hull.
trimaran model to specified altitude assignment
and do the next experiment when the water sur-
face. The process of an experiment is shown in
Figure 5.

4 EXPERIMENTAL PHENOMENA
AND RESULTS’ ANALYSIS

In the experiment, main hull of the model slammed

the water surface at first. The high speed caused
high waves which splashed to the side hulls. With
the immersed depth increasing, the waves splash to
Figure 8. The slamming pressure time history of one
center part of the side hull and the cross structure.
test point of the cross sturcturel (drop height 3.00 m).
Then the side hull fall into the water. At last the
cross structure fell into the water with waves
splashing in all directions.

4.1 Slamming pressure of main hull

and cross structure
Figure 6 is the time history graph of one measuring
point slamming pressure of the main hull, which
shows that the slamming pressure boost to the
summit, and last a extremely short time. Figure 7
shows the slamming peak pressure distribution of
the main hull. With the drop height increases, the Figure 9. The slamming peak pressure distribution
peak slamming pressure increases as the impact along the horizontal position of the cross structure.
water speed increases; slamming pressure peaked
in the B/10 to the vertical section of the main hull.
Figure 8 is a slamming pressure time history of appears second peak, which caused by the wave-
one test point of the cross structure at the drop induced sputtering of side hull.
height of 3.0 m. As the slamming occurs, the Figure 9 shows the slamming peak pressure
pressure increases rapidly to peak, the role of the distribution of the cross structure. With the drop
process is very short; followed by cross structure height increases, the peak slamming pressure

498

MARSTRUCT.indb 498 2/18/2011 5:52:39 PM

 increases as the impact water speed increases;
slamming pressure peaked in the the 5b/8 of the
cross structure, while the minimum value occurs in
the b/5∼2b/5 region of the cross wet deck.
4.2 Regression of slamming pressure equation
The slamming pressure of the trimaran is deter-
mined by the impact water speed, in addition to
other nonlinear factors such as air cushions, splash
effect and so on, which have great different influ-
ence on the slamming pressure in various impact
speeds. Theoretical and experimental studies have
shown that the slamming peak pressure is linear
essentially with the square of the impact speed, so
the slamming pressure of trimaran can be used the
formula as followed:
Pm k ρV 2 (8)
where:
Pm represents the slamming peak pressure, Pa
ρ represents the water density kg/m3;
V represents the impact speed, m/s
k the factor changed with speed, reflecting the
various nonlinear factors effect on the slamming.
In this paper, the regression analysis of the coef-
ficient k related with the slamming pressure of the
trimaran is completed based on a series of tests.
Figure 10 and Figure 11, respectively, show the
Figure 10. Relationship between k value and the impact
speed of main hull.
Figure 11. Relationship between k value and the impact
speed of cross structure.
499
MARSTRUCT.indb 499
relationships between the k values and impact speed
of the main hull and cross structure, reflecting
that the nonlinear factors, such as the air cushion,
splash effect, decrease the effect on the slamming
peak pressure as the impact speed increases.
Exponential regression of k values formula was
as follows:
Main hull:
k . . 8 e −v / 1.83776
−v / 1.83854
+ 33.05073e (9)
Cross structure:
k . . 8 8e −v / 0.97863
−v / 0.97864
+ 101.58418e (10)
Figure 12 and Figure 13, show k values regres-
sion curves of the main hull and cross structure,
reflecting that the k value would be exponentially
decreasing trend with the impact speed increases.
The k value decreases rapidly as speed increases
at low speed, while slow down at high speed. The
change trend of k value of the trimaran reflects
Figure 12. k values regression curve of main hull.
Figure 13. k values regression curve of cross structure.
2/18/2011 5:52:41 PM
 5 CONCLUSIONS

According to the experimental research, it comes

to the conclusions:
1. In terms of the experimental results and the
regression analysis, elicit the prediction formula
of slamming peak pressure of trimaran and
obtain the law: peak pressure is proportional to
the square of the impact speed, and the effects
of nonlinear factors, such as air cushion and
Figure 14. Typical cross section of CAO’s model. splashing, is quadratic declining with increasing
of impact speed.
2. Slamming pressure of main hull and cross struc-
Table 2. The dimensions comparison of the models.
ture increase with the increasing of the impact
speed. The pressure of main hull is greater than
Model Cao Peng cross structure.
3. The maximum of slamming pressure is at the
B (Width of the main hull, mm) 600 900 place which is at a distance of B/10 from longitu-
b (Width of the cross 400 450 dinal section in center plane, and the maximum
structure, mm) slamming pressure of cross wet-deck is at the
δ0 (Initial attack angle, degree) 35 10 place of 5b/8 in the width of cross structure.
Max pressure of the 0.51 0.23 4. The slamming pressure of the cross structure
main hull, v = 4.0 m/s (Mpa) (3B/8) (B/10) could be smaller with appropriate combina-
Max pressure of the cross 0.66 0.15 tion of cross structure’s width and the deadrise
structure, v = 4.0 m/s (Mpa) (6B/8) (5b/8) angles.

the effect on the slamming pressure coursed by
a variety of factors on different water impact.
Previous studies suggest that various factors effect-
ing on slamming pressure is constant, reflecting the
k value is constant in the calculation, which would
cause some errors in the prediction of slamming
pressure on different impact speed. So it must be
extremely careful to use this method to predict the
slamming pressure.
The pressure of the cross structure is lower
than the main hull, which is in contradic-
tion to the experimental result of Cao (2008).
Figure 14 shows the typical cross section of the
Cao’s model, and Table 2 compares the dimen-
sions and results of two models. From the scenes
at the moment of water entry, it may cause by the
cancellation of the spray induced by the main hull
and the side hull interaction, which to function
as the bulbous bow in reducing the wave-making
resistance.
REFERENCES
Anders Rosén, Karl Garme, & Jakob Kuttenkeuler.
2007. Full-Scale Design Evaluation of the Visby Class
Corvette. Ninth International Conference on Fast Sea
Transportation, Shanghai:583–588.
Cao Zhenglin. 2008. Research on the slamming load of
trimaran [PhD thesis]. Wuhan: Wuhan University of
Technology.
Chuang, S.L. 1966. Experiment on flat-bottom slam-
ming. Journal of Ship Research. 10(1):10–27.
Michael R, Davis, James R. 2007. Computation of wet
deck bow slam loads for catamaran arched cross
sections. Ocean Engineering. 34:C2265–C2276.
Ochi MK, Motter LE. 1973. Prediction of slamming
characteristics and hull responses for ship design.
SNAME.
Ochi MD, Bonilla-Norat J. 1970. Pressure-velocity rela-
tionship in impact of a ship model dropped onto the
water surface and in slamming in waves. AD-709071.
Zhao R, Falinsen O.M. 1996. Water entry of arbitrary two-
dimensional sections with and without flow separation.
Proc. 21st Symp. on Naval Hydrodynamics.

500

MARSTRUCT.indb 500 2/18/2011 5:52:43 PM

 Appendix 1. Dimensions of physical quantities and similar constants.

Physical quantities Dimension Similar constants

Material attribute Stress σ −2

[FL ] CE
Strain ε – 1
Modulus of elasiticity E [FL−2] CE
Poisson’s ratio μ – 1
Density ρ FL−4T2 CE/C1
Geometry Linear scaling l L C1
Thickness scaling δ L C1
Load Pressure P FL−2 CE
Bending moment M [FL]
Acceleration g [LT−2] 1
Velocity v LT−1 C11/2
Time t [T] C11/2
Rotary viscosity υ [L2T−1] C13/2
Kinematic viscosity μ [FL−2T]
Froude number Fr V / gl – 1
Reynolds number Re – 1

501

MARSTRUCT.indb 501 2/18/2011 5:52:45 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Failures mode analysis of corroded steel structures subjected

to compressive load

S. Saad-Eldeen, Y. Garbatov & C. Guedes Soares

Centre for Marine Technology and Engineering (CENTEC), Technical University of Lisbon, Instituto Superior
Técnico, Lisboa, Portugal

ABSTRACT: This paper analyses the failure modes based on experimental test results of multispan
corroded box-girders, representing amidships ship hull section and subjected to four-point loading.
Comparative study for three tested corroded box girder specimens with respect to collapse modes, strain
measurements, residual stresses, load-displacement and moment-curvature relationships and the effect of
different corrosion levels on structural integrity are performed.

1 INTRODUCTION carried out to analyse the behaviour of ship hull

Along the ship service life, the ship’s hull is exposed
to different types of loads and subjected to differ-
ent damage scenarios, including corrosion deterio-
ration that reduces the structural capacity.
Corrosion deterioration reduces the strength
capacity and needs to be evaluated accounting for
the long-term structural integrity under conditions
of high temperature, inappropriate ventilation,
high stress concentration and high stress cycling.
Very high rates of corrosion are achieved in spaces
such as ballast tanks and at specific structural
details such as horizontal stringers or longitudi-
nal and web frames. Depending on the location of
the ship structural elements, corrosion deteriora-
tion characteristics are different. Many corrosion
depth measurements have been published as for an
example by Loseth et al. (1994), Yamamoto and
Ikegami (1998), Paik et al. (2003a,b) and Garbatov
and Guedes Soares (2008, 2009).
In the past decade, there was a major world-
wide concern about the continuous loss of ships
and the major contributing factor to the cause is
considered to be structural failure. The hull girder
ultimate strength is an explicit control of the most
critical failure mode for large ships.
Many research works for intact marine struc-
tures have been carried out for analysing the
ultimate strength. One of the first works in this
field was performed by Caldwell, 1965, after that
a method for ultimate strength assessment was
developed by Smith, 1977, including the effect of
buckling collapse of compressed members.
Experimental results provide first-hand infor-
mation for understanding the collapse behaviours
of structures, therefore, intensive works have been
by a box-girder specimen loaded up to its ultimate
strength limit. The box-girder specimen has been
designed to reproduce the midship section of ship’s
hull subjected to a pure bending moment. Typical
elements of the box-girder are plates, stiffeners and
transfer frames.
Various box-girders have been tested during
the last decades. Reckling (1979) carried out col-
lapse tests on seven box girders models. Ostapenko
(1981) performed a test on three box girders models
subjected to combined bending, shear and torque.
Nishihara (1984) tested eight box girder models
representing conventional type of ships as tanker,
bulk carrier and container ship. All the specimens
were subjected to pure bending.
Dow (1991) performed a collapse test on
1/3-scale welded steel hull girder model of a frigate
under sagging bending moment. Mansour et al.
(1990) carried out experimental tests for two
large-scale models, which represented two types
of ships.
Gordo and Guedes Soares (2004, 2008) reported
experimental results of the ultimate strength of
box girders. Four-point loading was applied to
obtain a pure constant bending throughout the
whole specimens. The second study demonstrated
the behaviour of box girders made of different
materials but with the same arrangement to have
a good basis of comparison and to prove experi-
mentally the effect of material properties on the
collapse strength. The conclusion was that the
structural efficiency of the stiffened panel in com-
pression made of high-tensile steel is lower than the
one made of mild steel. However the high-tensile
steel model obtained a good global efficiency than
the mild steel box.

503

MARSTRUCT.indb 503 2/18/2011 5:52:45 PM

 A continuation of the previous works is the one
performed by Gordo and Guedes Soares (2009)
where three box girders subjected to pure bending
moment were tested. The specimens were made of
very high-tensile steel reinforced with bar stiffeners
of the same material. The box girder was subjected
to pure bending moment, inducing tension on the
bottom and compression on the top of the box.
All of the mentioned experimental works were
dealing with intact structures. The step forward is
the study of ultimate strength of aged box gird-
ers performed by Saad-Eldeen et al. (2010a), who
carried out a test of an initially corroded multi-
span stiffened box girder representing a midship
section of ship hull. The box girder was subjected
to four-point loading resulting to a uniform bend-
ing moment along the specimen and the ultimate
bending moment of the tested box girder has been
compared with empirical formulae showing a good
agreement.
An continuation of the previous study is the one
conducted by Saad-Eldeen et al. (2010b) where a
box girder made of mild steel subjected to moder-
ate corrosion was tested. The experimental results
and the collapse modes have been analyzed and
confirmed the expected the global deformation
shape and strength capacity as a result of corro-
sion deterioration and the reduction of the resid-
ual stresses.
An extension of the corroded box girder test was
presented by Saad-Eldeen et al. (2010c) where a
severely corroded box girder was also tested under
the same conditions. Corrosion thickness meas-
urements have been performed and the corrosion
wastage has been fitted to non-linear time variant
degradation model. It was concluded that the load
carrying capacity and ultimate bending moment
are highly affected by corrosion deterioration of
plating and the material properties changes.
The objective of this paper is to analyse the dif-
ferent failure modes based on experimental results
of multispan corroded box-girders, represent-
ing amidships ship hull section and subjected to
four-point loading. Comparative study for three
tested corroded box girder specimens with respect
to collapse modes, strain measurements, residual
stresses, load-displacement and moment-curvature
relationships and the effect of different corrosion
levels on structural integrity are performed.
2 EXPERIMENTAL SET UP
The experimental results of the ultimate strength
test presented here are for specimens shaped
as a box girder with length, breadth and depth
of 1400 mm, 800 mm and 600 mm respectively
(see Figure 1).
504
MARSTRUCT.indb 504
Figure 1. Box girder, experimental set up.
2.1 Corrosion test
Before the ultimate strength test, the box girders
were exposed to Baltic seawater without any
corrosion protection system (no coating). The
box girders were placed in large tank and seawa-
ter was pumped into the tanks continuously. The
temperature of seawater was increased and addi-
tionally oxygen depolarization sub process rate
was increased by agitation of seawater which
resulted in corrosion rate increase. Detailed infor-
mation about the corrosion set up is presented by
Domzalicki et al. (2009).
A survey and measurement of corrosion thick-
ness on the three corroded box girders plating
has been carried out to collect a data base for
the strength analysis. The corrosion data consists
of 636 measurements of corrosion thickness,
212 measurements from each box Saad-Eldeen
et al. (2010b).
The three deteriorated box girders have such a
configuration so that they may idealize the midship
section of tanker ships. The mean value of the meas-
ured corrosion depth of the deck plates (0.41, 2.31
and 2.62 mm) for initially, moderately and severely
corroded box girders respectively are compared
with the upper limit of the 95 percent confidence
interval of the corrosion depth defined by the
regression equations developed by Garbatov et al.
(2007), for deck plate of ballast tanks of real tanker
ships, revels that the initially corroded box match
the 0.2 year of deterioration, while the moderately
and severely corroded boxes are related to 17.9 and
23.3 years of deterioration respectively considering
that the coating life is 0 year.
2.2 Ultimate strength test
The box girders consist of 3 bays. The use of
multiple-bays instead of single bay allows to have
more realistic representation of ship hull and to
2/18/2011 5:52:45 PM
 Table 1. Principal characteristics (average) of box girders.

Initially Moderately Severely

Item corroded corroded corroded Units

Deck Plating 4.1 2.2 1.9 [mm]

Stiffener 25 × 4.4 24 × 4.1 23 × 2.7 [mm]
S.S. Plating 4.0 2.7 2.5 [mm]
Stiffener 25 × 4.3 24 × 4.1 21 × 3.1 [mm]
P.S. Plating 3.9 3.5 2.9 [mm]
Stiffener 25 × 4.3 24 × 4.1 21 × 3.2 [mm]
Bottom Plating 3.8 1.9 1.8 [mm]
Stiffener 25 × 4.4 24 × 4.2 21 × 2.5 [mm]

avoid the effect of the boundary conditions for
the central plates, which is due to the possible
eccentricity of load and the interference between
adjacent panels.
The deck panel is stiffened with five longitudinal
flat bars with a spacing of 150 mm. The side panel
is stiffened with two stiffeners on a distance of 300
and 500 mm respectively and the bottom panel
was stiffened with one stiffener in the middle, as
may be seen in Figure 1. The box girders are made
of normal ship building steel with yield stress of
235 MPa.
The geometry configurations of the analysed
box girders are given in Table 1. The thickness
measurement survey has been performed in the
way to find out the remaining thicknesses of the
box girders structural elements. The as-built thick-
ness of the plates and stiffeners are 4.5 mm. The
corrosion thickness is not uniformly distributed
which affects the response of the box girders.
The box girder has been mounted between two
stiff supporting arms, using bolt connections. The
box girder was subjected to four-point vertical load-
ing. The bottom part is subjected to tension and the
upper part, deck, is under compression. The bend-
ing moment is kept constant along the box girder,
between the bolt connections. There are four points
for transmitting the load, two are located at the
supports of the arms and two are on the boundary
between the box-girder and the supporting arms.
To avoid shear effects, the load is applied on
a certain distance away from the ends of the box
girder (the connections between the box girder and
the supporting arms). The load is applied using
a 700 kN hydraulic jack and it is transmitted to
the box girder through a horizontal beam, as are
shown in Figure 1.
3 EXPERIMENTAL RESULTS
During the service life of the ship, the struc-
ture deteriorates resulting in structural elements
thicknesses diminution, which reduces the ultimate
strength.
3.1 Modes of collapse
The collapse modes of the plate panels under pre-
dominately compressive loads usually undergo
three main types, namely plate buckling, inter-
frame flexural buckling (stiffener tripping) and
overall grillage collapse (plate and stiffener), see
Figure 2. In some cases there is an interaction
between two collapse modes, this interaction gen-
erally reduces the load-carrying capacity corre-
sponding to a single buckling mode.
In the pre-buckling stage, the plate’s response
to load follows the Hooke’s Law where its load
displacement relationship is linear. When com-
pressive loads reach the critical buckling load,
buckling occurs and the buckling profile of the
plate usually in the form of half-waves of approxi-
mately equal length.
Flat bar stiffeners provide very little torsional
restraint, therefore, if the plate between stiffen-
ers buckles, the tripping starts to take place in
the stiffener. Finally, the overall collapse mode
involves the global buckling of plating, stiffeners
and transverses.
The first collapse mode observed for three boxes
was overall pre-buckling of the plating and stiffen-
ers as one unit (see Figure 3), however, in this mode
each box behaves in different shape, due to the dif-
ferent level of corrosion deterioration present in
structures.
As can be seen from Figure 3(a), for initially
corroded box, only the bay no. 2 and the bay no. 3
are affected. However, the bay no. 3 in which the
achieved lateral buckling displacement in the
direction of the plate is bigger than the one in
the second bay in the direction of the stiffener.
Therefore, the global deformation was asym-
metrical with respect to the middle section of the
box girder, and this can be explained with the fact
that the corrosion depth distribution was non

505

MARSTRUCT.indb 505 2/18/2011 5:52:46 PM

 uniform around the bay no. 3 which makes it the
weakest one.
For the moderately corroded box (see Fig. 3b),
the collapse shape is well formed, the bay no. 1, 3
buckled towards the plating and the bay no. 2
towards the stiffener. The global deformation
was almost symmetrical with respect to the mid-
dle section of the box girder, which may explain
the diminution of the residual stresses during the
loading-unloading initial cycle. Additional to that
is the loss of the plating thickness at the bay no. 2,
which is much less than for other bays.
Before the progressive collapse takes place, some
local modes appear in each box girder. All the
Figure 2. Collapse modes. modes relate to interframe tripping lateral torsional
buckling of the stiffeners with the formation of the
plastic hinges in both plating and stiffeners in the
deck and side panels, as may be seen in Figure 4 for
all of the box girders.
The formation of the plastic hinges in all stiff-
eners is explained with the fact that they were
subjected to additional load and as a result of
that the stresses raised close to yielding. After the
local redistribution of loading associated with the
formation of the first plastic hinge, the load was
transferred to the nearest stiffener promoting the
simultaneous collapse of the panel.
This type of collapse is a typical column mode
Figure 3a. Overall pre-buckling mode, initially cor- collapse related to short span stiffened panels.
roded box.

Figure 4a. Interframe buckling mode (stiffeners trip-

Figure 3b. Overall pre-buckling mode, moderately cor- ping), initially corroded box.
roded box.

Figure 3c. Overall pre-buckling mode, severely cor- Figure 5a. Interframe buckling mode (stiffeners trip-
roded box. ping), moderately corroded box.

506

MARSTRUCT.indb 506 2/18/2011 5:52:46 PM

 The collapse of the compressed panel propagates
to the side structure, leading to the collapse of its
plating. Some plastic hinges also occurred in the side
stiffeners.
Due to the severe corrosion, plating perfora-
tion occurred in some locations in bottom and
deck, the severely corroded box responses in a
weak way. As the load is increasing, the formation
of the plastic hinges becomes visible especially at
the bay no. 3 (deck/port side connection), which
forced the plating to tear (see Figure 7). The reason Figure 8a. Overall collapse mode, initially corroded box.
for tearing from one side (port side) is that, there
was a localized severe corrosion in the deck plating
near to the port side, also the bottom plating was
perforated which leads to transmit the most of the
load to the deck panel. After the local tearing of
the deck plating occurred, the deck panel plating
lost the most of its stiffness resulting to increases
of stresses. The resulting higher stresses provoked
the simultaneous fracture of the stiffeners, as may
be seen in Figure 7.
Through the final load step, the box girders fold
up in a progressive collapse (overall grillage buck-
ling) defined with a great discharge of load accom- Figure 9b. Overall collapse mode, moderately cor-
panied by large deformation due to the formation roded box.
of plastic hinges in both plating and stiffeners at
the deck and at the deck/side connection as is pre-
sented in Figure 10.
It has to be pointed out that the net sections
did not remained plane during the plastic stage.
The axial and rotational deformations are more

Figure 10c. Overall collapse mode, severely cor-

roded box.

severe in the middle of both stiffeners and plating

Figure 6c. Interframe buckling mode (stiffeners trip-
ping), severely corroded box.
Figure 7. Severely corroded box, bay no. 3.
in the top and in the side panel of the bay no. 3
for initially and severely corroded and bay no. 2 for
moderately corroded box as shown in Figure 10.
3.2 Strain measurements
Twenty-two strain gauges have been connected to
each box. For the gauges mounted almost on the
same location in the three boxes around the cen-
tre line, i.e. at bay no. 3, the strain readings are
analyzed. Strain sign can be expressed according
to the deformation direction; therefore, positive
and negative signs belong to upward and down-
ward deflection of plates respectively. As shown in
Figure 11, for the initially corroded box, the gauges
record large stain with negative sign which reflect
the large downward deformation happened in the
bay no. 3 as may be seen in Figure 10(a).

507

MARSTRUCT.indb 507 2/18/2011 5:52:50 PM

 For the moderately corroded box the gauge
recorded a negative sign which reflect the downward
deformation happened in bay no. 3. For the severely
corroded box the gauges, in general, recoded posi-
tive values which gives an indication that the defor-
mation is upward as presented in Figure 10(c).
It is clear that, the general behaviour of each
gauge demonstrate the general collapse mode for
its box at bay no. 3, for example, for the moderately
box after certain loading the gauge start to record
a constant strain which gives an indication that the
bay started to be almost flat due to the shifting of
the buckling wave crest towards the central bay as
may be seen in Figure 10(b).

3.3 Residual stresses Figure 12. Initial loading cycles-residual stresses.

During the manufacturing, as a result of welding
processes, residual stresses are generated. Theses
Table 2. Load-displacement and moment-curvature
stresses can be released during the service life as a relationship.
result of loading/unloading processes. An attend
was performed to release these stresses before the Initial
ultimate strength experimental tests by conduct- cycles
ing initial loading cycles followed by total load area, Load-disp., Moment-cur.,
discharge in the elastic stress range. As reported in [kN.m] [kN, mm] [kN.m, rad/m]
Saad-Eldeen et al. (2010a–c), during the tests initial
loading cycles have been applied to release the resid- Box Area Load Disp. BM Cur.
ual stresses. The plot of the conducting initial load- Initial 4016 580.55 28.57 568.94 0.0059
ing cycles for the three boxes is shown in Figure 12. Moderate 1150 389.77 30.82 381.98 0.011
As may be seen from Figure 12, as the corrosion Severe 331 205.11 23.01 201.01 0.0075
level increases, the area under the load-displacement
curve decrease, which reflect the diminution of the
existing residual stresses, i.e. with respect to the ini-
tially corroded box girder, the area under the curve
decreases with 67.80% and 91.75% for moderately
and severely corroded box girders respectively.

Figure 13. Load-displacement relationship.

This gives an indication about the diminution of

the remaining residual stresses as a result of both
loading/unloading and corrosion.

3.4 Load-displacement relationship

Figure 11. Local strain vs. vertical displacement for The experimental results of the load-displacement
deck panel at bay no. 3. relationship are tabulated in Table 2. For the

508

MARSTRUCT.indb 508 2/18/2011 5:52:53 PM

 Table 3. Descriptors of structural tangent flexural rigidity.
Initial
Linear 1.89E + 05
7.34E + 04
Nonlinear −8.57E + 03
Figure 14. Moment-curvature relationships.
loading carrying capacity, it is evident that in the
elastic range (see Figure 13) at any certain load, the
displacement increases by increasing the corrosion
degradation level especially for severely corroded
box. Also by increasing the corrosion level the ulti-
mate carrying capacity that the box can withstand
decreases with 32.86% and 64.67% for moder-
ate and severe corrosion respectively as presented
in Table 2.
It may be observed from Figure 13 that there
is an initial drop in the load-displacement curve
for initially corroded box, which is following by
the collapse, the second drop down appears to be
associated with the spread of plasticity through
the second bay and the third bay. On the contrary,
the behaviour of the other boxes is different due
to the corrosion degradation. Furthermore, it is
noticeable that, the higher the corrosion level the
lower the elastic range.
3.5 Moment-curvature relationship
The experimental results of the moment-curvature
relationship are given in Table 2. In general,
before buckling and yielding occur, the moment-
curvature relationship is linear until the occur-
rence of any of them. For the tested box girders it
MARSTRUCT.indb 509
Moderate Severe Units
8.51E + 04 4.50E + 04 [kN.m2]
2.08E + 04 2.09E + 04
−7.38E + 03 −7.15E + 03
is noticeable that as the corrosion level increases,
the linear range becomes less.
As the vertical bending moment increases the
curvature increase as may be seen in Figure 14.
As a result of different corrosion levels, the bend-
ing moment that the box can withstand is highly
affected, with the presence of moderate and severe
corrosion the ultimate bending moment decreases
with 32.86% and 64.67% respectively.
It may be also observed from Figure 14 that,
the tangent flexural rigidity in both linear and
nonlinear region decreases as the corrosion level
increases. The calculated tangent flexural rigidity
for the three box girders are given in Error! Not a
valid bookmark self-reference. As a result of mod-
erate and severe corrosion, in the elastic range,
the tangent flexural rigidity decreases with 55%,
76.19% respectively.
4 CONCLUSIONS
Failure mode assessment based on the test results
of three corroded box girders has been presented.
The effect of corrosion deterioration on the col-
lapse modes and the overall behaviour of the box
girders have been analysed. Strain gauges reading
on several locations have been analysed and com-
pared for the three corroded boxes. The effect of
corrosion deterioration on reducing the existing
residual stresses has been observed and analysed.
With the presence of corrosion, the ultimate bend-
ing moment as well as the load carrying capac-
ity of the box girders decreased with 32.86% and
64.67% for moderately and severely corroded box
girders respectively.
ACKNOWLEDGEMENTS
The work reported here is a contribution to
the activities of the MARSTRUCT VIRTUAL
INSTITUTE, (www.marstruct-vi.com) in particu-
lar its Technical Subcommittee 2.3 on Ultimate
Strength.
The first author has been funded by the Portu-
guese Foundation for Science and Technology
(Fundação para a Ciência e Tecnologia—FCT)
under contract SFRH/BD/46790/2008.
509
2/18/2011 5:52:54 PM
 REFERENCES
Caldwell, J.B. 1965. Ultimate Longitudinal Strength.
Transactions RINA 107: 411–430.
Domzalicki, P. Skalski, I. Guedes Soares, C. &
Garbatov, Y. 2009. Large Scale Corrosion Test.
Analysis and Design of Marine Structures. Guedes
Soares, C. and Das, P.K. London, UK, Taylor &
Francis Group: 193–198.
Dow, R.S. 1991. Testing and Analysis of 1/3-scale Welded
Steel Frigate Model. Proceedings of the Interna-
tional Conference on Advances in Marine Structures,
Dunfermline, Scotland; 749–973.
Garbatov, Y. & Guedes Soares, C. 2009. Optimal Main-
tenance for Corroded Deck Tanker Ship Structures.
Reliability Engineering and System Safety 94(11):
1806–1817.
Garbatov, Y. & Guedes Soares, C. 2008. Corrosion
Wastage Modeling of Deteriorated Ship Structures.
International Shipbuilding Progress 55: 109–125.
Garbatov, Y. Guedes Soares, C. & Wang, G. 2007.
Non-linear Time Dependent Corrosion Wastage of
Deck Plates of Ballast and Cargo Tanks of Tankers.
Journal of Offshore Mechanics and Arctic Engineering
129(No 1): 48–55.
Gordo, J.M. & Guedes Soares, C. 2004. Experimental
Evaluation of the Ultimate Bending Moment of a
Box Girder. Marine Systems & Ocean Technology 1:
33–46.
Gordo, J.M. & Guedes Soares, C. 2008. Experimental
Evaluation of the Behavior of a Mild Steel Box Girder
under Bending Moment. Ships and offshore struc-
tures 3: 347–358.
Gordo, J.M. & Guedes Soares, C. 2009. Tests on Ultimate
Strength of Hull Box Girders Made of High-Tensile
Steel. Marine Structures 22: 770–790.
Loseth, R. Sekkesaeter, G. & Valsgard, S. 1994.
Economics of High—Tensile Steel in Ship Hulls.
Marine Structures 7: 31–50.
Mansour, A.E. Yang, J.M. & Thayamballi, A. 1990. An
Experimental Investigation of Ship Hull Ultimate
Strength Transactions SNAME 98: 411–439.
510
MARSTRUCT.indb 510
Nishihara, S. 1984. Ultimate Longitudinal Strength of
Midship Cross Section. Naval Architecture & Ocean
Engineering 22: 200–214.
Ostapenko, A. 1981. Strength of Ship Hull Girders
under Moment, Shear and Torque. Proceedings
SSC-SNAME Symposium on Extreme Loads
Response, Arlington, USA; 149–166.
Paik, J. Lee, J. Hwang, J. & Park, Y. 2003a. A Time-
Dependent Corrosion Wastage Model for the Struc-
tures of Single and Double Hull Tankers and FSOs.
Marine Technology 40: 201–217.
Paik, J. Thayamballi, A. Park, Y. & Hwang, J. 2003b.
A Time-dependent Corrosion Wastage Model for Bulk
Carrier Structures. International Journal of Maritime
Engineering 145(Part A2): 61–87.
Reckling, K.A. 1979. Behaviour of Box Girders under
Bending and Shear. Proceedings of the 7th Interna-
tional Ship and Offshore Structures Congress (ISSC),
Paris, France; II.2.46–9.
Saad-Eldeen, S. Garbatov, Y. & Guedes Soares, C.
2010a. Experimental Assessment of the Ultimate
Strength of a Box Girder Subjected to four-point
Bending Moment. Proceedings of the 11th Interna-
tional Symposium on Practical Design of Ships and
other Floating Structures (PRADS2010), ASME,
New York.
Saad-Eldeen, S. Guedes Soares, C. & Garbatov, Y. 2010b.
Compressive Strength Assessment of a Moderately
Corroded Box Girder. [submitted for publication].
Saad-Eldeen, S. Garbatov, Y. & Guedes Soares, C. 2010c.
Experimental Assessment of the Ultimate Strength of
a Box Girder Subjected to Severe Corrosion. [submit-
ted for publication].
Smith, C.S. 1977. Influence of Local Compressive Fail-
ure on Ultimate Longitudinal Strength of A Ship’s
Hull. Proceedings of 3rd International Symposium on
Practical Design in shipbuilding, Tokyo; 73–79.
Yamamoto, N. & Ikegami, K. 1998. A Study on the
Degradation of Coating and Corrosion of Ships
Hull Based on the Probabilistic Approach. Journal of
Offshore Mechanics and Arctic Engineering 120(3):
121–128.
2/18/2011 5:52:55 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Fatigue tests of butt welds and plates edges of 80 mm thick plates

H. von Selle & O. Doerk

Germanischer Lloyd SE, Germany

J.K. Kang & J.H. Kim

Daewoo Shipbuilding & Marine Engineering Co., Ltd., Korea

ABSTRACT: Within a joint development project between the Korean Shipyard Daewoo Shipbuilding &
Marine Engineering Co., Ltd (DSME) and Germanischer Lloyd (GL) fatigue tests of 80 mm thick plates
have been performed. The objective of the fatigue tests was to investigate the effect of different post
treatment methods on the fatigue performance. Test specimens were butt welds as well as plate edges in
untreated and post treated conditions respectively. While for the butt weld specimens ultrasonic impact
treatment (UP) was applied as post treatment method the plate edge specimens were treated by chamfering
and different combinations of edge grinding, surface grinding and UP surface treatment. Material was
higher tensile steel HT40 for all specimens. The results of the fatigue tests are discussed and compared
with the values and recommendations from the Rules and corresponding guidelines.

1 SCOPE OF INVESTIGATION The benefit which can be addressed for design

The increasing size of container vessels is
accompanied by an increase of the plate thicknesses
in the upper flange of hull girder, e.g. for the coaming
and coaming top plate as well as the extensive use of
higher tensile steel. Both, increasing plate thicknesses
as well as an increased yield strength, make that the
fatigue strength becomes a limiting factor for the
design. The application of special treatment meth-
ods that result in an improved fatigue strength seems
to be one possibility to overcome this challenge.
Within a Joint Development Project (JDP)
between DAEWOO Shipbuilding & Marine Engi-
neering Co., Ltd (DSME) and Germanischer
Lloyd (GL), fatigue tests were carried out in order
to investigate the effects of different treatment
methods on the fatigue strength of welded butt
joints as well as of cut edges. The material, the
plate thickness, the cutting and welding processes
as well as the post treatment procedures of the test
specimens have been chosen to be as close to the
shipbuilding practice as possible.
It is well known from literature, e.g. from
the IIW recommendations (Hobbacher, A.
2009, Haagensen, P.J. & Maddox, S.J. 2009) and
from various IIW documents (Weich, I. 2009,
Maddox, S.J. 2010, Marquis, G. & Björk, T. 2008)
that the fatigue strength of welded joints can be
significantly improved by post weld treatment
techniques. A widely spread technique is grinding.
The improving effect is the smoothening of the
notches, either of weld transition or of plate edge.
is given in Rules and Recommendations, e.g.
Hobbacher, A. (2009), Haagensen, P.J. & Maddox,
S.J. (2009), Germanischer Lloyd. (2010).
It has to be emphasized that the effect of post
weld treatment is mainly proven for constant ampli-
tude loading. In Marquis, G. & Björk, T. (2008) it
has been found the benefit decreases significantly if
structures are spectrum loaded with e.g. a straight line
spectrum which is typical for wave induced loads.
The ultrasonic hammer peening method UP
for welded joints is also well known, but it is not
so frequently used as grinding. Up to now it is not
included in Rules and Recommendations. Fields of
applications are old bridges with the aim of refitting
welded structures. In ship building industry it is not
common practice. Because the literature does not
cover the application to thick plates, this test pro-
gram has been established to overcome this lack of
information. Tools for UP are shown in Figure 1.
Figure 1. Tools for ultrasonic peening UP.

511

MARSTRUCT.indb 511 2/18/2011 5:52:55 PM

 The effect of the UP method is based on
two different mechanisms, the smoothening of
the weld toe radius and the undercut as well
as the degradation of tensile and introduction
of compressive residual stresses. Generally the
requirement for achieving a fatigue strength
improvement by post weld treatment techniques is
an effective application of the particular treatment
method within the manufacturing process.
The statements given before are also valid for
cut edges with the restriction that ultrasonic ham-
mer peening is rather uncommon and has not been
tested in the past.
The overall test program has been subdivided into
• butt weld (16 specimens)
• cutted edges (32 specimens)
The butt welds have been tested in as-welded
condition as well as in post weld treated condition
using ultrasonic peening.
For the cutted edges several treatment methods
have been investigated as corner treatment by
grinding on the one hand and cut surface treatment
by disc grinding as well as by ultrasonic peening on
the other hand.

2 BUTT WELD FATIGUE TESTS

The main objective of the butt weld fatigue tests
was to investigate the effect of ultrasonic impact
peening (UP) as a post weld treatment method on
the fatigue strength.
Within the JDP altogether 16 butt weld specimens
were tested, 8 specimens in as-welded condition
and 8 specimens in UP treated conditions. The tests
were carried out under constant amplitude axial
load on the 12 MN resonance testing machine of
the Department of Ferrous Metallurgy (IEHK) at
the Rheinisch Westfälische Technische Hochschule
(RWTH) in Aachen (Germany). The test frequency
was approximately 33 Hz.
To ensure test conditions that correspond to
the actual loading conditions of the coaming
and coaming top plate of large container vessels
the tests were carried out under axial loading and
stress ratio R = +0.1. The specimen thickness of
80 mm as well as the higher tensile strength steel
E40 and the electro gas welding (EGW) used for
the specimen fabrication are also typical for the
upper hatch coaming area of large container
vessels. Figure 2 shows a sketch of the butt weld
specimens. The data for the base material is given
in Table 1. Table 2 shows the parameters of the
welding process, consumables and condition.
In delivery conditions some specimen showed a
horizontal deflection and notches at the specimen
edges near the weld bead due to thermal cutting.
Figure 2. Geometry of the butt weld specimens.
In order to avoid horizontal vibrations during
the tests in the resonance testing machine and to
prevent crack start at the edge notches mentioned
above, the specimen edges were milled. This meas-
ure reduced the specimen breadth from originally
120 mm to approx. 100 mm. The welded and UP
treated specimens were rounded in the welded area
to force the crack initiation location towards the
centre of the specimens.
The axial and the angular misalignment of all
butt weld specimens were measured by laser scan-
ning at the Institute of Ship Structural Design and
Analysis of the Hamburg University of Technol-
ogy (TUHH) before testing. Figures 2 and 3 show
typical scans of the weld surface for a specimen in
as-welded condition and in UP treated condition
respectively. As a result of the UP treatment at the
weld toes rounded channels have been generated.
As expected for the plate thickness of 80 mm the
measured misalignments were comparatively low.
The maximum of the measured axial misalign-
ments of the specimens is 1% of the plate thickness
(mean value 0.4%) and the maximum angular dis-
tortion is 0.5° (mean value 0.14°). More details and
conclusions regarding the misalignment influence
are given in the following chapters 2.1–2.3.

512

MARSTRUCT.indb 512 2/18/2011 5:52:56 PM

 The misalignments mentioned before lead to

0.35
Ceq
additional stresses in the specimens when axially
loaded and they are superimposed by additional

0.012
static mean stress due to clamping of the specimen

TI
in the testing machine. To account for all stress
components during tests, all specimens were fitted

–
with strain gauges.
For all fatigue tests of the butt welded specimens
S-AL

0.045
the final rupture of the specimen was employed as
the failure criterion. In this way the comparability
0.018 between the test results in as-welded conditions
and the test results in UP treated conditions is
Nb

ensured. Furthermore for fatigue tests of small

scale specimens this failure criterion is commonly
0.0
V

used in the corresponding Rules and guidelines.

In order to get information of the crack initiation
Mo

0.0

location and the crack propagation phase all fatigue

tests of the butt weld specimens have been carried
out with a special two-level load history resulting in
0.03
Cr

beachmarks on the fatigue crack surfaces.

After a certain amount of load cycles, depending
0.02

on the load level, the applied stress range was

Ni

reduced to approximately 50% of the initial value for

a small number of load cycles, e.g. 5000. This was
0.01
Cu

realized by increasing the lower stress level while the

upper while the upper stress was kept constant.
0.02

These reduced load cycles do not contribute

S

notably to the damage sum of the specimen but

results in clearly visible beachmarks on the fatigue
0.08
Chemical composition [%]

crack surface which can be further evaluated.

P

Figure 4 shows a fracture surface with beachmarks

Actual material data for the base material of the butt weld specimens.

on the fatigue crack surface.

1.55
Mn

2.1 Tests of the butt weld specimens in as-welded

0.14
Si

conditions
The fatigue tests of the specimens in as-welded
0.08

condition have been carried out at different load

C

levels. This method, the so called pearl chain

Impact test

method, enables a reliable consideration of the

383 (AVE)
−40°C [J]

slope m of the regarding S-N curve. The stress

ratio of the tests was approximately R = 0.1.
At all specimen fatigue cracks initiated at one
of the upper weld toes close to but not directly at
E.L.

28%

the edges (see Fig. 4). This behaviour is typical for

butt weld specimens. Additional fine mesh finite
element investigation showed that there is a stress
512 MPa

peak close to the edges of the specimens which

determines the crack initiation location in the
T.S.

transverse direction of the weld toes. The specimen

shape in combination with the weld reinforcement
Tensile test

408 MPa

is responsible for this.

By means of a statistical evaluation of the
Y.P.

fatigue test results corresponding S-N curves of the

butt weld specimens in as-welded condition have
GL E40

been calculated (Fig. 5). Due to the small amount

Table 1.

Name

of test results with n < 10 a fixed slope of m = 3.0

has been assumed for the S-N curves according to

513

MARSTRUCT.indb 513 2/18/2011 5:52:57 PM

 Table 2. Welding process, consumables and conditions.

Current Voltage Speed Heat input

Process Brand name Position Diameter Manufacturer (A) (V) (cm/min) (KJ/cm)

DSW-50GTF Face side 1.6 mm 375 39 3.1 608

EGW KOBELCO
Tandem
DSW-50GTR Root side 1.6 mm 400 42

mainly due to the unintended high maximum load

stress in the range of the actual yield strength of
the material of approximately

σmax = 400 N/mm2.

In addition the contribution of high tensile

residual stresses to the overall load at the weld toes
has to be considered in this respect.
As a result of the statistical evaluation of the test
results the characteristic value ΔσR (or FAT-class) of
the S-N curve at N = 2 ⋅ 106 load cycles and a prob-
ability of survival of PoS = 97.7% is calculated to be

ΔσR = 106 N/mm2.

The scatter ratio between the number of cycles

Figure 3. Laser scan of typical weld surfaces in as-welded
(top) and UP treated condition (bottom).
Figure 4. Fracture surface with beachmarks.
for 10 and 90% probability of survival is TN = 1:1.89.
This is a relatively small value but within the usual
range observed at fatigue tests of welded joints.
From the basic S-N curve of the test results
(Fig. 5) a design S-N curve can be determined by
considering following influence factors:
• residual stress state,
• effect of the actual plate thickness,
• additional bending stresses due to misalignment.
According to the recommendations of the IIW
(Hobbacher, A. 2009) small scale fatigue tests
should be carried out at a high stress ratio of
R ≈ +0.5 to simulate high tensile residual stresses
as they are present in a real structure. In practise it
is normally not possible to fulfil this requirement
as the yield stress would be exceeded significantly
with respect to the load cycle range usually aspired

the IIW recommendations (Hobbacher, A. 2009).

The design S-N curve has been calculated as the
mean curve minus two standard deviations resulting
in a probability of survival of 97.7%. As it is com-
mon test practise S-N curves are also given for
probability of survival levels of PoS = 10%, 50% and
90%. The specimen U6 has been excluded from the
evaluation because the test showed low cycle fatigue
behaviour. During the test extensive yielding at the
weld toes accompanied by corresponding cyclic Figure 5. Basic S-N curve of the butt welds in as-welded
plastic deformations could be observed. This was condition.

514

MARSTRUCT.indb 514 2/18/2011 5:52:57 PM

 within fatigue tests. Therefore it is alternatively 2.2 Tests of the ultrasonic peened butt weld
recommended to reduce the fatigue strength at specimens
2 ⋅ 106 load cycles by 20% if the tests were carried
The fatigue tests of the ultrasonic peened (UP)
out at a stress ratio of R = 0.
butt weld specimens were carried out similar to
The standard FAT-classes or detail categories as
the fatigue tests of the specimens in as-welded
given in Germanischer Lloyd (2010) contain addi-
condition. Again the nominal stress ratio of the
tional bending stresses due to axial and angular mis-
tests was chosen to be R = +0.1.
alignment to a certain amount. For normal quality
The measured average stress magnification factor
butt welds a stress magnification factor of km = 1.3
due to additional bending stresses due to axial and
is already covered by the corresponding FAT-class
angular misalignments was km = 1.13. The observed
or detail category ΔσR = 80 N/mm2 (type A3).
stress ratio was R = +0.15 as a mean value of all
From the strain gauge measurements of the
tests. This means that both the stress magnification
specimens in as-welded conditions an average stress
factor as well as the stress ratio were in the same
ratio of the tests was evaluated to be R = +0.14.
range as for the specimens in as-welded condition.
This increase against the nominal stress ratio at the
Contrary to the fatigue tests of the specimens
tests of R = 0.1 is due to additional clamping and
in as-welded condition the fatigue tests of the
bending stresses of the misaligned specimens. The
UP treated specimens showed different crack
average stress magnification factor for the consid-
initiation locations. This can be interpreted as a
eration of misalignments already covered in the
result of the UP treatment. While at three speci-
fatigue tests was measured to be km = 1.1.
mens failure started at the weld toe and the weld
In the present test results of the 80 mm thick
root respectively, one specimen showed a failure
specimens the thickness influence for the actual
starting at a small surface defect beside the weld
thickness is already included. In the standard
area. The remaining two specimens showed a fail-
FAT class or detail category from the rules
ure initiated at the clamping. This means that for
ΔσR = 80 N/mm2, that is based on a reference thick-
five from eight specimens the weld transition was
ness of 25 mm, this effect has to be considered by
not the critical location in respect of fatigue. In
a separate thickness effect factor ft. According to
other words, due to post welded treatment other
Hobbacher, A. (2009) the magnitude of the thick-
notches became more severe than the weld toe
ness effect factor for 80 mm thick butt weld joints
undercut or weld root. An example for a failure
in as-welded condition is
starting at the clamping is shown in Figure 6.
The S-N curve of the fatigue tests of the
⎛ t ⎞n UP treated butt weld specimens is presented in
ft = ⎜ ⎟
⎝ 25 mm ⎠ Figure 7. For the same reason as for the as-welded
specimens a fixed slope of m = 3 has been assumed
for the evaluation. As a result of the statistical
with n = 0.2 for butt welds resulting in ft = 0.79.
evaluation of the test results the detail category or
Finally this means that the fatigue test results of the
FAT-class was determined to be
specimens in as-welded condition have to be modified
as follows, if it is intended to compare the test results ΔσR = 91 N/mm2.
directly with the corresponding FAT-class or detail
category ΔσR = 80 N/mm2 given in Hobbacher, A. It is commonly known that for UP treated
(2009) and Germanischer Lloyd (2010). specimens the slope of the S-N curve decreases
Factors on fatigue strength accounting for
• The stress ratio R: 0.87 (R = +0.14 in the tests →
R = +0.5 acc. to IIW recommendations)
• Misalignment: 0.83 (km = 1.1 in the tests →
km = 1.3 covered by FAT classes)
• Thickness effect: 1.26 (=1/ft) (t = 80 mm in the
tests → t = 25 mm reference thickness).
The product of these factors is 0.92, which
means that the basic test results have to be lowered
by 8% for a comparison with the basic FAT classes.
For the actual fatigue tests of the butt weld joints
in as-welded condition this results in a FAT-class
or detail category of
Figure 6. Failure starting at the clamping of an UP
ΔσR = 97 N/mm2. treated butt weld specimens.

515

MARSTRUCT.indb 515 2/18/2011 5:52:59 PM


Figure 7. S-N curve of the UP treated butt weld
specimens.
leading especially to a benefit in the high cycle
area. If the above mentioned limitations due to
the small number of specimens are not taken into
account the following results can be derived from
the evaluation of the UP treated butt weld tests:
Slope of the S-N curve obtained directly from
the tests: m = 8.23; detail category (FAT class):
ΔσR = 138 N/mm2.
One reason for this low characteristic value of the
S-N curve is the relatively large scatter TN = 1:4.21
of the test results due to the different failure mech-
anisms observed during the fatigue tests. It is ques-
tionable to treat all specimens statistically as one
population. Therefore it is not recommendable to
derive FAT-classes or detail categories ΔσR fulfilling
the requirements acc. to Hobbacher, A. (2009) from
these fatigue test results. In this respect it should be
noticed that there were no recommendations found
regarding the magnitude of the thickness effect for
UP treated butt welds. Up to now this topic has not
been investigated and is therefore a future task to
enable reliable fatigue strength assessment of UP
treated welds in thick wall structures.
When discussing the results it has to be con-
sidered that all specimens showing a clamping
failure are run out results in respect to the fatigue
assessment of the weld and the efficiency of the
UP treatment method. Modifications regarding
the stress ratio and the misalignment state were not
included in the test results presented in Figure 7.
Generally the test results of the UP treated speci-
mens can be divided into two parts. One part where
the UP treatment of the weld toes was effective and
another part where the UP treatment of the weld
toes was not effective. This is discussed in the fol-
lowing chapter more in detail.
2.3 Comparison of UP treated and as-welded
specimens
In order to investigate the effect of UP treatment on
the fatigue strength of butt weld joints the fatigue
tests results of the specimens in as-welded and UP
516
MARSTRUCT.indb 516
treated condition are condensed into one common
S-N diagram which is presented in Figure 8. Again
modifications of the test results due to the stress
ratio and the misalignment state have not been
considered here for both series.
This indicates that the post weld treatment proc-
ess was not stable in any case. In difference to this
UP treated specimens where the failure did not
initiate in the weld area but at the clamping or at
surface defects away from the weld seam showed
a significantly enlarged life time. In this respect it
should be mentioned that these test results have
to be assessed as run outs regarding the fatigue
strength of the weld.
If the characteristic design value of the S-N
curve of the specimens in as-welded condition is
directly compared with those of the UP treated
specimens the latter shows decreased fatigue
strength. At first view this is unexpected as the
mean values (PoS = 50%) do not show this tendency.
But it can be explained by the larger scatter band
for the UP treated series.
Basically the fatigue tests showed a behaviour
that can be divided into two different performance
groups. The first performance group consists of all
specimens in as-welded conditions and of those
UP specimens where the UP treatment was not
successful or effective. All UP treated specimens
that failed from the weld toe and the weld root
belong to this group.
On the other hand all UP treated specimens that
failed from the clamping or from surface defects
away from the weld which means that the UP treat-
ment was successful and effective are assigned to
the second performance group.
Generally it was not possible to make a decision
weather the UP treatment was successful or not by
visual inspection. So, in respect of the UP treatment
as a technique for the improvement of the fatigue
strength of butt weld joints the quality assurance
becomes one of the most challenging tasks. If it is
intended to utilize the beneficial effect of UP treat-
ment on the fatigue strength it has to be ensured
Figure 8. S-N diagram of butt weld specimens in as-welded
and UP treated condition.
2/18/2011 5:53:01 PM
 that the UP treatment is, under practical condi-
tions, applied in a successful and effective way.
3 CUT EDGE FATIGUE TESTS
Within the JDP 32 cut edge specimens have been
tested in total. The tests have been performed for
four series with 8 specimens each as follows (see
also Figure 9)
• 1C, corner chamfering specimens
• 2R, corner rounded specimens
• 2R + G, corner rounded + surface grinded
specimens
• 2R + UP, corner rounded + surface treated by
ultrasonic peening
The tests were carried out under constant ampli-
tude bending load on a 690 kN resonance testing
machine at Hamburg University of Technology
(TUHH), Ship Structural Design and Analysis.
The testing frequency was about 30 Hz. The test-
ing arrangement is shown in Figure 10. For mate-
rial data see Table 1.
The load has been chosen in a way that the
tension side has been loaded at a mean stress ratio
of R = 0.1.
Figure 9. Corner preparations by grinding.
Figure 10. Four point bending arrangement.
517
MARSTRUCT.indb 517
Due to the positive effect of compressive mean
stress on fatigue, crack initiation has been expected
on the tension side of the specimens.
The influence of the mean stress on the fatigue
strength acc. to Germanischer Lloyd (2010) is
shown in Figure 11.
To ensure a sound clamping in the testing
machine the specimens have been grinded in way
of supports. Strain gauges have been applied on
the tension side of the specimen which is the side
where crack initiation is likely to occur. The gauges
have been applied in the middle layer as well as
close to the specimen corners to measure addi-
tional bending stresses over the specimen height
(see Figure 10). The stress ranges at mid layer have
been adjusted to the target values according to
the test program. As the specimens and therefore
the testing arrangement was not fully symmet-
ric additional bending at top and bottom in the
range of +/−50 N/mm2 has been measured. This
results in an increased or decreased stress range
at the specimens corners respectively. As the test-
ing arrangement did not allow a visual inspection
of the specimen surface when the testing machine
was running the testing frequency as well as the
signals from strain gauges have been used as crack
initiation indicators. In most cases a frequency
decrease has been recognized up from a very early
testing state.
With respect to the testing arrangement the
tests have been stopped when the cracks have been
propagated through 2/3 of the specimen width. The
remaining life time can be expected to be small so
the number of cycles is taken as a failure criterion.
For all cut edge specimens the applied load levels
as well as the corresponding number of load cycles
within the fatigue tests are presented in the S-N
curve in Figure 12.
The slope has been derived from the tests of all
cut edge series, based on a common evaluation for
all post treatment methods. As it has been done
for the tests of the butt weld specimens the S-N
curves are given for probability of survival levels
of PoS = 10%, 50% and 90%. Again the so-called
design S-N curves corresponding to PoS = 97.7%
Figure 11. Bonus factor for compressive mean stress on
fatigue according to Germanischer Lloyd. (2010).
2/18/2011 5:53:02 PM
 Figure 12. S-N Curve of cut edge specimens—all series Figure 13. Evaluation of FAT classes for different treat-
(slope estimated from tests). ment methods and slopes.

and the associated FAT classes (value at 2 ⋅ 106 load Table 3. Results of the fatigue tests.
cycles) are presented in addition. To enable a direct
comparison of the test results with the actual fatigue FAT Class detail ΔσR [N/mm2]
assessment S-N curve for grinded cut edges from
the Rules of Germanischer Lloyd (2010) these are Series m = 3.5 m = 4.0 m from test
also included in Figure 12 (dotted curve). Here the
1C 169 182 180 (m = 3.9)
increase of the fatigue strength due to the application
2R 176 186 143 (m = 2.5)
of higher tensile steel has been considered by means
of the material influence factor fm. According to 2R + G 196 207 190 (m = 3.3)
the Rules of Germanischer Lloyd (2010) fm = 1.13 2R + UP 143 160 207 (m = 6.2)
for HT40 steel resulting in modified FAT class of All 168 181 176 (m = 3.8)
FAT 158 (m = 4.0) instead of FAT 140 (m = 4.0) for
mild steel.
As a summarization of the cut edge fatigue test
the fixed slopes at this test series. In this respect it
results the influence of the different investigated
has to be mentioned that the tests of the 2R + UP
post treatment procedures as well as the effect of
series shows a poor correlation and a relatively
different slopes assumed for the fatigue strength
large scatter even for the slope calculated directly
assessment are shown in Figure 13 allegorized by
from the tests (see Figure 13 TN = 1:6.73).
the appertaining FAT classes. Complementary to
Under consideration of the limitations described
Figure 13 the results of the different evaluations of
before the following conclusions can be drawn from
the fatigue tests are given numerically in Table 2. With
the fatigue tests of the cut edge specimens:
respect to the Rules of Germanischer Lloyd (2010)
the fatigue tests results have been evaluated with a • The results of the fatigue tests showed that the
fixed slope of m = 3.5 and m = 4.0. Furthermore S-N curves from the Rules lead to a somewhat
results for slopes directly obtained from the tests are conservative assessment of the fatigue strength
presented. It should be noted that the latter values of cut edges (see Fig. 12). Even if the already
are not very meaningful as the load levels are close described limitations of the test results are taken
together on the one hand and the number of speci- into account, a modified FAT class for high qual-
men is not very high on the other hand. ity treated cut edges seems to be reasonable.
Figure 13 shows that the fatigue tests results are • Series 1C and 2R show a very similar fatigue
very sensitive to the slope m of the S-N curve that performance—no positive effect of the more
has been used for the assessment. Especially for the time-consuming 2R treatment could be observed
test series 2R and 2R + UP. For the series 2R the from the tests.
(doubtful) slope of m = 2.5 obtained directly from • Surface grinding (series 2R + G) has a positive
the tests leads to a much lower detail category com- effect on the fatigue strength.
pared to the assumption of fixed slopes m = 3.5 and • UP surface treatment influences the fatigue
m = 4.0 according to the Rules. Contrary to this behaviour—especially the slope of the apper-
the application of the fixed slopes from the Rules taining S-N curve.
leads to significantly lower detail categories than • In general the additional positive effect of com-
obtained by applying the slope from the tests for plete surface treatment on the fatigue strength
the 2R + UP series. This is mainly due to the large was found to be relatively small compared to the
scatter bands that result from the enforcement of beneficial basic effect of corner treatment.

518

MARSTRUCT.indb 518 2/18/2011 5:53:03 PM

 • Test series 2R + UP shows increased fatigue
strength if the slope obtained from the tests is
applied. Due to uncertainties of the test results
especially regarding the evaluation of the slope
of the S-N curve it is not possible to draw relia-
ble conclusions for this post treatment method.
4 CONCLUSIONS
Within the JDP of DSME and GL the influence
of post treatment methods on the fatigue strength
of butt welds and thermally cut edges was inves-
tigated by means of fatigue tests. As expected the
results differ significantly between the butt welds
and the thermally cut edges. The S-N curves given
in the Rules of Germanischer Lloyd (2010) were
proven to result in a suitable fatigue assessment
including a necessary amount of conservatism. In
this respect the various uncertainties within the
actual fatigue tests and their effects on the validity
of the results have to be considered.
The fatigue tests of the butt weld specimens
showed a heterogeneous picture regarding to the
effects of UP treatment as a method for fatigue
strength improvement. In cases where the UP
treatment has been carried out in an effective way
the fatigue tests showed a shift of the critical crack
initiation location away from the weld toes and
roots towards other crack initiation points. The
latter were accompanied by a significant increase
of the fatigue strength.
At two UP treated butt weld specimens the
crack initiated at the weld toe, an indication for a
non effective UP treatment. The fatigue strength
of these cases was nearly the same as the fatigue
strength of the specimens in as-welded condition.
This means that the quality assurance becomes one
of the most important and most challenging tasks
in connection with the application of UP treatment
as a fatigue strength improvement method, espe-
cially since it was not possible to decide whether
the UP treatment was effective or not by visual
inspection. A qualification standard for workers
performing UP treatment similar to welder quali-
fication could solve this problem. Due to the lim-
ited number of specimens and the various failure
mechanisms of the UP treated specimens and the
limitation of the test program to constant ampli-
tude tests neither quantitative nor qualitative con-
clusions regarding the effects of UP treatment on
the fatigue strength under variable amplitudes
should be drawn from the actual project. To over-
come this lack additional investigations and tests
would be necessary.
519
MARSTRUCT.indb 519
The results of the cut edge fatigue tests are very
sensitive regarding the applied evaluation pro-
cedure and the slope utilized for the S-N curve
estimation.
As a general result of the of the cut edge tests it
was found that high quality post treated cut edges
showed an increased fatigue strength compared
to the Rules values. Based on this a new FAT-
class 150 has been introduced into GL-Rules,
edition 2010.
Unexpectedly the tests showed nearly the
same fatigue strength for the 1C series and the
2R series. In this respect the tests showed that
the effect of corner treatment is more distinct
than the effect of complete surface treatment.
However, 2R treatment is to be recommended
based on the experiences of a large number of
vessels in the past.
In comparison to the series 1C and 2R, where
only corner treatment has been carried out, grind-
ing of the complete surface area, as performed
at the 2R + G series, seems to lead to improved
fatigue strength. Due to the large scatter of the test
results and the small number of tested specimens
no reliable conclusions can be drawn from the tests
of the 2R + UP series. The influence of UP surface
treatment tends to affect mainly the slope of the
appertaining S-N curve.
REFERENCES
Germanischer Lloyd. 2010. Rules for Classification and
Construction, I—Ship Technology, Part 1—Seagoing
Ships, Chapter 5: Structural Rules for Container
Ships, Hamburg: self publishing.
Haagensen, P.J. & Maddox, S.J. Revision March 2009.
Recommendations on Post Weld Improvement of
Steel and Aluminium Structures. International Insti-
tute of Welding, Document XIII-2200r-07.
Hobbacher, A. (Editor) 2009. Recommendations for
Fatigue Design of Welded Joints and Components.
IIW doc.1823–07, Welding Research Council Bulletin
520, New York.
Maddox, S.J. 2010. Improving the Fatigue Strength of
Misaligned Butt Welds, International Institute of
Welding Document XIII-2328-10.
Maddox, S.J. 2010. Fatigue Life Improvement of Steel
Fillet Welds by Hammer, Needle or Shot Peening,
International Institute of Welding Document XIII-
2327–10.
Marquis, G. & Björk, T. 2008 Variable amplitude fatigue
strength of improved HSS welds, International Insti-
tute of Welding Document XIII-2224-08.
Weich, I. 2009. Edge layer condition and fatigue strength
of welds improved by mechanical post weld treatment.
International Institute of Welding Document XIII-
2265-09.
2/18/2011 5:53:05 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Study on the ultimate strength test of high speed trimaran

Weiguo Wu & Shuangxi Xu

School of Transportation, Wuhan University of Technology, Wuhan, Hubei, China
Key Laboratory of High Speed Ship Engineering (Wuhan University of Technology), Ministry of Education,
Wuhan, Hubei, China

Weiqin Liu, Xiaobin Li & Jingxia Yue

School of Transportation, Wuhan University of Technology, Wuhan, Hubei, China

ABSTRACT: In this paper, structural model test method was introduced to study the ultimate strength
of a model structure of high speed trimaran. Firstly, similarity theory with finite element method was
applied to instruct the structural model design. Then, to simulate the peak wave bending moment of
the high speed trimaran, three point bending was applied on the model by hydraulic jacks in the ultimate
strength test. Finally, The ultimate strength of the trimaran model was successfully determined from
the model test. The load-stress and load-displacement curves of the model, and the trimaran failure mode
were obtained.

Keywords: high speed trimaran; ultimate strength; structure model test; similarity theory

1 INTRODUCTION In this paper, a trimaran model test was built

for longitudinal ultimate strength test based on the
Trimaran is a special type of ship. Due to its excel- similarity theory. The obtained ultimate bearing
lent sea keeping and outstanding tactical perform- load from the model test can be analysis reference
ance, trimaran has a very wide prospects, such as to the same kind of high speed trimaran.
research ships, maritime shipping, maritime trans-
port, exploration and rescue boats, etc. Therefore,
the ultimate bearing capacity of its complex 2 STRUCTURE MODEL DESIGN
structure has been the concern of many researchers.
There are some common methods used to study 2.1 Similarity theory
the hull ultimate strength, such as a direct calcula-
For the thin-walled marine structure, main
tion method, the idealized Structural Unit Method
dimensions and material thickness have different
(SUM), gradually collapse method, nonlinear
scale levels, so it is impossible to strictly satisfy
finite element method and model test method. On
the geometrical similarity in the model test.
the other hand, to completely predict the ultimate
Theoretically, both the length L and the thickness
bearing capacity by numerical simulation is very
t are taken as basic physical parameters, which has
difficult because the ultimate strength problem
been verified by a large number of experiments.
involves geometric and material nonlinearities.
In this study, linear scale L, the thickness δ, and
It is obvious that different ship types have dif-
physical force F were chosen as the basic physical
ferent destruction modes. Wang, etc. proposed
quantities. Then the basic dimensionless system of
similarity theory to apply to the structural model
model test is [F, L, δ ].
test firstly (1986). Sun, etc. applied similarity
According to the dimensional analysis method
theory to design a model with large opening for
of the similarity theory, to simulate the same load
ultimate torsion strength test (2003). Liu, etc. also
type and boundary condition, the stress conversion
used similarity theory to design a SWATH ship
between model and actual structure should be:
model for transverse ultimate strength test (2009).
The destruction mode of trimaran structure is CM C
much more complex and there is not any trimaran Cσ = , Cσ = F (1)
ultimate strength test or study so far. CL2Ct CLCt

521

MARSTRUCT.indb 521 2/18/2011 5:53:05 PM

 Here: are conversion coefficients for structural Table 2. List of materials used in model.
stress, moment, force, main dimensions and plate
thickness respectively. Steel plate 2 mm, 2.5 mm, 3 mm, 3.5 mm, 4 mm
They can calculated by:
Angle steel L30 × 30 × 3, L40 × 40 × 4
Flat steel −30 × 3
Cσ σ r/ m, CL = Lr /Lm
(2)
CM M r /M
Mm CF = Fr /Fm , Ct tr /ttm

Here the subscripts r and m represent actual

structure and test model respectively.
Moreover, the section inertia moment and the
position of neutral axis should keep similar based
on the similarity theory as follows:

Ir er
= CL3Ct = CL (3)
Im em

Here, I is the section inertia moment of structure;

e is the distance between neutral axis and base line.
Three calculated sections and two middle sec-
tions in main hull are designed by use of the equa-
tions (1) to (3), which can satisfy the similarity
relation between test model and actual structure.
We can see from equation (3), The inertia moment
error means error between Ir and CL3Ct × I m ,the Figure 1. A typical transverse framework of model.
neutral axis error means error between er and
CL × em. The two errors between model structure
and trimaran are considerable small, the validity
of the model design was verified. Table 1 shows the
errors of neutral axis and section inertia moment
of the designed sections.

2.2 Model design

According to the test requirements and laboratory
conditions. The plate thickness ratio of actual ship
and model is 2:1; The linear dimension ratio is 4:1.
Steel with 235 Mpa yield stress was used to make
model. To reduce the welding stress and distor-
tion, the angle and flat steel instead of T section or Figure 2. A typical transverse bulkhead of model.
non-standard profiles were used. Some reinforce-
ments are reverse arrangement to ensure building Based on the similarity relation introduced
precision. The types of steel used in the model test above, the similarity ratio of neutral axis between
are listed in Table 2. test model and ship is 1:4; and the inertia moment
of section is 1:128. Structural plane and section of
model ship is shown in Figures 1 and 2.
Table 1. Section errors.

Errors
2.3 Similarity verification
Sections Neutral axis Inertia moment To examine the similarity of the test model, the
FE analysis for both the actual trimaran and its
Calculated section 1 0.6% 0.9%
scale model are carried out. Only linear static
Calculated section 2 0.1% 1.1%
Calculated section 3 2% 0.04%
analysis was performed. Half of FEA models
Middle section 1 0.2% 0.1%
were applied in the calculation of model structure
Middle section 2 0.1% 1.8% and trimaran, symmetry constraints were adopted
in the longitudinal section. The load condition

522

MARSTRUCT.indb 522 2/18/2011 5:53:05 PM

 of three point bending was determined to simu-
late longitudinal ultimate bending moment.
Appropriate constraints were simulated to simply
supported constraint at bow and stern, load force
was exerted in calculated section. FE model of
trimaran was established by 40406 shell elements
and 23678 beam elements; while the FE model of
test model is by 22176 shell elements only. The
Von Mises stress distribution of main structure
was compared to test the similarity of the longi-
tudinal strength. A number of point forces were
applied on the main deck.
According to equation (1) and the dimension
ratios, it can be assumed that when the stresses
on the real structure and the model are same, the
conversion coefficient should be 1/8.
Table 3 shows stress value of different stress
points. Figures 3 and 4 show that maximum stress
Table 3. Errors between model and trimaran.
Trimaran Model
Key stress points (MPa) (MPa) Error
Bottom of main hull deck 84.9 85.2
Center line of main hull deck 57.3 57.2
Side line of main hull deck 85.8 86
Figure 3. Von Mises stress contour of the ship.
Figure 4. Von Mises stress contour of the model.
MARSTRUCT.indb 523
Figure 5. Model machining.
and Von Mises stress distribution are similar
between the ship and the model. The position of
maximum stress points are same as well. The com-
parison indicates that the similar model design is
applicable and reasonable.
0.35% 3 MODEL MACHINING
0.17%
0.23%
Model shown in Figure 5 was assembled on the
highly precise machining platform in the Ship
Structure Laboratory, School of Transportation,
Wuhan University of Technology. The precision
of the test model is sensitive to the technologi-
cal process and processing methods. The follow-
ing techniques have been used during the model
building:
1. Shearing or flanging was preferred than flame
cutting and welding.
2. The prefabrication and correction of shell parts
and internal frames should be done first. Proper
assembly procedures were important to control
welding procedure strictly.
3. Linear lofting and frame should be located pre-
cisely. The welding length should be as small as
possible to ensure reliable connection and effec-
tive force transmission. High efficiency electrode
and the best welding current were chosen.
4. The size of all components should be checked in
each assembly steps, and they must be corrected
immediately if any problems happened.
4 MODEL TEST
4.1 Load model
Hydraulic loading system was used to pro-
duce the external force on the structural model,
see Figure 6. Firstly, the model was pre-loaded
3∼5 times to release residual stress. The maximum
523
2/18/2011 5:53:08 PM

Figure 6. Loading of model test.
load was close to the yield limit. Then a three-point
bending was applied to simulate the longitudinal
ultimate bending moment. 8 Jacks were put on the
calculated section of the main hull. Two jigs were
settled to simulate simply supported constraint at
bow and stern. 11 dial indicators were uniformly
distributed along the length direction to test the
longitudinal displacement.
There were eight loading steps in this test:
2t∼10t (2t in every step). The strain and displace-
ment were recorded at each step. Before the load
of each hydraulic loading system increased to 6t,
the strain and displacement data were recorded
per 2t. Then they were recorded per 1t or 0.5t until
the model collapsed. The test model was defined
as the failure limit state when a sudden and signifi-
cant permanent deformation was observed accom-
panied with continual load noise and noticeable
reduction of the hydraulic jack loads.
4.2 Station arrangement
Test workload on the work station was limited as
small as possible. FE analysis was carried out before
deciding the measurement points. According to
Figures 3 and 4, the surrounding region of calcu-
lated section is complex stress area.
Three-direction strain gauge was located at the
region with complex stress. Since the test model
was loaded symmetrically, strain gauges were
mainly arranged on the one side, a few points
on the another side were also arranged to check
the repeatability of the test data. A number of
strain gauges were arranged on the sections with
high stress near the calculated section bulkheads.
The strain gauges in a typical section is shown in
Figure 7.
To verify the property of the hull bending,
11 dial indicators were used to record bending
deformation (see Fig. 8).
524
MARSTRUCT.indb 524
Figure 7. Measurement points arrangement on model.
Figure 8. Testing longitudinal displacement point
arrangement of model.
5 TEST RESULTS
The test phenomena is follows: (1) occasional
sounds of fracture after 8t; (2) tearing firstly hap-
pened at the bottom keel after 9t companied by
unstable and wavy deck; (3) loading force reached
the limit value when the load of each jack increased
to 10t, and the record of the pressure sensor on
hydraulic jacks reduced suddenly and couldn’t
increased again. Therefore, the longitudinal ulti-
mate load Pt is 80t or 784.8 KN. Failure mode is
shown in Figures 9 and 10.
The Von Mises stress of key measure point
on a typical section is shown in Figure 11. The
stress of selected point is excellent in the linear
elastic range. Testing point 1 is tensile, when the
load of each jack is 7.2t, the model is yield. When
the load of each jack is 6.5t, testing point 2 is
pressed to yield. When the load of each jack is
8.3t, testing point 3 is tension to yield. When the
load of each jack is 9.5t, testing point 4 which is
arranged on the longitudinal beam is yield. The
station of the bottom is yield before the station
of the deck. The facture mode of trimaran model
is bottom tearing and deck instability with a wavy
fold. the bottom was tearing before the deck was
instability, this showed that the bottom structure
should be more easily damaged than the deck in
the longitudinal sagging ultimate condition. The
main reason is that the neutral axis of model is
much more closed to the deck since there are
cross structures and side hulls in the trimaran so
that bottom section modulus is smaller than deck
section modulus.
2/18/2011 5:53:10 PM
 Figure 9. Failure mode of ship bottom.
Figure 10. Failure mode of deck.
Figure 11. Load-stress curves of key testing point in
typical section.
From Figure 12, the curvature of load-deformation
curve in calculated section where is position of load-
ing approach zero, this indicate that 10t is ultimate
load of trimaran model. According to Figure 13,
the fairing of these curves are well, the data of test
which is good linearity is verified. When the load is
525
MARSTRUCT.indb 525
Figure 12. Load-deformation curves for test model.
Figure 13. Longitudinal displacement curve of main hull.
less than 6t, these curves are linear, the distance of
every curve is almost equivalent; when the load is
more than 6t, the trend of displacement is changed,
the distance of curves is increasing, this shown that
test model has been out of linear elasticity.
6 ANALYSIS AND CONCLUSIONS
The ultimate strength of trimaran model was stud-
ied by model test in this paper. The ultimate load
and destruction mode of model were obtained, the
gradual destructive process was recorded. Based on
the test, some results and collusions are obtained.
1. The static FE model of trimaran and test
model were analyzed by MSC.Nastran, which
verify the validity of similarity theory for model
design.
2. Test loading process were designed according to
the real distribution of wave bending moment
on the trimaran, and the method of longitudi-
nal ultimate strength test of trimaran model is
verified effectively.
2/18/2011 5:53:13 PM
 3. The longitudinal ultimate load of trimaran
model was obtained from the test. The destruc-
tion of trimaran model is bottom tearing and
deck instability with a wavy fold.
4. According to the test result, the bottom was
yield before the deck which shows that the
bottom structure of similar trimaran should
be more easily damaged than the deck in the
longitudinal sagging ultimate condition. The
main reason is that the neutral axis of model
is closed to the deck since there are cross struc-
tures and side hulls in the trimaran so that
bottom section modulus is smaller than deck
section modulus.
ACKNOWLEDGEMENTS
Sincere thanks are expressed to all members of struc-
ture department who participate in the structural
model test. Special thanks to Mr. Xueren Wu,
Mr. Xinghou Li and Mr. Zhaoxiang Yu who have
helped in the experiments.
526
MARSTRUCT.indb 526
REFERENCES
Bin Liu, Weiguo Wu. Ultimate Strength Test Study of
SWATH Ships. Proceedings of International Offshore
and Polar Engineering Conference, Osaka, Japan,
June, 2009.
Faxiang Wang, Zongliang Cheng, Junyu Ying, Wenbiao
Wang. Structure Design and Torsinal Strength Test
for Double Skin Barge Model. Shipbuilding of China.
1986.03.
Hai-Hong Sun, C. Guedes Soares. An experimental study
of ultimate torsional strength of a ship-type hull
girder with a large deck opening. Marine Structures,
16(2003)51–67.
Rahman, M.K. Ultimate strength estimation of ships’
transverse frames by incremental elastic-plastic finite
element analysis. Marine Structures, 11(1998)291–317.
Weiguo Wu, Duan Hong. Experiment Research of
Stresses along the Junction Lines Between the Super-
structure and the Deck. Journal of Ship Mechanics,
2002.06.
Xiaoqiong Huang, Li Cheng, Xionghui Yang, Xingtao
Zhou. Long Term Prediction and Analysis of Wave
Loadson Trimaran Cross Structure. Chinese Journal
Of Ship Research. August 2009.
2/18/2011 5:53:18 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

An experimental study on mechanical, fatigue and crack

propagation properties of IMO type B CCS materials
at room and cryogenic temperatures

C.H. Yoo, K.S. Kim & J. Choung

Inha University, Incheon, Republic of Korea

C.S. Shim
Mokpo national university, Mokpo, Republic of Korea

J.K. Kang
Daewoo Shipbuilding & Marine Engineering Co. Ltd, Goeje, Republic of Korea

D.H. Kim
Hyundai Heavy Industries Co. Ltd, Ulsan, Republic of Korea

Y.S. Suh
Samsung Heavy Industries Co. Ltd, Goeje, Republic of Korea

Y.L. Shim
ABS, Houston, USA

H.S. Urm
DNV Korea Ltd, Busan, Republic of Korea

M.S. Kim
Lloyd’s Register Asia, Busan, Republic of Korea

G.B. An
POSCO, Pohang, Republic of Korea

ABSTRACT: IMO type B CCS (cargo containment system) is considered for LNG FPSOs. For the
three materials of SUS304, 9% Ni steel and Al 5083-O alloy, regarded as possible candidates for IMO type B
CCS, extensive tensile, fatigue, fatigue crack growth rate and CTOD tests were carried out at room, inter-
mediate low (−100°C) and cryogenic (−163°C) temperatures in this paper. All materials show a tendency
for initial yield and tensile strengths to increase with decrease in the temperature, while change in elastic
moduli is not as remarkable. The largest and smallest increase ratios of the initial yield strengths due to
temperature reduction are seen in SUS304 and Al 5083-O alloy, respectively. In fatigue tests, fatigue
strengths increase due to the temperature decrease for three materials. The highest increase of fatigue
strength due to the temperature decrease is observed in Al 5083-O alloy, while 9% Ni steel shows lowest
increase. In fatigue crack growth rate test, SUS304 and Al 5083-O alloy show reduction in crack propaga-
tion rate, due to the reduction in temperature, while any visible improvement of da/dN is not witnessed in
case of 9% Ni steel. In CTOD test, CTODs tend to decrease in case of SUS304 and CTODs for Al 5083-O
alloy increase with reduction in temperature.

1 INTRODUCTION feasible with reduction in the production costs.

As LNG FPSO (Floating Production Storage &
Recently, significant increase in demands for LNG Offloading) concept attracts worldwide attention,
(Liquefied Natural Gas) is observed. Offshore CCS (Cargo Containment System) of LNG FPSO
development of deep sea gas fields is becoming should be evaluated extensively for ultra-low

527

MARSTRUCT.indb 527 2/18/2011 5:53:18 PM

 temperatures and LNG sloshing impact. It is noted 2.2 FT (Fatigue Test)
that independent tank type B (IMO type B) has
Fatigue tests with zero stress ratio (R = 0) were per-
best resistance against LNG sloshing among vari-
formed under load control at RT and CT. FTs were
ous CCS types. Stainless steel (SUS304), nickel-
carried out under axial loading and continued until
alloy steel (9% Ni steel) and aluminum alloy
final fracture, except for tests exceeding 1E7 cycles.
(Al 5083-O alloy) are being considered possible
Fatigue test specimens manufactured from
candidates for IMO type B CCS materials. Though
SUS304 stainless steel, 9% Ni steel and Al 5083-O
service temperature of CCS materials is cryogenic,
alloy are listed in Table 2 and presented in Figure 2.
tests at room temperature were also performed
Number of cycles to crack initiation, number of
because classification societies specify design refer-
cycles to failure were measured. Strains were meas-
ence temperature as a room temperature for IMO
ured throughout fatigue tests by using uniaxial
type B tank. This paper presents normalized dia-
strain gages to evaluate the hotspot stress and
grams and improvement ratios for three materials
crack initiation life. It is assumed that 5% reduc-
used for IMO type B tank. It is shown that the
tion in strain at hot spot location indicates crack
mechanical and fatigue properties of all materials
initiation (Matsuoka & Fujii 1995). Hotspot
are improved as the temperature decreases.
stresses based on test data (σhs) were calculated
using linear extrapolation by Equation 1 (IIW 1995)
for SUS304 and Equation 2 (DNV 2008) for 9% Ni
2 EXPERIMENT PROCEDURES
steel and Al 5083-O alloy. Hotspot stresses of butt
welded specimens include weld-induced initial
2.1 MTT (Monotonic Tensile Test)
bending stress (σbend). For this reason, concept of
Monotonic tensile tests were carried out at RT effective hotspot stress (σhs_eff) is introduced as
(Room Temperature), LT (Low Temperature, −100°C) defined by Equation 3 (DNV 2008).
and CT (Cryogenic Temperature, −163°C). Base
material and butt welded specimens were prepared σhs = 1.67σ0.4 − 0.67σ1.0 (1)
for stainless steel SUS304 (modified SUS304) and σhs = 1.5σ0.5 − 0.5σ1.5 (2)
9% Ni steel. Base and weld material specimens were
manufactured for Al 5083-O alloy. Specimen type where σx = stress at x mm from weld toe.
and dimensions are provided in Figure 1 and
Δσhs_eff = Δσmem + 0.60Δσbend (3)
Table 1.
Test speed is also provided in Table 1. MTT were where σmem = membrane stress.
conducted in accordance with ASTM E8M-04 and
elastic moduli, initial yield strengths based on 0.2%
offset method, engineering tensile strengths and Table 1. Thickness (B) or diameter (Φ) of specimen,
true tensile strengths/corresponding strains were specimen type and test speed.
obtained at each temperature.
B or Φ Test speed
Specimen
Material Metal mm type N/sec
SUS304 Base B:13 Plate 1070
Weld B:16 1020
9% Ni steel Base B:10 Plate 400
Weld B:10 400
Al 5083-O Base Φ:10 Round 200
alloy Weld Φ:10 200

Table 2. Weld type of specimen for each material.

Material Weld type

SUS304 Butt weld

Longitudinal attachment weld
Transverse attachment weld
9% Ni steel Butt weld
Longitudinal attachment weld
Al 5083-O alloy Butt weld
Figure 1. Geometries of specimens for monotonic Longitudinal attachment weld
tensile tests.

528

MARSTRUCT.indb 528 2/18/2011 5:53:18 PM

 (a) SUS304

Figure 2. Geometries of fatigue test specimens.

(b) 9% Ni steel

2.3 FCGRT (Fatigue Crack Growth Rate Test)

Base, weld and HAZ metal of SUS304 (modified
SUS304), 9% Ni steel and Al 5083-O alloy were
used in fatigue crack growth rate tests carried out
at RT, LT and CT. The dimensions of compact ten-
sion specimens are shown in Figure 3. The base,
weld and HAZ metal was taken from large welded
plate (refer to Fig. 4).
COD (Crack Opening Displacement) gage was
installed on front face next to the notch to measure
the crack length. Stress intensity factor range (c) Al 5083-O alloy, base metal
(ΔK) for compact specimen can be obtained from
Equation 4.

ΔP ( + )
ΔK = 3
B W
( − )2
× (0. 86 + 4.664α − 13 32α 2 + 14.772α 3− 5 6α 4 )
(4)

where α = a/W; a = distance between the load line

and crack tip; W = specimen width. Finally, fatigue (d) Al 5083-O alloy, weld and HAZ metal
crack propagation is characterized by Paris equa-
tion as shown in Equation 5. Figure 3. Geometries of FCGRT specimens.

da/dN = C (ΔK)m (5)

FCGRTs were performed under load control
with zero stress ratio (R = 0) at RT, LT and CT.
Paris constants (C, m) were evaluated at each
temperature.
2.4 CTODT (CTOD Test)
Base, weld and HAZ metals of SUS304 (modified
SUS304), 9% Ni steel and Al 5083-O alloy were
used in CTOD tests carried out at RT, LT and CT.
The base, weld and HAZ metal was taken from

529

MARSTRUCT.indb 529 2/18/2011 5:53:19 PM

 δ = K2
(1 − υ ) +
2
0.46 ( − )Vp
(6)
2σYS E 46W + 0 56a0 + z
0.46W

⎛ 1 ⎞
δ =⎜
⎝ mσY ⎟⎠
⎧ ⎫
⎪ (
⎪K2 − υ )+ η Ap ⎪
⎪
⎨ ⎬
⎪ E ⎡ ⎛ α +z ⎞⎤⎪
⎪⎩ ⎢B ( − ) 1+
⎝ ⎥
0.88a0 + 0 2W ⎟⎠ ⎦ ⎪
⎣ ⎭

Figure 4. Schematic view of specimen sampling.

(7)

where K = stress intensity factor; σYS = 0.2% proof

strength; σY = effective yield strength; υ = Poisson’s
ratio; E = elastic modulus; a0 = average original
crack length; B = specimen thickness; W = effective
width of specimen; and z = distance of knife edge
measurement point from load line.

(a) SUS304
3 RESULTS AND DISCUSSION

3.1 MTT (Monotonic Tensile Test)

Figures 6–8 show the engineering stress-strain and
true stress-strain curves at RT, LT and CT nor-
malized by yield stress and yield strain at RT.
Figures 9–11 show the improvement ratios of elas-
tic modulus, initial yield strength, engineering ten-
(b) 9% Ni steel sile strength, true tensile strength and true tensile
strain due to reduction in temperature for each
material. The mechanical characteristics are nor-
malized with regard to those at RT. Regardless of
type of material and metal (base or weld), reduction

(c) Al 5083-O alloy, base metal

(d) Al 5083-O alloy, weld and HAZ metal

Figure 5. Geometries of CTOD test specimens.

large welded plate, see Figure 4. Compact tension

specimens for CTOD tests are shown in Figure 5.
COD gage was installed on load line to measure
the CMOD (Crack Mouth Opening Displacement).
CTODs were calculated according to Equation 6 Figure 6. Normalized stress strain curves for SUS304
and 7 (ASTM E1290 & BS 7448, respectively). at RT, LT and CT.

530

MARSTRUCT.indb 530 2/18/2011 5:53:21 PM

 Figure 7. Normalized stress strain curves for 9% Ni
steel at RT, LT and CT.

Figure 9. Improvement ratios due to reduction in

temperature for SUS304.

Figure 8. Normalized stress strain curves for Al 5083-O

alloy at RT, LT and CT.

in temperature results in increase of initial yield

strength, engineering tensile strength and true ten-
sile strength.
In case of SUS304, true tensile strain which
can be good index for low temperature tough-
ness largely decreases with the temperature
reduction. As for 9% Ni steel, any visible
improvement of true tensile strain of base metal
is not witnessed, but significant increase of that
of weld metal was observed. The remarkable fea-
ture of Al 5083-O alloy is that true tensile strain
increases for both base and weld metal with Figure 10. Improvement ratios due to reduction in
reduction in temperature. temperature for 9% Ni steel.

531

MARSTRUCT.indb 531 2/18/2011 5:53:23 PM

 Figure 11. Improvement ratios due to reduction in
temperature for Al 5083-O alloy.

3.2 FT (Fatigue Test)

S-N curves normalized by yield stress at RT are
presented in Figures 12–14 where dashed black
lines are referred to IIW (IIW 2005). Solid and
dotted lines are from regression analyses of test
data using Equation 8 which uses same slope as
IIW guidance (IIW 2005). Solid and dotted lines
are regression lines of test data at RT and CT,
respectively. The data points with final fracture at
more than ten million cycles are excluded in regres-
sion analyses. In Figures 12–14, nominal stress (σn)
was applied to regression analysis.

log N = log a − 3log Δσ (8)

where N = number of cycles to failure; Δσ = stress

range; and log a = intercept of log N-axis by S-N
curves.
Table represents improvement ratios of fatigue
strengths according to temperature reduction. As seen
from Table 3, regardless of materials and weld
type, temperature reduction results in increases
of fatigue strength. The highest increase of fatigue
strength due to the temperature decrease is observed
in Al 5083-O alloy, while 9% Ni steel shows lowest
increase.
Figure 12. Nominal stress based S-N curves for SUS304
at RT and CT.
In case of butt welded specimens, Al 5083-O alloy
shows highest increase of fatigue strength, while
lower increase is observed in other two materials.
As for specimens with longitudinal attachments,
SUS304 and Al 5083-O alloy shows higher increase
of fatigue strength than 9% Ni steel.

532

MARSTRUCT.indb 532 2/18/2011 5:53:25 PM

 Figure 13. Nominal stress based S-N curve for 9% Ni Figure 14. Nominal stress based S-N curve for Al
steel at RT and CT. 5083-O alloy at RT and CT.

Table 3. Improvement ratios of fatigue strengths for

3.3 FCGRT (Fatigue Crack Growth Rate Test)
three materials according to temperature reduction.
da/dN-ΔK curves from experiments are presented
in Figures 15–17. It is seen from Figures 15, 17 σhs_eff
that at lower temperatures the threshold values of Weld σn or σhs
Material type based based
stress intensity factors are greater and crack prop-
agation rates are higher for SUS304 stainless steel SUS304 Butt 1.476 1.475
and Al 5083-O alloy. These tendencies are not Longi. 1.608 1.511
obvious for 9% Ni steel. Crack propagation occurs Trans. 1.684 1.638
in lower ΔK than other materials in case of Al 9% Ni steel Butt 1.342 1.437
5083-O alloy. Longi. 1.063 1.089
Al 5083-O Butt 1.838 2.018
3.4 CTODT (CTOD Test) alloy Longi. 1.634 1.875
Normalized CTODs or improvement ratios of
CTODs versus temperature are presented in
Figure 18. CTODs at LT and CT are normalized for HAZ metal increases at LT compared with
with respect to those at RT. For SUS304 CTODs that at RT and CT. CTODs for Al 5083-O
tend to decrease with reduction in temperature alloy tend to increase with the reduction in
and the largest decrease of CTODs was observed temperature. The largest CTOD value was
for base metal. For 9% Ni steel, base and weld observed at LT for base and weld metal, followed
metal show a similar tendency whereas CTOD by reduction at CT.

533

MARSTRUCT.indb 533 2/18/2011 5:53:26 PM

 Figure 15. da/dN-ΔK curves for SUS304 at RT, LT
and CT.
Figure 16. da/dN-ΔK curve of base, weld and HAZ
metal for 9% Ni steel at RT, LT and CT.
Figure 17. da/dN-ΔK curve of base, weld and HAZ
metal for Al 5083-O alloy at RT, LT and CT.
534
MARSTRUCT.indb 534
Figure 18. CTOD improvement ratios at LT and CT.
4 CONCLUSIONS
Three types of materials of SUS304, 9% Ni steel
and Al 5083-O alloy are popular candidates for
IMO type B CCS. Even if it is known that mechan-
ical properties at low temperature are superior to
those at room temperature, a lot of guidances or
regulations from classification societies still per-
sist to use properties at room temperature. This
paper presents experimental studies on mechanical,
fatigue and crack propagation properties of base
and weld metals at three levels of temperatures
of RT, LT and CT. Improvement ratios of the
properties are presented for LT and CT as com-
pared to RT.
2/18/2011 5:53:28 PM
 1. In monotonic tensile tests, regardless of type of
material and metal (base or weld), reduction in
temperature results in increase of initial yield
strength, engineering tensile strength and true
tensile strength.
In case of SUS304, true tensile strain which
can be good index for low temperature tough-
ness largely decreases with the temperature
reduction. As for 9% Ni steel, any visible
improvement of true tensile strain of base metal
is not witnessed, but significant increase of that
of weld metal was observed. The remarkable
feature of Al 5083-O alloy is that true tensile
strain increases for both base and weld metal
with reduction in temperature.
2. In fatigue tests, regardless of materials and weld
type, temperature reduction results in increases
of fatigue strength. The highest increase of
fatigue strength due to the temperature decrease
is observed in Al 5083-O alloy, while 9% Ni steel
shows lowest increase.
In case of butt welded specimens, Al 5083-O
alloy shows highest increase of fatigue strength,
while lower increase is observed in other two
materials. As for specimens with longitudinal
attachments, SUS304 and Al 5083-O alloy shows
higher increase of fatigue strength than 9%
Ni steel.
3. In fatigue crack growth rate tests, at lower tem-
peratures the threshold values of stress intensity
factors are greater and crack propagation rates
are higher for SUS304 stainless steel and Al
5083-O alloy. These tendencies are not obvious
for 9% Ni steel.
4. For SUS304, CTODs tend to decrease with
reduction in temperature and the largest decrease
of CTODs was observed for base metal. For 9%
Ni steel, base and weld metal show a similar ten-
dency whereas CTOD for HAZ metal increases
at LT compared with that at RT and CT. CTODs
for Al 5083-O alloy tend to increase with the
reduction in temperature. The largest CTOD
value was observed at LT for base and weld
metal, followed by reduction at CT.
535
MARSTRUCT.indb 535
ACKNOWLEDGEMENTS
This work was supported by DSME, HHI, SHI,
ABS, DNV, Lloyd’s Register and POSCO Research
Grants under project SNAMF. The experimental
work was conducted in the INHA UNIVERSITY,
Incheon, Republic of Korea and MOKPO
NATIONAL UNIVERSITY, Mokpo, Republic
of Korea.
REFERENCES
ASTM—The American Society for Testing and Materials.
2004. E8M-04. Standard Test Methods for Tension Test-
ing of Metallic Materials. HIS: West Conshohochen.
Baek J.H., Kim, Y.P., Kim, W.S. & Kho, Y.T. 2001.
Fracture toughness and fatigue crack growth proper-
ties of the base metal and weld metal of a type
304 stainless steel pipeline for LNG transmission.
International Journal of Pressure Vessels and Piping.
Vol. 78. pp. 351–357. Elsevier.
Choung, J.M. & Cho, S.R. 2008. Study on true stress
correction from tensile test. Journal of Mechanical
Science and Technology. Vol. 22. pp. 1039–1051.
DNV—Det Norske Veritas. 2008. Fatigue Assessment of
Ship Structures. Classification Notes. No. 30.7.
IIW—The International Institute of Welding. 1995.
Stress Determination for Fatigue Analysis of Welded
Components. IIS/IIW-1221-93.
IIW—The International Institute of Welding. 2005.
Fatigue Recommendations. XIII-1965-03/XV-1127-03.
Matsuoka, K. & Fujii, E. 1995. An evaluation method on
fatigue crack initiation life at welded joints in steel
structures. Journal of the Society of Naval Architects
of Japan. Vol. 178. pp. 513–522.
Park, W.S., Kang, K.Y., Kim, J.H. & Lee, J.M. 2010.
A comparative on mechanical behavior of low tem-
perature application materials for ships and offshore
structures. Proceedings of Autumn Meeting of SNAK.
Changwon. 21–22 Oct. pp. 1300–1312.
Shim, K.T., Kim, J.H., Lee, K.H., Ahn, B.W. &
Kim, Y.K. 2008. A study on the fatigue crack growth
behavior of 9% Ni steels. Proceedings of Fall Annual
Conference of KSME. Pyeongchang. 1–7 Nov. pp.
167–172.
2/18/2011 5:53:32 PM
 Materials and fabrication of structures

MARSTRUCT.indb 537 2/18/2011 5:53:32 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

A study on laser assisted friction stir welding of C-Mn steel plates

Pankaj Biswas & N.R. Mandal

Department of Ocean Engineering and Naval Architecture, Indian Institute of Technology, Kharagpur, India

ABSTRACT: A comprehensive model was developed for the calculation of the thermal profile based
on frictional heating from FSW tool and absorption of laser energy on the plate surface. The calculations
showed that using a suitable laser source 75% of the desired temperature level was attained through con-
duction mode and the rest was attained through frictional heat. Thus achieving desired plasticizing of the
plate material, the forces required to move the FSW tool through the butt will be considerably less result-
ing in less wear and tear and longer life of the FSW tools. The computed temperature profile due to laser
source agreed well with the corresponding experimental data. The good agreement demonstrates that the
comprehensive model can serve as a basis for development of a feasible FSW process for shipbuilding
quality normal strength and higher strength steels.

1 INTRODUCTION interest in extending the technology to other

Friction Stir Welding (FSW) is a solid state joining
technique. The welds are created by the combined
action of frictional heating and mechanical
deformation due to a rotating tool. The majority
of the heat generated from the friction, i.e. about
95%, is transferred into the workpiece and only 5%
flows into the tool (Chao et al. 2003).
FSW of Aluminum and its alloys is relatively
well established. Presently friction stir welds
have been successfully produced on many of
the important commercial Al alloys including
the 1xxx (Murr et al. 1997 ), 2xxx (Christner &
Sylva 1996, Li et al. 1998), 5xxx (Reynolds 1998,
Karlsson et al. 1998, Biswas & Mandal 2009), 6xxx
(Reynolds, 1998, Karlsson et al. 1998, Lienert et al.
1998, Murr et al. 1998, Sato et al. 1999) and 7xxx
(Mahoney et al. 1998, Jata et al. 2000) families of
alloys, as well as Al-Cu-Li alloys (Kinchen et al.
1999, Jata & Semiatin 2000). FSW permits join-
ing of most, if not all, Al alloys, including those
considered unweldable with fusion welding proc-
esses (Dawes & Thomas 1996). Moreover, fric-
tion stir welds often exhibit improved mechanical
properties relative to those of fusion welds on the
same material (Murr et al. 1998, Kinchen et al.
1999). Finally, residual stresses and distortion in
friction stir welds are typically lower than those
of fusion welds (Wang et al. 2000). The improve-
ments in mechanical properties, residual stresses,
and distortion relative to fusion welding are
thought to stem from the lower heat input inherent
with FSW.
While most of the FSW efforts to date have
involved joining of Al alloys, there is considerable
materials, including steels (Lienert & Gould 1999,
Merklein & Giera 2006). Merklein M. and Giera A.
(2006) studied FSW of thin steel aluminium hybrid
blanks.
Several numerical model of heat transfer analysis
of aluminum alloy by FSW are available in pub-
lished literatures. Like, Khandk et al. (2003) devel-
oped a three-dimensional thermal model where the
heat generation was modeled based on experimen-
tally measured torque distribution. Ulysse (2002)
modeled the effects of tool speeds, both rotational
and linear, on forces and plate temperatures dur-
ing FSW of thick aluminum plates, based on a
three-dimensional viscoplastic model. Nandan
et al. reported results of a three-dimensional mate-
rial flow and heat transfer model during FSW of
6061 aluminum alloy (Nandan et al. 2006). They
calculated the temperature fields, cooling rates and
the geometry of the thermomechanically affected
zone (TMAZ). The computed temperature fields
and TMAZ agreed well with the corresponding
independent experimental data.
Although several numerical models of FSW of
aluminum alloy have been reported in the literature,
most of these were not concerned with the FSW
of steel. There are a few exceptions. Nandan et al.
(2006a) reported results of a three-dimensional
material flow and heat transfer model during
FSW of 304 austenitic stainless steel (Nandan
et al. 2006b). Zhu and Chao (2004) proposed a
three-dimensional thermal model for 304L stainless
steel. Two-dimensional steady-state heat transfer
near the tool pin was modeled by Cho et al. (2005)
for the FSW of 304L stainless steel. Nandan
et al. (2007) reported on numerical analysis of

539

MARSTRUCT.indb 539 2/18/2011 5:53:32 PM

 three-dimensional material flow and heat transfer
during FSW of mild steel. In particular, they
examined the temperature fields and cooling rates
by solving the equations of conservation of mass,
momentum and energy in three dimensions with
appropriate boundary conditions.
Applications of FSW in mild steel and high ten-
sile steels as compared to aluminum alloys remained
limited due to problems in terms of required force
and tool life. The system requires high forces to
move the tool through the plasticized material
which, in turn, wears down the tool. Attempts are
being made to solve these problems by preheating
the workpieces with laser ahead of the tool shoul-
der (Fujii et al. 2008, Daftardar, S. 2009). Laser
assisted FSW will thus require less force to push
the tool through the material. It would result in
faster welding speed with less tool wear.
In this present study a comprehensive model
based on dual heat source, i.e. laser heat and tool
frictional heat was developed to calculate the ther-
mal history of the steel plates. Three-dimensional
transient finite element thermal analysis was car-
ried out to obtain the thermal profile. To reduce
the computation time, coarse mesh was generated
away from heating line and finer meshing was used
along and near the heating line. The analysis was
carried out for 6 mm thick C-Mn steel plates.
2 HEAT GENERATION
2.1 Heat generation from FSW tool
In friction stir welding heat is generated due to fric-
tion and plastic deformation at the tool-work piece
interface and at TMAZ. The heat generation at the
contact surfaces due to friction is the product of
frictional force and the tangential speed of the tool
with respect to the work piece. Whereas the heat
generated due to plastic deformation at the tool-
work piece interface is the product of shear stress
and the velocity of the work piece material sticking
to the tool as it traverses. This velocity is actually
the tangential speed of the tool.
The heat generation due to friction on an ele-
mental area dA at the tool-work piece interface
considering high rotational speed compared to
traverse speed of the FSW tool, is given by,
dQf = δωrμpdA (1)
In case of pure friction, i.e. no material is stick-
ing to the tool, δ = 1.
The heat generated due to plastic shear defor-
mation leading to work piece material sticking to
the tool is given by,
dQp = (1−δ )ωrτydA (2)
540
MARSTRUCT.indb 540
By Von Mises criterion τ y = σ y / 3
In case of heat generation only through plastic
shear deformation δ = 0.
The overall magnitude of heat generation due
to plastic deformation at TMAZ is rather small
(Zhu & Chao 2004), hence it was neglected.
Therefore total heat due to friction and plastic
deformation is given by,
dQ = dQf + dQp
= ω rdA [δ (μp − τy) + τy ]
Let τcontact = [δ (μp − τy) + τy] (3)
Therefore dQ = ωrdAτcontact (4)
There is no straight forward mechanism to
estimate the extent of slip. At the same time with
increase in temperature the yield strength of the
work piece material decreases resulting in reduc-
tion in heat generation from plastic deformation.
In such a situation it was felt more logical to con-
sider pure friction and neglect the heat generation
due to plastic deformation
Hence taking δ = 1, Eq. (3) reduces to
τcontact = μp (5)
Therefore from (4) and (5), the expression for
heat generation on an elemental surface area dA at
the tool-work piece interface is given by,
dQ = ω rμp dA
i.e. dQ = ω rdF (6)
where dF = μp dA
In this study all the three tool-work piece inter-
face surfaces, i.e. tool shoulder, tool pin side and
tool pin tip were considered. Q1, Q2 and Q3 are the
components of the respective heat generated from
these interfaces as shown in Figure 1. Therefore
the total heat generated is given by,
Qtotal = Q1 + Q2 + Q3
2.1.1 Tool shoulder—work piece interface
The expressions were derived considering a con-
cave shoulder surface. The purpose of this geomet-
ric feature is to act as an escape volume as the tool
pin is plunged into the plate during the welding
operation.
The concave shoulder surface is represented by
its vertical and horizontal projected surfaces as AV
and AH respectively. Therefore for an elemental
segment,
dAV = r dθ dz,
2/18/2011 5:53:32 PM
 2 π RS
Q1 ∫∫ μ ppr 2 ( ta α)) ddrd
dθ
0 RP
2
= π μp
3
(R 3
S )
− RP3 (1 + ) (9)

d?
2.1.2 Tool pin—work piece interface
Heat generated from a cylindrical tool pin con-
dr sists of two contributions; Q2 from the side surface
rd? and Q3 from the tip surface of the tool pin. Heat
generated due to friction of an elemental area on
the cylindrical pin surface is given by,
Q3
Rp dQV = ω r dFV
dz

Q2 dFV = μp dAV, now dAV = r dθ dz,

Hp

Q1
Rs Therefore dQV = ω r2 μp dθ dz.
Integrating dQV over the tool pin vertical side
surface one obtains,
a

2 π HP
Q2 ∫ ∫ μ ppR
RP2 ddzdθ
0 0
= 2 π μ p RP2 H P (10)

Similarly the heat generated due to friction of

the tool tip surface is given by,
Figure 1. Segmental areas and heat generation on
different parts of FSW tool. 2 π RP
Q3 ∫∫ μ ppr 2ddrdθ
0 0
2
Now dz = tan α dr, = π μ p RP3 (11)
Therefore dAV = r dθ tan α dr 3
and Therefore the total heat generation is given by,

dAH = r dθ dr. Qtotal = Q1 + Q2 + Q3

The forces acting on the tool shoulder surface =

2
3
μ pω ⎡⎣
πμ ( − )( + )+ 3
P + 3RP H P ⎤⎦
2
(12)
can be written as,
In case of a flat shoulder, the heat generation
dFs = μp(dAH + dAV) expression simplifies to,
= μp r dθ dr(1 + tan α) (7)

From Eq. (7) one can observe that the concave Qtotal
2
3
p (R 3
S 3RP2 H P ) (13)
shoulder surface actually contributes to increased
frictional area by a factor of tan α.
Therefore combining Eq. (6) and Eq. (7) the heat 2.1.3 Heat source model for FSW tool
generation from the elemental shoulder surface is The heat source modeling for FSW tool was done
given by, based on the following assumptions,
i. the heat input is linearly proportional to the
dQS = ω r dFS = ω r2 μ p dθ dr(1 + tan α ) (8) distance from the center of the tool,
ii. the plunging force applied to the plate surface
The heat generated through friction of tool by the tool creates a uniform pressure over the
shoulder with the plate surface is obtained by inte- shoulder surface,
grating Eq. (8) from pin root radius to the outer iii. the heat is generated from the work done by the
radius of shoulder surface. friction force only.

541

MARSTRUCT.indb 541 2/18/2011 5:53:33 PM

 The distribution of heat flux (Chao & Qi 1998)
over the plate surface is given as,
3Qtotal r
q( r ) = fo r ≤ r0
2π RS3
The tool geometry used in this present study is
shown in Figure 2.
2.2 Heat generation from laser beam
An important mechanism in laser beam heating is
the interaction of the laser beam with the material.
This interaction mechanism is influenced by many
parameters such as: the laser power Q, the inten-
sity distribution I of this power at the surface, the
heating speed Vb, the material properties and the
wavelength of the laser radiation λ.
The laser energy is absorbed by Fresnel absorp-
tion at the surface of the workpiece and can be
described by an absorption coefficient η, indi-
cating a fraction of laser power is absorbed in
the plate. For steel the absorption coefficient is
of the order of 0.4 for Nd:YAG laser radiation
(Sjoerd, P. 2003) in case of conduction mode of
welding. The rest of the laser energy is reflected.
Whereas in case of keyhole welding the laser
beam energy is absorbed on the walls of keyhole
through repeated reflections on the keyhole wall.
Because of this multiple reflections keyhole mode
of welding has a high absorption coefficient of
about ≥80%.
In the present investigation it being a case
of only heating, it is natural that laser material
interaction will be of conduction mode. Hence
the absorption coefficient will be on the lower
side. The laser energy is absorbed in a thin layer
(∼40 nm) (Sjoerd, P. 2003) at the surface of
the workpiece where it is converted into heat.
5
3
5.8
10
35
10
25
Figure 2. Details of tool geometry with cylindrical
tool pin.
MARSTRUCT.indb 542
The absorbed energy is then transported across
the plate thickness by means of heat conduction.
2.2.1 Heat source model for laser heating
(14) The power density distribution (Zhao, H. &
Debroy, T. 2001) of the laser beam was considered
Gaussian in nature and the distribution of the
absorbed laser power density on the metal surface
is given by,
2
⎛ r⎞
2ηQ −2⎜⎝ rb ⎟⎠
qb (r ) = e (15)
π rb2
where η is the absorption coefficient, Q is the laser
power, rb is the beam radius at the plate surface, r is
the radial distance from the beam axis. The beam
radius at various locations along the vertical axis
is given by,
1
⎡ ⎧ λ zzM 2 ⎫⎤ 2
rb = r0 ⎢1 + ⎨ 2 ⎬⎥
⎢⎣ ⎩ π r 0 ⎭⎥⎦
where r0 is the beam radius at the focal point, λ
is the beam wavelength, z is the beam defocused
length, i.e. the distance between the focal point to
the plate top surface, and M2 is a dimensionless
beam parameter. It is given by,
β
M2 π r0
λ
where β is the half angle of beam divergence.
To validate the heat source model as well as to
establish the absorption coefficient laser heating
was carried out on 5 mm thick mild steel plate
samples as shown in Figure 3. The heating was
Figure 3. Laser heating on mild steel plate.
542
2/18/2011 5:53:36 PM
 done using a Ytterbium Fiber (λ = 1070 nm) Table 4. Temperature dependent enthalpy
laser source (Model YLR-2000) with a maxi- for steel.
mum power of 2 kW. The laser beam was guided
Temperature (°C) Enthalpy (MJ/m3)
from the laser source to the workstation by an
optical fibre with output fiber core diameter 0 0
150 μm. Temperature profile was recorded at 100 360
two locations 10 mm and 20 mm away from the 200 720
heating line. 300 1100
Temperature recording was done using Agilent 400 1500
34970A data logger. The laser parameters used are 500 1980
given in Table 1. Using the heat source model as 600 2500
given by Equation (15) and considering material 700 3000
properties as given in Tables 2 to 4, the tempera- 800 3700
ture profile was calculated and compared with the 900 4500
measured data as shown in Figure 4. >1000 5000
Here one can observe a very good correlation
between the calculated and experimentally meas-
ured temperature data. 300 Calcutaed temperature distribution
Measured temperature distribution
250

Table 1. Laser parameters. Temperature ( C)

0
200

Parameters Values 150

100
Laser beam radius at plate surface (mm), rb 2.0
Laser power (W), Q 1000 50
Laser beam speed (mm/min), Vb 200
0
Absorption coefficient, η 0.63 0 50 100 150 200 250 300
Time (s)

Figure 4. Measured and calculated temperature profile

Table 2. Temperature dependent material properties of 10 mm away from heating line.
C-Mn steel.

Temperature Thermal conductivity Specific heat 3 MATERIAL PROPERTIES

(Celsius) (Watt/m°K) (J/Kg°K)
In the present analysis, the material properties of
0 51.9 450 C-Mn steel were used as given by Brown & Song
100 51.1 499.2
(1992). Table 2 shows the temperature depend-
300 46.1 565.5
ent material properties used for the transient heat
450 41.05 630.5
transfer analysis. Temperature dependent convec-
550 37.5 705.5
tion coefficients (Adak & Mandal 2003) for steel
600 35.6 773.3
surfaces are given in Table 3. Temperature depend-
720 30.64 1080.4
800 26 931
ent enthalpy (Mandal & Adak 2001) of the C-Mn
1450 29.45 437.93
steel are given in Table 4.
The friction coefficients used in this study were
calculated by considering the relative velocity
between the tool and the workpiece. The relative
Table 3. Temperature dependent convection coeffi- velocity increases from zero at the axis of rotation
cients for steel surfaces. (static condition) to ωRS at the periphery of the
tool shoulder (dynamic condition). Experimental
Temperature (°K ) Convection coefficient (W/m2 K) evidence (Nandan et al. 2006b) suggests that the
frictional co-efficient μ can be expressed as,
56 1.815
278 9.079 μf = μ0 exp(−κ δ ω r)
556 18.15
2778 52.66 In case of pure friction, i.e. no material is stick-
ing to the tool, δ = 1. The value of μ0 was taken

543

MARSTRUCT.indb 543 2/18/2011 5:53:39 PM

 as 0.4 and constant κ was 1 s/m (Nandan et al.
2006b). r is the radial distance from the tool axis to
the point in consideration.
4 THREE-DIMENSIONAL FINITE
ELEMENT MODEL
A three dimensional finite element transient ther-
mal model was used to determine the thermal
history on the work piece based on the dual heat
source given by Equations (14) and (15). The fol-
lowing assumptions were made in the analysis,
i. Linear Newtonian convection cooling was con-
sidered on all the surfaces.
ii. 95% of the frictional heat was transferred to the
work piece.
The governing differential equation is:
5 RESULTS AND DISCUSSION
The temperature profile caused by laser assisted
FSW process was simulated by considering a dual
heat source comprising of heat delivered from laser
energy and tool friction as discussed above.
Transient thermal analysis using FEM was car-
ried out. Brick elements with fine meshing in the
weld zone were considered. The FE meshing view
of the plate is shown in Figure 5.
The laser beam was considered moving ahead
of the FSW tool. The various parameters of FSW
and laser heat source as used in the thermal analy-
sis are given in Tables 5 and 6 respectively.

∂ ⎡ ∂T ⎤ ∂ ⎡ ∂T ⎤ ∂ ⎡ ∂T ⎤ ∂T
K + ⎢K ⎥+ K = ρC
∂x ⎢⎣ ∂x ⎥⎦ ∂y ⎣ ∂y ⎦ ∂z ⎢⎣ ∂z ⎥⎦ ∂t

The following boundary conditions were applied

in the present FE model.
Initial condition: A specified initial temperature for
the entire plate,
T = T∞ for t = 0
First boundary condition: A specific heat flow act-
Figure 5. FE model and meshing view.
ing over weld surface for t > 0 representing the
FSW heating source,
qn = −qsup Table 5. FSW parameters for welding of 6 mm thick
normal strength steel plates.
The quantity qsup is the heat supplied to the plate
due to friction at the tool-plate interface and laser FSW tool Laser beam
beam. qn is the normal component of heat flux position
vector. Here qsup is given by Eqs. (14 and 15). Rotational Traverse ahead of
speed speed Plunging FSW tool
Second boundary condition: qconv represents the Sl. No. (rpm) (mm/min) force (N) (mm)
heat loss due to convection from the plate surface
at temperature T. Considering hf as convection 1 1000 80 6500 15
2 1000 80 6500 20
coefficient, the heat loss qconv is given by,
3 1000 80 6500 25
4 1400 80 5000 20
qconv = hf (T − T∞)
5 2000 80 3800 20
Considering heat loss due to convection over
top and bottom surface (Newton’s law of cooling)
Table 6. Laser parameters.
∂T
qn qconv or −k = h f (T − T∞ ) fo t > 0.
∂n Parameters Values

Laser beam radius at plate surface (mm), rb 2.0

It should be noted that heat flux qsup and Laser power (W), Q 650
convection loss qconv do not occur over the same Laser beam speed (mm/min), Vb 80
boundary segment at the same time. Heat flow into Absorption coefficient, η 0.63
the boundary is taken as positive.

544

MARSTRUCT.indb 544 2/18/2011 5:53:40 PM

 The isotherms as obtained using dual heat
source are shown in Figure 6 for the welding
parameters as given in Table 5 (Sl. No. 1) and
Table 6. The laser source produced a temperature
of about 900°C as shown in Figure 7. At this tem-
perature the yield stress of low carbon mild steel
reduces to about 10 MPa as shown in Figure 8
Figure 6.
welding.
1400
1200
1000
Temperature ( C)
Isotherms in laser assisted friction stir
Temperature rise
due to frictional heat
(Clausen, H. B. 2000). Thus it plasticizes the mate-
rial sufficiently and thereby would reduce the tool
plunging force. Subsequent friction between the
tool and the plate surface provided the additional
heat to raise the temperature to about 1200°C as
shown in Figure 7.
This would enable forward movement of
the FSW tool resulting in required stirring of
the plate material to achieve welding along the
plate butt. It can also be observed from Figure 7
that the time, Δt8−5 taken to cool over the range
800–500°C was about 25 s. Whereas in case of
manual metal arc welding process with a heat
input of about 1.3 kJ/mm, it is around 12 s
(Nandan et al. 2007).
The effect of variation of laser beam posi-
tion ahead of the FSW tool on thermal history
is shown in Figure 9. As expected with the laser
beam closer to the FSW tool, the rise in peak tem-
perature was more. The maximum temperatures
achieved for different laser beam positions keep-
ing other parameters constant is shown in Table 7.
Here one can see that the increase in temperature
by moving the laser beam closer by 10 mm was
only about 42°C. Hence for a FSW tool having
shoulder diameter 30 mm, laser beam position was
taken 20 mm ahead of the centre line of the FSW
tool in subsequent analysis.
The effect of increasing tool rpm, keeping the
traverse speed and the laser beam position fixed
at 80 mm/min and 20 mm ahead of FSW tool

respectively, is shown in Figure 10. The maximum

o

800 Temperature rise temperatures achieved with the dual heat source
due to laser heating
600 for different tool rpm is shown in Table 8.
The combination of weld parameters as shown
400 in Sl. No. 3 of Table 8 will lead to a temperature
200 level of 1227°C. At this temperature it will be very
much feasible to carryout FSW of steel without
0
0 25 50 75 100 125 150 175 200 excessive wear and tear of the FSW tool.
Time (s)

Figure 7. Transient temperature distribution at the

Laser position 15mm ahead
center of weld line. 1200 Laser position 20mm ahead
Laser position 25mm ahead
1000
Temperature ( C)

300 800
o
Yield Stress (MPa)

250
600
200
150 400

100 200
50
0
0 0 50 100 150 200
0 200 400 600 800 1000 Time (s)
Temperature (C)
Figure 9. Temperature distribution for different
Figure 8. Variation of yield stress with temperature. position of laser beam ahead of FSW tool.

545

MARSTRUCT.indb 545 2/18/2011 5:53:41 PM

 Table 7. Welding parameters, laser beam position and peak temperature.

FSW tool
Laser beam Maximum
Plate thickness Rotational Traverse speed Plunging position ahead temperature
Sl. No (mm) speed (rpm) (mm/min) force (N) of FSW tool (mm) rise (οC)

1 6.0 1000 80 6500 15 1105

2 6.0 1000 80 6500 20 1082
3 6.0 1000 80 6500 25 1063

Table 8. Welding parameters, laser beam position and peak temperature.

FSW tool

Plate Rotational Traverse speed Laser beam position Maximum

Sl. No. thickness (mm) speed (rpm) (mm/min) ahead of FSW tool (mm) temperature (°C)

1 6.0 1000 80 20 1082

2 6.0 1400 80 20 1151
3 6.0 2000 80 20 1227

Tool rpm 1000 iii. The laser source would heat up the plate to
1400
Tool rpm 1400 about 900°C reducing the yield stress of the
1200 Tool rpm 2000 material to around 10 MPa. It would result in
lesser tool plunging force with less tool wear.
1000 iv. Laser heating with a laser power source of
Temperature ( C)

650 kW along with the frictional heat of FSW

o

800
600
400
200
0 about 1.3 kJ/mm.
0 50 100 150
Time (s)
Figure 10. Temperature distribution for dual source
with varying tool rpm at a welding speed of 80 mm/min.
6 CONCLUSIONS
The combined heating due to a laser source and
the frictional effect of FSW tool was investigated
through experiments and computer modeling. The
major findings are as follows:
i. The major part of the total heat was generated
through the laser source in the proposed laser
assisted FSW process of C-Mn steel plates. Thus
sufficient plasticizing of the plate material could
be achieved through laser preheating which
would reduce the wear and tear of FSW tool.
ii. The laser energy absorption coefficient for
shot blasted steel plates coated with zinc based
primer was found to be around 0.63.
would result in a temperature level of about
1200°C at the plate surface.
v. The time, Δt8−5 taken to cool over the range
800–500°C as observed from above analysis was
about 25 s. It is almost double to that of the
manual metal arc process with a heat input of
200 250 300 vi. The above study indicates a strong possibility of
successfully carrying out laser assisted FSW of
steel plates with reduced tool wear and tear.
ACKNOWLEDGEMENTS
The results reported in this paper are from a
research project carried out under funding from
Naval Research Board vide their sanction number
DNRD/05/4003/NRB/169 dated 08/10/2008. The
authors gratefully acknowledge NRB’s financial
support to successfully carry out the investigation.
NOMENCLATURE
α shoulder concavity angle,
β half angle of beam divergence,
δ extent of slip,
λ laser beam wavelength,
ρ density of plate material,

546

MARSTRUCT.indb 546 2/18/2011 5:53:44 PM

 η laser energy absorption coefficient,
μ coefficient of friction,
τy Shear yield stress,
ω tool angular rotational speed,
c specific heat,
dA elemental area,
k thermal conductivity,
p tool plunging force per unit area,
r radial distance along tool radius/beam
axis,
r0 laser beam radius at the focal point,
rb laser beam radius at the plate surface,
z laser beam defocused length,
AH horizontal projected surfaces area,
AV vertical projected surfaces area,
M dimensionless laser beam parameter
Q heat input to the work piece,
Q1 heat generated under the tool shoulder,
Q2 heat generated at the tool pin side,
Q3 heat generated at the tool pin tip,
Qtotal total heat generated,
RP pin root radius,
RS outer radius of shoulder surface,
T∞ ambient temperature.
REFERENCES
Adak, M. & Mandal, N.R. 2003. Thermo-mechanical
analysis through a Pseudo-linear Equivalent Con-
stant stiffness System, Journal of Engineering for the
Maritime Environment, Vol. 217, No. M1, pp.1–9.
Biswas, P. & Mandal, N.R. 2009. Experimental Study
on Friction Stir Welding of Marine grade Aluminum
Alloy, Journal of Ship Production, Vol. 25, No. 1,
pp.1–6.
Brown, S. & Song, H. 1992. Implication of three-
dimensional numerical simulations of welding large
structures, Welding Journal, 71(2), pp.55s–62s.
Chao, Y.J. & Qi, X. 1998. Thermal and Thermo-
Mechanical Modeling of Friction Stir Welding of
Aluminum Alloy 6061-T6, Journal of Materials
Processing & Manufacturing Science, 7, pp.215–233.
Chao, Y.J., Qi, X. & Tang, W. 2003. Heat transfer in fric-
tion stir welding–Experimental and numerical studies,
Transaction of the ASME, 125, 138–145.
Cho, J.H., Boyce, D.E. & Dawson, P.R. 2005. Modeling
strain hardening and texture evolution in friction
stir welding of stainless steel, Mater Science and
Engineering, A 398, pp.146–163.
Christner, B.K. & Sylva, G.D. 1996. Friction stir weld
development for aerospace applications. Proceed-
ings of ICAWT 1996, pp.311–320. EWI, Columbus,
Ohio.
Clausen, H.B. 2000. Plate Forming By Line Heating,
PhD thesis, Published in Denmark by Department
of Naval Architecture and offshore Engineering
Technical University of Denmark.
Daftardar, S. 2009. Laser Assisted Friction Stir Welding:
Finite Volume Method And Metaheuristic Optimiza-
tion, A Thesis Submitted to the Graduate Faculty of
547
MARSTRUCT.indb 547
the Louisiana State University in partial fulfillment of
the requirements for the degree of Master of Science
in Industrial Engineering.
Dawes, C.J. & Thomas, W.M. 1996. Friction stir process
welds aluminum alloys. Welding Journal 75(3): 41–45.
Fujii, H., Tatsuno, T., Tsumura, T., Ueji, R.,
Nakata, K. & Nogi, K. 2008. Laser assisted hybrid
friction stir welding of carbon steel, 7th International
Friction Stir Welding Symposium, Awaji Island,
Japan, May 20–22.
Jata, K.V., Sankaran, K.K. & Ruschau, J.J. 2000.
Friction-stir welding effects on microstructure and
fatigue of aluminum alloy 7050-T7451. Metallurgical
and Materials Transactions 31A: 2181–2192.
Jata, K.V. & Semiatin, S.L. 2000. Continuous dynamic
recrystallization during friction stir welding of high-
strength aluminum alloys. Scripta Materiala 43:
743–749.
Karlsson, L., Svensson, L.-E. & Larsson, H. 1998. Char-
acteristics of friction stir welded aluminum alloys.
Proceedings of the Fifth International Conference on
Trends in Welding Research. pp.574–579. Materials
Park, Ohio: ASM International.
Khandkar, M.Z.H., Khan, J.A. & Reynolds, A.P. 2003.
Prediction of temperature distribution and thermal
history during friction stir welding: input torque
based model, Science and Technology of Welding and
Joining, 8(3), pp.165–174.
Kinchen, D.G., Li, Z. & Adams, G.P. 1999. Mechani-
cal properties of friction stir welds in Al-Li 2195-T8.
Proceedings of the First International Conference on
Friction Stir Welding. Session 9, Paper 2. Cambridge,
England: TWI.
Li, Z.X., Arbegast, W.J. & Hartley, P.J. 1998. Microstruc-
ture characterization and stress corrosion evaluation
of friction stir welded Al 2195 and Al 2219 alloys.
Proceedings of the Fifth International Conference on
Trends in Welding Research, pp.568–573. Materials
Park, Ohio: ASM International.
Lienert, T.J. & Gould, J.E. 1999. Friction stir welding
of mild steel. Proceedings of the First International
Conference on Friction Stir Welding. Cambridge
England: TWI. Session 10, Paper 3.
Lienert, T.J., Grylls, R.J., Gould, J.E. & Fraser, H.L.
1998. Deformation microstructures in friction stir
welds on 6061-T651. Proceedings of Hot Deforma-
tion of Aluminum Alloys, pp.159–167. TMS.
Mahoney, M.W., Rhodes, C.G., Flintoff, J.G.,
Bingel, W.H. & Spurling, R.A. 1998. Properties of
friction stir welded 7075 T651 aluminum. Metallurgi-
cal and Materials Transactions 29 A: 1955–1964.
Mandal, N.R. & Adak, M. 2001. Fusion Zone and HAZ
prediction through 3-D simulation of welding thermal
cycle, Journal of the Mechanical Behavior of Materi-
als, Vol. 12, No. 6, pp.401–414.
Merklein, M. & Giera, A. 2006. Friction stir welding of
thin steel aluminium hybrid blanks, The 9th Interna-
tional Conference on Material Forming, ESAFORM
2006, Glasgow, United Kingdom, April 26–28,
pp.859–862.
Murr, L.E., Liu, G. & McClure, J.C. 1997. Dynamic
recrystallization in friction stir welding of alumi-
num alloy 1100. Journal of Materials Science 16:
1801–1803.
2/18/2011 5:53:45 PM
 Murr, L.E., Liu, G. & McClure, J.C. 1998. A TEM study
of precipitation and related microstructures in friction
stir welded 6061 aluminum. Journal of Materials Sci-
ence 33: 1243–1251.
Nandan, R., Roy, G.G. & DebRoy, T. 2006. Numerical
simulation of three-dimensional heat transfer and
plastic flow during friction stir welding, Metal Mater
Trans A 37(4), pp.1247–1259.
Nandan, R., Roy, G.G., Lienert, T. & DebRoy, T. 2006.
Numerical modelling of 3D plastic flow and heat
transfer during friction stir welding of stainless steel,
Science and Technology of Welding and Joining,
11(5), pp.526–537.
Nandan R., Roy G.G., Lienert T.J. & Debroy T. 2007.
Three-dimensional heat and material flow during
friction stir welding of mild steel, Acta Materialia,
Vol. 55, No. 3, pp.883–895.
Reynolds, A.P. 1998. Mechanical and corrosion perform-
ance of GTA and friction stir welded aluminum for
tailor welded blanks: alloys 5454 and 6061. Proceed-
ings of the Fifth International Conference on Trends
in Welding Research, pp.563–567. Materials Park,
Ohio: ASM International.
Sato, Y.S., Kokawa, H., Enomoto, M. & Jogan, S. 1999.
Microstructural evolution of 6063 during friction stir
welding. Metallurgical and Materials Transactions
30A: 2429–2437.
548
MARSTRUCT.indb 548
Sjoerd, P. 2003. Weld Pool Control in Nd:YAG Laser
Welding, ISBN 90–77172–05-X, Copyright © 2003
by S. Postma, Enschede, The Netherlands, Printed
by Oce′.
Thomas, W.M., Threadgill, P.L. & Nicholas, E.D. 1999.
Feasibility of friction stir welding of steel. Science and
Technology of Welding and Joining 4(6): 365–372.
Ulysse, P. 2002. Three-dimensional modeling of the
friction stir-welding process, Int. J. Machine Tools
Manufacture, 42, pp.1549–1557.
Wang, X.L., Feng, Z., David, S.A., Spooner, S. &
Hubbard, C.R. 2000. Neutron diffraction study of
residual stresses in friction stir welds. Submitted for
publication in Proceedings of The International Con-
ference on Residual Stresses-6.
Zhao, H. & Debroy, T. 2001. Weld Metal Composition
Change during Conduction Mode Laser Welding of
Aluminum Alloy 5182, Metallurgical and Materials
Transactions B, Vol.32b, pp.163–172.
Zhu, X.K. & Chao, Y.J. 2004. Numerical simulation of
transient temperature and residual stresses in fric-
tion stir welding of 304L stainless steel, J. of Material
Process Technology, 146, pp.263–272.
2/18/2011 5:53:45 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Steel-concrete-steel sandwich structures in ship and offshore

engineering

T.J. Grafton & J.R. Weitzenböck

Det Norske Veritas AS, Technical Advisory Ship and Offshore, Høvik, Norway

ABSTRACT: Concrete is commonly used in many civil engineering structures, often re-enforced in some
way with steel, either as tensioning rods or as part of a composite structure. Composite steel-concrete
structures have been used in bridge building for many years, with rules for design freely available (British
Standards Institution, 2008a,b), and has been a topic of research for maritime structures, e.g. for arctic
structures (Matsuishi et al. 1985) or impact resistance (Iwata & Hattori 1994).
This paper explores the potential of steel-concrete-steel sandwich in floating structures and exposes
the advantages, disadvantages and challenges of using this material through two design studies. The use
of light weight concrete (with a density as low as 1000 kg/m3) is compared with more traditional higher
strength concretes (density 2500 kg/m3) and it is found that the lighter weight concrete is desirable if the
sandwich is to be competitive with steel designs on weight. The paper concludes by proposing where this
technology may have potential and suggests areas for further exploration.

1 INTRODUCTION option for worn structures (Kennedy et al. 2003).

The use of concrete as a building material
for floating structures appears to have been
limited primarily due to the increased weight in
comparison with highly optimised steel struc-
tures and the lack of available facilities with
experience in construction of large concrete
floating structures at reasonable cost. Concrete
ships have been built and traded successfully
(Fougner 1922) but interest in concrete ships
appears to have peaked when the cost of steel has
been high and/or when there has been a lack of
availability of steel plate, for example in wartime
(Morgan 1975).
The next period of growth in the use of
concrete for a construction material at sea was
with the discovery of oil and gas deposits in the
North Sea in the 1970s. Concrete has been used
for spar platforms and floating LNG terminals
(Gerwick Jr. 1984); for floating production units
(Lanquetin et al. 2008), and for ocean going barges
(Sare & Yee 1977).
In recent years the idea of using an alternative
material, sandwiched between steel plates, to
make a composite structural element has been
bought from a topic of research to an established
technology through the efforts of Intelligent
Engineering with their patented Sandwich Plate
System, SPS, (Brooking & Kennedy 2004, and
Lloyd’s Register 2006). This uses a polymer as the
core material and has been popular as a repair
Such composite structures offer an opportunity
to reduce or even eliminate structural ele-
ments such as secondary stiffening and primary
girders.
The success of such structures relies on a
reduction in the complexity of the structure,
leading to easier construction, more cost efficient
construction, or both. For most seagoing applica-
tions the structural weight of the new composite
member must compare favourably to an optimised
stiffened steel plate solution.
Concrete, used in combination with re-enforcing
materials such as steel rods, is the material of choice
for large civil constructions due to the ability to
mix from base materials remotely using locally
available contractors and materials. For this type
of construction no ‘factory’ is needed, even for very
large structures. All construction materials can be
delivered to the building site along with machin-
ery required for mixing the concrete and assembly.
The main infrastructure required to build in con-
crete is form work, to make the shape, and heavy
machinery to place re-enforcing members and to
assist with pouring and settling of the mix. This
versatility has seen concrete dominate the civil
construction industry.
So, taking two established materials, used
everyday for large engineering constructions,
accepted readily by the industries they are used
in and supported by design rules and verification
processes is appealing, especially as concrete is a

549

MARSTRUCT.indb 549 2/18/2011 5:53:45 PM

 relatively low cost material, being around 2.3 times 3 STRUCTURAL OVERVIEW OF A STEEL-
cheaper than standard shipbuilding grade steel. CONCRETE-STEEL (SCS) SANDWICH
STRUCTURE

The proposed steel-concrete-steel beam is a struc-

2 CAN CONCRETE BE USED FOR THE
tural sandwich consisting of five elements as shown
CORE MATERIAL OF A SUCCESSFUL
in Figure 1.
LAYERED STRUCTURE AT SEA?
This constitutes a sandwich panel where the faces
primarily carry in-plane loads and bending and
Layered structural elements, with a low weight
the core primarily carries transverse shear loads.
core material sandwiched between a stronger
Sandwich panels are particularly effective in carrying
face materials have been developed for use in
transverse loading and providing flexural stiffness.
the maritime industry since the 1970s. The most
The relevant theory is described in any textbook on
widely used example is foam cored composite as a
sandwich structures, for example (Zenkert 1995).
construction material for the hulls of small craft.
To maintain the sandwich effect there must be
These structures are often referred to as sandwich
sufficient adhesion in the bondlines (indicated in
structures.
Figure 1) where the shear stress at the interface
Sandwich structures are often used to reduce
does not exceed the shear strength of the concrete
weight, a notable engineering example being the
core. In practice this can be difficult to achieve with
development of GLARE for the Airbus A380
a bond of concrete to a smooth steel surface. In the
aeroplane, a sandwich of thin elements of alumin-
civil engineering industry it is common to use an
ium and fibreglass (Beumler 2004).
additional bonding agent, for example epoxy, or
For a successful sandwich, the mechanical
mechanical fastenings (Raina 1994).
properties of the face plates and core must be well
Concrete is a much weaker material than steel,
matched to ensure both are utilised by a similar
both in compression and especially in tension. This is
amount under load, so avoiding one part of the
highlighted by the material properties given below in
sandwich reaching a limit state long before another,
Table 1, where the ‘best’ high strength concrete and
and to give the lowest total sandwich weight for a
lightest low strength concrete are used as examples.
given load carrying capacity.
It will be apparent from this table that under
Considering the above argument the marriage
load the concrete core is significantly weaker than
of steel with conventional concrete in a sandwich
the steel face plates and so if the steel plate were uti-
structure is unlikely to lead to a light weight struc-
lised to the maximum permissible the concrete core
ture as their strength properties are not particularly
may crack. This cracking can be characterised as
well matched and because concrete is not particu-
due either to Bending, Vertical Shearing or due to
larly light when compared with other sandwich
shear transfer between the face plates and the core.
core materials.
Steel-concrete-steel as a sandwich structure
using a novel light weight concrete core has 3.1 Bending cracks
been researched in some depth by Bergan et al.
If the core consists of a material that is weak in ten-
(Bergan & Bakken 2005 and Bergan et al. 2006).
sion, the tensile bending stresses may cause cracking
They propose the use of a concrete with a density
of the core at a lower value of applied bending
around 1000 kg/m3 instead of the heavier conven-
moment than yielding of the faces. If this happens
tional concrete with density around 2500 kg/m3.
the part of the core that experiences tensile strains
This reduction of density comes with an associated
will undergo extensive cracking and contribute
reduction in strength as light weight aggregates
are added to the mix to keep the weight down.
Bergan et al. show that the adoption of this light 1 2
weight concrete significantly improves the compet- 3
itiveness of the sandwich with regards to overall 5 4
weight and give examples where it could be used
for construction of ships hulls. 1. Face 1: steel plate.
So, the review of existing research shows that 2. Bondline 1 between face 1
and core: adhesion of concrete to steel.
high structural weight has meant that steel and
3. Core: concrete.
concrete sandwich structures have not been seri- 4. Bondline 2 between core and
ously considered for floating marine constructions. face 2: adhesion of concrete to steel.
However, the use of a low density concrete core 5. Face 2: steel plate.
appears to increase the competitiveness of the
structure with regards to weight. Figure 1. Steel-concrete-steel beam.

550

MARSTRUCT.indb 550 2/18/2011 5:53:45 PM

 Table 1. Material properties of concrete and steel.

Characteristic cylindrical
Density compression strength Mean tensile strength Youngs modulus

Type of concrete kN/m3 MPa MPa GPa

Light weight 9.8 9 1.1 5.4

High strength 25 90 5 44
Characteristic yield Characteristic ultimate
Density strength strength Youngs modulus

kN/m3 MPa MPa GPa

Normal steel for 78.5 235 360 206
ship construction

little to the bending stiffness. This will cause a shift
of the neutral axis towards the compressive face of
the panel which should be accounted for in design.
3.2 Transverse shear failure
Transverse shear is the force acting to transit the
load from the face plate through the bonds into
the core. The bond between the steel and concrete
must be strong enough to take this load otherwise
debonding will occur. Such debonding can lead to
a catastrophic failure of the sandwich and is akin
to loosing the flange of a steel I-beam.
3.3 Vertical shear cracks
These cracks occur when the shear force transferred
from the face plate through the bond into the core
is greater than the tensile capacity of the concrete.
These cracks will eventually extend through the
core between the faces following the direction of
the shear forces at 45 degrees.
3.4 Summary
The sandwich structure can be designed using
traditional sandwich theory suited to concrete
sandwich structures, modified to take account of
cracks in the concrete core.
This cracking is clearly undesirable and should
be avoided for ‘normal’ expected load conditions.
However, it is important to understand the crack-
ing behaviour of the core material for any ultimate
limit state analysis.
4 DESIGNING A SCS SANDWICH
STRUCTURE
4.1 Design codes
Because steel and concrete are widely used engi-
neering materials many standards exist to guide the
designer, these include composite steel and concrete
structures for civil applications (British Standards
Institution 2008a,b). In addition, Eurocode 2 on
the design of concrete structures (British Stand-
ards Institution 2007) has sections on lightweight
concrete and on design of unreinforced concrete.
The Steel Construction Institute published the
“Design Guide for Steel-Concrete-Steel Sandwich
Construction—Volume 1: General principles and
rules for basic elements” (Narayanan et al. 1994).
This is the only document which was found to
address the design of elements of an SCS panel.
It provides rules for the design of sandwich ele-
ments considering ultimate and service limit states.
It does not cover the fatigue limit state.
These codes can be used with Class Society Rules,
guidance documents and standards (Det Norske
Veritas 2003, 2007, 2008 and 2010a,b) to assess an
SCS structure for use in the marine environment.
To use the above combination of standards
connectors must be provided to improve the bond
between the concrete and the steel face plates,
see Figure 2. These standards effectively forbid
adhesive bonding between layers, for example in
section 6, paragraph A502 of DNV-OSS- C502
(Det Norske Veritas 2007).
This rule set does not consider in any detail
how to join structural elements together, or how
to join structural elements to traditional stiffened
steel structures. The civil engineering standards
assume a homogeneous structure is built in
‘one shot’ rather than by assembling prefabricated
SCS panels. Thus, joining of structural elements
remains an area for further detailed investigation.
4.2 Connecting stud design
Connecting studs provide the important bonding
between the face plates and the concrete core and
transfer the shear loads into the core. The shear
stress transferred into the core should be limited

551

MARSTRUCT.indb 551 2/18/2011 5:53:47 PM

 so that the concrete surrounding the stud does not
crush due to insufficient compression strength in
the concrete. This can be a problem for light weight
concretes where the compressive strength may only
be around 9.0 MPa, see Table 1. To avoid this crush-
ing the surface area of the stud must be increased.
Studs should be placed so that any potential
shear crack in the concrete, whose direction is
assumed to be 45 degrees from the longitudinal
axis, is crossed by a stud, see Figure 2. The stud
spacing should also be chosen to avoid wrinkling
of the face plates due to in-plane loads (sometimes
referred to as buckling in texts on sandwich
structures), see Figure 3.
The use of these studs offers one other practi-
cal advantage in that the bond of concrete to steel
between adjacent studs does not need to be good for
the structure to behave as intended. This damage
tolerance allows welding to the top plate without
concern over damage to the bond due to heat input.
As concrete has very low thermal conductivity any
such welding will not transfer heat to the opposite
steel face and therefore will not damage any coating
on the opposite face plate. Whilst suitable weld
procedures must be tested and developed, this
advantage offers opportunities to modify the build
Cylindrical
studded connector
Figure 2. SCS beam with shear studs. Maximum spacing indicated to avoid complete cracking of the core between
the steel faces.
Figure 3. SCS beam with shear studs. Maximum spacing to avoid wrinkling of compression plate when subject to
in-plane loads.
552
MARSTRUCT.indb 552
process for the structure allowing late additions
of equipment to be included without the need for
re-coating or, possibly, additional supporting foun-
dations (assuming the structure is strong enough to
account for the reaction forces and moments).
4.3 Tensioning of the concrete core
It is common in concrete construction to compen-
sate for the poor tensile strength of concrete by the
use of pre or post stressing. The advantage of pre
or post stressing can be seen in Figure 4 for a SCS
beam with a high strength concrete core.
The y-axis shows the axial force used to tension
the concrete core. Plotted against the bending
moment capacity of the SCS beam this gives a
curved surface from which the optimum stressing
force can be obtained. Where the curve crosses
the x-axis (marked with a cross) is the maximum
bending capacity without applying a stressing force.
The knuckle in the curve to the right of the graph
shows the possible increase in bending capacity
by pre or post stressing. The applied load must sit
within the boundary of the curve allowing for any
factor of safety placed on the maximum bending
capacity. The graph shows that a 55% increase in
2/18/2011 5:53:47 PM
 CAPACITY CURVE - SANDWICH WITH HIGH STRENGTH CONCRETE

16 Capacity curve
Applied load

AXIAL FORCE (+VE) COMPRESSION [MN] 14 Floating Point

12
Increase in bending
10 capacity of ≈ 55%
8

6 Pre-stressing force
4

0
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7
-2

-4

-6
BENDING MOMENT [MNm]

Figure 4. Capacity curve for a SCS beam with a core of high strength concrete. Core thickness is 200 mm sandwiched
between two steel face plates each 10 mm thick.

CAPACITY CURVE - SANDWICH WITH LIGHT WEIGHT CONCRETE

6 Capacity curve
Applied load
AXIAL FORCE (+VE) COMPRESSION [MN]

Floating Point
Pre-stressing force
4
Increase in bending
capacity of ≈ 9%
2

0
0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 0,45 0,5

-2

-4

-6
BENDING MOMENT [MNm]

Figure 5. Capacity curve for a SCS beam with a core of light weight concrete. Core thickness is 200 mm sandwiched
between two steel face plates each 10 mm thick.

bending capacity can be achieved for the highest Knowledge gained from these studies could there-
strength concrete. fore indicate what span was feasible to cover as well
However, the addition of light weight aggre- as the expected self weight of the panels. Based on
gates to make light weight cements reduces the this knowledge two typical marine applications
advantages of pre or post stressing considerably as were selected to compare original steel design with
shown in Figure 5, where the additional bending an SCS sandwich solution.
capacity is reduced to only 9%. It can therefore be
concluded that the additional effort of pre or post 5.1 RoRo ship
stressing the concrete core is not worthwhile for
light weight concretes. The main deck above the machinery space of a
large RoRo car carrier was chosen. This deck was
designed to carry a range of vehicles and for a uni-
5 DESIGN STUDIES form load of 1.0 tonnes/m2. The deck is supported
fore and aft by bulkheads (spaced over 100 metres
Parametric studies were conducted to give an apart), with longitudinal and transverse girders
indication of SCS panel performance for differ- supported periodically by pillars.
ent cross sectional parameters, loads and bound- Two different options are considered for the
ary conditions (Weitzenböck & Grafton 2010). comparison, the first, number 1 in Figure 6, is

553

MARSTRUCT.indb 553 2/18/2011 5:53:48 PM

 1 2
Figure 6. RoRo deck structural girder system. Shaded areas show where SCS panels would replace existing steel
structure.
an 11.4 metre simply supported panel between
longitudinal girders. The second and third aim to
optimise the design by using SCS panels over three
spans as a continuous beam (of approximately
30 metres span). The continuous beam will reduce
the bending moment and may allow a lighter
panel to carry the load. This option is number 2 in
Figure 6. The second longer span panel is designed
with both thick steel plates and thin core (2a) and
thin steel plates and thick core (2b). The light
weight concrete described previously is used to
obtain the lightest weight SCS structure.
SCS panel design is controlled by ultimate
limit state, i.e. bending and shear caused by the
local static loads from cargo and self weight.
The fatigue design life is checked for dynamic
axial forces, acting in the longitudinal direction
of the vessel, caused by global hogging/sagging
moment. As a reference, the maximum deflection
at midspan is calculated as a serviceability limit
state. The maximum deflection was found to be
75 mm, which was considered acceptable. Note
that racking loads, which can influence the size of
the transverse girders for this type of ship, have
not been considered. The concrete core and con-
necting studs are assumed sufficient to withstand
the local loads acting through the tyres of the
vehicles.
Table 2 shows the SCS panels proposed to replace
the existing steel structure. Options 2a and 2b, with
a continuous panel over three spans, are the most
advantageous with respect to weight and material
costs. Option 2a, with thick steel plates and thin
core, has the lightest weight and option 2b, with thin
steel plates and thick core, has the lowest material
costs. The material costs in Table 2 are based on a
material cost estimate of 300 USD pr ton concrete
554
MARSTRUCT.indb 554
• Transverse Girder
• Longitudinal Girder
• Pillar
and 700 USD pr ton steel. The material costs used
in the comparison are based on discussion with
shipyards in Asia and are only used only to give
an indication of the cost. The material costs for the
existing steel structure, plates, stiffeners and trans-
verse girders, is 115 USD/m2.
The study revealed the following advantages and
challenges for replacing the existing steel structure
with a continuous SCS panel.
5.1.1 Advantages
− Total depth of deck is decreased from 600 to
170 mm. This 72% reduction in deck height
gives a potential to install an extra deck, with
economical benefits, if the same reduction
applies for several decks.
− Elimination of stiffeners and some girders,
reducing fabrication complexity and area of deck
for coating and fire protection (where needed).
− With a well-established welding procedure the
welding of the studs will be very efficient. The
total amount of welding will also be reduced
compared to the existing steel structure.
− The panel is very robust and has a large potential
to absorb energy from impact loading or rough
handling. This is especially useful for decks sub-
jected to wheel loading.
− The panel will be easier to maintain due to the
flat surfaces.
5.1.2 Challenges
− Total weight of the panel is increased by 101%,
from 162 to at best 326 kg/m2.
− Estimated material costs are increased by 33%,
from 113.4 to 150.7 USD/m2.
− Assembly, connection and production of the
panels have to be investigated further.
2/18/2011 5:53:50 PM
 Table 2. SCS panel design for the RoRo deck.
Simply supported
Boundary conditions (Option 1)
Depth of concrete core (mm) 200
Thickness of steel plates (*) (mm) 10
Studs: 83 mm diameter,
13 per m2
Weight of concrete (kg/m2) 194
Weight of steel plates and 173
studs (kg/m2)
Total weight of panel (kg/m2) 367
Material cost—concrete (**) (USD/m2) 58.2
Material cost—steel and 121.1
studs (**) (USD/m2)
Total material cost (**) (USD/m2) 179.3
Maximum deflection (mm) 63
(*) Normal steel with 235 MPa yield strength; (**) Per metre squared of deck area.
5.1.3 Summary—RoRo deck
For a replacement of parts of the steel structure
with SCS panels the weight and material cost pen-
alty are likely to be prohibitive. However, consid-
ering that the panel is much simpler to produce,
without intersections between stiffeners and gird-
ers, and no lugs and brackets, the total fabrication
cost may be reduced. Fabrication cost can be 25%
or greater of the total cost of the ship (depending
on ship type) with steel cost only 5–10%. Therefore,
there is scope to improve the cost competitiveness
of the SCS structure through simpler fabrication.
5.2 FPSO process deck
The RoRo deck study showed that SCS solutions
would struggle to be competitive if used for parts of
weight critical ship structures. Therefore, an alterna-
tive structure was chosen to see if SCS could offer
advantages on a structure where weight was less
critical. A Process Deck for a Floating Production,
Storage and Offloading unit (FPSO) was selected.
This is a deck module located on the upper part
of an FPSO containing equipment for processing
petrochemicals prior to offloading. Process decks
are located in open air, due to the risks from fire or
explosion, and are generally prefabricated and par-
tially outfitted prior to final lift on to the FPSO.
It was hoped that the following advantages of
SCS could be utilised:
− Simple build of deck as one large SCS panel,
supported by girders below.
− Minimise, or eliminate other underdeck sup-
porting structure for foundations of heavy
equipment.
− Thermal barrier between process decks in case of
fire and limited protection in case of explosion.
555
MARSTRUCT.indb 555
Continuous Continuous
(Option 2a) (Option 2b)
150 235
10 6
110 mm diameter, 66 mm diameter,
16 per m2 16 per m2
144 227
182 118
326 345
43.2 68.1
127.4 82.6
170.6 150.7
75 52
Further, the design was to be as weight
competitive as possible with the steel equivalent.
So, once again the light weight concrete was used
in preference to the stronger heavier concrete.
Loads and dimensions for comparison are
extracted from an existing process deck on an FPSO
vessel. The static loads on the deck are caused by
operational loads and self weight from structural
elements, equipment, piping, valves etc. Large point
loads, from the operating separators, make the proc-
ess deck heavily loaded in some areas. The compari-
son is performed for a 2.5 m wide SCS panel at the
heaviest loaded area. The dimensions of the process
deck are 18.4 m by 27.7 m. A schematic of the deck
is shown in Figure 7.
The static model for the process deck is indicated
in Figure 8, spanning in the shortest direction as
shown in Figure 7.
Because of the heavy loads and large span for the
process deck, a higher steel grade (yield 355 MPa) is
chosen for the steel plates. Two different panels are
chosen to replace the existing steel structure, one with
thin steel plates and thicker core and the second with
thick steel plates and smaller core, see Table 3. The
design of the proposed panels is controlled by bend-
ing and shear capacity in ultimate limit state. As a
reference the maximum deflection is calculated.
Table 3 shows the SCS panels proposed for the
steel deck. As before, the material costs in table 3
are based on a cost estimate of 300 USD pr ton
concrete and 700 USD pr ton steel. The steel weight
of the existing steel structure is 359 kg/m2, giving
an estimated material cost of 251 USD/m2.
In addition to the general benefits already
outlined in the RoRo deck study the following
advantages and challenges for replacing the
existing steel structure with the lightest SCS panel
(panel 1 in Table 3) were identified.
2/18/2011 5:53:51 PM
 • Girders
2.5 metres • Stiffeners
• SCS Panel

Figure 7. Structural arrangement of the process deck, showing underdeck stiffeners and girders. SCS panel shown
shaded. Total SCS panel assumed to be constructed of interconnected panels placed atop the existing girder system.

Figure 8. Static model of the FPSO process deck, showing point loads for heavy equipment.

Table 3. SCS panel design for the FPSO process deck.

SCS panel 1 SCS panel 2

Depth of concrete core (mm) 360 460

Thickness of steel plates (*) (mm) 10 8
Studs: 37 mm diameter, 25 per m2 30 mm diameter, 25 per m2
Weight of concrete (kg/m2) 345 444
Weight of steel plates and studs (kg/m2) 207 178
Total weight of panel (kg/m2) 553 622
Material cost—concrete (**) (USD/m2) 103.8 133.2
Material cost—steel and studs (**) (USD/m2) 144.9 124.6
Total material cost (**) (USD/m2) 248.7 257.8
Maximum deflection (mm) 89 67

(*) High strength steel with 355 MPa yield strength; (**) Per metre squared of deck area.

5.2.1 Advantages − For the SCS panel, welding directly on the

− Material costs competitive with original steel
structure.
− Increased flexibility for the location of the
equipment because:
− For the existing steel structure, welding directly
on one side of the plate will not damage the
steel treatment on the other side of the plate.
top side of the upper plate will only damage
the bond between the concrete and the plate.
However, this concrete to steel adhesive bond is
not taken into account since studs are provided
to transfer the shear stress between the layers.
− Thus, the steel treatment on the bottom surface
of the lower plate is not affected by welding.

556

MARSTRUCT.indb 556 2/18/2011 5:53:51 PM

 − Better performance under explosion.
− Improved heat insulation.
− With a well-established production and assembly
procedure it is possible to be competitive on costs.
5.2.2 Challenges
− Total weight of the panel is increased by 54%,
from 359 to 553 kg/m2.
− Assembly, connection and production of the
panels have to be investigated further.
5.3 Limitations
The limitation of these comparisons is that the
main structural system used for both the conven-
tional steel structure and the SCS panels is the same.
This system is the result of an optimization process
meant for a steel structure. In the studies SCS
panels are used in place of parts of the structure
and only the SCS panels are optimised, no changes
are made to the existing steel structure.
6 HOW TO EXPLOIT SCS STRUCTURES
Optimise the structure for lowest weight:
− Limit panel span to avoid low utilisation of the
core. For longer spans the section modulus for
bending determines the core thickness giving a
thicker core than needed to withstand the shear
forces.
− Arrange panel supporting structure (walls,
girders, etc.) to enable use of the most weight
efficient panel sizes.
− Take advantage of reductions in deck height and
structural complexity to make a smaller, simpler
structure.
Minimise build costs:
− Uplift methods from the civil construction
industry. For example, a ‘one shot’ build where
the steel structure is assembled at the construc-
tion site and the concrete is poured in one go to
make the SCS structure.
− Minimise the need for specialist infrastructure
at the build site. For example, prefabricated steel
plates with interconnecting circular sectioned
shear studs could be procured in the required
sizes and delivered to the build site. Only con-
creting equipment and the ability to lift and
weld the steel panels would then be needed at
the build site.
− Minimise joins between SCS panels and between
SCS panels and other structure. Reduces weld-
ing and the need for new joining methods to be
qualified.
557
MARSTRUCT.indb 557
− Focus on smaller and simpler structures that
can be produced using locally sourced materials,
equipment and labour.
Take advantage of other benefits:
− Low thermal conductivity of concrete. Reduces
requirements for insulation.
− Better impact protection (Iwata & Hattori 1994).
May eliminate the need for additional structural
protection such as fenders.
− Flat interior walls. Reduces coating area.
7 FURTHER WORK
Before competitive SCS structures can be built
some areas of further research are needed:
− Studies need to be completed to demonstrate
cost effectiveness, for both initial cost and
through life cost.
− Joints between structural elements need to be
carefully designed, qualified and tested, using
for example the DNV technology qualification
process (Det Norske Veritas 2001 and 2006).
8 CONCLUSIONS
This paper has demonstrated that Steel-Concrete-
Steel structure can be used in ships and offshore
structures however the challenge will be to remain
competitive on weight and cost.
To be competitive on weight light weight con-
crete must be used in place of more traditional
heavier concretes.
Cost competitiveness is likely to depend on
developing simple build processes that minimises
weld length and structural joints.
The most promising applications for this tech-
nology appear to be for smaller ships or barges
where local build using local resources offers sig-
nificant cost advantages compared with building
at a remote shipyard. Ship types should be selected
where build and operation costs are not highly sen-
sitive to overall weight.
Concrete has disadvantages when used in con-
junction with steel in a layered plating system. Main-
taining a good bond between the steel and concrete
is critical and, using today’s accepted standards, can
only be obtained by using studs extending between
the two face plates. Large scale cracking of the
core due to tensile loading must also be avoided,
although with a studded solution there will be some
tolerance to cracks in the core material.
Unless a construction process can be devel-
oped that allows for the pouring of the concrete
into prefabricated steel plates representing the hull
2/18/2011 5:53:52 PM
 shape a large amount of joins will be needed to
connect prefabricated SCS panels to each other
or to conventional steel structure. As these joins
remain unproven some development work will be
necessary. The range of joints requiring verification
and qualification will be reduced with a ‘one shot’
method for pouring concrete into a steel form.
If SCS structures are to be taken further a full
technology qualification process needs to be fol-
lowed relevant to the intended application and the
loads it may experience.
REFERENCES
Bergan, P.G. & Bakken, K. 2005. Sandwich design:
A solution for marine structures? International confer-
ence on computational methods in marine engineering,
MARINE 2005, P. Bergan et.al. (eds.), Barcelona.
Bergan, P.G., Bakken, K. & Thienel, K-C. 2006. Analysis
and design of sandwich structures made of steel and
lightweight concrete. European conference on com-
putational mechanics, solids, structures and coupled
problems in engineering. C.A. Mola Soares et.al.
(eds.), Lisbon, Portugal.
Beumler, T. 2004. Flying GLARE. A contribution to
aircraft certification on strength properties in non-
damaged and fatigue damaged GLARE structures.
Thesis for Doctor of Engineering, University of
Delft, Delft university press, The Netherlands, ISBN
90-407-2481-4.
British Standards Institution. 2007. Eurocode 2: Design
of concrete structures, part 1–1: General rules and
rules for buildings and civil engineering structures. BS
EN 1992-1-1:2004.
British Standards Institution. 2008a. Eurocode 4: Design
of composite steel and concrete structures, part 1–1:
General—common rules and rules for buildings. BS
EN 1994-1-1:2004.
British Standards Institution. 2008b. Eurocode 4:
Design of composite steel and concrete structures,
part 2: General rules and rules for bridges. BS EN
1994-2:2004.
Brooking, M.A. & Kennedy, S.J. 2004. The performance,
safety and production benefits of SPS structures for
double hull tankers. International conference on the
design and operation of double hull tankers, Royal
Institution of Naval Architects, London.
Det Norske Veritas. 2001. Recommended practise
DNV-RP-A203, qualification procedures for new
technology.
Det Norske Veritas. 2003. Offshore standard DNV-OSS-
C501, composite components.
Det Norske Veritas. 2006. Offshore service specification,
DNV-OSS-401, Technology qualification management.
Det Norske Veritas. 2007. Offshore standard DNV-OSS-
C502, offshore concrete structures.
Det Norske Veritas. 2008. Offshore standard DNV-
OS-C101. design of offshore steel structures, general
(LRFD method).
Det Norske Veritas. 2010a. Recommended practise DNV-
RP-C203, fatigue design of offshore steel structures.
558
MARSTRUCT.indb 558
Det Norske Veritas. 2010b. Rules for the classification of
ships, newbuildings, hull equipment and main class,
part 3, chapter 1: Hull structural design, ships with
length 100 metres and above.
Fougner, N.K. 1922. Seagoing and other concrete ships.
London: Henry Frowde and Hodder & Stoughton, The
Lancet Building, 1 Bedford Street, Strand, W.C.2.
Gerwick, Jr., B.C. 1984. Implications of recent develop-
ments in concrete ships and floating structures. The
Society of Naval Architects and Marine Engineers
(SNAME) STAR symposium, Los Angeles, California,
April 11–13: pages 199–209.
Iwata, S. & Hattori, Y. 1994. Toughness of composite
steel-concrete structure of sandwich system. Pro-
ceedings of the fourth International Offshore and
Polar Engineering Conference, Osaka, Japan: pages
355–361.
Kennedy, S.J., Bond, J., Braun, D., Noble, P.G. &
Forsyth, J.D. 2003. An innovative “No Hot Work” app-
roach to hull repair on in-service FPSOs using sandwich
plate system overlay. Offshore Technology Conference
(OTC), OTC 15315, Houston, Texas, U.S.A.
Lanquetin, B., Dendani, H., Collet, P. & Esteve, J. 2008.
Lessons learnt from 12 years of operations of a
huge floating production unit made of pre-stressed
high performance concrete. International petroleum
technology conference, IPTC 12546, Kuala Lumpur,
Malaysia: pages 1966–1980.
Lloyd’s Register. 2006. Provisional rules for the applica-
tion of sandwich panel construction to ship structure.
April 2006 edition.
Matsuishi, M., Hattori, Y., Iwata, S., Ishihama, T.,
Yamamoto, T. & Kitazono, M. 1985. Ultimate strength
of composite steel-concrete structure of sandwich sys-
tem. Journal of the Society of Naval Architects of
Japan (JSNA), vol. 157. Reprinted in the English lan-
guage in Naval Architecture and Ocean Engineering,
vol. 24, 1986: pages 233–240.
Morgan, R.G. 1975. History and experience with con-
crete ships. Conference on concrete ships and floating
structures, Berkley, California: pages 3–12.
Narayanan, R., Roberts, T. & Naji, F. 1994. Design
guide for steel-concrete-steel sandwich construction,
volume 1: general principles and rules for basic ele-
ments. Steel Construction Inst., Ascot, UK. ISBN
1-85942-011-7.
Raina, V.K. 1994. Concrete bridges: inspection, repair,
strengthening, testing and load capacity evaluation.
Tata McGraw-Hill publishing company limited, ISBN
0-07-462349-4.
Sare, P.N. & Yee, A.A. 1977. Operational experience
with pre-stressed concrete barges “Concrete Afloat”.
Proceedings of the conference on concrete ships and
floating structures organised by The Concrete Society
in association with the Royal Institution of Naval
Architects, London, UK.
Weitzenböck, J.R. & Grafton, T.J. 2010. Assessment of
the INCA steel-concrete-steel sandwich technology—
public report. Report number 2010–1284, Det Norske
Veritas AS, DNV Research & Innovation: 148 pages.
Zenkert, D. (ed.). 1995. The handbook of sandwich con-
struction. Engineering Materials Advisory Services
Ltd. (EMALS), U.K., ISBN 0947817778, 278 pages.
2/18/2011 5:53:52 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Welding of large gaps of block joints in ships using different

techniques

S. Zacke & W. Fricke

Institute of Ship Structural Design and Analysis, Hamburg University of Technology (TUHH)

ABSTRACT: When welding block joints of ships it may occur that the gaps between them are relatively
large. Furthermore high stresses due to the constraint conditions may result in high residual stresses.
Therefore an investigation was performed to determine if gap widths up to 30 mm fulfil the requirements
regarding the fatigue and fracture strength. The welding of butt joints at 250 mm wide plates was per-
formed under definite constraints presenting the surrounding ship structure. Among other parameters
the welding method (string-bead, weaving technique) was varied. During the welding the reaction force
was measured and it was noticed that it was remarkably higher for the weaving technique compared to the
string-bead technique. Both welding techniques lead to different fracture and fatigue results. The weaving
technique seems to be more adequate when welding large gaps. A numerical welding simulation using the
weaving technique showed compressive plastic strains in the root layer that maybe reasonable for the good
results. Generally the investigations indicated a clear influence of welding technique on the fatigue and
fracture strength. The gap width did not have a significant influence.

1 INTRODUCTION

The welding of large gaps at block joints in ship-

building is regulated by the classifications socie-
ties regarding the welding procedure. At present
the maximum permitted gap amounts to 25 mm.
Furthermore this gap width has to be welded with
the string-bead technique. Additionally high con-
straints due to the surrounding ship structure can
support the development of high and unfavorable
residual stresses and distortions. For this reason
different welding joint variants at 15 mm thick
plates were fabricated to examine their fatigue
and fracture strength. Several parameters were
Figure 1. a) Geometry of the specimens. b) Welding
investigated; three of them are presented in the techniques.
following:
− gap width (8 to 30 mm)
− welding technique (string-bead technique, weav-
ing technique)
− Preloading and stress ratio R
The geometry of the specimens as well as the
investigated welding techniques are presented in
Figures 1 and 2.

2 WELDING OF THE SPECIMENS

The procedure for each investigated specimen

was the following: At first the two plates were Figure 2. Restraining welding set-up.

559

MARSTRUCT.indb 559 2/18/2011 5:53:52 PM

 butt-welded under definite constraint conditions It is assumed that the shrinkage stress of both
in a restraining welding set-up, see Figure 3. welding techniques is very different due to the high
Afterwards the welded plate was released and cut heat input expected for the weaving technique.
for the fatigue tests. Therefore the reaction force transverse to the
The constraint achieved with the welding set-up welding direction was measured during and after
can be interpreted as spring stiffness that was welding. This load becomes part of the fatigue
obtained by several onboard measurements at tests as a preload or mean stress.
block joints within previous investigations (Savu Figure 3 shows the reaction force Fy, which
and Huismann, 2003). ranges between 350 and 700 kN. The upper curve
The main properties of the investigated mate- characterises the weaving technique. Generally
rials are summarized in Table 1. The presented the curve of the string-bead technique is below
results were obtained with the higher strength that of the weaving technique. All curves depend
shipbuilding steel D36. on the gap width, whereas the gradient is steeper
for the weaving technique, overall the gradient is
relatively small.
Reaction Force Fy [kN]

600

400 3 FATIGUE TESTS

200
weaving D36 string-bead D36 The butt welds having a width of 250 mm were cut
0 to 40 mm wide specimens, see Figure 5.
8 15 30 For the fatigue tests altogether three different
gap width [mm]
load types were used, two of them are described
Figure 3. Reaction forces measured after cooling.
in this paper. The difference concerns the stress
ratio R meaning the ratio between the lower and
the upper stress applied in the tests. For load
type I R is equal zero, for load type II R depends
on the measured reaction forces, see Table 2.
The reason for this procedure is to take into
account the heat-induced residual stresses that
differ for both welding techniques.

Figure 4. Testing set-up and geometry of the specimens.

Table 1. Material properties.

Base Filler
material material

Steel grade D36 FILARC PZ 6113 Figure 5. Finite element model for the string-bead tech-
nique and computed stresses.
Yield stress fy 355 MPa For protective gas
Actual yield stress 390 MPa MISON 18:
460 MPa Table 2. Load types for the fatigue tests.
Tensile stress 520 MPa For protective gas
Load types I II
MISON 18:
460 MPa
Δσ [MPa] 270 270
Ultimate tensile strain 29% Min. 22% σupper [MPa] 270 Max. 400
Welding procedure Automatic MAG Stress ratio R [−] 0 Max. 0.325

560

MARSTRUCT.indb 560 2/18/2011 5:53:54 PM

 180 weaving technique
R=0
97.7% with a slope exponent m = 3 to 2⋅106 load
string-bead technique
cycles, the scatter bands shown in Figure 6 are
Fatigue Strength at 2E+06 Load

Ps=10%
R=0,27
140
R=0 obtained. The figure shows that the scatter bands
R=0
are very different for the welding techniques. For
Cycles [MPa]

R=0,32
100
85
60
8 mm gap width
Figure 6. Scatter bands of the fatigue strength.
Generally one can predict or compute the
crack initiation point due the specimen’s bend-
ing distortion resulting from the weld process.
For the specimens welded with the weaving tech-
nique the predictions matched with the tests, for
the string-bead welded specimens, however, they
did not. The expected crack initiation point is at
the top layer, but the crack started from the root
notch. There are several possible reasons for that,
e.g. a worse notch geometry at the root notch.
Due to this the notch stresses were computed with
the finite element program ANSYS, taking into
account the actual notch geometry as well as the
bending distortions that were measured in terms
of angular misalignments.
When comparing the maximum stresses in
Figure 6 for the root and the top layer it can be
seen that both reach a similar stress level. The
unsymmetrical stress distribution on the left and
right side results from the misalignment. The flank
angles of the string-bead technique specimens
were often smaller than 135° and therefore critical.
But there were also specimens having larger flank
angle but they, too, cracked from the root. That
means the large flank angles are a possible reason
for the unexpected crack initiation location, but
due to the fact that all specimens independently
from the flank angle cracked from the root a fur-
ther influence is assumed to be reasonable, i.e. the
material properties in the heat affected zone can be
an important issue.
4 EVALUATION OF THE FATIGUE TESTS
For the evaluation the nominal stress as well as two
local stress approaches were used (Niemi, 2006),
R=0,27 Ps=50%
R=0
111 Ps=90%
the weaving technique the influence of the raised
107 R=0,32
99 96
Ps=97,7% stress ratio R clearly leads to a reduced fatigue
78
85
76
strength. For the string-bead technique, however,
almost no influence regarding the stress ratio could
30 mm gap width
be found. Generally, for small gaps the weaving
technique seems to be adequate, for larger gaps the
string-bead technique has a better fatigue strength.
Regarding the influence of the gap width itself no
clear tendency could be found. It can be concluded
that the influence of the increased gap width can
be neglected.
The local approaches include geometrical
and distortion effects. The two mainly applied
approaches are the structural and the notch
stress approach. Distortion effects as the angular
misalignment are involved in both procedures. The
notch stress approach additionally includes the
local notch geometry. For the local approaches as
well as for the nominal stress approach the clas-
sification societies require a minimum fatigue
class (FAT) to fulfil (Germanischer Lloyd, 2009)
(Hobbacher, 2009):
− nominal stress approach—FAT90
− structural stress approach—FAT100
− notch stress approach—FAT225
Figure 7 compares the resulting FAT classes
from the nominal and local stress approaches for a
gap width of 30 mm.
For each approach and welding technique a rec-
tangle is illustrated representing as upper or lower
boundary the result for the different stress ratios
that were tested. Referring to this it can be seen
that for the string-bead technique the FAT classes
for R > 0 are higher than for R = 0. The weaving
technique has opposite results considering the
stress ratio’s influence, as expected.
With regard to the rule’s requirements for the
three concepts it can be stated that the results of
the weaving technique fulfill the requirements.
For the string-bead technique specimens, however,
300
R=0
30 mm gap width
FAT classes [MPa]

S=String-bead technique S
200 P=Weaving technique
(Radaj, 2006). The evaluation based on the nomi- R>0 R>0
nal stress approach is independent of the local R>0
P R=0 R=0
P
R=0 S
weld geometry and the imperfections of the speci- 100 R>0
R=0 R>0
mens, the result only refers to the applied load and S
R>0
R=0 P
the number of cycles to failure. Therefore all local
0
effects affecting the failure of the specimens are nominal
1 stress
2 structural
3 stress
4 notch5 stress r = 16 mm
implicitly included in this procedure.
When extrapolating the fatigue strength for Figure 7. Comparison of the FAT classes derived from
the survival probabilities Ps = 10%, 50%, 90% and the nominal and local approaches for 30 mm gap width.

561

MARSTRUCT.indb 561 2/18/2011 5:53:56 PM

 only the evaluation with the nominal stress
approach corresponds to the required value, the
local approaches show significantly lower fatigue
strengths than necessary.
Altogether, the results for both welding tech-
niques are very different regarding the local con-
cepts. The reason for the bad fatigue behaviour of
the string-bead technique specimens is the angular
misalignment. This misalignment has two effects:
Due to the downwards-bending of the string-bead
technique specimens an according bending stress
occurs when fixing the specimens in the pulsator for
the fatigue tests. Thereof resulting tensile stresses
in the top layer indicate that the crack initiation
point is there. The second and more significant
effect is the additional or secondary bending that
increases the nominal stress range during loading.
This is caused by the excentric loading due to the
misalignment and results in a bad fatigue strength
of the root.
Altogether the two main questions resulting
from the tests are on the one hand why the string-
bead technique specimens cracked from the root
instead of the top and are there any reasons for
their bad fatigue results.
5 MATERIAL CHARACTERISTICS
One reason for the bad fatigue behaviour could
be the material properties in the weld that espe-
cially influence the fracture behaviour. First of
all hardness measurements and micrographs were
investigated.
The micrographs (Figure 8) show significant dif-
ferences when comparing the welding techniques.
The microstructure of the weaving technique spec-
Figure 8. Micrographs of specimens with 30 mm
wide gap.
562
MARSTRUCT.indb 562
imens is finer and more homogeneous. That of the
specimens welded with the string-bead technique
is coarser.
The fine-grained microstructure results from a
diffusion-controlled phase transformation during
cooling, where large quantities of austenite could
be transformed. Especially for the 30 mm wide gap
specimens welded with the weaving technique a
very homogeneous micro-structure can be seen in
the root layer, this results from the high heat input
per unit length of the weld due to the large weaving
amplitude. The straight-line movement of the
string-bead technique however causes a continu-
ous heat input for every pass having faster cooling
rates. That is why the rough phase austenite is not
totally transformed.
According to the coarse structure of the string-
bead technique specimens their hardness values
are higher than those of the weaving technique
specimens. The hardness measurements presented
in Table 3 do not indicate any abnormalities and
fulfil the requirements of the classification societies.
The comparison shows an increased hardness for
the string-bead specimens by 25HV–30HV.
It can be seen that there is a decreasing trend of
hardness with increasing gap width. Furthermore
the root layer has lower values than the top layer,
this results from the multiple heating of the root
according to the number of passes. This is also
the reason for the decreased hardness for the large
gap width, because the number of passes increases
with the gap width and therefore the root layer is
warmed up more often. In general the results indi-
cate a favourable influence of increasing gap width
regarding the hardness.
Further investigations involve a chemical analy-
sis to exclude possible inclusions or similar welding
defects. Within the scope of the chemical investiga-
tion the base and the weld material were examined.
When comparing the carbon content of the
weld for both welding techniques, it was noticed
that the string-bead technique specimens have a
higher mass content of carbon than the weaving
technique specimens. This confirms the results of
Table 3. Comparison of hardness measurement results
HV10.
8 mm 30 mm
Gap width Weld HAZ Weld HAZ
Weaving (top) 200 178 176 165
String-bead (top) 222 205 207 211
Weaving (root) 180 172 149 157
String-bead (root) 205 201 171 186
2/18/2011 5:53:58 PM
 the hardness measurements, since a higher pro-
portion of carbon explains the higher hardness
values of the string-bead technique specimens.
Overall, the requirements for material composition
of Germanischer Lloyd (2009) are fulfilled for all
specimens. Inclusions or other crack influencing
contents were not found.
As a main indicator for the fracture behaviour
the Crack Tip Opening Displacement values
(CTOD) were also determined. The CTOD val-
ues can be interpreted as the displacement of
the surfaces of a crack normal to the original
(undeformed) crack plane at the tip of the fatigue
precrack. Therefore three point bending tests
were performed using small specimens (cross
section 15 × 15 mm). The CTOD values were
measured with δ5-clips (Eren et al., 2010).
Figure 9 presents the maximum CTOD val-
ues as a result of the bending tests. It was found
that the base material has the highest fracture
toughness CTOD-values within a scatter band of
0.85–1.05 mm. The fracture resistance of the filler
material welded with the weaving technique is in
the range of about 0.7 mm, the CTOD-values of
the string-bead technique are listed in the lower
part of the figure as columns separately for the
different gap widths. On average, these specimens
only reach half the fracture resistance of those
welded with the weaving technique, some values
were even critical.
The material characteristics were investigated
to search for any reasons that maybe have caused
the unexpected crack initiation point as well
as to evaluate the fracture behaviour. In general the
investigations confirm better material properties
for the weaving technique, especially in the root.
The partly martensitic structure and the higher
hardness of the string-bead technique specimens
1 temperature distributions for all models.
Base Material D36
0,8
Weld Material Weaving Technique
0,6
0,4
0,2
Weld Material String-beadTechnique
0
8 mm gap width 30 mm gap width
Figure 9. Maximum CTOD values of A235 and
D36 steels welded by the string-bead and weaving
technique.
563
MARSTRUCT.indb 563
in the top notch indicate a higher strength there.
The crack started in the root notch where a lower
strength is assumed. Due to this, the structure
of the weld is a possible reason for the fatigue
results.
6 SIMULATION OF THE RESIDUAL
STRESSES
To investigate further the influence of the differ-
ent welding techniques several welding simulations
were done with the finite element program ANSYS.
The computation of residual stresses that maybe
influence the fatigue characteristics of a specimen
needs a detailed modelling of the welding process
and especially the welding technique. Due to this
both welding techniques, string-bead and weaving
technique, as well as a simple single-layered butt
weld process were computed. The geometrical
model was the same for all variants, see Figure 10.
The centrepiece in welding simulations is the
heat generation process. The computation subjects
range from modelling of heat generation and weld
pool phenomena, heat flow to thermal stresses and
deformations. The possibilities regarding models
for microstructure evaluation, cracking and other
phenomena that are determined by the tempera-
ture and deformation history of the material are
almost unlimited. The most limiting factor to
date is the technical capability of the operational
computer. The following computation results
are based on a relatively simple model, excluding
phase transformation. The thermal and the struc-
tural part are uncoupled. The heat input refers to
the welding parameters used in the tests for both
welding techniques. To investigate the influence of
the number of weld layers a butt weld with only
one layer was also computed. The heat input for
this model was estimated with the aim of similar
Figure 10. Finite element model, cp. Figure 1.
2/18/2011 5:54:00 PM
 As described before the plate was not fully fixed
during welding. That means the plate could move
perpendicular to the welding direction. This elastic
restraint has also to be assigned to the finite ele-
ment model. Therefore link elements as 3-D spar
elements with the stiffness of 1.250 kN/mm that
was realised in the tests were used. To evaluate
this influence on the residual stress distribution
the fixed and the elastic restrained state was com-
puted. An overview of the computed variants is
presented in Table 4.
To validate the simulation a comparison with
the measured reaction forces was done. Figure 11
presents the results that indicate a good accordance
for both welding techniques. Furthermore the fixed
models show significantly higher reaction forces
pointing out a clear influence of the boundaries.
As already illustrated in Figure 4 the string-bead
technique reaches lower reaction forces compared
with the weaving technique maybe resulting of the
lower energy input per unit length.
A main investigation issue of the project was
the influence of the constraint condition due to
the surrounding ship structure when welding block
joints. Though the elastic constraints of the plates
are relatively rigid due to the high stiffness range
of the link elements the influence on the stress and
strain distribution is noticeable. The transverse
stresses along a center path perpendicular to the
welding direction at half of the plate thickness are
presented in Figure 12.
Three different stress distributions are illus-
trated, the upper ones are simulations realizing the
Table 4. Computed variants.
Boundary condition
Fixed Elastic
String-bead (1 layer) ⌧ technique simulations are significantly affected by
String-bead (20 layers) ⌧
Weaving (5 layers) ⌧ ⌧
Figure 11. Measured and computed reaction forces
after cooling.
564
MARSTRUCT.indb 564
400
Transverse Stress [MPa]
300
200
100
weaving fixed weaving elastic single layer
0
-200 -100 0
Distance to welding line [mm]
100 200
Figure 12. Transverse stresses of weaving technique
(fixed, elastic) and a single-layer computation along
centre path.
0,05
Transverse Plastic Strain [-]
0
single layer
-0,05
weaving fixed
weaving elastic
-0,1
-200 -100 0 100 200
Distance to welding line [mm]
Figure 13. Transverse plastic strains of weaving technique
(fixed, elastic) and a single-layer computation along top path.
weaving technique specimen with five layers, the
lower curve belongs to the single-one layer welding
simulation.
First of all the stress range of the single-layer
model is about 200 MPa below that of the multi-
layer simulations. The stresses are illustrated along
a path in the centre plane of the plate and accord-
ingly they are membrane stresses. Obviously the
single-layer welding is hardly stressed regarding
the transverse stresses. Due to the repeated and
slow-acting heat input the five layered weaving
transverse loading. This was also observed in the
tests as the weaving technique specimens had less
angular distortion on the one hand and high reac-
tion forces on the other hand.
The most decisive difference between the mod-
els concerns the plastic strain distribution. To get
an impression of how significant it is to model the
actual welding technique instead of only loading
a model with a temperature distribution Figure 13
describes the transverse plastic strains along the
same cross path as before on the top of the plate.
The single-layer model has compressive strains
in the weld especially at the notches which mean
the two strain peaks. These peaks are also vis-
ible for the other two models but with a very
low strain range compared with the single-layer
2/18/2011 5:54:01 PM
 model. The global plastic strains, however, are
tensile for the weaving models and zero for the
single-layer model. This is due to the bending
upwards deformation of the weaving technique
specimens that was seen in the simulation as well
as in the tests. The strains of the fixed model are
generally higher than the strains of the elastic
constrained model.
The results confirm that the welding technique
must not be neglected in a welding simulation.
The location and the sign of the plastic strain
is very important for the evaluation of fatigue
results, for example. Furthermore the deforma-
tion behavior of the specimens depends on the
welding technique. Figure 14 presents the vertical
displacement for the string-bead, the single-layer
and the weaving model. The most significant dif-
ference is the deformation direction. Both string-
bead welded models bent downwards and the
weaving technique specimen shows a bending
upwards.
When examining the maximum deformation
values for the string-bead and the weaving mod-
els it seems that the latter has a higher vertical
distortion. Thereby it has to be considered that
these high values are in the weld meaning that the
weld itself is compressed and bent upwards.
The global deformation is lower than that of
the string-bead technique. In the tests the angular
distortions of the fatigue specimens were measured
and the outcome was that the string-bead speci-
mens have about twice as large deformation as the
weaving specimens. Considering the global bending
this is confirmed by the simulation.
Because of the different bending behaviour and
reaction forces the stress distribution transverse
to the weld direction is also dissimilar. Figure 15
shows the transverse stresses along a path at half
the plate thickness.
Similar to the stresses of the single-layer simu-
lation in Figure 12 the string-bead model only
1 layer and tensile strains in the root for the string-
single-layer
Vertical Displacement [mm]
weaving fixed
string-bead elastic
0 are assumed to have a negative influence on the
-1
-200 -100
Distance to welding line [mm]
Figure 14. Vertical Displacement of weaving, string-
bead technique and single-layer computation along
bottom path.
MARSTRUCT.indb 565
400
weaving elastic
Transverse Stress [MPa]
string-bead elastic
200
notch
locations
0
-200 -100 0 100 200
Distance to welding line [mm]
Figure 15. Transverse stresses of weaving and string-
bead technique along centre path.
reaches a stress level below 100 MPa meaning there
is hardly a membrane stress in this plate. Both stress
curves explain the differences of the reaction forces
in Figure 11 as the weaving specimens are mainly
stressed in-plane with only a small bending part.
The string-bead technique specimens show large
bending deformation and less membrane loading.
This effect is traced back to the welding proce-
dure regarding the movement of the heat source
and the heat input per unit length. Furthermore
the weaving movement connects the flanks of the
plates immediately so that the whole shrinkage and
elongation force can be distributed in both parts.
For the string-bead technique, however, at least in
the root layer there is no connection between the
plates and therefore the plate is able to deform eas-
ily due to missing constraint in the weld. The weld-
ing techniques also lead to a different distribution
of the strains. Especially plastic strains develop
mainly during the welding process and result from
the high shrinkage and elongation when the mate-
rial is heated-up or cooled-down. In combination
with the deformation state at this point of time
the plastic strains are generated. Figures 16 and 17
present the transverse plastic strains along a top
and a bottom path.
In summary the plastic strain results in the
weld show small compressive strains in the top
bead technique. Regarding the fatigue results
this means that there are tensile plastic strains at
the crack initiation location (bottom path) that
fatigue strength. The weaving technique model
shows compressive stresses in the whole weld. This
is due to the bending-up of the weld as shown in
Figure 14.
0 100 200 The dimension of the tensile strains is about 1%
and the goal is that this welding technique leads to
unfavourable plastic strains in the root. Therefore the
residual strains are a plausible reason for the crack
initiation at the root and the bad fatigue results.
565
2/18/2011 5:54:03 PM
 Transverse Plastic Strain [-]
-0,002
-0,004
-0,006
-0,008
Figure 16. Plastic strains of weaving and string-bead
technique specimens along top path.
0,02
Transverse Plastic Strain [-]
-0,02
-0,04
-200
Figure 17. Plastic strains of weaving and string-bead
technique along bottom path.
7 CONCLUSIONS
The aim of the investigations described was to eval-
uate the weaving technique in comparison to the
string-bead technique used in shipbuilding practice
as well as the applicability of both welding techniques
for large gap widths welded under unfavorable con-
straints. The focus was on the strength properties
with respect to fatigue and fracture strength, exem-
plified by 15 mm thick steel specimens.
The fracture investigations regarding the material
properties pointed out that the weaving technique
leads to a more homogeneous microstructure, less
hardness values and good crack resistance values
compared to the string-bead technique. An influ-
ence of the gap width could not be found.
With respect to the fatigue tests this result
can be confirmed due to the sufficient and rule
confirming fatigue strength of the weaving tech-
nique specimens. Welding by the string-bead tech-
nique, however, leads to a surprisingly low fatigue
strength for the local approaches. The reason is
a possibly unfavorable combination of material
MARSTRUCT.indb 566
0,002 properties and tensile residual plastic strains in
0 the root. A welding simulation for both welding
notch
techniques shows that the latter effect only occurs
locations for the string-bead technique.
Generally the investigations indicate a clear
weaving influence of the welding technique on the fatigue
string-bead and fracture strength of a weld. The gap width,
-0,01 however, has only marginal effects. In the future
-200 -100 0 100 200
Distance to welding line [mm]
general investigations of different welding tech-
niques could be done with simulations tools as the
welding simulations confirmed the experimental
results.
ACKNOWLEDGEMENTS
The investigations were performed within the
0 project “Strength of block joints welded with wide
gaps in shipbuilding” which was funded with pub-
lic means within the programme “Industrial Coop-
erative Research” by the German Federal Ministry
weaving
of Economics and Technology via the AiF and
string-bead
was coordinated by the Centre of Maritime Tech-
-100 0 100 200 nologies (CMT) in Hamburg. Comprehensive
Distance to welding line [mm] fracture tests and investigations were done by
GKSS Research Centre in Geesthacht, Hamburg.
REFERENCES
Eren, S.E. & Zacke, S. & Fricke, W. & Kocak, M.: Fatigue
and Fracture Strength of Ship Block Joints Welded
with Large Gaps. IIW doc. XIII-2336-10/X-1680-10.
IIW Conference, Istanbul, 2010.
Germanischer Lloyd: Classification Rules I—Ship Tech-
nology, Part 1; Seagoing ships, Section 20 Fatigue
Strength. Self-Publishing, Hamburg, 2009.
Germanischer Lloyd: Classification Rules II—Materials
and Welding, Part 1; Metallic Materials, Chapter 2: Steel
and Iron Materials. Self-Publishing, Hamburg, 2009.
Hobbacher, A.: Recommendations for Fatigue Design
of Welded Joints and Components. IIW doc.1823-07,
Welding Research Council Bulletin 520, New York,
2009.
Niemi, E. & Fricke, W. & Maddox, S.J. 2006, Fatigue
Analysis of Welded Components—Designer’s Guide
to Structural Hot-Spot Approach, Cambridge:
Woodhead Publ.
Radaj, D. & Sonsino, C.M. & Fricke, W.: Fatigue
Assessment of Welded Joints by Local Approaches.
Woodhead Publ., 2nd. Ed., Cambridge 2006.
Savu, D. & Huismann, G. Welding of Block Joints with
Large Gaps. Final Report BMBF-Research Project
03SX133B/ Part P3.1. Universität der Bundeswehr.
Hamburg, 2003.
566
2/18/2011 5:54:05 PM
 Methods and tools for structural design and optimisation

MARSTRUCT.indb 567 2/18/2011 5:54:07 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Scantling optimization of ship structures considering fatigue at the early

design stage

A. Amrane & P. Rigo

ANAST—University of Liège, Chemin des chevreuils, Liège, Belgium

ABSTRACT: Fatigue aspects are leading criterion for the scantling optimization of ships structures.
LBR-5 software is an integrated package which performs, at the early design stage, cost and/or weight opti-
mization of ships structures (scantling). This software takes into account structural constraints (yielding,
buckling, displacement, ultimate strength). Until now, the fatigue failure issue was not implicitly included in
the optimization loop. At the early design stage, limited information and details are available. Therefore sim-
plified methods have to be applied. LBR-5 software provides the nominal stress, so chosen procedure uses the
nominal stress and the Miner’s rule. This approach requires a library of stress concentration factors for pre-
defined typical structural details. The goal of the present work is to develop a tool for fatigue assessment that
can be integrated on LBR5 optimization process, to realize global optimization taking into account the fatigue
issues. Until now, the integration of the fatigue tool is not finished. Only fatigue verification on the optimized
structure is possible. On the present paper, the procedure adopted is presented and a longitudinal scantling of
the mid-ship section of a LNG is optimized with LBR-5. Production cost is considered as objective functions.
The optimized scantling is checked by the fatigue tool. The fatigue criterion is not respected on some panels.
Corrections are performed on these panels to have an optimized scantling without fatigue problems.

1 LBR-5 PRESENTATION ○ Primary stress (longitudinal) due to hull

LBR-5 is the French acronym of “Stiffened Panels
Software” version 5.0. This software is an integrated
package which performs, at the early design stage,
cost and/or weight of stiffened ship structures,
allowing:
• linear 3D analysis of prismatic structures
(generally a cargo hold),
• the calculation of the most relevant limit states
of the structure in an analysis based on the gen-
eral solid-mechanics,
• optimization of the scantlings (plate thickness,
profile sizes, dimensions and spacing),
• including the unit construction costs and the
production sequences in the optimization proc-
ess (through a production-oriented cost objective
function).
LBR5 uses an analytical method (based on dif-
ferential equations of stiffened plates) to compute
the overall response of the hull structure, Rigo
(2005). This method is a direct analysis of the stress
and strain of the prismatic part of the ship or a
cargo hold. LBR5 methodology for the response
calculation can be summarized as follows:
• Primary and secondary stresses
○ Direct analysis of the stress and strain of the
prismatic part of the ship or a cargo hold or a
midship section
bending moment
○ Secondary stress (transverse bending of
frames) due to waves and cargo
• Tertiary stress
○ plate bending (simplified model—Hughes
approach, Hughes (1983)
○ stiffener bending (simplified model—beam
model)
LBR5 allows considering and assessing the
structural constraints: yielding, buckling, displace-
ment and ultimate strength. Until now, the fatigue
failure issue was not implicitly included in the opti-
mization loop.
With LBR5, only the prismatic (cylindrical) part
of the ship structure can be investigated and opti-
mized. Generally, the model is limited by two trans-
versal bulkheads (or cofferdams). The hull is divided
in several interconnected panels (elements), consid-
ered simply supported by the bulkheads. Figure 1
shows a typical LBR-5 stiffened cylindrical shell ele-
ment and a typical stiffened plate element.
LBR-5 is composed of 3 basic modules
(OPTI, CONSTRAINT and COST), (Rigo 2001,
Rigo & Fleury 2001). The user selects the rele-
vant constraints (geometrical and structural con-
straints) in external databases. When the
optimization deals with least construction costs,
unitary material, welding, cutting and labor costs
must be specified by the user to define an explicit

569

MARSTRUCT.indb 569 2/18/2011 5:54:08 PM


Figure 1. Stiffened cylindrical shell and plate element
for LBR5.
objective function (not empirical). For least
weight, these unitary costs are not used and the
objective function depends only on the geometri-
cal parameters. Using all these data (constraints,
objective function and sensitivity analysis), the
optimum solution is found using an optimization
algorithm CONLIN based on a convex lineariza-
tion of the non-linear functions (constraints and
objective functions) and on a dual approach,
(Fleury & Braibant 1986, Rigo & Fleury 2001).
Independent of the number of design variables
and constraints, the number of iterations requir-
ing a complete structural re-analysis is limited to
10 or 15.
For each panel one can associate up to 9 design
variables (XI). These 9 design variables are
respectively:
• Plate thickness,
• For longitudinal members (stiffeners, crossbars,
longitudinals, girders, etc.):
− web height and thickness,
− flange width,
− spacing between 2 longitudinal members.
• For transverse members (frames, transverse
stiffeners, etc.):
− web height and thickness,
− flange width,
− spacing between 2 transverse members
(frames).
Since 1980 the FEM has become a standard to
evaluate constraints on stress, displacement and
ultimate strength at each iteration. With FEM,
570
MARSTRUCT.indb 570
structure analysis of a large structure is quite
demanding and thus represents the major portion
of computing time. LBR-5 does not have the capa-
bility of a finite element analysis and is restricted
to prismatic structures and linear 3D analysis. But,
on the other hand, LBR-5 uses explicit exact first
order sensitivities (derivatives of the constraint
and objective functions by the hundreds of design
variables). Heavy and time consuming numerical
procedures are not required. Sensitivities are
directly available as the method is based on an ana-
lytic solution of the differential equations of cylin-
drical stiffened plates using Fourier series
expansions. So, sensitivity formulations are known
analytically. Due to the efficient CONLIN mathe-
matical optimization algorithm (convex lineariza-
tion and dual approach), optimization of the full
structure can be performed with hundreds of
design variables and constraints using less than
10–15 global structure re-analysis (iterations).
2 FATIGUE BACKGROUND
Crack propagation and fatigue aspects are leading
criterion for the scantling optimization of ships
and FPSO structures, which are submitted to
cycling loading-unloading operations. Fatigue
damage is an extremely localized phenomenon.
In particular, the welded joints are regions of
weakness in a structure for their low fatigue
strength, due to the presence of crack-like defects,
high stress concentration effects, and tensile resid-
ual stress fields caused by thermal expansion.
In general, there are several approaches for fatigue
assessment. In Figure 2, (Radaj et al., 2006), the
most important basic variants of the methods are
plotted. The simplest approach is the Nominal stress
approach. It necessitates only the knowledge of the
main dimension of the structure. But it’s not very
efficient when we consider real complex structures.
On those situations, other methods more relevant
can be used. These methods are structural hot-spot
stress and notch stress approaches. The fracture
mechanics approaches with J-integral or stress
Figure 2. Method for fatigue assessment, (Radaj et al.,
2006).
2/18/2011 5:54:08 PM
 intensity factors are not commonly used due to the
extremely time-consuming structural analysis.
Nominal stress is the stress calculated in general
utilizing elementary theories of structural mechan-
ics based on linear-elastic behaviour (beam or plate
theory) or coarse mesh FEM in the sectional area
under consideration, taking into account the stress
raising effects of the macro geometric shape of the
component in the vicinity of the joint buy disre-
garding the local stress raising effects of the welded
joint, such as e.g. large cut-outs. Overall elastic
behaviour is assumed. The Hot Spot stress is a
local stress at the hot spot taking into account the
influence of structural discontinuities due the
geometry of the connexion but excluding the effect
of the weld. The notch stress is the total (peak)
stress that we found at the notch (toe of the weld,
edge of cut out, …). Additionally to all the affects
considered by the hot spot stress, the notch stress
considers the effects of the weld.
The S-N curves are the most common way to
represent the capacity of the structure. It gives the
relationship between the nominal stress ranges “S”
and the fatigue life or number of cycles to failure N.
Fatigue cracks in welded structures appear on the
welded joints or flame-cut edges. S-N curves are
therefore given for welded joints and flame-cut
edges. The fatigue strength of structure is assessed
by evaluation of fatigue damage ratio. The damage
sum is calculated using a linear cumulative damage
theory and Palmer-Miner rule.
3 PRESENTATION OF THE PROCEDURE
ADOPTED
Due to the complexity of the ship structural details,
the notch stress approach is chosen. The notch
stress method can cover different structural
geometries and details defining fatigue strength
with one design curve.
The fatigue module has to be linked with
LBR-5 software to calculate, at the early stage
design, the fatigue damage on critical connections
of the ship structures. LBR-5 software provides
only the nominal stress. The hot spot stress will be
calculated by multiplying the nominal stress by a
hot spot stress concentration factor. In similar way,
the notch stress can be calculated by multiplying
the hot spot stress by a notch factor.
The S-N curve considered is derived from
Bureau VERITAS rules (1998), for non-tubular
joints in the as-rolled condition with no flame cut
edges and correspond to the curve B provided by
the U.K. department of Energy (1993). This basic
curve is corrected to take into account the influ-
ence of static and residual stress, the effect of
571
MARSTRUCT.indb 571
compressive stress, the influence of thickness and
the Haibach effect.
3.1 Fatigue-critical locations in ship structures
Even if LBR5 software don’t consider the entire
ship structure (only the prismatic (cylindrical) part
at the mid-ship structure), a big amount of fatigue
crack occurs on this area. So, generally, maximum
stress is situated in this area because the hull girder
moment is a maximum there.
A lot of Structural details and connections are
similar between different ship types. But, there are
some special structural details witch depend on the
ship type.
In general, fatigue cracks are observed on geo-
metrical discontinuities, such as connection between
longitudinal and transverse structures.
The end of longitudinal stiffeners, particu-
larly brackets and cut-outs are critical details.
Connections between stiffened plates, for example
at the hopper tank in LNG carriers and cofferdam
area is also critical.
Additionally, on some cases, the ends of pillars,
girders and web frames are fatigue-critical.
3.2 Calculation of stresses at critical locations
The stress evaluation is performed within the
LBR5 software. So, it’s important to extract the
relevant stress for each critical detail.
Depending of the method chosen to perform
the fatigue assessment, three kinds of stresses have
to be taken into account: nominal stress, structural
hot-spot stress and notch stress, Figure 3.
3.2.1 Nominal stress
For fatigue-critical structural details nominal stress
to be considered are the principal stresses at the
Figure 3. Nominal, hot spot and notch stresses, Bureau
VERITAS (2009a).
2/18/2011 5:54:09 PM
 hot spots which form the smallest angle with the σh: Nominal hull girder on the flange of the
crack rising surface. The relevant stress compo- stiffener
nents depend on critical detail considered. σl: Nominal local stress on the flange of the
The number of critical details is relatively large. stiffener
So it’s necessary to gather these details into catego-
ries to obtain a generic approach for the early 3.2.1.2 Critical details situated on the
design stage. plates connections
The critical details are divided into 5 categories: On the plate, there is biaxial stress state. LBR5 soft-
• Critical details situated at the connections of ware can provide normal and shear stresses on
longitudinal ordinary stiffeners with transverse both directions and on top and bottom surfaces of
primary elements the plate. So, the stress to consider is the principal
• Critical details situated on the plates connections stress which is approximately in line with the per-
• Critical details situated at the plate/frames pendicular to the weld toe, i.e. within a deviation
connections of ±60°. According to IIW recommendations,
• Critical details situated at frames flanges Hobbacher (2007), see Figure 6.
• Critical details situated at the girders flanges
Of course, other details categories will be added
σ nominal ( σ1 , σ 2 ) (2)
to this list on the future.
The relevant nominal stresses associated with where σ1 and σ2 are principal stresses
each detail category are described on the following σ x( p + σ y( p
) )
paragraphs. σ 1,2 =
2
3.2.1.1 Connection of longitudinal ordinary 2
stiffeners with transverse primary ⎛ σ x( p ) σ y( p ) ⎞ 2
elements
± ⎜
⎝ 2
⎟⎠ + τ xy ( plat
p e)
(3)
The critical hot spots are shown on Figure 4. The
nominal stress at the hot-spot point is the normal The σx( plate), σy( plate) and τxy( plate) are normal and
stress which is perpendicular to the weld seam. In shear stresses at top or bottom surface of the deck
this case, the maximum principal stress is equal to plate.
the normal stress.
3.2.1.3 Critical details situated at the
σnominal = |σh + σl| (1) plate/frames connections
Lot of cracks may initiate on plate/frames connec-
tions. The stress to be considered is the maximum
is the principal stress at the hot spot.

σ nominal ( σ1 , σ 2 ) (4)

where σx1 and σx2 are principal stresses

σ x( p ) + σ y( p /frame )
σ 1,2 =
2
Figure 4. Typical connection of longitudinal ordinary 2
⎛ σ x( p ) σ y( p /frame ) ⎞ 2
⎟⎠ + τ xy ( platee fframe )
stiffeners with transverse primary elements, GERMA- ± ⎜ (5)
NISCHER Lloyd (2007). ⎝ 2

Figure 5. Biaxial stress at weld toe, Hobbacher (2007). Figure 6. Stresses on plate/frames connections.

572

MARSTRUCT.indb 572 2/18/2011 5:54:09 PM

 The σx( plate) is normal stresses at top or bot-
tom surface of the deck plate at the intersection
with the frame, see Figure 6.
σx( plate/ frame) and τxy( plate/ frame) are normal
and shear stresses at frame web/plate junction.
3.2.1.4 Critical details situated at frame flange
In certain cases, the ends of frames can be a critical
areas. The hot spots are situated on the frame
flanges (Figure 7). The nominal stress is equal to
the normal stress on the frame.
σ nominal σ y( fra
f meflange ) (6)
3.2.1.5 Critical details situated at the girders
For the ends of girder, the hot spots are situated on
the girder flange (Figure 8). The nominal stress is
equal to the normal stress on the flange of the
girder.
σ nominal σ y(Girderflange) (7)
3.2.2 Hot spot stress
After calculating the nominal stress on the criti-
cal locations, the structural hot-spot stress range
Δσhs is defined using hot-spot stress concentration
factor KG.
Δσ hs Δσ nominal ⋅ KG , (8)
The value of the hot-spot stress concentration
factor depends on the type of detail. This approach
adopted requires a library of stress concentration
factors for predefined typical structural details.
Figure 7. Stresses on frames ends.
Figure 8. Stresses on girders ends.
573
MARSTRUCT.indb 573
Values of some critical locations based on litera-
ture review (Bureau VERITAS 1998, 2009b, D.N.V.
2005, Remes et al., 2009) are presented on Table 1.
The values can be updated in the future based on
parametric FE—analysis. The user can also modify
and add new details to the library.
3.2.3 Notch stress
The notch stress range can be from the following
formula, Bureau VERITAS (1998):
Δσ notch 0. F Δσ hs , (9)
KF is the notch stress concentration factor. Its
value depends on the weld geometry and shape.
For flame-cut edges, it depends on the cutting
quality, post treatment and control quality.
For welded elements, the fatigue notch factor,
equal to:
θ
kF = λ ⋅ , (10)
30
λ: Coefficient depending on the weld configura-
tion, and given in Bureau VERITAS (2009a),
Pt B, Ch 7, Sec 4.
θ: Mean weld toe angle, in degrees, without being
taken less than 30°. Unless otherwise specified,
θ may be taken equal to:
• 30° for butt joints
• 45° for T joints or cruciform joints
For flame-cut edges, Kw may be taken equal to
the values defined in Table 2 , depending on the
cutting quality, post treatment and control
quality.
3.2.4 Fatigue damage
The damage sum is calculated using a linear cumu-
lative damage theory and Palmer-Miner rule,
Bureau VERITAS (2009a):
(11)
where
N Number of load cycles during design life
Kp Constant of the design S-N curve
Δσnotch Notch stress range
ξ Weibull shape parameter
μ Coefficient taking into account of the
change in slope of the S-N curve
ΓC Complete Gamma function
The Weibull distribution witch is defined by the
Weibull shape parameter ξ, is used as a probability
density function to describe the long term stress
2/18/2011 5:54:15 PM
 Table 1. Hot-spot stress concentration factor KG.
Concentration
Description of joint Configuration
factor
with non-watertight collar plate:
Connection of longitudinal ordinary stiffeners KG = 1.8
with stiffeners of transverse primary support-
ing members - No bracket with full collar plate (watertight):
KG = 1.7

Connection of longitudinal ordinary stiffeners

with non-watertight collar plate:
with stiffeners of transverse primary support-
KG = 1.65
ing members - With bracket

Critical details situated on the plates connec-

tions KG = 1.8

Critical details situated at the plate / frames

connections KG = 1.8

Critical details situated at frames and girders

flanges- No bracket KG = 1.9

Critical details situated at frames and girders

KG = 1.7
flanges- Vertical bracket

Critical details situated at frames and girders

KG = 1.65
flanges- Horizontal bracket

Critical details situated at frames and girders

KG = 1.5
flanges- Rounded horizontal bracket

distribution. The two slope S-N curve is defined by
the Constant Kp.
Ship structures are subjected to various types of
fluctuating loads. For fatigue assessment, the main
critical loads are induced by the wave action.
The fatigue loadings (moments and pressures)
were defined in LBR5 software by an automatic
transfer of the fatigue loads calculated by Bureau
Veritas Rules from MARS software (BV) to LBR5
optimization tool.

574

MARSTRUCT.indb 574 2/18/2011 5:54:19 PM

 Table 2. Fatigue notch factor for flame-cut edges,
Bureau VERITAS (2009a).

Flame-cut edge description KF

Machine gas cut edges, with subsequent 1,4

machining, dressing or grinding
Machine thermally cut edges, corners removed, 1,6
no crack by inspection
Manually thermally cut edges, free from 2,0
cracks and severe notches
Manually thermally cut edges, uncontrolled, 2,5
no notch deeper than 0,5 mm

On the ship’s life, many loading conditions may

occur. However, many ships are navigating most of Figure 9. 3D view of the “Standard” design.
their life on two loading conditions: full loaded and
ballast loading conditions. Consequently, unless Table 3. Technological limits.
specific situations, fatigue analysis may be limited to
full loaded and ballast loading conditions. Lower limit Upper limit
With each loading condition, four Load cases Variable [mm] [mm]
“a”, “b”, “c” and “d” are associated. Each load
case represents a combination of hull girder loads, Plate thickness 6 25
external sea pressures and internal inertial and Frames web thickness 5 20
fluctuating loads, Bureau VERITAS (2009a). Frames spacing 1000 3500
For each load case, the damage is calculated Stiffeners web height 80 460
(equation (11)) and then a total damage is calcu- Stiffeners web thickness 5 20
lated by the combination of damage of each load Stiffeners flange width 10 100
case, like following described on Bureau VERITAS Stiffeners spacing 400 900
(2009a), Pt B, Ch 7, Sec 4.

The initial scantling is characterized by a weight

4 TEST AND VALIDATION
The fatigue tool as described before calculates the
fatigue damage by using the stress calculated by
the LBR5 software. To be involved on the optimi-
zation process, it implies:
− to calculate sensitivities of the Fatigue module
to the LBR5 design variables;
− to integrate these into LBR-5;
− to validate the integration by executing several
tests;
Until now, this work is not finished. Only
fatigue verification on the optimized structure is
possible.
A longitudinal scantling of the midship section
of a LNG is chosen to perform the optimization
with LBR-5. Production cost is considered as
objective function.
One central tank is considered in all analysis.
The structure is symmetrical, and for facility rea-
sons only half of the tank is modelled. The tank
has a length of 40.5 meters. A 3D view of the
model is presented in Figure 9.
of 18054.74 kN and a cost of 3164759.61 €. These
values are considered for a half of tank having
40.5 m of length.
Before optimization, we have the next weights:
− the net weight = 16715.12 kN
− the corrosion weight = 906.11 kN
− the gross weight = 17621.24 kN
− total weight = 18054.74 kN
The total cost estimated is 3164759.61 €.
First, an optimization is performed without tak-
ing into account fatigue. Different structural con-
straints were used (Von Mises stress ≤ 175 MPa,
ultimate strength of the beam column, plate thick-
ness yielding/buckling and sloshing constraints).
Also, Geometrical (Plate thickness ≤ 2 × stiffener
web thickness, …) and equality (All web-frame
spacing is equal, …) constraints are imposed. The
technological limits used for the optimization are
shown in Table 3.
After production cost optimization, the weight
of the structure becomes 16627.84 kN and the cost
3001482.47 €. Thus, the cost’s gain compared to
the initial scantling is approximately 5.25 %.

575

MARSTRUCT.indb 575 2/18/2011 5:54:20 PM

 Concerning the weight, the next values are
found:
− the net thickness = 15313.81 kN
− the corrosion thickness = 880.52 kN
− the gross thickness = 16194.34 kN
− total weight = 16627.84 kN
After optimization, the optimum scanting is used
to verify if it respect the fatigue criteria. For fatigue
assessment, different panels have been considered:
• Connection of side longitudinal ordinary stiff-
eners with stiffeners of transverse primary sup-
porting members
• Connection of bottom longitudinal ordinary Figure 11. Comparison LBR5/VeriSTAR: intersection
stiffeners with floors inner bottom with cofferdam.
• Connection of inner bottom longitudinal ordi-
nary stiffeners with floors
• Connection of inner bottom with transverse
cofferdam bulkheads
• Connection of inner bottom with hopper tank
sloping plates
• Connection of hopper tank sloping plates with
inner side plates
A comparison is done in order to validate the
fatigue results obtained with LBR-5 code by FE
results using VeriSTAR software provided by
Bureau VERITAS (Figures 10–12). For the early
design stage, the results of LBR5 are acceptable.
On Table 4, are presented the damage of:
Figure 12. Comparison LBR5/VeriSTAR: knuckles.
− the scantling before optimization;
− the scantling after optimization without fatigue
constraints;
− the scantling after optimization after fatigue On the scantling after optimization without
corrections to avoid fatigue damage fatigue constraints, there are fatigue problems on
the panels 8, 24, 25, 26. Stresses on those panels
increase because plate thicknesses and stiffeners
scantlings are decreased after optimization.
Therefore, corrections have to be done on those
panels to avoid fatigue cracks. To decrease the
damage values, the inertia of the stiffener with the
attached plate have to be increased for the hot
spots situated on stiffeners. For the hot spots situ-
ated on the plates, the plate thickness has to be
increased.
By correcting the scantling after optimization
process, the value of the production cost is
3015410.64 €, and the gain compared to the initial
scantling is approximately 4.58%.
The correction to avoid fatigue problems
increases the cost and weight comparatively to the
scantling after optimization process by 0.51% and
1.13% respectively (Table 7).
Material cost represents the most important
part of the increase on the total. This is logical
because only the plate thickness and stiffeners
Figure 10. Comparison LBR5/VeriSTAR: stiffeners. scantling was modified.

576

MARSTRUCT.indb 576 2/18/2011 5:54:21 PM

 Table 4. Fatigue damage on the different scantlings.

Before Optimization (without Optimization

optimization fatigue constraints) (fatigue corrections)

Damage Damage Damage

Connection of bottom stiffeners with floors

Panel 3 Departure Node 0.44 0.40 0.38
Panel 3 Arrival Node 0.41 0.37 0.36
Panel 4 Departure Node 0.40 0.37 0.36
Panel 4 Arrival Node 0.36 0.34 0.33
Panel 5 Departure Node 0.35 0.34 0.33
Panel 5 Arrival Node 0.32 0.33 0.31
Panel 6 Departure Node 0.32 0.31 0.30
Panel 6 Arrival Node 0.37 0.36 0.35
Connection of side stiffeners with transversal frames
Panel 8 Arrival Node 0.58 7.60 0.76
Panel 9 Departure Node 0.7 0.67 0.66
Panel 9 Arrival Node 0.36 0.34 0.34
Panel 59 Departure Node 0.38 0.35 0.34
Connection of inner bottom stiffeners with floors
Panel 24 Departure Node 0.43 13.46 0.65
Panel 24 Arrival Node 0.49 14.06 0.73
Panel 25 Departure Node 0.35 12.38 0.51
Panel 25 Arrival Node 0.42 13.29 0.63
Panel 26 Departure Node 0.33 11.61 0.46
Panel 26 Arrival Node 0.35 12.43 0.51
Connection of inner bottom with transverse cofferdam bulkheads
Panel 24 Arrival Node 0.47 1.83 0.80
Panel 25 Departure Node 0.45 1.81 0.77
Panel 25 Arrival Node 0.48 1.85 0.81
Panel 26 Departure Node 0.21 1.92 0.83
Connection of inner bottom with hopper tank sloping plates
Panel 28 Departure Node 0.21 0.20 0.19
Connection of hopper tank sloping plates with inner side plates
Panel 30 Arrival Node 0.15 0.11 0.11

Table 5. Dimensions of the panels with fatigue problems (net scantling).

Optimization (without Optimization

Dimensions (mm) Before optimization fatigue constraints) (fatigue corrections)

Panel 8
Plate thickness 17.61 13.17 13.2
HP320 × 12

Web height 305 230.7 287.2

Web thickness 10 5.3 10.0
Flange breath 59 92.3 55.8
Flange thickness 35 10.7 32.8
Panel 24
Plate thickness 19 16.55 20.0
HP320 × 12

Web height 305 182.8 287.2

Web thickness 10 6.8 10.0
Flange breath 59 80.6 55.8
Flange thickness 35 10.3 32.8

(Continued)

577

MARSTRUCT.indb 577 2/18/2011 5:54:23 PM

 Table 5. (Continued )

Optimization (without Optimization

Dimensions (mm) Before optimization fatigue constraints) (fatigue corrections)

Panel 25
Plate thickness 19 16.6 20.0

HP320 × 12
Web height 305 182.8 287.2
Web thickness 10 6.8 10.0
Flange breath 59 80.6 55.8
Flange thickness 35 10.3 32.8
Panel 26
Plate thickness 22 16.6 20.0

HP320 × 12
Web height 305 182.8 287.2
Web thickness 10 6.8 10.0
Flange breath 59 80.6 55.8
Flange thickness 35 10.3 32.8

Table 6. Initial, optimized without fatigue constraints stress concentration factors. The difference of the
and fatigue corrected scantling comparison. results compared to VERISTAR can be explained
by the fact that LBR-5 does not have the capability
Mass Gain in Cost Gain of a finite element analysis and is restricted to pris-
Scantling [tons] mass [M€] in cost
matic structures and linear 3D analysis. For the
Initial 1840.44 3.16 early design stage, the results of LBR5 are accept-
Optimization 1694.99 7.90% 3.00 5.16% able and help the designer to choose the best
(without fatigue solution (scantling). The remain work is finishing
constraints) the integration of the fatigue tool inside LBR5
Optimization 1714.13 6.86% 3.02 4.58% optimization loop.
(after fatigue
corrections)
ACKNOWLEDGEMENTS

Table 7. Optimized without fatigue constraints and
fatigue corrected scantling comparison.
Scantling Mass [tons]
Optimized (without 1694.99
fatigue constraints)
Optimized & Corrected 1714.12
fatigue
Difference 19.14
% 1.13%
5 CONCLUSIONS
A fatigue tool has been developed in the frame-
work of the IMPROVE Project. A methodology is
chosen by taking into account of the capabilities
of the LBR5 software and the need to be used early
deigns stage. Some tests were performed by com-
paring the results to the results obtained with
VERISTAR software provided by Bureau VER-
ITAS. This comparison shows that the LBR5
results are not perfect and further tests must be
performed to calibrate more the fatigue module
and to add more critical details on a library of
The authors thank the IMPROVE PROJECT
(FP6) and the associated partners as part of the
present paper has been developed in the framework
Cost [M€] of this project.
3
REFERENCES
3.02
Bureau VERITAS, 1998. Fatigue strength of welded ship
0.02 structures. Paris, Bureau Veritas.
0.51% Bureau VERITAS, 2009a. Bureau VERITAS rules, Pt B,
Ch 7, Sec 4, Paris, Bureau Veritas.
Bureau VERITAS, 2009b. Bureau VERITAS rules Pt B,
Ch 12, Sec 2, Paris, Bureau Veritas.
D.N.V., 2005. Fatigue assessment of ship structures.
Classification notes No. 30.7. Hovik, Norway.
Fleury, C. & Braibant, V., 1986. Structural Optimization:
A New Dual Method using Mixed Variables. Int
J Numer Methods Eng; 23:28.
GERMANISCHER Lloyd, 2007. GERMANISCHER
Lloyd rules.
Hobbacher, A., 2007, IIW Joint Working Group XIII–XV.
Recommendations for fatigue design of welded joints
and components. IIW document XIII-2151-07/
XV-1254-07. Paris, France.
Hughes, O.F., 1983. Ship Structural Design: A Rationally-
Based, Computer-Aided, Optimization Approach,
John Wiley & Sons, NY.

578

MARSTRUCT.indb 578 2/18/2011 5:54:23 PM

 Paik J.K. & Mansour A.E., 1995. A simple formulation
for predicting the ultimate strength of ships, Journal
of Marine Science and Technology Volume 1,
Number 1, 52–62.
Radaj D. et al., 2006. Fatigue assessment of welded joints
by local approaches (second edition). Cambridge:
Abington Publishers.
Remes, H. et al., 2009. Rational models to assess fatigue
at the early design stage. EU FP6 project IMPROVE-
Final Conference; 1:51– 52. Dubrovnik, Croatia.
Rigo, P., 2001. A module-oriented tool for optimum design
of stiffened structures–Part I, Marine structures.
579
MARSTRUCT.indb 579
Rigo, P. & Fleury, C., 2001. Scantling optimization based
on convex linearizations and a dual approach–Part II.
Marine Structures.
Rigo, P., 2005. Differential Equations of Stiffened panels
of Ship Structures and Fourier Series Expansions.
Ship Technology Research; 82–100.
U.K. department of Energy, 1993. Proposed Revision to
Fatigue Guidance.
2/18/2011 5:54:23 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Underwater noise emissions: Another challenge for ship design

M. André
Technical University of Catalonia, Barcelona, Spain

T. Gaggero & E. Rizzuto

University of Genoa, Genoa, Italy

ABSTRACT: The issue of the environmental impact of shipping activities is treated in this work from
the specific viewpoint of underwater noise emissions from the vessels. There has been recently a rising con-
cern about the negative effects that this kind of emissions can have on the marine wildlife in general and
in particular on marine mammals. In these animals, acoustic communication and perception has acquired
a privileged role compared with other senses and other zoological groups. The paper examines the role in
this scenario of noise emissions from shipping and explores how their impact can be quantified, in order
to establish a target for the control of such emissions. The standards recently issued for the characterisa-
tion of underwater noise signature by commercial vessels are also reviewed. The paper builds on the first
results of the project SILENV, funded with the 7th Framework programme of the European Union.

1 INTRODUCTION This aspect of the problem has been considered

The problem of anthropogenic noise emissions in
the sea has been assessed only in recent years. The
marine fauna lives in a medium which poorly trans-
mits light but through which sound propagates very
well, even over long distances. In particular marine
mammals are acoustic specialists and depend heavily
on sound for survival to communicate, to coordinate
their movements, to navigate, to exploit and investi-
gate the environment, to find food and mates and to
avoid predators and other threats. These capabilities
are the result of an evolution of the marine fauna to
adapt to the sea habitat, whose acoustic aspects are
of a paramount importance. Anthropogenic noise
is a threat for this habitat, with negative effects that
can even imply injuries or death for single animals
directly exposed to high sound pressure levels emit-
ted by a specific strong source (sonar, explosion).
In less acute cases, single sources can disturb the
activity of a specific population in a given area,
altering the behaviour and/or implying forced
migrations. On a larger scale, the progressive incr-
ease of the background noise in the oceans due to
a diffused network of different emitting sources
can impact the long-range communication system
of large marine mammals, with implications on the
survival capabilities of some species.
A major attention has been paid in the last decade
to acute consequences like those connected with the
impact of submarine explosions for geological explo-
rations or the operation of powerful military sonars,
but a growing attention is now being devoted to less
acute but more long-ranging types of impact.
mainly at a regional level, in particular for restricted
areas where a high concentration of protected spe-
cies of marine mammals or fishes is present.
The growing attention towards these topics is
reflected in the number of national and interna-
tional associations and treaties which deal with
this subject, among which: The International
Union for the Conservation of Nature (IUCN),
The International Whaling Commission (IWC),
The International Council for the Exploration
of the Sea (ICES), The Agreement on the Con-
servation of Small Cetaceans of the Baltic and
North Seas (ASCOBANS 1992), The Agreement
on the Conservation of Cetaceans in the Black
Sea, Mediterranean Sea and contiguous Atlantic
area (ACCOBAMS 1996). All these associations
and agreements have the aim of increasing the
public attention and concern about the conser-
vation of the marine life. Among other kind of
threats, they refer explicitly to noise, underlin-
ing the need of reducing emissions of noise in
water. The issued documents, however, do not
give explicit limit levels or quieting strategies for
the various sources.
For what specifically concerns noise emissions
from ships, the topic has become officially an object
of discussion within the United Nations Interna-
tional Maritime Organisation (IMO) since 2001.
In the last years the Correspondence Group on
“Noise from commercial shipping and its adverse
impact on marine life” issued several documents,
containing also a rank of quieting strategies (IMO
2009a, 2009b, 2010).

581

MARSTRUCT.indb 581 2/18/2011 5:54:23 PM

 2 UNDERWATER NOISE SOURCES: correlation between the shape and intensity of the
THE SHIP radiated spectra of underwater noise and a few
macro parameters characterising the ship.
Marine life is threatened by the degradation in the
− Dimensions: in general increasing dimensions
habitat caused by intense human activities such as
generate increased sound levels (Mitson 1995).
fisheries, ship traffic, pollution, and coastal devel-
In general the maximum of the spectra shifts
opment. In addition to being affected by chemi-
towards higher frequencies with the decrease of
cal pollution, cetaceans and other vertebrates
the dimensions (Kipple et al. 2004).
are affected by noise pollution. At present, noise
− Speed: the noise in general increases with increas-
represents a ubiquitous form of marine pollution,
ing speed (as the main sources on board operate at
especially in areas of heavy maritime traffic and
a higher energy level) but it is important to under-
developed coasts. Intense underwater noise levels
line that usually the design speed corresponds to
are generated by airguns, widely used for geophysi-
optimal sound level conditions. Off design condi-
cal explorations for oil and gas industry as well as
tions (for example sailing at reduced speed near
for academic and administrative purposes. Other
the coast) can cause higher emissions from the
sources are represented by high power sonar, either
propulsive plant. This is important because it
military or civil, by ship traffic, by shoreline and
underlines that reducing the speed may not always
offshore construction works, and by a series of
be the right strategy to reduce radiated noise.
other commercial, scientific, military and indus-
− Loading conditions: usually ships in ballast
trial emissions. With the most powerful sources
conditions generate higher levels due to a larger
(airguns, sonars, and explosions) this may lead to
cavitation phenomena on propellers associated
direct injures to animals in the proximity of the
to lower hydrostatic pressures on the blades.
source. General ship traffic, heavy industries on
− Type of propeller (fixed or controllable pitch
the coast and other human activities, on the other
propellers, pods): CPP seem to give more prob-
side, generally do not generate such intense noise;
lems as regards underwater noise. There is evi-
but this different acoustic pollution is constant
dence that noise level can increase dramatically
over time and may affect large areas. It may affect
especially when pitch is altered instead of modi-
seriously not only individual animals, but entire
fying the rotational rate to change the vessel
populations. The increased background noise, in
speed. This because the angle of attack is modi-
fact, affects underwater life as airborne noise does
fied differently in the profiles at different radii
on terrestrial animals, human beings included.
because of the rigid rotation of the blade. This
However, considering that sound in water travels
makes the blade work very far from the design
five times faster than in air, and that the density of
conditions and in some cases generates even a
the water transmits acoustic energy very efficiently
negative angle of attack on a few sections, with
over distances much greater than in air, the under-
possible face cavitation phenomena.
water noise effects may extend over very large water
− Ship maintenance: underwater radiated noise
volumes. The knowledge that man-made noise can
seems not to be heavily dependent on ship
affect marine life, marine mammals in particular,
maintenance. Propellers affected by fouling or
and the need for a regulatory system to mitigate
damaged may experience increased cavitation
such effects, has increased over the past few years,
patterns, but are usually cleaned or repaired
mainly with reference to specific types of sources,
during dry-docking (IMO 2009a). Also other
like military sonars and acoustic emissions for seis-
phenomena like misalignment of the shaft are
mic surveys.
usually detected and corrected in a short time.
An increasing concern is however given nowa-
days to all types of noise pollution, and the atten- In the following single sources on board that
tion extends to other zoological groups such as contribute to the whole underwater signature will
fishes and invertebrates. Shipping noise in par- be analysed in more detail.
ticular is considered to be a major source for the
increase in the overall background noise levels in
2.2 Propellers
the oceans. In the following, the main sources of
noise emissions from ships will be analysed. Propellers are probably the most important sources
of underwater noise (IMO 2009a,b). As recalled
in Carlton 2007, there are four main mechanisms
2.1 Macro-parameters influencing ship
which contribute to the noise emission from pro-
underwater emissions
pellers. Some of them are typical of cavitating pro-
When considering the noise emissions from a pellers, other ones occur also on propellers working
ship as a whole, experimental surveys suggest a in non-cavitating conditions.

582

MARSTRUCT.indb 582 2/18/2011 5:54:23 PM

 The fluctuating pressure field generated by the
blade profiles moving in water can be seen as the
sum of two effects:
− The displacement of the water by the propeller
blade profile (thickness effect).
− The pressure difference between the suction and
pressure surfaces of the propeller blades (lifting
surface effect).
Propellers normally operate behind the hull,
which affects the flow field at the propulsor’s posi-
tion in a different way across the propeller’s disk.
Accordingly, the blade profiles work in circumfer-
entially varying hydrodynamic conditions, repre-
sented by water velocities fluctuating in terms of
magnitude and direction. Both the above men-
tioned components of pressure vary along the
circular trajectory of the blades surfaces, thus
generating a fluctuating pressure field at the blade
passing frequencies and their harmonics.
In addition, a broad-band noise at higher fre-
quencies is associated to the turbulence both in the
incident flow and at the trailing edge of the sec-
tion of the blade. The resulting total emission for
propellers in non-cavitating condition is composed
of tones associated to the blade-related frequencies
and a broad band noise at higher frequencies.
As known, cavitation phenomena are related to
the possible occurrence of pressure values below
the vapour pressure and to the inherent formation
of vapour bubbles in the fluid region close to the
blade (usually on the suction side). Such bubbles
exist until the pressure rises again above the vapour
limit and feature during their lifetime a varying
volume. The cavitation volume fluctuates because
of variations in the hydrodynamic components of
the pressure field and also because of the variation
in the hydrostatic component, linked to the blade
distance from the water free surface. Both these
effects change along the circular trajectory of the
blade profiles, making in general cavitation phe-
nomena more evident in the upper region of the
blade motion (close to the stern counter).
Due to these phenomena, two more components
are added to those previously mentioned for non
cavitating propellers, referring respectively to:
− The periodic fluctuation of the cavity volumes
− The sudden collapse process associated with the
implosion of cavitation bubbles when the pres-
sure rises again over the vapour value.
Once again, these phenomena have a periodicity
linked to blade passing through the same region
(in the same hydrostatic and hydrodynamic condi-
tions), so a part of the cavitation noise emission is
generated at the blade rate and its harmonics. The
collapse of cavitation bubbles, on the other hand,
583
MARSTRUCT.indb 583
creates shock waves and hence essentially “white
noise” covering a frequency band up to 1 MHz
(Carlton 2007). Above the cavitation inception
speed, shape and frequency range of the wideband
cavitation spectrum do not change appreciably
with ship speed.
Other types of noise radiation related to propel-
lers are:
− Induced vibrations: the propeller can induce
vibrations on the structure of the hull (through
the shaft connections and also via the pressure
field impinging on the closer hull areas), causing
therefore indirectly an additional emission of
waterborne noise.
− Propeller singing: This is a typical phenom-
ena affecting marine propellers. It is related
to the periodic detachment of vortices, which
induce a vibration of the blade at their reso-
nance frequencies (tonal component). It can be
overcome by modifying the trailing edge of the
blade itself.
As the experience with naval vessels suggests,
a careful propeller design can achieve impor-
tant improvements in the radiated pressure field.
Military technology, if transferred to the commer-
cial ships, can bring a significant reduction of the
emission of this major source on board and there-
fore of the underwater noise pollution from ship-
ping activities.
2.3 Machinery
Other important and powerful sources of noise on
board are represented by the propulsion and aux-
iliary engines.
Most of the seagoing ships are equipped with
two or four stroke diesel engines for the main pro-
pulsion and about all the ships are equipped with
four stroke diesel engines for what concerns elec-
tric power generation. Both are important sources
of noise because of the structure-borne vibra-
tions induced in the hull through various types
of connections. This mechanical energy feeds in
turn the waterborne radiation from the hull. At
lower vessel speed, induced vibrations generated
by the operating machinery, are believed to be the
predominant noise sources at frequencies below
100 Hz (Southall et al. 2007).
Tonal components corresponding to the engine
rotational speed multiples (firing frequency) and
submultiples are present in the engine emission.
They correspond to all those phenomena that
occur within the engine itself: explosions in the
combustion chambers, inertial forces arising from
the alternate motions of pistons in the cylinders.
Higher frequency components are due to the
2/18/2011 5:54:23 PM
 “piston slap”, that is the impact of the piston important results, even though with a significant
against the walls of the cylinder caused by reversals increase in costs.
in the direction of the piston side force. If installed,
reduction gear may also contribute significantly to
2.4 Flow noise
the noise radiation.
Diesel Electric propulsion systems are typically Noise due to the flow around the hull and in par-
adopted on passenger ships and specific military ticular on the appendages, is usually overcome by
units (submarines, mine-hunters). They offer the other types of underwater radiated noise. It may
advantage of a mechanical decoupling between become an important component in high speed
the electric engine mounted on the shaft and the vessels for radiation inside the ship. In general flow
prime mover connected to the generator set. This noise is confined to a region near the hull and does
allows some more freedom in the choice of the not transmit to far distances.
engine location and removes one of the most effec-
tive connections in the noise transmission path to
the hull. Additional frequency components arise 3 THE SOUND FIELD AT SEA
from the electric engine, depending on the quality
of the driving electric signal produced by convert- 3.1 Natural background noise
ers, which may change as a function of load and
Even in the absence of anthropogenic noise sources,
rotational speed of the shaft.
the sea ambient it is not quiet due to other natu-
Gas turbines are sometimes mounted on fast
ral sources. Natural background noise is mainly
ships for their high power/weight ratio, while
composed by: wind, sea state, rain and biological
vapour turbines are nowadays applied mainly on
sounds. As shown in Figure 1, natural background
gas-carriers to utilize the cargo boil off. In general,
noise could reach relatively high sound levels,
turbines operate at high rotational speeds (with
comparable to those from some anthropogenic
emissions at higher frequency if compared to die-
sources.
sel engines) and large reduction gears are necessary
to operate the shaft, thus representing potential
3.2 Propagation
strong emission sources.
Noise components from auxiliary machinery The knowledge of the phenomena governing
and other shipboard equipment may contribute sound propagation is a fundamental aspect for
to the ship’s noise overall signature, but usually at studying underwater noise and its impact on the
lower levels than the propulsion systems (because marine fauna. When characterizing a noise source,
of a lower power and of better possibilities of pre- we need to know how the signal is modified travel-
venting transmission of noise to the surrounding ling from the source to the transducer. On the other
structure). Examples of noise-generating auxil- hand, studying the impact of a source, we need to
iary equipment include: air conditioning plants; know how the signal spectrum is modified before
hydraulic and oil-dynamic alternative or rotating reaching the receiver, which, in the specific case of
pumps; ventilation fans. marine mammals, could be very far away.
A reduction of noise generation and propaga- Sound propagation in the sea is a very complex
tion to water from main items of machinery can be phenomenon influenced by the characteristics of
achieved in general by actions on the sources or on the medium (water) as well as of the boundaries
the transmission path. Belong to the first category (free surface and bottom).
the selection of a quieter type of motor or actuator
or of a quieter model of the same type; balancing
interventions on machinery can also reduce noise
at source. On the other hand the actions that can
be put in practice on the transmission path are:
− isolation of the source from the supporting
structure via resilient mountings
− dissipation of mechanical energy into heat (use
of extensive damping treatments)
− application of enclosures to air-borne noise
sources (isolation)
− use of sound absorbing materials (damping of
air-borne noise pressure)
Also in this case the application of military Figure 1. Natural background noise vs. ship traffic
standards to the control of noise can achieve noise (Urick 1983).

584

MARSTRUCT.indb 584 2/18/2011 5:54:24 PM

 0 efforts have been devoted to the definition of pro-
500 cedures for the quantification of such radiation. In
1000 the following, a couple of recent standards will be
1500 briefly reviewed.
2000
Depth [m]

2500

3000 4.1 ANSI/ASA S12.64-2009

3500
A standard for measurements of ship radiated
4000
noise has been issued by the American National
4500 Standards Institute and by the Acoustical Society
5000
1490 1500 1510 1520 1530 1540 1550 1560
of America (ANSI/ASA 2009).
Sound Speed [m/s] The document covers measurements of under-
water sound pressure levels from ships in prescribed
Figure 2. Example of a sound speed profile. operating conditions. It applies to all kind of sur-
face vessel (with no limitations in size) transiting at
As regards the medium, the transmission of a speeds not greater than 50 kt. Possible aims for
sound is mainly influenced by the celerity profile the measurements mentioned in the document are:
which corresponds to the spatial distribution of to show compliance with contract requirements, to
the propagation velocity. Water pressure, tempera- enable periodic signature assessments and research
ture and salinity affect the speed profile, which var- and development issues.
ies following zones and seasons. A typical celerity The standard offers three grades of measure-
profile is shown in Figure 2. ment (A, B, C) which differ in terms of uncer-
Sound tends to converge to zones with a mini- tainty, complexity and repeatability. It contains
mum in the velocity creating channels in which indications on various subjects:
sound travels for very long distances. For this
− Instrumentation
reason, the path followed by sound could be very
− Signal acquisition
complex, making difficult to evaluate sound lev-
− Test site requirements
els far from the source. Another important aspect
− Sea surface conditions
in sound propagation in the sea is represented by
− Hydrophone deployment
the absorption due to the conversion of acoustic
− Test course and vessel operating conditions
energy into heat or to chemical effects involving
− Test sequence
the salts dissolved in the water. The amount of
− Post processing
energy loss due to absorption is highly dependent
on frequency, in fact high frequencies tend to van- In particular the layout suggested for the hydro-
ish much earlier than low frequencies. phones (grades A and B measurements) is reported
Finally, the sound in the sea can be reflected by in Figure 3.
boundaries (free surface and bottom). Reflections The use of multiple hydrophones is fundamen-
from the free surface are influenced by several tal to capture the source directivity in the vertical
parameters, including: the depth of the source, the plane.
sound frequency, the height of the waves.
The depth of the source influences the so called
Lloyd frequency, below which the phase difference
between the direct and the surface-reflected paths
is less than π. This effect is maximized if the free
surface is flat and perfectly reflecting. Actual con-
ditions include diffraction effects by the perturbed
sea surface which limit the impact of the Lloyd’s
mirror effect. Reflections by the bottom are influ-
enced not only by the shape but also by the com-
position of the bottom layers (i.e. by their capacity
of absorbing sound).

4 MEASUREMENT OF SHIP
UNDERWATER NOISE

In parallel with the increased attention for the

impact of noise radiation from ships, growing Figure 3. Hydrophone geometry (ANSI/ASA 2009).

585

MARSTRUCT.indb 585 2/18/2011 5:54:24 PM

 The typical test for each operating condition
consists in a passage with the buoy at the side. In
Figure 4 the sketch of the test course is reported.
The time at which data start to be recorded is
important for the characterization of the source
directivity in the horizontal plane. In fact the time
duration of records (usually performed symmetri-
cally around the Closest Point of Approach (CPA)
influences the range of variation of the angle
formed between the buoy-ship line and the course
of the ship. The longer the time record is, the
wider is the range of angles, thus including data
in a larger number of different reciprocal positions
between the hull and the hydrophones during the
ship advance.
4.2 DNV Silent Class Notations
In DNV Silent Class Notations (DNV 2010) a
different measurement procedure for underwater
noise is suggested. As it can be seen from Figure 5,
reporting the general layout for tests, the proce-
dure is based on a single hydrophone, placed in the
proximity of an inclined bottom. A correction of
−5 dB is suggested when the hydrophone distance
Figure 4. Test course (ANSI/ASA 2009).
Figure 5. Sketch of the measurement arrangement in
DNV 2010.
586
MARSTRUCT.indb 586
from the bottom is less than 0.2 m to account for
reflections from the bottom itself.
No reflection from the sea surface is taken into
account explicitly, but the adoption of a transmis-
sion loss of 18 log(r) to refer measured quantities
to 1 m is suggested (r = distance of the hydrophone
from the ship). The coefficient multiplying loga-
rithm is therefore between perfect spherical (20)
and perfect cylindrical (10) spreading.
The source directivity is not considered in the
procedure.
A suggestion is contained to investigate differ-
ent operating condition; in particular in addition
to straight course also measurements in thruster
operating conditions are mentioned.
By comparing the two above mentioned proce-
dures for measurements, the ANSI/ASA one seems
to be more effective in characterizing the source,
while the DNV one seems to be prone to provide
results quite site-dependent, valid for comparisons
with other measurements performed in the same
site, but not for a full characterisation. On the
other hand, some of the requirements contained
in (ANSI/ASA 2009) about the test site (e.g. the
minimum depth under the keel) may be difficult
to be fulfilled. No information is provided in the
standard about how to proceed in the case of no
fulfillment (e.g. corrections for shallow water).
This limitation is however acknowledged in the
text, and extension of the text on the subject of
measurements in shallow water is foreseen.
Neglecting the source directivity may have
drawbacks when assessing the interaction with the
surrounding ambient during normal operation.
Directivity plays an important role in particular
when considering special aspects, like the problem
of collisions between ships and cetaceans possibly
due to ‘acoustic shadow areas’ around the hull.
Considering other operating conditions in addi-
tion to normal navigation, as suggested in DNV
(2010), may be of interest, because other condi-
tions may have a larger acoustic impact. Thruster
operation, however, seems not to be representative,
as it occurs usually in ports and is not expected to
impact the marine fauna significantly.
5 THE RECEIVER: MARINE FAUNA
5.1 The perception of sound by Marine mammals
A key aspect in studying the impact of underwa-
ter noise pollution is the sensibility of the receiver,
i.e. how the marine fauna, in particular mammals,
perceive sound. Characterising the hearing capabili-
ties of marine mammals is not a trivial task. First
of all, many species exist, each of them with dif-
ferent hearing sensibilities. For this reason a large
2/18/2011 5:54:25 PM
 number of tests need to be carried out. These tests
are performed in specific pools, subjecting the ani-
mals to tonal noises and recording their reactions
even at neural levels. This allows deriving a curve
of sensibility (audiogram) for the range of frequen-
cies and amplitudes in which the animal can hear.
These audiograms are important means to evalu-
ate the harassment due to noise. It is clear that such
audiograms can be surveyed only on species with a
small size that can be constrained for the purpose
of the test in a small space. For this reason we do
not have audiograms for large mysticetes and we can
only make hypotheses on the frequency range where
they feature a higher sensibility. This information
can be indirectly derived by the frequency ranges on
which they emit, which supposedly correspond to
the peak of sensitivity in the audiograms (troughs in
the curves). This correspondence is verified in many
cases, including human beings. In Figure 6 audio-
grams and communication ranges of odontocetes
are reported together with communication ranges of
mysticetes.
It is clear that mysticetes communicate at fre-
quencies significantly lower than odontocetes.
This consideration could support the hypothesis
that the maximum sensitivity of such mammals is
shifted towards low frequencies.
5.2 The effects of anthropogenic noise
Noise can severely interfere with the life of marine
mammals. Noise pollution can cause marine mam-
mals to abandon their habitat and/or alter their
behaviour by direct disturbance or by masking
their acoustic signals over large areas (Hildebrand
2005); higher sound levels could directly affect their
hearing capabilities by producing either temporary
or permanent hearing losses (Gordon et al. 2004).
All these effects may be critical for the survival
of marine mammals. High energy sound sources
can have immediate impacts and even trigger fatal
events, as recently evidenced by several dramatic
and well documented atypical mass stranding
events (a mass stranding is defined as 2 or more
Figure 6. Audiograms and communication ranges of
odontocetes and communication range of mysticetes.
587
MARSTRUCT.indb 587
animals stranded in the same area) of beaked
whales (e.g. Greece 1996, Bahamas 2000, Canary
Islands 2002), see e.g. Frantzis (1998). In some
cases, anthropogenic high power sound sources (up
to 250 dB re 1 μPa at 1 m distance) radiate low- to
high-frequency sound and individual animals are
exposed to high levels of sound (>160 dB re 1 μPa)
over relatively short periods of time (acute expo-
sure), e.g. in some military sonar operations. In
other cases, potential exposure to high noise levels
can occur for longer periods, weeks and months, as
in the case of seismic surveys or some construction
works (i.e. pile driving for construction of ports,
bridges). Seismic surveys and low-frequency naval
sonar, other than high levels close to the source,
may radiate low-frequency sound over very large
areas thereby exposing populations to lower sound
levels (<160 dB re 1 μPa) over relatively long peri-
ods of time (chronic exposure). Continuous expo-
sure to low frequency sound is also the effect of
distant shipping noise, multiple distant seismic
surveys or construction works.
Although atypical mass stranding events repre-
sent the most dramatic class of incidents related to
acute sound exposure, at least for certain marine
mammal species (beaked whales) (Frantzis 1998), it
should be remembered that the effects of repeated
non lethal exposures and of increased background
noise levels are generally unknown but may poten-
tially have both significant short and long term
effects. Furthermore, the biology of “disturbance”
and the effect of noise on the fecundity of marine
mammals and their prey species are not well
understood. Noise may not always have an imme-
diate impact on cetacean populations but may
indirectly affects them through its effects on prey
abundance, behaviour and distribution.
5.3 Masking
The process known as auditory signal masking
occurs when noise reduces, partially or completely,
the capacity to hear sound or signals. The amount
of the interference depends on the spectrum and
the temporal-spatial relationship between the sig-
nals and the masking noise, among other factors.
(Southall et al. 2007).
In particular a key point is the bandwidth of
the masker and of the masked signals and their
relative position in frequency. The situation for the
case of two relatively narrow signals is depicted in
Figure 7, where the threshold shift due to masking
is shown.
A key parameter in the quantification of the
phenomenon is the critical band, which is the
range of frequency in which the auditory system
of a given species cannot distinguish two tonal
components.
2/18/2011 5:54:25 PM
 6 CRITERIA FOR SETTING LIMITS
TO SHIP’S NOISE EMISSIONS

In general, limits on the waterborne noise emis-

sions from ships can be set on the basis of two dif-
ferent aims, corresponding to a minimisation of
the impact on the marine fauna in terms of:
− behavioral changes in single individuals or
− reduction in communication between members
of the same species
In order to deal with the problem of fixing
limits to noise emissions, the characteristics of
Figure 7. Masking effects of pure tone. sources, receivers and propagation paths have to be
modelled.
10
4 In all the cases the characterisation of the sur-
1/3 Octave Bandwidth rounding ambient (in terms of celerity profile,
Critical Bandwidth
bottom geometry, bottom material and natural
background noise) is needed to take properly into
account sound propagation. The absolute posi-
Critical Bandwidth [Hz]

10
3
tions of sources and receivers are needed, too.
In a simple case, one ship, one emitter and one
receiver can be considered, while in a real case mul-
tiple ships, multiple receivers and multiple emitters
2
can interact with each others.
10
A further simplification consists in omitting the
characterisation of both the emitting and receiving
individuals and concentrating on the disturbance
10
2
10
3
10
4 (ship radiated noise spectrum). In this case it can
Frequency [Hz]
be set as a target that the ship at a certain distance
does not contribute more than the (natural) back-
Figure 8. Critical bandwidth vs. 1/3 octave bandwidth
(for humans).
ground noise (or, similarly, the maximum distance
from the ship at which this occurs can be set). This
approach is effective for preventing both com-
Accordingly, the power contents featured by a munication and behavioural changes, but is quite
masker within the critical bandwidth sum up in conservative; in fact it is plausible that noise levels
affecting the capabilities of perceiving single tonal higher than the background noise could even not
components of sound in the same range. On the be heard in a certain range by animals or, if heard,
contrary, masker signal components acting out of could not interfere with communication.
critical bandwidth of the signal do not affect the Improving the accuracy of the model, further
perception of the signal itself. A typical situation elements could be introduced in the analysis in
is represented by a tonal signal corresponding to a addition to the ship radiated noise.
marine mammal vocalisation and a masker effect As regards communication aspects, if the char-
due to the broadband noise emitted by a ship. acteristics of the emitting individual are available
Critical bandwidths vs. the signal frequency are (vocalisation spectrum), the target can be set in
shown in Figure 8 for humans. The figure shows maximising the area where the signal emitted over-
that the critical bandwidth follows quite closely comes the ship-dominated background noise. The
the line corresponding to a third-octave bandwidth emitted vocalisation results to be ‘perceivable’ in
in the frequency region above 400–500 Hz, while, such area. This condition, however, is necessary
below that region, the critical bandwidth tends to but not sufficient to ensure that the sound is actu-
be constant (and larger than a 1/3 octave). In pro- ally perceived by a receiving individual placed in
portion, the masking effect appears therefore to be that area, because this depends also on the receiv-
potentially larger for low frequencies. ing capabilities.
What above, however, refers to the human ear If the focus is on the prevention of behavioural
and is not applicable for other species. Further, it is changes, in addition to the ship radiated noise, the
noted that a direct quantification of critical band- hearing sensibilities (audiograms) of the receiver
widths is, for large cetaceans, probably even more could be characterised in order to evaluate the har-
difficult than a survey of audiograms. assment due to noise.

588

MARSTRUCT.indb 588 2/18/2011 5:54:25 PM

 Table 1. Categorisation of models.

Level Impact Ship Emitter Receiver

1 Behavioural x
Communication x
2 Behavioural x x
Communication x x
3 Communication x x x

Increasing further the complexity of the analy-

sis of communication problems, the characteris-
tics of the receiver in terms of audiogram and of
critical bandwidth could be added to the informa-
tion about the ship and the emitting animal. With
this approach, the space in which an individual
can actually perceive the signals from another
one, in the presence of the disturbing ship, can be
evaluated. The ratio between the volume in which
communication can take place in the presence of
the ship and the (larger) volume where communica-
tion could occur without the ship can be used as an
indicator for the intrusiveness of the ship’s noise.
On the basis of what above the following cat-
egorisation of possible models could be derived,
following the complexity of data requirements:
The three levels correspond to an increasing
model complexity, but the main difficulties are
represented by the characterisation of the receiver.
In fact for a large number of species, the hearing
capabilities are still unknown, in particular those
of mysticetes. It is also recalled that these charac-
teristics vary for each species.
In the following examples of the models
described above, present in literature, will be
analysed.
6.1 ICES (level 2: behavioural impact analysis)
The International Council for the Exploration of
the Sea (ICES) was the first institution that fixed
limits for the underwater noise radiated from ships.
Starting from the cod audiogram, the ICES pro-
posal considers the lowest point (the frequency of
maximum sensibility: 200 Hz) of the curve at 30 dB
over the threshold sensitivity (Fig. 9); the curve
was interpreted as the limit over which behavioural
effects (escaping) started to appear. This reaction
was further associated to a target distance from the
vessel of 20 m.
The point (20 m, 105 dB) was then represented
on a level vs. distance graph (Fig. 10). The slope
of line derives from the spherical dispersion law,
in fact the level increases of 6 dB when the dis-
tance halves. By this law it was found that, to fulfill
this limit, the vessel noise at 1 m must not exceed
132 dB (Fig. 10).
Figure 9. Hearing threshold for cod is shown, with indi-
cation of how this was utilised (modified from Mitson
1995).
Figure 10. Determination of the ship noise specifica-
tion (at 200 Hz) from the proposed maximum acceptable
fish reaction range of 20 m. The line follows spherical
propagation [modified from (Mitson 1995)].
Finally, a limit curve in the frequency domain
(for the source level at 1 m) is produced, according
to the following procedure. The point (0.2 kHz,
132 dB), corresponding to the previous elabora-
tion (maximum allowable noise level at 1 m) is
adopted as the knuckle of a bilinear limit curve.
The slope of the line for lower frequencies is not
zero to minimize the difficulties to designers and
builders in meeting the essential noise require-
ments. For high frequencies (above 1 KHz) it was
decided to use the law 130 dB-22log(f kHz) with
the aim to stay at about 6 dB over the ambient
noise (Fig. 11).
Fulfilling this limit should ensure that a ship can
approach fish stocks (in particular cods) at 20 m
without scaring them. It is noted that a justifica-
tion in this sense of the curve is available only for
the point at 1 kHz, while the rest of the curve is set
on notional bases.
An extension of the concept could be to take
into account not only the minimum in the audio-
gram of a single species, but the whole sensitivity
curve (or even better the lower envelop of the audi-
ograms for several species).

589

MARSTRUCT.indb 589 2/18/2011 5:54:27 PM


Figure 11. Proposed underwater radiated noise speci-
fication at 11 knots free running for all vessels used in
fisheries research (ICES 1995).
6.2 Communication masking evaluation
As mentioned, masking effects take place in cases in
which a sound cannot be listened or distinguished
because of a disturbance generated by another
sound. The amount of masking effects depends on
the characteristics of both masking and masked
sounds, in particular: frequency range, level of
sound and relative phases between sounds.
In the case of marine mammals communication
we can consider the broadband noise due to ship
traffic as masking noise and the noise vocalizations
of cetaceans (more concentrated in frequency) as
the potentially masked signal. In order to evaluate
the effect of masking due to ships traffic we can
consider that sound and noise are simultaneous.
The model suggested in Clark 2009 takes into
account in principle the whole communication
problem, including, even though in a simplified
way, both the characteristics of the emitting and
receiving individual. The underlying idea is to con-
sider the difference between the communication
space in a pre industrialization natural background
noise condition and the communication space in
nowadays background condition. To this aim, the
sonar equation can be used in order to evaluate
the signal to noise ratio in the frequency range of
interest, which varies from species to species.
To evaluate if a vocalization in a given band can
be heard, the energy of the two signals (ship noise
and mammal emissions) in the band of interest can
be evaluated. In the comparison, the previously dis-
cussed concept of the critical bandwidth has to be
taken into account. In Clark (2009) a 1/3 bandwidth
is used as a first approximation, but the actual criti-
cal bandwidth for mysticetes is unknown.
7 CONCLUSIONS AND FUTURE
PERSPECTIVES
The noise emitted by ships in water has certainly
a responsibility in raising the background noise in
590
MARSTRUCT.indb 590
the oceans. This global problem affects in particular
large marine mammals in their long-range com-
munication capabilities, based on low frequency
signals. Disruption in intra-species communication
may lead to problems in mating and, on the long
range, to impairment of the survival possibilities
of such species.
On the other hand, at a more local level, the
whole marine fauna of a given location can be
affected by noise emissions from intense ship-
ping activities in that area (as well as from other
local anthropogenic sources). These impacts can
reflect, too, in behavioural changes of whole popu-
lations in specific locations and to a local loss of
biodiversity. This is seen to be a problem f.i. in
areas of particular naturalistic interest, where, in
addition to the normal traffic passing by, a higher
concentration of cruise ships and pleasure crafts
tends to be present.
A concern has risen in the international com-
munity about these negative effects and about
how to limit them. The strategies to be followed
for the two classes of problems (global and local)
are probably different, the former one pointing at
a reduction in the emissions of the world’s fleet,
the latter focusing on more selected classes of
ships and locations. In both cases, however, more
refined means for the prediction of sound trans-
mission in the sea are needed, together with an
improved insight in the actual sensitivity of the
marine fauna in general (and specifically of mysti-
cetes) to the ship-generated broadband noise.
A proper quantification of the impact of ships
noise could be used to assess the benefits con-
nected to a given reduction in the noise emis-
sions and would allow to set rational targets. Such
rational targets should in principle be set on the
basis of a cost benefit analysis, weighting properly
the reduced environmental impact of ships and the
increase of costs to achieve it. A viable alternative
to a proper cost benefit analysis is, on the other
hand, to set objectives for the noise containment in
a notional way, based f.i on the existing best prac-
tice and on a progressive reduction of limits.
In any case, new challenges are at the horizon
for the shipbuilding industry as regards the limita-
tion of underwater noise emissions, in view of the
reduction also of this aspect of the ships’ environ-
mental impact.
ACKNOWLEDGEMENTS
This work was developed in the frame of the col-
laborative project SILENV—Ships oriented Inno-
vative soLutions to rEduce Noise & Vibrations,
funded by the E.U. within the Call FP7-SST-2008-
RTD-1 Grant Agreement SCP8-GA-2009-234182.
2/18/2011 5:54:28 PM
 REFERENCES
ACCOBAMS 1996. Final Act of the Negotiation Meet-
ing to adopt the Agreement on the Conservation of
Cetaceans of the Black Sea, Mediterranean Sea and
Contiguous Atlantic Area, Monaco.
ANSI/ASA 2009. ANSI/ASA S12.64-2009/Part 1, Quan-
tities and Procedures for Description and Meas-
urement of Underwater Sound from Ships- Part1:
General Requirements.
ASCOBANS 1992. Agreement on the Conservation of
Small Cetaceans of the Baltic And North Seas, United
Nations, New York.
Carlton, J.S. 2007. Marine Propellers and Propulsion.
Burlington: Butterworth-Heinemann.
Clark, W.C., Ellison, W.T., Southall, B.L., Hatch, L., Van
Parijs, S.M., Frankel, A. & Ponirakis, D. 2009. Acous-
tic Masking in marine ecosystems: intuitions, analysis,
and implication. Marine ecology progress series 395:
201–222.
DNV 2010. Det Norske Veritas: Rules for Classification
of Ships, Silent Class Notation, Part 6, Chapter 24.
Frantzis, A. 1998: Does acoustic testing strand whales?
Nature, 392:29.
Gordon, J., Gillespie, D., Potter, J., Frantzis, A.,
Simmonds, M., Swift, R. & Thompson, D. 2004: The
effects of seismic surveys on marine mammals. Marine
Technology Society Journal, 37:16–34.
Hildebrand, J.A. 2005. Impacts of Anthropogenic
Sound. In: Reynolds, J.E. et al. (eds.). Marine Mam-
mal Research: Conservation beyond Crisis. The Johns
Hopkins University Press, Baltimore, Maryland. pp
101–124.
591
MARSTRUCT.indb 591
IMO 2009a. Document MEPC 59/19, Noise From Com-
mercial Shipping and its Adverse Impacts on Marine
Life.
IMO 2009b. Document MEPC 60/18, Noise From Com-
mercial Shipping and its Adverse Impacts on Marine
Life.
IMO 2010. Document MEPC 61/19, Noise From Com-
mercial Shipping and its Adverse Impacts on Marine
Life.
Kipple B., Gabriele C. 2004. Galcier Bay Watercraft
Noise—Noise Characterization for Tour, Charter, Pri-
vate and Givernment Vessels, Naval Warfare Center,
Technical Report NSWCCD-71-TR-2004/545.
Mitson, R.B. 1995. Underwater noise of research vessels:
review and recommendations. ICES Cooperative
Report 209.
Southall, B.L., Bowles, A.E., Ellison, W.T., Finneran, J.J.,
Gentry, R.L., Greene Jr., C.R., Kastak, D., Ketten, D.R.,
Miller, J.H., Nachtigall, P.E., Richardson, W.J.,
Thomas, J.A. & Tyack, P.L. 2007. Marine mammal
noise exposure criteria: Initial scientific recommenda-
tions. Aquatic Mammals, 33(4):411–522.
Southall, B.L. & A. Scholik-Schlomer. 2008. Final report of
the NOAA International Conference: Potential Appli-
cation of Vessel-Quieting Technology on Large Com-
mercial Vessels, 1–2 May 2007. Silver Spring, USA.
Urick, R.J. 1983. Principles of underwater sound.
New York: McGraw-Hill.
2/18/2011 5:54:29 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Normative framework for noise emissions from ships: Present situation

and future trends

A. Badino, D. Borelli, T. Gaggero, E. Rizzuto & C. Schenone

University of Genoa, Genoa, Italy

ABSTRACT: The work addresses the problem of the ship noise emissions towards the inner part of
the vessel and towards the external space. In the former case, the focus is on the air-borne and structure-
borne transmission to living and working spaces on board. Limits and regulations are devoted primarily
to the preservation of the worker’s health and integrity and to the attainment of comfortable conditions in
resting and recreational spaces for crew and passengers. As regards airborne noise emissions from the ship,
the main concern is for the inhabitants of the living areas near ports and along the coast in case of heavy
ship traffic. Finally, noise emissions in water are considered; in this case, the concern is about their negative
impact on the marine wildlife. The present situation in terms of national, regional and international
requirements and the trends in the normative framework for the various emissions are highlighted.

1 INTRODUCTION while the problem of noise emitted from ships in

The noise produced by transportation means is a
general problem of today’s life, which takes always
the double aspect of emissions towards the inner
part of the vehicle and towards the external space.
This is a common feature for road vehicles, trains,
airplanes and ships. In the specific case of surface
marine vehicles, external acoustic pollution takes the
shape of airborne as well as waterborne emissions,
while the internal transmission is particularly
complicate, due to the complexity and dimensions
of the ship structures, which feature shapes and
scantlings differ from the other vehicles.
The problem of noise impact on board ships
was addressed for the first time in the 80’s. The
first document formalising provisions regarding the
control of noise exposure for workers on board was
the “Code on noise levels on board ships”, issued
by the International Maritime Organisation (IMO)
in 1981. This document was born with the aim to
stimulate and promote noise control on board ships
focusing on workers’ health. Subsequently, in the 90’s,
the main Classification Societies have addressed the
problem from the viewpoint attaining passengers’
comfort too. This latter aspect plays today an
important role in the design and construction of
cruise ships, ferry boats and leisure crafts.
The interest on noise pollution from ships is
today quite high at European level, prompted
by complaints from inhabitants of densely
populated areas near ports, channels and coasts.
The subject however features a less defined
normative framework: at the moment only Directive
2003/44/EC for recreational crafts is available,
port areas is covered only by requirements issued
by local authorities in a rather inhomogeneous
way. Noise emissions in water are a comparatively
new aspect which has not been so far addressed in
detail as regards commercial ships; but the rising
concern about the safeguard of marine fauna and
biodiversity pushed international organism to con-
sider the impact of marine traffic on sea inhabit-
ants, in particular on marine mammals which base
their whole life on acoustic communications.
The state of the art of the knowledge of the phe-
nomena involved in the various kinds of emission
is quite different, also as a result of the time elapsed
since their negative effects were realised and there-
fore studied and regulated. The degree of the
present knowledge is reflected also into the number
and the effectiveness of the indicators that are
available to quantify the impact of such emissions
and, accordingly, to enforce limits and regulations
to keep these effects within acceptable values.
In the following an analysis of the main norms
concerning noise emissions from ships will be
provided.
2 NOISE AND VIBRATION INSIDE
THE SHIP
2.1 ILO (International Labour Organisation)
The International Labour Organization (ILO) is the
tripartite UN agency that brings together govern-
ments, employers and workers of its member states
in common action to promote an improvement of
working conditions throughout the world.

593

MARSTRUCT.indb 593 2/18/2011 5:54:29 PM

 Given the very broad mission of the organization,
the aspect of noise and vibration for workers is
treated in very general documents, covering a large
number of issues, not in details. Some of these
documents are inherent to workers in general,
some to specific classes of workers.
2.1.1 ILO MLC Maritime Labour Convention
For what concerns specifically workers in the ship-
ping sector, the following documents were issued by
the ILO within the Maritime Labour convention.
2.1.1.1 Guidelines on the prevention of N&V
General suggestions on the countermeasures to
be taken in order to reduce the noise impact on
board are provided (e.g. accommodation spaces
as far as possible away from machinery, use of
acoustic insulations, soundproof control rooms in
machinery spaces).
2.1.1.2 Guideline on exposure to noise
Guideline B4.3.2 suggests the competent
authorities, ship-owners and seafarers to address
the problem of noise on board ships with the
objective of improving the protection of seafarers
from the adverse effects of exposure to noise.
2.1.1.3 Convention on the work in the fishing
sector
The ILO Convention N. 188 (ILO 2007a) on the
work in the fishing sector suggests the compe-
tent authority to take measures to limit excessive
noise and vibrations in accommodation spaces in
accordance with relevant international standards.
In particular it suggests (for vessels of 24 metres
in length and over) to ensure adequate protection
to fishers from the effects of noise and vibrations
including the effects for N&V induced fatigue.
2.1.2 Recommendation (work in the fishing
sector)
The ILO Recommendation N.199 (ILO 2007b)
suggests that noise levels for working and living
spaces, which are established by the competent
authority, should be in conformity with the guide-
lines of the International Labour Organization on
exposure levels to ambient factors in the workplace
and, where applicable, the specific protection rec-
ommended by the International Maritime Organi-
zation, together with any subsequent amending
and supplementary instruments for acceptable
noise levels on board ships.
2.2 IMO normative framework
2.2.1 IMO Resolution A.468(XII) “code on
noise levels on board ships”
The IMO (1981) was the first document issued by
IMO regarding noise limits on board; it constitutes
594
MARSTRUCT.indb 594
the basis of all subsequent regulations regarding
noise on board. For this reason, it is a fundamental
document and will be analysed in more details.
The purpose of the code is to ensure suitable and
safe conditions for workers on board preventing
consequences of noise exposure. The code covers
various aspects of the noise control:
− Measurement procedures for noise levels,
including: survey positions, equipment specifi-
cations, signal processing, environmental and
operating conditions for the vessels
− Limits on maximum acceptable noise levels for all
spaces to which seafarers normally have access
− Characterisation of sound insulation perform-
ances and procedures for the experimental
evaluation
− Characterisation of the performances for
personal ear protection devices and inherent
minimum limits.
2.2.1.1 Application
The code applies to ships in service (at port or at
sea) with seafarers on board. It does not apply to
passenger spaces. Ships covered are:
− New ships with gross tonnage ≥1600 tons
− Existing ships with GT ≥1600 ‘as far as reason-
able and practicable to the satisfaction of the
Administration’
− New ships <1600 GT ‘as far as reason-
able and practicable to the satisfaction of the
Administration’.
2.2.1.2 Measurements
Specifications contained in the code cover, as men-
tioned, the various aspects of the measurement
procedure. In particular there are sections devoted
to the identification of:
− Measurement locations in the various spaces on
board
− Equipment characteristics [reference to stand-
ards IEC 651 (1979), IEC 179 (1973)]
− Aspects of signal processing [IEC 225 (1966)]
− Documentation of the surveys
− Provisions for the safety of personnel perform-
ing measurements.
The provisions of this part suffer from the tech-
nical obsolescence of the reference standards.
2.2.1.3 Operating conditions
A key point is also represented by the identification
of the ship conditions during the surveys of noise
levels. Two classes of conditions are covered: at sea
(navigation) and at port. They will be reviewed in
more details in the following:
Conditions at sea: The identification of oper-
ating conditions at sea include environmental
2/18/2011 5:54:29 PM
 conditions, general aspects of the whole ship Table 1. Limits on sound pressure levels imposed by
(loading conditions, propulsion power) and spe- IMO Res. A.468(XII).
cific information about single plants and inherent
running conditions as well as regards the outfitting Limits
Space
of the spaces. In more details, typical conditions type Spaces dB(A)
correspond to:
− Wind force 4 and sea state 3 should not be Work spaces Machinery spaces 90
(continuously manned)
exceeded
Machinery spaces (not 110
− Loaded or ballast condition continuously manned)
− Machinery running at design service shaft speed; Machinery control rooms 75
CPP in normal seagoing position Workshops 85
− All auxiliary machinery, navigation instruments, Non-specified work spaces 90
radio and radar sets, etc., normally, or likely Navigation Navigating bridge and 65
to be, in use at any one time should operate spaces chartrooms
throughout the measurement period Listening post, 70
− Measures in spaces containing diesel engine- including navigating
driven generators, fire pumps or other emergency bridge* wings
equipment should be taken with the equipment and windows
operating Radio rooms 60
− Mechanical ventilation and air conditioning in (with radio equipment
normal operating conditions operating but not
producing audio signals)
− Doors and windows in general closed
Radar rooms 65
− Spaces should be furnished with all necessary
Accommodation Cabins and hospital 60
equipment. spaces Mess rooms 65
Recreation rooms 65
Conditions at port: The identification of operating
Open recreation areas 75
conditions at port includes: Offices 65
− Cargo handling equipment operating Service spaces Galleys, without food 75
− For vehicle carriers noise levels should be meas- processing equipment
ured in cargo spaces during loading and dis- operating
charging operations Serveries and pantries 75
− Measures in machinery spaces with auxiliary Spaces not specified 90
operating.
2.2.1.4 Noise limits
Limits on noise are imposed in terms of A-weighted 2
acceptable levels. The use of A-weighted levels 1 ⎛ pA ( ) ⎞
(which is a standard procedure for industrial
Leq ( ) = 10log10 ∫
24h 24 h ⎜⎝ p0 ⎟⎠
dτ (1)
broad-band noise) is however balanced by further
prescriptions if in accommodation spaces there
are “subjectively annoying low frequency sound or where pA = instantaneous A-weighted sound pres-
obvious tonal components”. In that case the ISO sure; p0 = reference pressure.
noise rating (NR) number should also be deter- In spaces where the level exceeds 85 dB(A) the
mined and the limits specified may be considered use of ear protection and/or the limitation of
as satisfied if the NR number does not numerically exposure in terms of time are prescribed. A detailed
exceed the specified A-weighted level minus 5. representation of the exposure limits is given in
A similar prescription is given for machinery spaces Figure 1. As it can be noted, the partition of the
if a 105 dB(A) level is exceeded. graph zones follows the mentioned two criteria
Maximum acceptable sound pressure levels dB(A): combining values of the two quantities:
In this part the maximum noise levels in dB(A) − Time duration of exposure: none (forbidden)/
allowed in a space are reported in Table 1. A note occasional/daily
specifies that noise levels well lower than the maxi- − Ear protection: none/muffs/plugs/muffs + plugs
mum levels are desirable.
Moreover on the y-axis the maximum acceptable
Time exposure limits: The philosophy of the noise noise levels (corresponding to an exposure time of
exposure limits is to ensure that seafarers will not zero minutes) are reported, while moving along the
be exposed to an Leq (24) exceeding 80 dB(A). The x-axis the time variable takes over and limits are set on
Leq (24) is defined in Equation (1). the exposure time allowed for a certain noise level.

595

MARSTRUCT.indb 595 2/18/2011 5:54:29 PM

 Table 2. Insulation index required by IMO (1981).

Insulation index

Space dB

Cabin to cabin la = 30
Mess rooms, recreation rooms la = 40
to cabins and hospitals

Table 3. Ear protectors characteristics imposed by IMO

(1981).

Reduction
Type of ear
protection dB(A)

Ear plugs −20

Ear muffs −30
Figure 1. Time exposure limits given by IMO (1981).
Ear plugs + −35
ear muffs
The slope of the curves ensures that the energy Octave band centre frequency
perceived by the seafarer never overcomes the
amount given by a time of exposure of 24 hours kHz
at a level of 80 dB(A): in fact the level increases of 0.125 0.25 0.5 1 2 3.15 4 6.3
about 3 dB(A) halving the time of exposure (and
5 dB(A) by dividing it by three). Ear plugs 0 5 10 15 22 22 22 22
It is here noted that the Leq (24) level has been Ear muffs 5 12 20 30 30 30 30 30
more recently confirmed by the World Health
Organisation to be strongly correlated with the
ability of the noise hazard to damage human hear- dB(A) reduction and dB reduction in frequency
ing (WHO 2004). (Octave bands).
2.2.1.5 Acoustic insulation between
accommodation spaces 2.2.2 IMO resolution A.343
In the code there are indications not only regarding (IX)—Recommendation on methods
the receiver (the seafarers) but also the transmission of measuring noise levels at listening posts
path. In particular limits for the airborne insulation This recommendation (IMO 1975) reports further
index for bulkheads and decks are given. The noise instructions on measuring strategies and instru-
insulation index is defined in Equation (2). mentation. In particular it refers to International
Electrotechnical Commission (IEC) standard 179
W1 and 225 for what concerns the sound level meters
la l 10
log [dB] (2) and it gives indications on the environmental con-
W2 W3
ditions at the point of measure (wind between 5
and 10 knots). It suggests limits on the background
where W1 = incident acoustic power; W2 = acoustic
noise levels in the listening posts during measures.
power transmitted to the receiving ambient through
the wall; W3 = acoustic power transmitted to the
2.2.3 SOLAS international convention
receiving ambient through boundary elements or
for the safety of life at sea
other components.
As known, the SOLAS Convention issued by IMO
According to ISO Standard R717, insulation
(IMO 1974) in its successive forms is generally
index should comply at least with the values given
regarded as the most important of all international
in Table 2.
treaties concerning the safety of merchant ships.
2.2.1.6 Ear protection and warning information Provisions as regarding noise control are con-
In the document there are also prescriptions on tained in Chapter II-1 part C Reg. 36 with title
ear protectors and warning notices that should be “Protection against noise” which makes directly
placed on board. reference to IMO (1981) and that is integrally
Protectors have to ensure the following inser- reported below: “Regulation 36 Protection against
tion loss (see Table 3), expressed in terms of overall noise”.

596

MARSTRUCT.indb 596 2/18/2011 5:54:29 PM

 “Measures shall be taken to reduce machinery
noise in machinery spaces to acceptable levels as
determined by the Administration. If this noise
cannot be sufficiently reduced the source of exces-
sive noise shall be suitably insulated or isolated or
a refuge from noise shall be provided if the space
is required to be manned. Ear protectors shall
be provided for personnel required to enter such
spaces, if necessary”.
2.2.4 Comments and future trends about IMO
noise control framework
As it can be noticed, the IMO noise control frame-
work is still based, through IMO (1974) Chapter II-1
Reg. 36, on the IMO (1981) reported above. Despite
the age, the Code still represents a valid reference
from a normative point of view even though in the
preamble of the Code it is stated that: “The Code is
not intended for direct incorporation by reference
or reproduction in national legislation although it
could provide a basis for such legislation”.
The code has been actually largely referred to by
national requirements on noise on board ships and
is deemed to cover comprehensively the subject of
noise control on board. The technical part regarding
equipments and, to some extent, the signal processing
is, on the other hand, obsolete. At IMO, a debate
on the update of the general framework for dealing
with noise aspects is undergoing in the Maritime
Safety Committee through the sub-Committee on
Ship Design and Equipment [see documents IMO
(2007b) and IMO (2009a)].
In the proposal currently discussed (formulated
by all European Countries plus the EU Commis-
sion: see the mentioned documents) the update
regards mainly:
− A reformulation of SOLAS Chapter II-1 Reg.
36, aimed at making the noise limits mandatory
through the insertion of a more explicit refer-
ence to IMO (1981).
− A reformulation of parts of IMO (1981), in
particular:
− Introduction in Chapter 2 of the requirement
of noise predictions at design stage, to be veri-
fied after construction (or modification) and
with regular controls during the ship life.
− New noise limits in Chapter 4.2, with some
minor changes in the definitions of the classes
of spaces on board and a reduction of 5 dB of
the limits in most of them.
− In Chapter 5, it is proposed to introduce noise
exposure limits by taking into account ISO
(1990).
− In Chapter 6, it is proposed to include more
severe requirements for airborne sound insula-
tion (about 5 dB increase, with slight changes
in the definitions of the cases).
597
MARSTRUCT.indb 597
Other updates, regarding more technical aspects
concerning e.g. measurements and equipment are
not yet formalized, but they are foreseen as output
from future activities of working and correspond-
ence groups within the Ship Design and Equip-
ment Sub-Committee.
2.3 Class societies: comfort class notations
In the 90’s, Classification Societies introduced in
their Rules additional, not mandatory, class nota-
tions dealing with comfort on board ships. These
COMF notations are based on standards and
guidance notes relating to noise and vibrations
criteria. The aim of these additional notations is
to evaluate the shipboard habitability and to assess
with an independent certification the comfort of
crew and passengers on board all kind of ships. In
general, three grades of comfort are present in the
Comfort Classes, each of them with different limit
values. Comparing these limits with the IMO ones
(as regards crew spaces, the only ones treated in
IMO 1981) it is noted that the limits of the lower
comfort grade are close to the IMO original limits
while in the higher grades limits are lower. The pro-
visions of Comfort Classes cover mainly:
− Measurement procedures
− Limits on noise and vibration for crew
− Limits on noise and vibration for passengers
− Limits on acoustic insulation
− Limits on impact noise levels.
2.3.1 Criteria for fixing limits
For what concerns measuring procedures, the rules
of the Classification Societies, with some excep-
tions, suggest to carry out measures in standard
conditions, i.e. on straight course at design speed
or 85% of MCR. Low speed or manoeuvring con-
ditions are not taken into account.
Noise limits are fixed in dB(A); this means a limit
on the total power perceived (i.e. filtered by means
of the A-filter). Ships are subdivided in different
categories as a function of type, length and speed.
The general criteria used to assign a limit value
to a space are:
− The type of use of the space
− Private (cabins)
− Public
− The noise level we expect to find
− High (discotheque)
− Medium (restaurant)
− Low (libraries)
− The time spent by passengers in the space
− Long stay (cinema, theatre, cabin)
− Medium stay (restaurant)
− Short stay (shop)
− Passage (corridor).
2/18/2011 5:54:30 PM
 As regards the noise insulation index and the
impact noise level, limits are fixed in function of
the characteristics of the spaces separated by the
wall. For instance, if on one side there is a disco-
theque and on the other one a cabin, the insulation
index must be higher than in the case of cabin to
cabin separation.
In general the Comfort Classes focus on the same
aspects of the IMO (1981) but they contain lower
limits as a result of the availability of more mod-
ern techniques to reduce noise impact and of new
standards of comfort requested by passengers.
2.4 Comments on requirements for noise onboard
For what concerns noise inside the ship, limit levels
are mainly fixed, for the various locations, on the
perceived power [instantaneous levels in dB(A)].
By instantaneous is here meant a value that is
obtained from a short term average [of the order of
minutes]. In the documents devoted to workers’ health
(in particular IMO 1981), limits are fixed on the total
amount of perceived energy during the working shift.
In fact, requirements are set in terms of both sound
pressure levels and exposure time. Limits are fixed
in order not to overcome reference equivalent levels
corresponding to a continuous exposure to a given
level for a given duration [f.i. in IMO (1981): 80 dB(A)
for 24 hours (LAeq(24) = 80 dB(A)]. Problems in apply-
ing this approach are represented by the difficulty in
evaluating precisely exposure times and levels for the
various categories of workers.
Both methods described above do not capture
possible concentrations of energy in time or fre-
quency. Therefore, the annoyance due to inter-
mittent sounds or tonal components may not be
adequately reflected in the indicators so far used
in the rules.
Moreover, the general trend towards a reduction
of broad-band noise levels has probably enhanced
the effect of these components concentrated in
time and/or in frequency.
3 NOISE EMISSIONS IN AIR
Air noise pollution from ships affects people living
near ports or channels interested by intense ship-
ping traffic. When the ship is characterised as a
whole source, it must be considered as a moving
source during navigation in ports or channels and a
fixed source at quay. Within the ship, main sources
of airborne noise are: funnels, air intakes, air dis-
chargers and in general all those openings which put
the internal sources (ventilation systems, engines) in
communication with the external environment. In
the following a brief review of the normative frame-
work existing at different levels will be provided.
598
MARSTRUCT.indb 598
3.1 ISO standards
The following international standards apply to
the measurements for assessing the radiated noise
from ships:
− EN ISO 2922:2000 “Measurement of airborne
sound emitted by vessels on inland waterways
and harbours”
− EN ISO 14509-1:2009 “Small craft—Airborne
sound emitted by powered recreational craft—
Part 1: Pass-by measurement procedures”
− EN ISO 14509-2:2007 “Small craft—Airborne
sound emitted by powered recreational craft—
Part 2: Sound assessment using reference craft”
None of these standards, however, contain limits
or requirements to be fulfilled.
3.2 European regulations
As regards the European legislative framework,
the acoustic indicators currently adopted for
assessing environmental noise are mainly focused
on noise classified in two large categories: mobile
sources (like road-traffic, rail-traffic and aircraft
around airport) and stationary sources, like
those represented by industrial plants or whole
settlements.
Ships are not specifically mentioned; in princi-
ple, however, they belong to both categories as said
before. Two main European directives deal with the
problem of airborne noise: the European Directive
2002/49/EC (EU 2002) and the European Directive
2003/44/EC (EU 2003b). The first one applies to
environmental noise to which humans are exposed
in particular built-up areas, in public parks or other
quiet areas in urban agglomerations, in quiet areas
in open country, near schools, hospitals and other
noise-sensitive buildings and areas. The indicators
used are Lden and Lnight (day-evening-night levels),
that are equivalent levels defined as follow:
1 ⎛ ⎞
Lday Lev . + 5 Lnight
i + 10
Lden l
log ⎜ 12 ⋅ 10 10 + 4 ⋅ 10 10 + 8 ⋅ 10 10
⎟
24 ⎝ ⎠
(3)
where Lday, Levening, Lnight are A-weighted long-term
average sound level as defined in ISO (1987).
The aim of these indicators is to correlate the
measurement of sound pressure level and the per-
centage of people who have negative effects on
their health due to a prolonged exposure to an
examined noise source.
In EU (2003b) limits for airborne noise are given
for recreational crafts in function of the engine
power (see Table 4). Such limits are to be measured
according to, ISO (2007), ISO (2009).
2/18/2011 5:54:30 PM
 Table 4. Airborne noise limits given by EU
directive 2003/44/EC.
Maximum sound
Single engine power pressure level @ 1 m
kW dB
PN ≤ 10 67
10 < PN ≤ 40 72
PN > 40 75
3.3 National regulations
The Italian legislation follows the EU (2002) direc-
tive issuing the Legislative Decree no. 194/2005,
defining the same acoustic indicators and the same
tools of acoustic planning. Currently, the Italian
legislation has not yet defined the sound limits as
required by EU (2002) directive.
Regarding the assessment of noise emitted by
recreational crafts, the Italian legislation follows
the European Directive 2003/44/EC issuing the
Legislative Decree no. 171/05, defining the same
limits for the maximum sound pressure level emit-
ted by this type of vessels. No indications are avail-
able for other units.
Also the Spanish legislation follows the EU
(2002) directive. In Royal Decree 1367 the noise
limits are defined for the three day periods (day,
evening and night) but not for Lden.
Finnish, Swedish and English legislations have
different regulations from EU (2002), but none of
them give specific limits for airborne noise.
3.4 Comments on airborne noise
As above recalled, only recreational crafts, small
crafts and watercrafts are covered by norms. Limits
for ships are not provided, but measurement pro-
cedures for all kind of ships on inland waterways
and harbours are given in ISO (2002).
The main problems in defining limits for air-
borne noise radiated from merchant and passen-
ger ships are due to the complexity of the source
characteristics and of the transmission path.
In addition to the mentioned problem of the
moving source, ship dimensions are often compa-
rable with (if not larger than) the distance from the
receiver. The propagation path is dominated by 3D
effects and in particular by the local morphology
of the surroundings: orography, building geom-
etry. This situation of strong coupling between
the source and the environment, which in addi-
tion differs from case to case, makes it difficult to
characterise the source as such and therefore set
requirements on it. On the other hand, whether
limits set in the receiving positions comply or
599
MARSTRUCT.indb 599
not with specifications, strongly depends on the
particular boundary conditions (f.i. the harbour
geometry and orography). The approach of setting
limits on the receiving position is however followed
in the case of other moving sources, like in the case
of railway transportation, in which pertinence
belts, centred on the source line (railway), define
different limits, decreasing with the distance.
A further problem, in urban areas of ports, is
the superimposition of ships’ airborne noise on
noise from all other sources: the choice to consider
ships as a category of urban noise sources or to
submit them to a specific regulation can conduce
to different environmental policies. The manage-
ment of such a matter, together with the definition
of the authority responsible for its application, is
a fundamental aspect to get an effective control of
port noise.
4 NOISE EMISSIONS IN WATER
The problem of noise emissions in water from
ships has been treated in military ships for the pur-
pose of limiting the acoustic signature and then
the detectability of the ship. More recently, investi-
gations about the impact of military sonars on the
marine wildlife have been carried out as a conse-
quence of the correlation of massive beaching of
mammals to military activities in the same area.
As regards commercial ships, the impact of
noise on marine fauna, considered only in the last
few years, is mainly represented by long range and
long term effects.
At IMO level some recent documents (described
below) regarding underwater noise from ships and
its impact on marine fauna have been issued.
4.1 IMO framework
IMO deals with the problem in the following docu-
ments in chronological order:
− IMO (2001). The purpose of these Guidelines
is to provide guidance to Contracting Parties to
the International Convention for the Prevention
of Pollution from Ships, 1973, as modified by
the Protocol of 1978 in the formulation and sub-
mission of applications for the designation of
Special Areas under Annexes I, II, and V to the
Convention. In the Guidelines noise is explicitly
mentioned in the list of pollutants.
− IMO (2007a), IMO (2008). This information
paper sent both to the Marine Safety Committee
and to the Marine Environment Protection Com-
mittee advises of the issue of noise generated by
international shipping and its potential adverse
impact on marine life. It requests Member
2/18/2011 5:54:30 PM
 Governments to note this information; informs
all interested entities, in particular those from the
shipping industry, shipyards, and ship builders
of this issue, and invites them to participate in
the ongoing dialogue regarding identification of
potential adverse impacts associated with vessel
noise and the potential mitigation of those
impacts.
− IMO (2009a). This document assesses the inci-
dental introduction of noise from shipping
Figure 2. Proposed underwater radiated noise speci-
activities in the oceans. Main underwater noise
fication at 11 knots free running for all vessels used in
sources from ships are analyzed. fisheries research (Mitson 1995).
− IMO (2009b). This document is the report of the
Correspondence Group on the issue of “Noise
from commercial shipping and its adverse
impact on marine life”. The Correspondence
Group was established to identify and discuss
ways to minimize the incidental introduction of
noise from commercial shipping operations into
the marine environment and to reduce potential
adverse impacts on marine life. In the document
the potential contribution for each of the ship
components to the radiated sound power are
analysed. Figure 3. Abstract of limit curves by DNV Silent Class
− IMO (2010). This document is the report of the Notation for different ships.
Correspondence Group on the issue of “Noise
from commercial shipping and its adverse impact
on marine life”. In the document it is established
to treat underwater noise sources according to
their relative contribution. In particular “since
shipping noise in characterised by spectral sig-
nature which clearly relates to propellers” the
committee agreed to concentrate on propeller
cavitation and to link it to ship speed, loading
conditions etc. Moreover “The Correspondence
Group addressed the demand for further
research activities” and drew up a ranking of
quieting strategies.
Figure 4. Schematic drawing of measuring situation
for surface vessel by DNV Silent Class Notation (DNV
4.2 ICES: international council for the 2009).
exploration of the sea
The very first attempt to fix limits of underwater limits to underwater emissions. This document
noise emissions from ships came from the Inter- gives indications on both measurements proce-
national Council for the Exploration of the Sea dures and limits of underwater emissions. Limits
in its Cooperative Research Report 209 issued in are given in function of vessel typology; in Figure 3
1995. The document contains a limit curve (Fig. 2) an abstract of limit curves is reported.
for fishing research vessels derived from the cod’s In the regulation a measurement technique
audiogram. The aim of this limit curve is to select for underwater noise is described (Fig. 4). In this
vessels that can come close to fishes without dis- configuration measurement could be affected by
turbing them. reflection from the free surface and the bottom.
A correcting factor is suggested to take into account
these effects but they are not given in function of
4.3 Classification societies
bottom slope, bottom composition, sea state and
As regards Classification Societies a very recent reg- other parameters influencing reflection. This need
ulation issued by Det Norske Veritas (DNV Silent that measurements in different locations could give
Class Notation) represents the first attempt to fix different results.

600

MARSTRUCT.indb 600 2/18/2011 5:54:30 PM

 Other standards for underwater noise emissions
measurements have been issued, such as the recent
ANSI-ASA S12.64-2009/Part 1 “Quantities and
Procedures for Description and Measurement of
Underwater Sound from Ships—Part 1: General
Requirements” in which a complete procedure to
characterise the acoustic underwater signature of
ships is given.
4.4 Comments on underwater noise
As regards underwater noise, no limits or norms
exist so far, even if the introduction of local limits on
underwater noise emissions, particularly in special
protected areas, is foreseen in the next future, due
to an increasing concern about the conservation of
marine mammals. Only ICES tried to fix limits for
ships starting from the characteristics of a specific
receiver (cod), but the wide differences between the
hearing sensibility of different species suggest to
consider a more varied target. The prediction of
noise levels at a distance from the source is further
complicated by the transmission of sound in the
ocean, which is influenced by a great number of
parameters often unknown. Moreover, the real
impact of noise on the marine fauna is not yet
clarified in all its aspects, mainly as regards long
term effects.
5 CONCLUSIONS AND FUTURE TRENDS
5.1 Noise inside the ship
As regards noise inside the ship, the present
regulatory framework is to a large extent still based
on the IMO (1981) which focuses on the limita-
tion of the total perceived acoustic power [dB(A)],
even if the possible relevance of tonal components
is also recognised. Moreover time duration of the
exposure is considered for crew. This is the so called
“noise dose”. With this approach the attention is
focused on the receiver and in particular on the total
amount of energy perceived during working hours
by the seafarer, who can be subjected to different
levels for different periods. The “noise dose” con-
cept is currently widely used in the industrial field
and could be systematically extended to ships.
Comfort classes focus mainly on the same aspect
of limiting the total perceived acoustic power but
lowering limits. The “noise dose” is not treated.
About the noise exposure for workers, the evo-
lution of industrial regulations towards lower limit
values and the attention to the peak levels of noise
(impulsive component), as in case of the European
Directive 2003/10/EC (EU 2003a), indicates a trend
that will probably be extended to the shipping
sector in a near future.
601
MARSTRUCT.indb 601
To increase the acoustic comfort for passengers
and crew, an improvement on regulations should
move in the direction of a more specific evaluation
of the sound characteristics. This means focusing
on the characterisation of the effects of:
− frequency spectra
− tonal components
− intermittent sound
These elements deeply influence the sound well-
ness on ships. To evaluate and analyse the corre-
late noise annoyance, several methods are in use
in the civil field (Beranek, 1988), whose use can
be extended to ships. Some of them are reported
below:
− Noise Criteria (NC) method
− Balanced Noise Criteria (NCB) method
− Room Criteria (RC) method
− Room Criteria Mark II (RC Mark II) method
− Room Noise Criteria (RNC) method
Without going into details about these methods,
their main aim is to evaluate the noise harassment
characterising the “quality” of the signal on the
basis of its frequency distribution.
Another important improvement in regulations
could be the enclosure of non-standard measuring
conditions. As reported in Chapter 2.3, COMF
classes regulations, prescribe to carry out measures
in standard conditions, leaving the possibility to
make measurements in additional different condi-
tions to the owner’s discretion. Experience actually
suggests that often low speed navigation (e.g. coast-
ing for cruise ships) and manoeuvring (e.g. thrust-
ers operations) are more noisy situations because
the whole propulsion chain works far from design
conditions.
5.2 Noise in air
Specific regulations fixing limits for noise radiated
in air from ships do not exist. However, a growing
attention to noise pollution is forcing international
bodies to deal with the problem. In the future,
measurement procedures and limits for air radiated
noise should be fixed. The ship is a complex 3D
source which cannot be easily modelled because
sources of air noise (see Ch. 3) could be located at
higher levels in the case of funnels (about 50 metres
and over for biggest ships) or at lower levels in
the case of ventilation systems. Moreover, during
navigation in ports or channels the sources are in
movement. All these aspects increase the difficulties
in defining limits and measurement procedures.
In fact, accurate measurement procedures should
include surveys at altitudes comparable to altitudes
of sources (e.g funnels) and receivers (e.g. upper
floors of buildings), in positions therefore not
2/18/2011 5:54:32 PM
 easily reachable. Furthermore measures should
also be carried out while the ship is travelling.
5.3 Noise in water
The topic of underwater radiated noise is the newest
in the field of noise pollution from ships. The
main problem in fixing limits for such emissions
is the characterisation of the receiver. In fact, the
species living in the oceans have extremely different
sensitivities due to different goals for which they use
sound (for more details see e.g. Andrè et al. 2011);
moreover it is very difficult to collect sufficient
data to compute audiograms especially for species
that spend the major part of time submerged.
As regards, for instance, mysticetes, it is known that
they are able to produce sound at very low frequen-
cies (about 20 Hz or less) in order to communicate
with other individuals at long distances, but do not
exist enough data to compute an audiogram of sen-
sitivity. On the other hand, odontocetes produce
sounds at very high frequencies (until 100 kHz) for
echolocation and due to their size, which allows
laboratory experiments, many audiograms are
available. For these reasons a unique approach for
fixing limits it is not still reached.
ACKNOWLEDGEMENTS
This work was developed in the frame of the
collaborative project SILENV—Ships oriented
Innovative soLutions to rEduce Noise & Vibrations,
funded by the E.U. within the Call FP7-SST-2008-
RTD-1.
Grant Agreement SCP8-GA-2009-234182.
REFERENCES
André, M., Gaggero, T., Rizzuto, E. 2011. Underwater
Noise Emissions: Another Challenge for Ship Design.
MARSTRUCT 2011.
Beranek, L. 1988. Noise and vibration control. Institute
of Noise Control Engineering.
DNV 2009, Det Norske Veritas: Rules for Classification
of Ships, Silent Class Notation, Part 6, Chapter 24.
EU 2002, Directive 2002/49/EC of The European
Parliament and of The Council, 25 June 2002.
EU 2003a, Directive 2003/10/EC of the European
Parliament and of the Council, 6 February 2003.
EU 2003b, Directive 2003/44/EC of The European
Parliament and of The Council, 16 June 2003.
ILO 2006. ILO MLC Maritime Labour Convention.
602
MARSTRUCT.indb 602
ILO 2007a. ILO Convention No.188 concerning work in
the fishing sector.
ILO 2007b. ILO Recommendation No.199 concerning
work in the fishing sector.
IMO 1974. SOLAS International Convention for the
Safety of Life at Sea.
IMO 1975. Resolution A.343(IX) Recommendation on
Methods of Measuring Noise Levels at Listening
Posts.
IMO 1981. Resolution A.468(XII): Code on Noise Levels
on Board Ships.
IMO 2001. Resolution A.927 (22). Guidelines for the
Designation of Special Areas Under Marpol 73/78
and Guidelines for the Identification and Designation
of Particularly Sensitive Sea Areas.
IMO 2007a. Document MEPC 57/INF.4 Shipping noise
and Marine Mammals.
IMO 2007b. Document MSC 83/25/13, Proposal for
protection against noise on board ships.
IMO 2008. Document MSC 84/INF.4, Shipping Noise
and Marine Mammals.
IMO 2009a. Document DE 53/10, Proposals for the
development of amendments to SOLAS regulation
II-1/36 and a revision of the Code on noise levels on
board ships.
IMO 2009b. Document MEPC 59/19, Noise From
Commercial Shipping and its Adverse Impacts on
Marine Life.
IMO 2009c. Document MEPC 60/18, Noise From
Commercial Shipping and its Adverse Impacts on
Marine Life.
IMO 2010. Document MEPC 61/19, Noise From Com-
mercial Shipping and its Adverse Impacts on Marine
Life.
ISO 1987, ISO 1996-2, Acoustics—Description, meas-
urement and assessment of environmental noise Part
2: Determination of environmental noise levels.
ISO 1990, ISO 1999, Acoustics—Determination of
occupational noise exposure and estimation of noise-
induced hearing impairment.
ISO 2000, UNI EN ISO 2922, Measurement of airborne
sound emitted by vessels on inland waterways and
harbour.
ISO 2007, UNI EN ISO 14509-2, Airborne sound emitted
by powered recreational craft.
ISO 2009, UNI EN ISO 14509-1, Airborne sound emitted
by powered recreational craft.
Mitson, R.B. 1995. Underwater noise of research vessels:
review and recommendations. ICES Cooperative
Research Report, 209.
Parson, Chris et al. 2004. Oceans of noise 2004. WDCS
Science Report.
WHO 2004. Occupational noise: Assessing the burden
of disease from work-related hearing impairment at
national and local levels. World Health Organization.
Geneva.
2/18/2011 5:54:32 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Methods and criteria to manage airborne outdoor ship noise

M. Biot & L. Moro

Division of Naval Architecture and Marine Technology, Department of Mechanical Engineering
and Naval Architecture, University of Trieste, Trieste, Italy

ABSTRACT: The present study has been performed by the University of Trieste within the BESST
European Project with the aim to design an operational procedure for predicting ship’s outdoor noise
in harbour. Procedure is here outlined in all its steps from the pre- to the post-processing phase and its
effectiveness is proved. Method is implemented by application to a comprehensive case study. Numerical
simulations are performed by exploiting a selected commercial software suited for acoustic mapping on
noise emission in open field conditions. Results are then discussed with the aim to check operational
procedure and computational tool. Basing on the same results, suggestions on the proper technique to
approach the topical subject of ship’s outdoor noise in harbour are given. Paper starts with an introduction
covering the matter of ship’s noise pollution in harbour and how it is today managed.

1 THE PRESENT FRAMEWORK

1.1 The ship as a noise source
In past decades European ports have become
increasingly complex industrial areas. From simple
piers for loading and unloading cargo and people
embarking and disembarking, ports have developed
in areas with many industrial activities that are
sources of noise. Moreover, in the last years, several
ports have been expanded, becoming closer to
residential areas. This is causing a constant increase
in noise pollution for residents in those areas.
Ships are one of the main noise sources in ports.
The increasingly dominant role played there by
ships is due to both growth of the maritime traffic
and power installed on board. On board we can
find different noise sources, the primary ones are
the ventilation/conditioning plants and especially
all machinery in engine rooms.
The main part of the outdoor noise during
harbour stay comes from the Diesel generator
engines exhaust placed at the top of the funnel.
In second place, there are the inlets/outlets of
each station of the HVAC (Heating, Ventilation
and Air-Conditioning) system, which radiates
noise from outer openings usually placed, for the
majority of merchant vessels, on the upper deck.
HVAC inlets/outlets contributing significantly to
the outdoor noise are those related to the engine
rooms: intakes for air supply to diesel engines and
auxiliaries and outlets of extractor fans.
Among the different types of merchant ships, those
requiring running a lot of engines and equipments
during their stay in harbour are of major concern as
far as noise pollution is considered. That is the case
of cruise and ro-ro/pax vessels, where large systems
need to be supplied to guarantee to passengers a lot
of onboard duties, like the huge air conditioning
system, which outer openings placed on the ship’s
sides give a hard contribution in increasing outdoor
noise in the surroundings.
Onboard mechanical systems which are essential
for the operation of ship at quay may not be totally
silenced. Sources of transient noise like transverse
propulsion units may likewise regarded as inevitable,
being them related to the ship’s manoeuvring.
Or may not, as ships could be assisted by tugs, but
which solution is less noisy for people living in the
residential areas close to port is an open question.
With regard to noise, the time at berth is usually
significantly longer than the maneuvering time,
so it may be considered more significant.
Secondary noise sources, like hydraulic pumps
and winches, are those generating low noise power
levels; they may also work with frequent starts and
stops. Due to inherent difficult in treating them by
limits and rules, they are commonly neglected as
individual noise sources and are considered as a
lump source.
Other noise sources may also generate airborne
noise, like ancillary onboard devices and activities
(e.g. public address systems, organised events on the
open deck of cruise vessels). Anyway, ship’s designers
are not directly involved in attempt of silencing them.
That even holds in cases of noise due to quay
based appliances for cargo handling or to the
wheeled cargo itself, which could be a major cause of
radiating noise from the ship to the surroundings.
An important study on the present state of ship’s
noise pollution in port has been carried out by

603

MARSTRUCT.indb 603 2/18/2011 5:54:32 PM

 Lloyd’s Register ODS for the Danish Environmental
Protection Agency (Lloyd’s Register ODS 2010).
As for the engine exhausts, it states that the sound
power levels vary between 135 and 142 dB(A), while
the ventilation fans radiate with levels from 81 to
110 dB(A). Engine exhaust noise typically varies
about 10 dB from full load to no-load condition, on
the other hand, if a fan is not operated at the design
conditions the sound power will generally increase.
1.2 The case of cruise ships
The typical positions of the noise sources on
board a cruise ship are shown in Figure 1. Most
noise sources are located near the funnel: in this
area are placed the openings for both the engine
and auxiliary rooms ventilation systems (and for
the combustion air supply of the main engines
and auxiliaries). Moreover, at the level of funnel
foundation, the engine rooms exhaust air fans are
sited. Other important noise sources are the air
conditioning openings usually located along the
ship’s side for about the whole ship’s length.
The Diesel engine exhausts, often placed at the
top of the funnel, significantly affect the level of
noise propagating even at large distances in the
surroundings. Funnel’s top is indeed the highest
and not shielded ship’s radiating source. Moreover,
the great part of the energy content of the radiating
noise generated by Diesel generator engines is
related to the lower frequencies, which are very
hard to be silenced. The frequency content of the
exhaust noise from a given Diesel generator engine
mainly depends on factors related to the engine
itself (the number of cylinders) or to the generator
(the fixed speed of the alternator shaft). As for the
last factor, because AC-current generated onboard
is either 50 Hz or 60 Hz and because marine alter-
nators onboard large cruise ship’s commonly have
pole pairs in number of 4 to 7, the typical diesel
generator engine speeds range from about 500 to
750 rpm (slower speed engines in larger vessels).
The most significant sound pressure levels
are associated to frequencies equal to the firing
Figure 1. Position of the noise sources onboard a
cruise ship.
604
MARSTRUCT.indb 604
frequency and its harmonics, so the majority of the
engine exhaust noise is at low frequencies, that is in
the frequency range below the 250 Hz octave band.
In general, the energy content of sound radiated by
fans is centered on frequency bands higher than those of
the engine (from 250 to 2000 Hz band). So, noise radi-
ating from ships to surrounding areas is mainly due to
engine exhaust in the low to medium frequency bands
and to air fans in the medium to high frequency
bands—and the latter is more easily silenced.
By summing up, outdoor airborne noise pol-
lution due to a large cruise vessel staying in port
is very invasive for a series of reasons: the huge
power delivered by engines for ship’s equipment,
the great amount of air treated by the onboard
HVAC systems, and the high energy content at the
lower-frequencies sound bands.
1.3 How to characterize onboard noise sources
Noise emanating from ships in harbours has about
the same characteristics of an industrial noise and,
as far as outdoor airborne noise is considered, ship
may be treated like an industrial noise source.
From an engineering point of view, the most
practical methods are the ones based on pressure
measurements, while determination of onboard
sources’ sound power levels using sound intensity
are not common (ISO standard reference: EN ISO
9614).
The Directive 2002/49/EC called “Environmental
Noise Directive” (EP&C 2002) suggests to mapping
harbour areas by measurement methods applied
for industrial noise in accordance with one of
the following ISO standards: ISO 8297:2006, EN
ISO 3744:2009 and EN ISO 3746:2009. The first
is applied when several sources are emitting, the
second is used in free-field environments over
a reflecting plane and the third is effective for
noise level measurements of sources emitting in a
partially reverberant surrounding.
Among them, the EN ISO 3744:2009 “Acoustics—
Determination of sound power levels of noise using
sound pressure—Engineering method in an essen-
tially free field over a reflecting plane” (ISO 2009)
appears the most suitable standard procedure to
derive the ship’s noise source power levels. The
standard is permitted for measurements over a
reflecting plane outdoors, just like in the case of a
ship berthed at a quay.
In the EN ISO 3744:2009 standard a method is
specified for estimating the sound power level pro-
duced by a source in terms of SPL measures taken
on a measurement surface enveloping the source. In
cases when the hypothesis of free field is not strictly
fulfilled, proper corrections are considered.
Methodology hereinafter described may used
for characterizing one by one each onboard noise
2/18/2011 5:54:32 PM
 source. On the other hand, the same procedure
may be applied to the ship as whole, that is
by considering the ship as a single large noise
source.
First of all, a reference box considering
possible reflecting planes is defined as an enve-
lope of the source, and the box diagonal d0 is
considered as the source characteristic dimension.
The microphones lie on a measurement surface
(a hemispherical surface or a rectangular parallele-
piped) which envelops the reference box at a proper
distance d. The A-weighted SPL value, averaged
from frequency band pressure levels taken on the
measurement surface, is given by
⎡ ⎤
LApi
⎢1 N 10 ⎥ noise. The standard defines the position of
LAp = 10 log ⎢ ∑i =1 10 ⎥ (1)
⎢ N ⎥ from the ship’s side and at several points around
⎣ ⎦
where LApi is the A-weighted SPL measured at
the i-th microphone position and N is the number
of microphone positions. The same expression is
valid for calculating background noise level LApB.
In the case of ships, as strong discrete frequency
component are emitted by main noise sources,
it is recommended to perform calculation using
one-third-octave band A-weighted values. The LAp
needs to be corrected for background noise and for
actual test environment:
⎡ LAp B LAp ⎤
⎢ 10 ⎥
LAp, corrected LAp + 10 log ⎢1 − 10 ⎥ − K 2A
⎢ ⎥
⎣ ⎦
(2)
where K2A is the near-field correction to account
for the influence of reflected or absorbed sound
on the surface sound pressure level, calculated as
suggested in the standard. Finally, the sound power
level LW of the source is given, according to the
far-field law, as
S pass over.
LW LAp corrected + 10 log
Ap (3)
S0
where S is the area of the measurement surface
and S0 is the reference area equal to one metre. The
rule sets the positions of the key microphones and
of those required in case of high directivity sources
(e.g. like in the case of the high frequency noise
radiated by ship’s funnels).
On the basis of the measures taken by the array
of microphones, also the directivity index of the
source may be assessed.
605
MARSTRUCT.indb 605
Sound power level values estimated according
to the EN ISO 3744:2009 standard are suitable for
characterizing the onboard noise sources with the
aim of performing numerical analyses, so opening
the door to the opportunity of the prediction of
ship’s noise pollution in real ports and harbours.
Other ISO standards are specific for the
assessment of airborne sound emitted by vessels
in inland waterways and in ports and harbours.
The one valid for ships is the EN ISO 2922:2001
“Measurement of airborne sound emitted by vessels
on inland waterways and harbours” (ISO 2001).
For ships staying in port the significant noise
parameter is the A-weighted sound pressure level
LAeq measured under the validity of far-field law
hypothesis and roughly corrected for background
microphones, to be placed 25 metres (±2 metres)
the vessel.
By summing up, the EN ISO 2922:2001 standard
proposes to characterize the noise pollution by the
SPL values measured by an array of microphones
placed along the ship’s side. No other information
may be drawn from the test, so making that
ineffective when the noise levels at some distances
need to be estimated, like in ship’s noise pollution
evaluations. Anyway, by applying the standard in
a proper way, a systematized collection of data
may be obtained, which may be used to create a
statistical correlation between noise emitted by
ships and noise pollution in port and harbour.
1.4 Which role for ships’ designers?
Other than the subject of noise as loss of comfort
inside the ship’s spaces, the noise radiated to air
and water is matter of great concern just in recent
years. The framework in which designers are
moving is actually very poor. Standards and rules
are not so structured to effectively support design
stage. The lack of a well-established approach to
control noise pollution, the missing of shared limits
and finally the scarce interest given to the subject
by shipowners is being decisive to lead designer to
The need to assist design process is clear in the
mandate of recent European projects about noise
pollution, where ship comfort designers carved out
a prominent space for themselves, aware of the
importance of getting predictive tools for treating
noise pollution in the early design phase. Their aim
is to be ready to comply any forthcoming regulation
based on realistic acceptance limits.
Designers of ship’s comfort are often involved
as people the more experienced in managing the
noise concepts and the more knowledged with
onboard noise sources and noise path mechanism.
2/18/2011 5:54:32 PM
 So, comfort designers are today entering the subject
of outdoor noise pollution. More often, instead
to perform prediction analyses, they are involved
in trouble-shooting and control measures after
delivery of ships. That is the case of a contingent
liability to comply with local rules if ship is the
cause of noise pollution to surroundings residential
areas when staying in a harbour.
Standards defining noise limits and assessment
criteria are expected. A series of essential prerequi-
sites are to be fulfilled by the new standards to be
of practical use, that are the ease identification of
the exciting sources, the repeatability and reliability
of measurements, and the manageability of the
assessment procedure. That is what ship’s designers
demand.
In the meantime, the only option for designers
to get ready to face the expected rules about ship’s
noise pollution in harbour, is to learn lesson from
the sailing ships recently built.
Two actions are compulsory: to make on field
measurements of noise emitted by ships berthed
in ports, and to try to predict the noise levels in
the ship’s surroundings by means of numerical
analyses—which need to be set by direct meas-
urements. Commercial software tools are indeed
available for open field noise simulations (such as
CadnaA and LMS Raynoise): they just need to be
implemented for carrying out specific simulations
of the environmental noise radiated from ships.
2 THE PREDICTION PROCEDURE
2.1 The choice of the theoretical approach
In predicting the outdoor noise of a ship staying
in port, a mathematical model to correctly identify
the characteristics of the specific acoustic field
around the ship needs to be defined.
Since analytical approaches are available only
for a very few number of acoustic problems, those
involving noise radiating in exterior and interior
fields with simple geometrical shapes, to study a
complex acoustic field, as that around the ship, it is
necessary to resort to numerical methods.
Several numerical prediction techniques are
available, each of them is appropriate only for
a limited number of situations, depending on a
number of general factors.
To make a correct choice of the most suitable
method for a given real problem many aspects
have to be considered. First of all, the maximum
wavelength of the sound phenomenon must
be compared with the main dimensions of the
radiation space. If the radiation space is much
larger than the sound wavelength, we talk about
an open field condition problem. Under the open
606
MARSTRUCT.indb 606
field assumption, neglecting the wave nature of
sound leads to accurate simulations, and the sound
simulation may be carried out using geometrical
acoustics methods.
In the geometrical acoustics methods it is
assumed that sound waves behave like sound rays,
exactly as in geometrical optics, where light waves
behave like light rays (Desmet & Sas 1966).
The outdoor ship noise may be classified as
an exterior, open field condition, time-harmonic
problem of an uncoupled vibro-acoustic system.
The open field condition hypothesis for the
cruise ship’s outdoor noise is justified by the
assumption that the maximum sound wavelength
radiated by the ship’s noise sources is related to
the firing frequency of diesel engines, which usu-
ally falls within the 63 Hz octave band. In effect,
as that is the lowest frequency octave band of the
sound spectrum with significant energy content, a
maximum significant wavelength is expected at no
less than about 15 meters in standard environmen-
tal conditions, much less of the main dimensions
of the acoustic field, generally of the same order
of the ship’s length.
Among the different methods used in geomet-
rical acoustics, the most powerful and efficient
are the hybrid ones that merge the best features of
the Mirror Image Source Method (MISM) and of
the Ray-Tracing Method (RTM).
The hybrid methods are the so called Beam
Tracing Methods (BTM). The main advantage of
beam tracing is its intrinsic geometric coherence,
since each beam represents an infinite number of
potential ray paths emanating from the source.
BTM does not suffer from the sampling deficiency
of the ray tracing, as receivers are points, not
volumes. Computational complexity due to image
source generation is balanced by simplicity of the
tracing approach. The triangular beam method
has to be preferred (Funkhouser et al. 2004).
2.2 The computation scheme
The geometrical acoustics concepts are the core
of the software tool (LMS Raynoise) hereinafter
employed in the outdoor airborne ship noise
computation. The software uses a hybrid BTM
based algorithm. Its typical applications comprise
industrial noise control.
Resorting to geometrical acoustics, the first
action is to generate the geometry model by
making use of a commercial CAD software tool.
Geometry model need generally to be set up by
level surfaces.
Noise field active elements are then generated
within the CAE software tool on the basis of the level
surface geometry model. They are called polygon
elements and are obtained by associating to each
2/18/2011 5:54:33 PM
 level surface a set of frequency-dependent material
properties (e.g. transmission and absorption coeffi-
cients). Diffraction edges are simply accounted for
by the software by searching the shortest broken
line between source and receiver (first order
diffraction path). In case, environmental charac-
teristics are also set out.
It is worth pointing out that role of user is decisive,
as goodness of outcomes is very dependent on
numerical simulation setting. The most important
calculation parameters in numerical geometrical
acoustics are the reflection order, and the number
and type of rays.
The most appropriate values for these very
important parameters depend on the general char-
acteristics of acoustic field. For instance, in an exte-
rior acoustic field, a low number of rays leads to a
poor mapping, whereas a too high number of rays
could lead to numerical complications. Conical or
triangular beams may be used and no significant
difference is generally observed in the outcomes in
case of open field conditions.
Finally, on the basis of the input data and set-
tings, a mapping of the acoustic field for each
frequency octave band is obtained. The acoustic
quantities, such as SPL values, are computed on
given field points planes.
2.3 Remarks on the operational procedure
The proposed procedure to predict outdoor
airborne ship noise in harbour is detailed in the
flow chart of Figure 2. Main phases of procedure
are hereinafter discussed.
Noise field study starts from the problem defini-
tion. In the first phase of the analysis, the main
onboard sources should be identified in terms
Figure 2. Operational procedure flow chart.
607
MARSTRUCT.indb 607
of nature and location. At the same time, the
main characteristics of the ship’s outer geometry
(i.e. ship’s side layout, upper deck arrangement and,
in case, funnel geometry) along with the layout of
the harbour (buildings, different obstacles, other
ships) need to be defined.
The following action is the determination of
sources’ sound power level and directivity based on
octave frequency band. They may be determined
according to EN ISO 3744:2009 or ISO 9614-2:1998
(ISO 1998). Measures taken in accordance with the
EN ISO 3744:2009 standard are generally more
practicable.
The following phase is the generation of the
3D geometry model and of the noise field active
polygons, which describe the behavior of bodies
lying in the acoustic field.
For each body that part of the outer surface
which interacts with sound radiation needs to
be generated with proper accuracy. The layout
of the surface model of the ship’s side (or sides)
overlooking the harbour area should reproduce
the actual athwartship extent of each deck, as that
strongly affects the noise field pattern close to ship.
Diffraction, which is a very important component
of the outdoor noise field close to the ship, needs
to be taken into account.
A too accurate modelling could even be del-
eterious, as for instance in the cases of air gratings
and openings on bulwarks, which should generally
be neglected.
Once noise field model has been fully settled,
the consistency between the actual acoustic field
and the numerical model needs to checked.
First of all, each noise source should be set by
checking correspondence between calculated and
measured SPL values on proper field points planes.
2/18/2011 5:54:33 PM
 For each noise source, adjustments of the sound
power level octave-band spectrum (and, in case, of
directivity and emission angle) have to be recur-
sively carried out till SPL corrections become neg-
ligible, within the lower frequency bands at least
(Moro 2010).
If simulation results are judged not to be accurate
enough, an in deep setting is then required, which
could lead to change noise field parameters (e.g. by
adding diffraction edges) or ray tracing parameters.
In general, fineness of mesh does not affect noise
field mapping if not in spatial accuracy.
3 THE CASE STUDY
3.1 General description
To show effectiveness and accuracy of the above
mentioned procedure, a case study has been
developed. A large passenger ship has been consid-
ered as the only noise source in the whole acoustic
field; for the outdoor noise prediction, the typical
onboard noise sources have been accounted. The
ship is considered to be moored in an Italian
harbour, at different berthing posts along the quays
of a dock. Around the ship, for an extent of about
a ship’s length measured in all directions, the quays
are free from obstacles except for a small built-up
area. Small houses where noise pollution has to be
checked are within a distance that ranges from 0,5
to 1,5 ship’s length.
To set the sources, data available to the ship-
builder have been used together with data collected
by near-field measurements aimed at supplement
the shipbuilder data, and taken according to
EN ISO 3744:2009. In other words, in case study
discussed, sound power levels have been calculated
from sound pressure level measures. Noise sources
characteristics have been set by considering proper
read-out points in the near field, where direct
measurements were taken.
As for the properties of materials, absorption
coefficients were defined by assuming standard
values derived from scientific literature, while the
transmission properties have been considered not
necessary to solve the specific acoustic field.
Among all the noise sources which take part in
the ship’s overall noise, those have been disregarded
which, because of their low sound pressure levels,
negligibly contribute to the overall noise field in
the ship’s surroundings.
3.2 Preliminary noise field simulations
Before to proceed to the simulations of the case
study, a preliminary analysis has been carried
out to determine the contribution of each
source to the overall noise generated by the ship
in harbour. The study is a very useful tool for a
608
MARSTRUCT.indb 608
better understanding of a complex field noise as
that generated by many sources. The considered
ship’s sources are: the “Diesel generator engine
exhaust” source, located on the tip of the funnel’s
wing, the “engine rooms exhaust air fans (outdoor
openings)” source located at the base of the
funnel and the “air conditioning system (outdoor
openings)” source distributed along the ship’s side.
Preliminary analysis was carried out near the
ships side, with the ship moored at a quay free from
buildings. The field-points plane lies on the ship’s
transverse section at the middle of the funnel and
it extends for 100 meters from the ship’s side into
the quay (see Figure 3).
Outcomes of the simulation of noise field gener-
ated by the Diesel generator engine exhaust source are
shown in Figure 3. SPL calculated at about 5 meters
from the duct upper outlet is 71 dB(A). SPL values
reduce moving away from the funnel because of the
sound divergence from the source. Specifically, in
the area near the ship’s side at quay level the general
noise field is modified because the ship’s superstruc-
ture, scattering the sound waves, creates a shadow
area. Inside the shadow area, near the quay edge,
SPL values lower to a minimum of 36.7 dB(A). The
transition from the shadow area to the main field
area is very localized and the SPL gradient is very
high; at about 5 meters away from the ship’s side
at the quay level, SPL values grow very quickly to
41 dB(A). At the reference point located 2,5 meters
in height and 40 meters away from the ship’s side,
the SPL value is equal to 46.8 dB(A).
The shadow area due to ship superstructures
may be properly identified by setting the diffraction
parameters. A further simulation has been carried
out neglecting the diffraction due to the ship’s super-
structure: if the diffraction edges are not defined,
the shadow area is not properly characterized and
the SPL values drop to 7.8 dB(A), that is a physical
nonsense.
Figure 3. Mapping of the SPL values (noise source:
exhaust).
2/18/2011 5:54:34 PM
 Simulations carried out on the other sources
show that a similar noise field pattern is given by the
two sources placed at the funnel (marked shadow
area at the quay edge), while the air conditioning
system source generates a homogeneous noise field.
The SPL values relevant to all the ship’s sources
are shown in Figure 4. SPL octave-band spectra
are calculated at 2,5 meters in height from the
quay level, at the distances of 10, 25 and 50 meters
from the edge of the platform. Curves of SPL
spectra clearly show extension of shadow area.
The difference in shape of calculated SPL spectra
is due to lacking of direct field-noise components
near to quay edge, where only diffuse components
are calculated. Reduction of sound energy is espe-
cially associated with high frequencies.
A series of comparisons has been made between
single source SPL spectra and overall SPL spectrum
at the reference point placed at 10 meters from the
quay edge (see Figure 5). One of the most important
Figure 4. Computed SPL values (noise source: all,
air fans).
Figure 5. Computed SPL values (various noise sources).
609
MARSTRUCT.indb 609
results is that comparing overall SPL spectrum and
that referring to the air conditioning system source,
at the lower frequencies, just at 10 meters form ship’s
side contributions of upper sources prevail—while
from 4 to 8 kHz, the energy contribution is almost
entirely given by air conditioning system source. As
the latter may be disregarded in the noise field far
away the ship, the contribution of air conditioning
system source may be neglected in predicting out-
door noise generated by the ship in harbour.
Contribution of the diesel engines exhaust
gas source to the overall outdoor noise was then
estimated. A clear predominance of the engine
rooms exhaust air fans source over the engines
exhaust gas source (difference in overall SPL equal
to 24.5 dB(A) @ 50 meters from quay edge) has been
derived. Figure 5 shows the all sources SPL spectrum
compared with the engine rooms exhaust air fans
SPL spectrum at 50 metres. That allows considering
negligible the energy contribution of the latter source.
This result is not in line with commonly experience,
which points out that exhaust noise is predominant.
3.3 Noise field simulations
In the simulations carried out for the overall out-
door prediction, the field points plane has been
placed in front of a building. Even tough in a real
case study prediction, analysis should be carried
out by collecting and discussing SPL outcomes
on the whole area of interest as, for instance, the
entire front of one or more buildings, in what
follows, SPL outcomes will be considered on a
single read-out point, in order to make more clear
explaining the procedure.
Outdoor noise predictions have been carried out
while the cruise ship is moored in a dock at different
berthing posts. The most interesting berthing post is
that for which direct measurements have been taken
(Station 1). Case 1 simulation gives at the analysis
point an overall SPL value of 46.4 dB(A).
The read-out plane is partially covered by build-
ings and so the noise level results to be attenuated
on the entire plane. The distance effect is pre-
vailing on that barrier effect and the SPL values
decreases according to distance from the ship. At
the right side of the building face a maximum of
47.30 dB(A) has been computed. The latter value
is slightly lower than the true value obtained by a
direct measurement, which gave 48,5 dB(A). The
little discrepancy may be due to a higher level of
the high-frequency background noise measured on
site, probably due to very local sources.
3.4 Final considerations
By summing up the outcomes for the different
berthing positions (from Station 1 to Station 7), a
trend of outdoor noise generated by the ship may
2/18/2011 5:54:35 PM

Figure 6. Computed SPL values, ship berthed at stations
1 to 7.
Figure 7. Layout of ship and surroundings area.
be outlined, so defining which of the berthing posts
gives the lowest noise pollution on the buildings
in the surroundings residential area. In Figure 6
the overall SPL values as drawn from numerical
simulations are shown, distances are measured
from ship’s aft perpendicular to the reference
building (the port layout is shown in Figure 7). If a
noise limit has been defined, permissible berthing
posts may be identified.
4 CONCLUSIONS
The main focus has been addressed to assess
the impact of the mechanical noise sources of
passenger ships, and to identify noise parameters
that would correlate noise levels with community
annoyance in the harbour proximity.
610
MARSTRUCT.indb 610
A prediction procedure has been outlined
and comparison made between calculated and
measured SPL values proves accuracy of proposed
method and validates it as a powerful tool for
reliable predictions.
In the case study, SPL values relevant to each main
source were derived from on field measurements.
On the other hand, when SPL spectra are avail-
able as a shipyards proprietary database collecting
data of each significant onboard noise source,
the procedure can be effectively applied just in
the design phase assuming that type, power and
arrangement of plants have been defined. That
allows estimating effectiveness of different noise-
sources configurations and, when simulation gives
to high overall SPL at fixed reference read-out
locations around the ship, corrective actions may
be studied.
REFERENCES
Desmet W. & Sas P., 1966. Introduction to numerical
acoustic. Proceeding of the International Seminar on
applied acoustics. Leuven, Belgium.
European Parliament and Council, 2002. Directive
2002/49/EC of the European Parliament and of the
Council of 25 June 2002 relating to the assessment and
management of environmental noise. Official Journal
of the European Communities L 189, 12–25.
Funkhouser T. et al., 2004. A beam tracing method
for interactive architectural acoustics. Journal of
Acoustical Society of America 115 (2): 739–756.
International Organization for Standardization, 1998.
ISO 9614-2:1998—Acoustics—Determination of sound
power levels of noise sources using soundintensity—
Part 2: Measurement by scanning. ISO, Geneva,
Switzerland.
International Organization for Standardization, 2001.
EN ISO 2922:2001—Acoustics—Measurement of
airborne sound emitted by vessels on inland waterways
and harbours. ISO, Geneva, Switzerland. Interna-
tional Organization for Standardization, 2009. EN
ISO 3744:2009—Acoustics—Determination of sound
power levels of noise using sound pressure—Engineering
method in an essentially free field over a reflecting
plane. ISO, Geneva, Switzerland.
Lloyd’s Register ODS, 2010. Noise from ships in ports.
Possibilities for noise reduction. Danish Environmental
Protection Agency, Environmental Project No. 1330.
Moro L., 2010. Setting of on board noise sources
in numerical simulation of airborne outdoor ship
noise. Proceeding of the 9th Youth Symposium on
Experimental Solid Mechanics, Trieste, Italy: 92–96.
2/18/2011 5:54:36 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Shape design optimization of fluid-structure interactions using SPH

and geometrically exact interfaces

Hyun-Seok Kim, Min-Geun Kim & Seonho Cho

National Creative Research Initiatives Center for Isogeometric Optimal Design and Department of Naval
Architecture and Ocean Engineering, Seoul National University, Seoul, Korea

ABSTRACT: Fluid-Structure Interaction problems are solved by applying a smoothed particle hydro-
dynamics method to a weakly compressible Navier-Stokes equation as well as an equilibrium equation
for geometrically nonlinear structures in updated Lagrangian formulation. The geometrically exact
interface, consisting of B-spline basis functions and the corresponding control points, includes the high
order geometric information such as tangent, normal, and curvature. The exactness of interface is kept
by updating the control points according to the kinematics obtained from response analysis. Under the
scheme of explicit time integration and updated Lagrangian formulation, the required shape design
velocity should be updated at every single step. The update scheme of design velocity is developed
using the sensitivity of physical velocity. The developed sensitivity analysis method is further utilized in
gradient-based shape optimization problems and turns out to be very efficient since the interaction pairs
of particles determined in the response analysis can be directly utilized.

1 INTRODUCTION and design independent properties of SPH kernel

The FSI problems are multi-physics problem
between structure and fluid. It is important to
consider exact interface between structure and
fluid because the inaccuracy of traction, pressure,
and velocity condition at interfaces may cause
failure in analysis.
The Smoothed Particle Hydrodynamics (SPH)
method, a particle based Lagrangian method, is
appropriate for large deformation problems since
it is very efficient for transient dynamic analysis;
all the equations of motion can be constructed by
the summation of nearby particles on evaluation
position instead solving algebraic system
equations. In this paper, SPH method considering
the exact geometric interface is employed for the
response analysis and developing the shape Design
Sensitivity Analysis (DSA) for the transient
dynamic FSI problems.
To impose exact kinematic and dynamic boundary
condition, we construct CAD based interface such
as B-spline curves, since grid based approximation
has the limitation of geometric information such
as tangent and normal vector. Since the interface
can be updated directly from the response analysis,
no more approximation to determine the interface
forces is necessary. Once the interface is defined at
initial configuration, it can be moved by directly
updating the control points of B-spline curves
using the interface velocities. The collocation
function easily update the design velocity using the
velocity of particles. The developed DSA method
is very efficient since the particle interaction pairs
determined in the SPH response analysis can also
be utilized in the DSA. The analytical design sensi-
tivity using the SPH method can reduce the numer-
ical noise and consequently yield accurate design
sensitivity.
2 SPH FORMULATION
The SPH method is applied to a weakly com-
pressible Navier-Stokes equation as well as an
equilibrium equation for geometrically nonlinear
structures (Antoci et al. 2007). In the SPH method
the domain is represented by a set of arbitrarily
distributed particles. Each discrete particle has the
representative material properties, such as mass m,
density ρ, etc., and its function value is approximated
by averaging the function values at all particles
within the compactly supported domain weighted
by a suitable kernel function. The compactly sup-
ported domain is characterized by the smooth-
ing length scale h which is usually a constant in
incompressible media. In this paper, the the dim-
dimensional cubic spline function is employed as
the kernel function for the computational efficiency
(Monaghan & Lattanzio 1985). The index ‘dim’
represents dimension of the system.

611

MARSTRUCT.indb 611 2/18/2011 5:54:37 PM

 Wab ≡ W ( ab , h)

⎧ 3 ⎛ | ab | ⎞ 2 3 ⎛ | ab | ⎞ 3 |x |
⎪1 − ⎝ ⎠
+
⎝ ⎠
, 0 ≤ ab ≤ 1,
⎪ 2 h h h
= α ddim ⎪⎪ 1 3
⎛ | x ab | ⎞ |x |
dim ⎨ ⎝2 − ⎠
, 1 ≤ ab ≤ 2,
h ⎪ 4 h h
⎪ | x ab |
⎪0, > 2,
⎪⎩ h

(1)
where (•)ab = (•)a − (•)b. For two-dimensional sys-
tem, the dimensional parameter αdim is defined as
α2 = 10/7π. Figure 2. Geometrically exact initial interface.

3 GEOMETRICALLY EXACT INTERFACE However, in this paper, a geometrically exact

initial interface is directly defined, Figure 2, by the
3.1 Definition of initial interface B-spline curve of CAD geometry as Equation 3.
In the SPH approach, the interface is approxi-
n
mately constructed using the distribution of par-
ticles as shown in Figure 1, where the red and
x int a ∑ , p (ξ )Pint
int i , (3)
i =0
green dots are respectively the initial interface and
pseudo boundary particles that are the outmost where Ni,p(ξa) are pth-degree B-spline functions
ones of solid domain. The approximated interface and Pinta are the corresponding control points.
particle xinta is defined as a constant distance in the ξ is called knot vector and it is defined in
direction from the solid particles on the pseudo parametric space. When we utilize pth order
boundary Γ s to the fluid domain as Equation 2. B-spline curve, we are able to obtain (p − k) times
(the index “inta” means particle ‘a’ that is defined continuous differentiability at a knot of multiplic-
on the interface). ity k (Piegl & Tiller 1997). Therefore, employing
geometrically exact interface, we can immediately
x int a = x a + 12 dnn a , (2) obtain the tangent vector at the interface by
using the derivative of basis function defined as
where xa is the position of particle ‘a’ at pseudo Equation 4.
boundary; d is the minimum distance between the
solid and fluid particles at the initial configuration; p
∇ξ i p (ξ )= Ni p −1(ξ )
na is the approximated normal vector evaluated at ξi + p − ξi
the pseudo boundary particle xa. p
− Ni +1 p −1(ξ ). (4)
ξi + p +1 − ξi +1

3.2 Update of interface

In the method of approximated interfaces a con-
stant distance d is obtained at initial configuration.
The initial interface evolves to a deformed con-
figuration, where a pseudo boundary is defined
at every time step. A normal vector on the pseudo
boundary is obtained by a liner approximation
using the information of neighboring particles.
Finally, the interface is updated for the deformed
configuration. Note that the fixed distance d and
the approximated normal vector could cause
troubles in the process of time integration, which
Figure 1. Approximated initial interface. will be discussed in numerical example section.

612

MARSTRUCT.indb 612 2/18/2011 5:54:37 PM

 In this our research, the initial interface is directly where τ denotes the shape design parameter
defined by the B-spline curve in Equation 3 without defining the transformation T. The shape design
the pseudo boundary. And it is updated using the velocity is defined as Equation 11.
kinematics obtained during the response analysis.
The velocity of interface particles is obtained by n
d xτ dT (
n
,τ )
n
T ( x, )
the SPH interpolation of the solid particles as
n
Λ( ) ≡
n
= = . (11)
dτ dτ ∂τ
Equation 5. τ =0 τ =0 τ =0

∑ b∈Ω ρ vbWh ( x int

mb
xb ) Since the update Lagrangian formulation is
a
v int a = . employed, the design velocity has to be updated
b
s
(5)
∑ b∈Ω ρ Wh (x
m b
( x intt a
xb ) at each configuration. The perturbed position at
configuration (n + 1) can be written as
s b

Using the relation of velocities between the

n+1 t
interface particle and the control point, we obtain n +1
the velocity of control interface as Equation 6. xτ = n xτ + ∫n t vτ dt. (12)

P int N −1 ⋅ v int , (6) Using the definition of Equation 11 and the

shape sensitivity of velocity, the updated design
where N is a (n + 1) × (n + 1) matrix consisting of
velocity is written as
B-spline components as Equation 7. (the index "int"
means a vector that is defined on the interface).
n +1
d ⎡ n+1t
n+1 n
d( ) d( ) ⎤
∫n vτ dt ⎦⎥
n +1 τ τ
⎡ N0, p (ξ ) N1 p (ξ0 )  N n p (ξ0 ) ⎤ Λ≡ = +
dτ dτ dτ ⎣⎢ t
⎢ ⎥ τ =00 τ =0 τ =0

⎢ N0, p (ξ ) N1 p (ξ1 )  N n p (ξ1 ) ⎥ (7) n+1

N=⎢ ⎥.
t
∫nt
n
= Λ+ v ′ dt (13)
   
⎢ ⎥
⎢⎣ N0, p (ξn ) N , p (ξn ) N n, p (ξn ) ⎥⎦
If a central differencing scheme is employed,
A new interface can be updated using the veloc- the updated design velocity is obtained as
ity of interface particle vint as Equation 8. Equation 14.
τ
nt + τ v int.
x int = x iint n +1 n+ 12
(8) Λ = nΛ + v′Δt (14)
Using Equation 6 and the fact that xint = N ⋅ Pint, The shape sensitivity of rate of change in density,
the new interface can be rewritten in terms of the acceleration for fluid particles and acceleration for
control points as Equation 9. solid particles are shown through Equation 15 to
17. The details are in Ha and Cho (2010).
 ) τ N ⋅ P int . (9)
intt τ Piint
x int N⋅ Pint N ⋅ (Pin nt int
d ⎡ n ⎛ ∂ρ a ⎞ ⎤
⎜ ⎟ = ∑ mb ( n ′a n ′b ) ∇ a nWab
dτ ⎢⎣ ⎝ ∂t ⎠ ⎥⎦τ = 0 b
4 SHAPE DSA + Δt∑ mb ( n v a − n v b ) ⋅ (∇ a nWaab )′
b
Due to the highly nonlinear characteristic of
transient dynamics, the finite difference sensitivity (15)
(Haug et al. 1986) requires very small perturbations
d ⎡ n ⎛ ∂v a ⎞ ⎤
fluid
i
of design. Nevertheless, the design sensitivities ⎢ ⎥
become inaccurate due to the accumulation dτ ⎣ ⎜⎝ ∂t ⎟⎠ ⎦
τ =0
of numerical noise induced from very small
⎛ n
pa′
n
pb′
n
pa ⎞
perturbation. Therefore, an analytical shape DSA n
ρa′ ⎟
⎜− n
ρa2
− n
ρb2
+ n
ρa3
method using a direct differentiation method (Haug
= ∑ mb ⎜ ⎟ l ⋅ ∇ a nWaab
et al. 1986) is developed and utilized in a gradient- ⎜ n
pb n n ⎟
based shape optimization for the FSI problems.
b
⎜+ ρb′ + Π ′ab ⎟
⎝ n
ρb3 ⎠
Consider the shape variation of a configuration nΩ
with the domain mapping T: nx → nxτ(nx), nx ∈ nΩ ⎛ n
pa
n
pb ⎞
+ ∑ mb −
n n n
as Equation 10. − + Π ab l ⋅ (∇
(∇ a Wab )′ + ( f a′ )
b ⎝ n
ρa2 n
ρb2 ⎠
n
xτ T ( n x,τ ) n
x τ n Λ ( n x ), (10) (16)

613

MARSTRUCT.indb 613 2/18/2011 5:54:40 PM

 d ⎡ n ⎛ ∂v asolid ⎞ ⎤
⎢ ⎜ ⎟⎥
d τ ⎣ ⎝ ∂t ⎠ ⎦ τ = 0
⎛ n σa′ n
σ ′b
n
σa n ⎞
⎜ n ρ 2 + n ρ 2 − n ρ 3 ρa′ ⎟
= ∑ mb ⎜ a b a
⎟ ⋅ ∇ a nWab
⎜ ⎟
n
b σb n n n
⎜⎝ − n ρ 3 ρ b′ + Π ′ab l + Θ ′ab ⎟⎠
b

⎛ n σa n σb n ⎞
+ ∑ mb ⎜
n n n
+ + Π ab l + Θ abb ⋅ (∇
( ∇ a Wabb )′ + ( f ′ a )
b ⎝ ρa
n 2 n 2
ρb ⎠

(17)

5 NUMERICAL EXAMPLES

5.1 High order geometric information

In the previous section, it was mentioned that due Figure 4. Approximated method.
to the inaccurate normal vector obtained at the
interface may cause trouble. Now we will show this
phenomenon through a numerical. Let us consider
two-dimensional model consists of a solid domain
that is fixed at its top and bottom and a fluid
domain example as show in Figure 3. For each solid
and fluid domain, we used 606 particles and 2,121
particles. The time step is set as 5 × 10−6 second. The
solid domain has the following properties; Poisson
ratio of 0.4, bulk modulus of K = 4 MPa, and shear
modulus of G = 0.854 MPa. The fluid domain is
assumed inviscid and incompressible, which has
the properties; density of ρf = 1,000 kg/m3, artificial
compressibility modulus of ε = 2 × 106 N/m2,
instead of actual one ε = 2 × 109 N/m2.
The performances for the same example and at
the same time step of the approximated method,
Figure 4, and the proposed method, Figure 5,
are compared. Initially, both methods provide

Figure 5. Proposed method.

accurate normal vectors at the interface. However,

after certain time integrations, the approximated
method, Figure 4, shows inaccurate normal vectors
while the proposed method, Figure 5, still shows
the accurate ones. Due to the inaccurate normal
vectors, the interface forces are also calculated
inaccurately. Thus, some fluid particles are pen-
etrating to the solid domain in the approximation
method whereas no penetration of fluid particles is
observed in the proposed method.

5.2 Importance of analytical sensitivity

Now, we will introduce the importance of analytical
sensitivity. Let us consider a water-jet impacting a
Figure 3. Two-dimensional example consist of solid plate made of rubber material. All of other prop-
and fluid domain. erties are same as in Section 5.1, except the bulk

614

MARSTRUCT.indb 614 2/18/2011 5:54:46 PM

 modulus, K = 1 MPa, for solid domain to be a 5.3 Shape optimization of corrugation panel
rubber material. The thickness (0.005 m) of plate is Consider a simplified two-dimensional container
selected as a design parameter and the x-directional tank filled with fluids as shown in Figure 8. The
velocity of each solid particle as a performance simplified part of corrugation structure and fluids
functional. In Figures 6 and 7 the analytical shape are selected as the design domain of interest.
variations of Vx in solid particles, Figure 7, are The fluid velocity is also simplified to v = (−2.5,
compared with finite difference ones, Figure 6, 5.0) m/sec and the top and bottom of panel are
as varying the perturbation amount, which varies assumed fixed. 12 control points are used for the
from 10−2 % to 10−12 % of the initial design. Note SPH model design of corrugation panel (Fig. 9).
that, to reduce the effect of numerical noise due 1,806 particles (6 × 301) are used for the solid
to the small amount of perturbations, the shape domain and 3,171 particles (151 × 21) for the fluid
variation instead of shape sensitivity is selected as domain. A constant time step of 1.5 × 10−6 sec
a comparison measure. is used.
From numerical experiments, the following The objective function is to minimize the strain
remarks are obtained; Firstly, the agreement energy of corrugation panel,
between the analytical variations and finite
difference variations is improved as reducing N
the perturbation amount from 10−2 % to 10−6 %.
However, further reduction to 10−12 % could result
Min ψ ( ) ∑ ∑ Δ (n a)⋅(
n
a) . (18)
n a =1
in poor agreement. Secondly, the distribution and
value of shape variations by finite differencing are where N is the number of particles and n is the
not symmetric even for symmetric design problems number of time integration. And there are 8 con-
and it significantly depends on the perturbation straints used to limit the coordinates of control
amounts. On the other hand, those of analytical points,
shape variations are symmetric and show similar
tendency regardless of perturbation amounts. 2 2
( xi − xi ) + ( yi yi )
Therefore, if the shape optimization is performed gi = − 1.0, ( ≤ i ≤ ) (19)
with the finite difference sensitivity, it could result 0.010636
in a premature result or even experience conver-
gence difficulty due to the inaccurate sensitivity. and
However, the analytical shape sensitivity always
provides a correct design direction. ( xi − xi + 3 )2 + ( yi yi + 3 )2
gi = 2 2
− 1.0, ( ≤ i ≤ )
0.010636
(20)
The design variables are the x coordinates of
control points 1, 2, 3, 8, 9, and 10,

b = (x1, x2, x3, x8, x9, x10)T (21)

As a result of shape design optimization,

the strain energy in the corrugation panel is sig-
nificantly decreased from 4.56745 × 10−02 to
Figure 6. Finite difference variations. 3.65608 × 10−02. The optimization history of the
objective function and constrains are shown in
Figures 10 and 11.
The deformed shapes and x-directional nor-
mal stresses in the initial and optimal designs are
shown in Figure 12. As the result of the shape
design optimization, not only the strain energy
has been decreased but also the maximum of
x-directional normal stress decreases by 16.59%,
the stress concentration at the circled area in the
initial design is released in the optimal design as
shown in Figure 13.
Furthermore, the same reciprocal properties are
Figure 7. Analytical variations. applied between the particles by using the SPH

615

MARSTRUCT.indb 615 2/18/2011 5:54:49 PM

 Ωf Ωf

Ωf
Ωs Ωs Ωs

Figure 8. Simplified corrugation panel model.

Figure 12. Comparison of deformed shapes and

x-directional normal stresses.

Figure 9. SPH corrugation panel model.

Figure 10. Optimization histories. Figure 13. Release of stress concentration.

Table 1. Computation time for response and

sensitivity analysis.

Response analysis Sensitivity analysis

2476.4 sec 338.6 sec

kernel function. Therefore, we can use the same

pairs of interacting particles for sensitivity analysis
that are already defined in response analysis. The
computation time for each response and sensitivity
Figure 11. Optimization histories. analysis are shown in Table 1.

616

MARSTRUCT.indb 616 2/18/2011 5:54:52 PM

 6 CONCLUSIONS
The FSI problems are solved by applying a smoothed
particle hydrodynamics method to a weakly com-
pressible Navier-Stokes equation as well as an
equilibrium equation for geometrically nonlinear
structures in updated Lagrangian formulation. The
CAD-based geometrically exact interface includes
the high order geometric information such as tan-
gent, normal, and curvature. The exactness of inter-
face is kept by updating the control points according
to the kinematics directly obtained from the solid
particles. In the SPH framework, the analytical
shape sensitivity based on the direct differentiation
is derived from the strong form, considering the
design dependency of the kernel function and the
interface conditions. Under the scheme of explicit
time integration and updated Lagrangian formula-
tion, the required shape design velocity should be
updated at every single step. The update scheme of
design velocity is developed using the sensitivity of
physical velocity. Through some numerical exam-
ples, the necessity and efficiency of the analytical
shape DSA method is discussed. The developed
sensitivity analysis method is further utilized in
gradient-based shape optimization problems and
turns out to be very efficient since the interaction
pairs of particles determined in the response analy-
sis can be directly utilized.
617
MARSTRUCT.indb 617
ACKNOWLEDGEMENTS
This research was supported by Basic Science
Research Program through the National Research
Foundation of Korea (NRF) funded by the
Ministry of Education, Science and Technology
(Grant Number 2010-18282). The support is grate-
fully acknowledged.
REFERENCES
Antoci, C.A., Gallati, M., & Sibilla, S. 2007, Numerical
simulation of fluid-structure interaction by SPH.
Computers and Structures, 85: 879–890.
Haug, E.J., Choi, K.K., & Komkov, V. 1986, Design
Sensitivity Analysis of Structural Systems. Orlando:
Academic.
Ha, Y.D., & Cho, S. 2010, Efficient design sensitivity
analysis of incompressible fluids using SPH projection
method. Structural and Multidisciplinary Optimization,
40(1): 307–318.
Monaghan, J.J., & Lattanzio, J.C. 1985, A refined parti-
cle method for astrophysical problems. Astronomy and
Astrophysics, 149: 135–143.
Piegl, L., & Tiller, W. 1997, The NURBS Book. 2nd edn.
New York: Springer.
2/18/2011 5:54:55 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Finite element-based shape optimization of an asymmetric steel

sandwich panel joint

D. Polić, D. Frank & A. Klanac

as2con—alveus ltd., Rijeka, Croatia

S. Ehlers
Aalto University, School of Science and Technology, Marine Technology, Espoo, Finland

ABSTRACT: The continuous demand for efficient structures contributes to the development of light-
weight steel structures, such as steel sandwich panels. However, these thin steel sandwich panels need to
be joined to one-another and to conventional structures. Hence, an appropriate joint shape needs to be
identified with an as low as possible equivalent stress level due to the joints’ sensitivity to fatigue. There-
fore, the aim of this paper is to identify the optimum joint shape using a finite element-based optimisation
procedure together with a B-spline based parametric joint shape. The optimization is performed with two
objectives, i.e. to minimize the stress and mass of the joint. As a result, a series of optimum joint shapes
and their characteristics will be presented and discussed, followed by an outlook of future work aspects
related to the joint fabrication and welding.

1 INTRODUCTION of the sandwich joints was investigated by Kujala

Lightweight structures, such as steel sandwich
panels, have been developed in recent years to
account for the continuous demand for efficient
ship structures. Furthermore, the utilization of
lightweight steel sandwich panel structures results
in a lower centre of gravity of the ship and contrib-
utes to the modularization of the production proc-
ess (Roland and Reinert 2000, Kujala and Roland
2002). However, this utilization of lightweight,
respectively thin, structures introduces new obsta-
cles concerning the joining process of such panels,
either to each other or to adjacent conventional
structures (see Figure 1).
Therein, it can be seen that the integration of
the lightweight structures need to enable a flat sur-
face on one side, e.g. a deck structure, which can be
joined by conventional welding from a single side.
However, the resulting asymmetric joint shape
causes an increase in stress due to the introduction
of secondary bending. Hence, the fatigue strength
Sandwich panel Asymmetric joint Conventionally stiffened panel
Figure 1. Concept of a sandwich panel joint.
et al. (1998), Ehlers (2004) and Fricke et al. (2009).
However, these investigations concentrated on the
fatigue life estimation of geometric discontinuities
due to welding, respectively local stress increases,
for pre-existing joint shapes.
Therefore, there exists a need to identify the most
suitable joint shape at first and thereby to reduce
the global stress in the joint due to the asymmet-
ric loading distribution. In this respect, the actual
weld geometry will not be considered during the
optimization procedure, because it will be assumed
that the increase in the weld will be similar for all
joint shapes, thus allowing a qualitative compari-
son only.
Hence, a finite element based shape optimization
procedure will be presented in this paper to identify
the most suitable asymmetric joint shape. The joint
geometry will be represented by a B-spline based-
parametric model containing the variables to be
optimized using the ANSYS parametric design
language. This approach will result naturally in a
smooth and continuous shape of the joint; whereas
an element removal based topology optimization
would require extensive post-processing.
The genetic algorithm (Goldberg, 1989) VOP
(Klanac and Jelovica, 2009) will be used for the
optimization procedure. In the current study, the
conflicting objectives will be the mass of the joint
and the stress in the joint. The elastic stress in
the joint will be evaluated with the finite element

619

MARSTRUCT.indb 619 2/18/2011 5:54:55 PM

 program ANSYS. The optimisation procedure will
be validated with an initial study of a triangular
joint, which will converge to optimum solution
known a priori for the given boundary conditions.
As a result, a series of optimised asymmetric
joint shapes will be presented and compared with
each other in terms of stress and mass. Further-
more, the robustness of the optimisation proce-
dure will be discussed and future implementation
aspects of the optimised joint are presented.
2 THE FINITE ELEMENT-BASED SHAPE
OPTIMISATION
2.1 The parametric joint shape
The asymmetric joint models are created with
B-spline curves defined by three points (start, cen-
tre and end) as presented by the parameters shown
in Figure 2.
The centre point of the B-spline curve presents
the radius of curvature and the coordinates of the
point are defined based on the origin, which is in
the top right corner of the joint. The start- and
end curvature is defined as the tangent to the glo-
bal coordinate system, e.g. the top B-spline curve
is defined as a tangent to the global axis at origin.
This approach allows defining the B-spline curva-
ture as concave, convex or S—type joint shapes in
the whole rectangular domain.
The internal shape of the joint is obtained by
offsetting the outer curves according to the vari-
able thicknesses of the corresponding section. As
a result of the variable thickness of the joint sec-
tions, the joint shape is found to be geometrically
feasible if the internal curves intersect within the
external joint shape, otherwise the geometry is
infeasible. Speaking concretely, geometry will be
considered infeasible if ANSYS cannot generate
area and/or mesh.
All edges in the joint, including the internal
edges, are rounded with a radius of 0,5 mm. Hence,
the joint parameters can occupy any form within
the rectangular domain (dotted line) defining the
length and the height of the joint (see Fig. 2).
Figure 2 shows the principal joint shape together
with the optional section marked in grey, L and H
present joint length and height, while s1x, s1y, s2x,
s2y, s3x and s3y are coordinates of the B-splines.
The thickness of each section is defined with t1, t2
and t3 and the internal radius of curvature fr. Joint
discretization and boundary conditions
The discretized domain contains the joint steel
joint depicted in Figure 2 ideally connected to
a faceplate pair of a sandwich panel on the left
hand side and to a conventional plate section on
the right hand side. Thus, allowing a reasonable
application of the boundary conditions as shown
in Figure 3.
The right hand side of the joint is subjected
to a unit displacement of 0.1 mm in x-direction
while the displacement in y-direction is prohib-
ited. The left hand side of the joint, respectively
the faceplate pairs, are fixed in x- and y- direc-
tion. As a result of these symmetrical boundary
conditions, the appearance of secondary bending
is enabled.
The joint and the connecting structure are mod-
elled primarily with eight-nodded quadrilateral
plane-strain elements, with parabolic interpolation.
The element size is 0.5 mm throughout the whole
model. The module of elasticity is set to 210 GPa
and the Poisson ration is set to 0.3, both param-
eters are kept constant during the optimisation.
2.2 The optimisation procedure
The objective of the optimisation procedure is to
minimize maximum Von Mises stress at the element
level and to reduce the mass of the joint as a result
of the change in shape of the joint. To illustrate
the procedure, the length L and the height H are
kept constant at 150 mm and 60 mm respectively to
allow a straightforward comparison of the results.

L
s3x
s2x
t1+t3

t1+t3
y

x
s3y

3
s2y

s1y

fr
t3
H

2
t1

1
t2

s1x

Figure 2. The parametric joint shape (grey section is Figure 3. Discretized joint and applied boundary
optional). conditions.

620

MARSTRUCT.indb 620 2/18/2011 5:54:56 PM

 Hence, the optimisation problem consists of 12
y
variables, namely H, L, s1x, s1y, s2x, s2y, s3x, s3y, x
t1, t2, t3 and fr according to Figure 2. No other
constraint is considered except the geometric fea-
sibility, if the design is infeasible, a fixed large con-
stant is assigned for its objective value.
The optimisation procedure comprises several
software tools: the optimization algorithm VOP,
the finite element software ANSYS to obtain the
critical Von Mises element stress in the joint, sub-
y
routines to calculate mass of the joint analytically, x
the results data storage application and a routine to
control the overall optimisation process. Therefore,
a ‘VB.NET’ application is created to sequentially
execute the genetic algorithm VOP followed by the
Figure 5. Shape transform from solid triangular joint
ANSYS structural evaluation and the storage of to the optimal solid rectangular with constant thickness.
the design data into an MS EXCEL-file while the
optimization process is ongoing (see Fig. 4).
In detail, VOP GA evaluates the fitness of the
Table 1. Parameter range definition for test joint and
design alternatives per generation and creates a new optimal values [mm].
generation by mutation and crossover of existing
designs from the previous generation. The result- H slx sly s2x s2y s3x s2y
ing joint variables are passed to the ANSYS input
file using the parametric design language to obtain Min 0 11 0 0 0 11 0
the critical Von Mises in the joint. The mass of the Max 60 138 60 126 60 138 60
joint is than assessed analytically by multiplying its Optimum 4 85 4 0 3 60 0
cross-sectional area with a steel density of 8 t/m3.
The weight factor for the stress is set to 0.9 and to
0.1 for the mass which means that stress minimiza- 1
tion is nine—time more preferred than mass. As 0.9
mass
a result, the variables of each design are stored in 0.8 stress (min mass)

the MS EXCEL-file and the optimum solution can 0.7 mass(min stress)
stress
0.6
be obtained. 0.5

0.4

0.3

2.3 Optimization procedure validation 0.2

0.1
The optimization procedure is validated with a 0

solid version of the joint shown in Figure 2, thus 0 10 20 30 40 50

Generation
60 70 80 90 100

having no interior edges. The objective of this

optimization is to convert the initial solid trian- Figure 6. Mass- and stress progression during the
gular joint into a solid rectangular with constant optimisation process.
thickness, which is known to be the best solution
(see Fig. 5). Therefore, the joint height, H, of the
joint is kept as a variable together with the B-spline This optimization for the purpose of validation
variables s1x, s1y, s2x, s2y, s3x and s3y. converged within 100 generation using a popula-
tion of 40 design alternatives. The initial population
was generated by a random selection of variables
VB.NET within the range defined in Table 1 in discrete 1 mm
application steps. As a result, the optimum design variables are
shown in Table 1, where the B-spline curves con-
verged to straight-line segments. Furthermore, it
can be seen in Figure 6 that the mass and the stress
Results.xls ANSYS VOP
were reduced simultaneously during the optimisa-
tion, thus indicating that the choice of weight fac-
tors results in a set of non-dominated minimum
mass and stress design alternatives for this joint
Figure 4. The optimization procedure. shape optimisation.

621

MARSTRUCT.indb 621 2/18/2011 5:54:57 PM

 To visualize the progression of the optimisation
procedure, the Von Mises stress and the mass of
the joint is normalized with the maximum occur-
ring stress and mass respectively. Furthermore,
Figure 6 includes the progression of the objective
stress (solid black line) and the corresponding mass
(dash-dot line), as well as the objective mass (solid
grey line) and the corresponding stress (dotted
line).
Figure 8. Minimum stress alternative (case 1).

3 THE OPTIMAL JOINT

Using the optimization procedure described

above, the asymmetrical joint shapes (Fig. 2) are
optimised including the grey section (Case 1) and
excluding the grey section (Case 2). The initial
population was generated by a random selection
of pre-defined variables for each case, see Table 2.
The B-spline variables are altered in discrete steps
Figure 9. Minimum mass alternative (case 1).
of 2 mm and the section thicknesses in discrete
steps of 1 mm. Furthermore, the variable fr is kept
constant at 0.5 mm for case 1.
3.2 Case 2
3.1 Case 1 The shape optimization converged within 110 gen-
The shape optimization converged within 110 gen- erations with a population of 40 design alterna-
erations with a population of 40 design alterna- tives. The minimum stress alternative was found in
tives. The minimum stress alternative was found in generation 97 and the minimum mass alternative
generation 71 and the minimum mass alternative was found in generation 95, see Figures 10, 11 and
was found in generation 94, see Figures 7, 8 and 9 12 and Table 4.
and Table 3.
3.3 Discussion on the optimised joint shapes
The principal finite element based shape opti-
Table 2. Variable range definition for joint at case 1 and misation resulted in the joint shapes shown in
case 2 [mm] (gray values are used only in the case 1- gray Figures 8, 9, 11 and 12. Because the actual loading
section). condition was simplified to a unit displacement,
slx sly s2x s2y s3x s3y t1 t2 t3 fr the purpose of this optimisation lies in the visuali-
zation of the rational assessment of the influence
Min 11 0 0 0 11 0 1 1 1 1 of the shape on the stress and mass reduction. The
Max 138 60 126 60 138 60 8 8 8 16 choice of boundary conditions allows some asym-
Step 2 2 2 2 2 2 1 1 1 1 metric loading distribution due to the asymmetric
geometry, but it remains for future work to iden-
tify if this simplification sufficiently represents
1 the panel-to-surrounding-structure connection.
0.9

0.8
mass However, due to the asymmetric support second-
stress (min mass)

0.7 mass (min stress) ary bending is invoked resulting in the ‘swan-neck’
0.6
stress
type of joint shape.
0.5

0.4
Considering the very simple boundary/loading
0.3
condition, the B-spline n°2 is probably almost
0.2 completely unaffected by the optimization process.
0.1

0
The similarity of the shape for both minima alter-
0 10 20 30 40 50 60
Generation
70 80 90 100 110
natives in each case confirmed this. Thus indi-
cating that case 2, without the additional grey
Figure 7. Optimization progress through generation section, is the preferred joint shape. Additionally,
(case 1). the results clearly indicate that the joint shape with

622

MARSTRUCT.indb 622 2/18/2011 5:54:59 PM

 Table 3. Optimal variables solutions (case 1).

Generation slx sly s2x s2y s3x s3y t1 t2 t3 Mass Stress

71 122 60 14 38 64 50 1 4 1 0.270 0.014

94 88 46 12 38 98 26 1 1 1 0.181 0.024

0.9
4 CONCLUSIONS
mass
0.8 stress(min mass)

0.7 mass (min stress)

stress
This paper presented a finite element-based shape
0.6

0.5
optimisation procedure for steel sandwich panel
0.4 joints. Therefore, the optimisation procedure was
0.3
described, validated and used to identify joint
0.2

0.1
shapes having a reduced stress and mass. As a
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140
result, the obtained joint shapes are presented and
Generation technical feasibility is discussed.
Furthermore, the identification of the optimised
Figure 10. Optimization progress through generation joint shape illustrates the basic principles of the pro-
(case 2). cedure. In the future, the procedure can be applied
to determine also the principal joint dimensions
in practical applications, namely the joint length
and height, considering shape limitations and vari-
ous boundary conditions. Additionally, further
research should concern the production complex-
ity and production cost to obtain optimized and
industrially feasible steel sandwich panel joints.

Figure 11. Minimum stress alternative (case 2). REFERENCES

Ehlers, S., 2004, “Design of Steel Sandwich Panels Joints

with Respect to Fatigue Life”, STG Summer Meeting
in Szcezin, pp. 65–68.
Fricke, W., Paetzold, H. & Zipfel, B., 2009, “Fatigue
Tests and Numerical Analyses of a Connection of
Steel Sandwich Plates”, Welding in the World,, Vol. 53,
No. 7/8, IIW-1954-08 (ex-doc. XIII-2244r2-08).
Goldberg, D.E., 1989, “Genetic Algorithms in Search, Opti-
mization and Machine Learning”, Addison-Wesley.
Klanac, A. & Jelovica, J., 2009, Vectorization and
Figure 12. Minimum mass alternative (case 2). Constraint Grouping to Enhance Optimization of
Marine Structures, Marine Structures, Vol. 22, No. 2,
pp. 225–245.
Kujala, P., Kotisalo, K. & Kukkanen, T., 1998, “Fatigue
Table 4. Optimal variables solutions (case 2) [mm]. of All Steel Sandwich Panels—Applications on Bulk-
head and Deck Structures of a Cruising Ship”, Pro-
Generation slx sly s3x s3y t1 t3 fr Mass Stress
ceedings of Practical Design of Ships and Mobile
Units, The Hague, pp. 879–887.
97 136 58 86 48 1 1 1 0.095 0.007
Kujala, P. & Roland, F., 2002, “Recent Developments in
95 72 56 78 40 1 1 1 0.073 0.016 Design and Production Methods of Steel Sandwich
Panels”, Proceedings of 3rd International Shipbuild-
ing Conference, St. Petersburg, Krylov Shipbuilding
the minimal mass is most feasible from a produc- Research Institute, pp. 134–142.
tion and technological point of view and may be Roland, F. & Reinert, T., 2000, “Laser-welded Sandwich
assembled from two curved plates or obtained by Panels for the Shipbuilding Industry”, Proceedings
extrusion. Furthermore, the geometric- and mate- of Lightweight Construction—Latest Developments,
rial discontinuities should be included to assess the London, pp. 1–12.
stress increase at these local hot spots.

623

MARSTRUCT.indb 623 2/18/2011 5:55:01 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Hull/superstructure-interaction in optimized passenger ships

H. Remes, J. Romanoff, P. Varsta, J. Jelovica & A. Klanac

Aalto University School of Science and Technology, Espoo, Finland

A. Niemelä
STX Europe, Turku, Finland

S. Bralic
Brze Vise Bolje d.o.o, Zagreb, Croatia

H. Naar
MEC Insenerilahendused, Tallinn, Estonia

ABSTRACT: The paper investigates the interaction between the hull and the superstructure in opti-
mized passenger ships when exposed to bending loads. The investigation is done using the Coupled Beams
theory which extends the basic beam theory to account vertical and shear stiffness between various decks.
Optimization of passenger ship with respect to weight and Vertical Centre of Gravity (VCG) is carried
out to create a set of Pareto-optimal solutions. The responses of these designs are compared in detail. The
investigation shows that the vertical and shear coupling between different decks significantly affects the
response of the passenger ships and changes load carrying mechanism of the hull girder. In the weight
optimal design the vertical bending moment is shared equally by the hull and the superstructure, while
in VCG optimal design the neutral axis approaches bottom plating of the ship considerably increasing
the share of load carrying of the superstructure. This means that the global response evaluation needs
to include vertical and shear coupling along whole length of the ship, and thus the simplified 2D section
models are not adequate for the conceptual design of passenger ship structures.

1 INTRODUCTION correcting the normal strain distribution along the

The size and structural complexity of passenger
ships has been increased dramatically during last
30 years. Thus, the design of a novel passenger ship
is often restricted by the limited reference database
and rational methods (see for example Hughes,
1980, 1983) to assess ship structural performance
must be applied even in concept design stage. For
structural design of passenger ships this means
that Finite Element (FE)—based analyses tech-
niques are needed (ISSC, 1997). These approaches
are time-consuming and do not allow large changes
in the steel general arrangement of the ship with-
out constructing and analyzing new FE-model.
Therefore methods have been developed for struc-
tural design of passenger ships which utilize both
speed of analytical methods and flexibility of
numerical methods.
Structural optimization in conceptual design
stage of hull girder obeying classical beam theory is
typically done with 2D-section models (e.g. Klanac
and Jelovica, 2009; Rigo, 2001). Recently, this
approach has been extended to passenger ships by
height of the hull girder (e.g. Andric and Zanic,
2010; Richir et al. 2007a&b; Zanic et al. 2007,
2010; Caprace et al. 2010). This update is done
using separate 3D FE analysis of the ship’s initial
design and by assuming that the shape of the
normal strain distribution does not change consid-
erably between different designs. As optimization
has moved from single-objective optimization
towards multi-attribute optimization, the differ-
ent Pareto-optimal designs can differ consider-
ably by scantlings from each other. This means
that the coupling between the hull and the super-
structure can be also very different. Therefore, this
approach to make correction to the 2D-section
model to assess primary response during multi-
objective conceptual design optimization becomes
questionable.
One major issue in passenger ship primary
response is shear deformation between differ-
ent decks (e.g. ISSC, 1997; Fransman, 1988;
Naar 2004). In case of multi-deck structures this
problem could be solved using so-called zigzag-
beam theories, see for example Reddy (2004),

625

MARSTRUCT.indb 625 2/18/2011 5:55:02 PM

 Carrera (2003, 2005). However, in passenger ships
the different decks and hull do not necessarily have
equal deflections (Bleich 1952, Naar 2004) mean-
ing that so-called layerwise theory of Reddy (2004)
needs to be applied. Naar et al. (2004) developed
and validated a method from this category of
beam theories; they included both shear and ver-
tical coupling between decks to model bending
behaviour of multi-deck passenger ships. However,
they did not study the load carrying mechanisms
when the scantlings of the ship are changed during
the structural optimization.
Therefore, the present paper aims through a
case study to investigate the differences on hull-
superstructure interaction between prismatic ships
that have been optimized for different objectives.
The designs considered are weight and Vertical Cen-
tre of Gravity (VCG) optimal, where the difference
could be assumed to be considerable by intuition;
while weight optimum aims to distribute the mate-
rial far from neutral axis, the VCG optimal design
positions material towards bottom of the ship. The
primary response is calculated by Coupled Beams
(CB) theory developed and validated with FEM by
Naar et al. (2004) and added to the local response
and strength models by Niemeläinen (2007) and
Mantere (2007). This structural analysis model
is coupled with genetic algorithm by Klanac and
Jelovica (2009) to establish the Pareto front; this
is done within ConStruct-software environment
(Remes, 2009). The hull-superstructure interaction Figure 1. Structural hierarchy used for conceptual
structural analysis with CB method.
is studied by comparing the deflections, curvatures,
shear strains and normal stresses of different decks.
In addition the deck forces, neutral axis variation The discretization of hull girder to coupled
along the length of the ship and bending moments beams is done by defining hard points in the
carried out by the hull and the superstructure have cross-section, i.e. typically the intersections bet-
been investigated. ween decks and bulkheads/girders. These hard
points act as reference axis of beam property
calculations. The extent of the beam in hori-
2 STRUCTURAL OPTIMIZATION zontal (y) and vertical (z) direction is defined by
the mid-nodes of each strake. After these points
2.1 Global response are defined the cross-sectional properties of the
beams, i.e. area (A in Fig. 1) and second moment
The global response of the ship is evaluated using
of area (I in Fig. 1), can be calculated from the
the Coupled Beams method developed by Naar
macro-element definitions. In addition to these the
et al. (2004), see Figure 1. The main idea is to divide
shear and vertical spring stiffness between beams
the hull girder to number of longitudinal beams that
can be defined. The hull girder is then exposed to
can deform both by bending and shear and which do
loading, p(x), at one of the coupled beams (CB1
not need to deflect the same amount. These beams
in Fig. 1). The coupling springs between various
can have in length direction one or several different
beams distributes this loading to entire hull girder
sections making it possible to have changes on the
solving the problem for deflections and nominal
cross-sectional properties along ship length. Each
stresses at each beam.
of these sections is composed of macro-elements,
whose properties are obtained from pre-defined
2.2 Local response and strength
material and profile tables (Niemeläinen, 2007).
The coupled beams are connected to each other by The tertiary stresses arising from the loads applied
shear and vertical springs to model slip or vertical on decks and bulkheads are calculated using the
deformation between the decks. basic Euler-Bernoulli beam and Kirchhoff plate

626

MARSTRUCT.indb 626 2/18/2011 5:55:02 PM

 theory and their analytical solutions, the details
are given in (Mantere, 2007). These stresses are
added to those induced by hull girder bending and
compared with the analytical strength criteria for
yielding and buckling (Niemeläinen, 2007). Thus,
vibratory response and comfort criteria are not
considered in the optimization.

2.3 Optimization with genetic algorithm

The division of hull girder to several macro-
elements results in large number of design variables.
In addition, the design space of multi-attributes
has a non-convex shape. Therefore, evolution-
based optimisation algorithms are most suitable.
The ConStruct platform uses the Genetic Algo-
rithm with vectorization and constraint grouping
(Klanac and Jelovica, 2009) to create the Pareto
surface for the given optimization problem.

2.4 Case description and selected optimum

designs
The load carrying mechanism of weight and VCG
optimal designs is studied with prismatic Post-
Panamax cruise ship having length L = 275.73 m,
B = 42.68 m, and H = 43.75 m; the main frame is
presented in Figure 2 and the ship is the same as
presented in Naar et al. (2004).
The web frame spacing is S = 2730 mm
and the longitudinal girders are y = 4000 mm,
y = 11480 mm and y = 17600 mm form the centre
line. The loads are selected according to DNV rules
(2007) and it covers the vertical bending moment Figure 2. The main frame of the case ship with decks
(Mt,s = −4400 MNm and Mt,h = 8400 MNm), exter- loads.
nal pressure loads and deck loads; see Figure 2.
23
This information is used to create the Pareto front
W opt.
Vertical centre of gravity [m]

which is presented in Figure 3. The differences in 22

plate thicknesses for weight and VCG optimum
21
designs is presented in Figure 4, while the dimen-
sions and materials of optimum designs are sum- 20
marized in Table 1 together with the objectives. 19
From Figure 3 it is seen that the VCG opti-
mum design has 19% lower VCG than the weight 18

optimum. The VCG optimum has 26% higher 17

weight than the weight optimum. Figure 4 and VCG opt.
16
Table 1 show that the scantlings of weight and
40 45 50 55 60
VCG optimal designs are considerably different.
While in VCG optimal design only the inner thick Weight [ton/m]
bulkhead forms the backbone of the superstruc-
Figure 3. 2D-presentation of the Pareto front for the
ture, in the weight optimal design also the outer selected case.
bulkhead has considerable thickness. The decks
and the double bottom of VCG optimal design
at the hull are much thicker than in weight opti- thickness of 9 mm. The active constraints in both
mal design. The main difference from the view designs are yielding and buckling due to shear at
point of hull superstructure interface, i.e. recess, x = L/4 and due to normal stress at x = L/2. Thus,
is that the VCG optimal design has plate thick- these two alternatives have considerably different
ness of 13 mm while the weight optimal design has main frames and are therefore good alternatives

627

MARSTRUCT.indb 627 2/18/2011 5:55:02 PM

 Figure 4. Comparison of the weight and VCG optimum designs.

Table 1. The weight and VCG optimal designs. First line indicates the plate thickness, stiffener spacing
and yield strength of the material, while the second line indicates the bulb flat dimensions and material. The
units are mm and MPa.

628

MARSTRUCT.indb 628 2/18/2011 5:55:04 PM

 two compare the differences in responses between
different Pareto-optimal designs.
3 COMPARISON OF WEIGHT AND VCG
OPTIMUM DESIGNS
3.1 General
The comparison is carried out by analyzing first
the responses (deflections, curvature, shear strains
and stresses) at deck level and then by analyzing
the hull and the superstructure as a whole.
forces of each deck and location of the neutral axis
can be calculated; neutral axis is considered as the
location of zero stress, is associated with vertical
bending and the distance is measured from the
base line of the ship. These can be used to calcu-
late the bending moments carried out by the super-
structure and the hull.
It has to be noted that the hull girder under con-
sideration is prismatic meaning that all the varia-
tions in response at longitudinal (x) direction are
due to beam bending behaviour where shear and
vertical flexibility are taken into account. All the
results are presented for hogging.

3.2 Displacements and stresses

In Timoshenko beam theory the beam can deflect
due to bending moment, M, and shear force Q.
The total deflection, W, is then
W = WM + WQ (1)
where WM and WQ are the deflections due to bend-
ing moment and shear force respectively. In beam
theory it is commonly assumed that all fibers
(i.e. decks) deflect the same amount. According to
Bleich (1952) this assumption is not necessarily valid
and therefore the total deflection between different
decks is selected for item to be investigated.
The normal strain, ε, is related to the curvature,
κ, of the beam by relation:
3.2.1 Vertical and shear flexibility
Figure 5A presents the comparison of total deflec-
tion, according to Eq. (1), of the double bottom,
deck at recess area (D5) and the upper most deck
(D11) for the weight and VCG optimum designs.
Figure 5B shows how this total deflection develops
between different decks. Figure 6A presents the
curvature and 6B the shear strain for double bot-
tom, recess (D5) and top deck (D11).
Figure 5A shows clearly that the double bottom,
recess area (D5) and the superstructure (D11)

d 2w
ε κ = −z M
(2)
dx 2

where z is vertical coordinate. Similarly the shear

strain is defined as

dw
γ =
Q
(3)
dx

Expressions (2) and (3) are used to identify the

differences between various decks on curvatures
and shear strains. This way the vertical and shear
coupling between different decks can be analyzed.
The normal stress at the hull girder can be obtained
from Eq. (2) with Hooke’s law. In present case the
material is considered to be the steel with the same
Young’s modulus for the entire ship. Therefore,
strain comparison can be skipped and only normal
stresses are compared. Then, the vertical distribu-
tions of normal stresses of the hull girder are ana-
lyzed at various locations along ships’ longitudinal
axis. From these stresses the loads carried out by Figure 5. A) the total deflection of double bottom,
each deck can be calculated with the knowledge recess (D5) and top deck (D11). B) relative deflection dif-
from Table 1. From these stresses the membrane ference for various decks.

629

MARSTRUCT.indb 629 2/18/2011 5:55:05 PM


Figure 6. A) Curvature and B) shear strain of double
bottom, recess (D5) and top deck (D11).
deflect differently; although the general shape is
the same the magnitude of deflection varies
considerably. Figure 5B also reveals that the deflec-
tion from double bottom up to deck D5 or D6
decreases gradually, but after this it remains con-
stant at the superstructure; in weight optimal design
the deflection is constant from D6-D11 while in
VCG optimal design this happens between D7-D11.
Thus, the vertical coupling between these parts of
the entire hull girder is clearly seen and varies based
on the design.
Figure 6A shows that the curvatures of the two
optimal designs vary considerably. It also shows
that the curvature is not constant within decks
of one design; at the recess the curvature is con-
siderably lower than in bottom or top deck. The
shear strains in Figure 6B show that the recess area
deforms under shear considerably more than in the
bottom or top decks. It is also seen that the shear
response is affected considerably by the design,
i.e. weight optimal design has higher shear strains
than the VCG optimal design. This is supported
by the fact that the plate thickness at the recess
area is considerably higher in VCG than in weight
optimum design.
630
MARSTRUCT.indb 630
3.2.2 Normal stresses, forces and moments
Figure 7 shows the vertical variation on the
normal stresses at x = L/8, L/4, and x = L/2.
Multiplying these stresses with areas of the decks
given in Table 1 the decks forces can be obtained.
Figure 8 shows the deck forces for each deck at
these same longitudinal locations and Figure 9
presents the variation of neutral axis along ship
length. Figure 10 shows the variation of the bend-
ing moment along ship length.
Figure 7 shows that the normal stress response
varies considerably between different longitudinal
sections of the ship; the same can be seen when the
weight and VCG optimal designs are compared.
While at x = L/2 and x = L/8 the normal stress has
piecewise linear normal stress along z-direction, at
x = L/4 the normal stress is nearly linearly distrib-
uted. It is also seen that the weight optimal design
has higher stresses at the bottom and top deck (D11)
than the VCG optimal design. When the deck forces
are looked at from Figure 8 and the position of neu-
tral axes from Figures 7 and 9, it is apparent that
the top most deck and the bottom contribute the
most to the global bending moment. From Figure 9
it is clearly seen that the neutral axis is constant in
weight optimal design and located at the intersection
between the hull and the recess area where also the
vertical centre of gravity is located. In VCG optimal
design the neutral axis if found to vary along ship
length, differs from VCG and is located at maxi-
mum 9 m below that of the weight optimum design.
Figure 10 shows that the bending moment is evenly
distributed between the hull and the superstructure
in weight optimal design, but in VCG optimal design
the hull carries about 30% of the moment at the mid-
ship while the contribution varies along ship length.
Figure 7. The normal stress at x = L/8, L/4 and L/2 for
VCG (dotted line) and weight (dashed line) optimum
designs.
2/18/2011 5:55:06 PM

Figure 8. Deck forces at x = L/8, L/4 and L/2 for VCG
(dotted line) and weight (dashed line) optimum designs.
Figure 9. Comparison of neutral axis along ship longi-
tudinal (x) axis.
Figure 10. Bending moment variation along ship lon-
gitudinal (x) axis.
631
MARSTRUCT.indb 631
4 DISCUSSION
The purpose of the present investigation was to
study the interaction between the hull girder and
the superstructure of optimized, prismatic pas-
senger ships. Coupled beam method developed
by Naar (2004) was used to assess the structural
response while optimization was carried out by
Genetic Algorithm developed by Klanac and
Jelovica (2009) utilizing the strength criteria from
Niemeläinen (2007). The optimization was per-
formed for weight and vertical centre of gravity.
The interaction was considered by looking at the
deflections, curvatures and shear strains, normal
stresses and axial forces carried out by each deck of
the prismatic ship. Also the variation of the bend-
ing moment along ship length was considered.
The investigation shows clearly that the verti-
cal and shear coupling are important phenomena
to consider during the optimization. Depending
on the scantlings of the structure this coupling
can change considerably. The investigation also
shows that the coupling is different at different
longitudinal sections of the ship. This means that
the entire ship length needs to be modelled in the
optimization and the approaches using section
models and pre-defined deck efficiencies cannot be
used when multiple-objective design optimization
is concerned. The investigation also shows that in
weight optimal design the neutral axis converges to
the vertical centre of gravity. This way the global
bending moment is carried out 50/50 by the hull
and the superstructure; in present case the main
dimensions are such that about 50% of the cross-
sectional area is in the superstructure and 50% at
the hull. In VCG optimal design the neutral axis
approaches bottom plating which leads to situa-
tion where the superstructure carries out consid-
erably larger share of the global bending moment
than the hull; in present investigation this division
is 70/30. In addition it is seen that the neutral axis
varies along ship length even though the ship is
prismatic.
The present investigation was carried out on
idealized, i.e. prismatic, ship and optimized with-
out considering fatigue issues or bending respo-
nse of secondary structures, i.e. web frames and
girders. To remove the prismatic assumption from
the design the theory by Naar (2004) would need
further development to account non-prismatic hull
beams. Alternative is to use 3D-FEM which is still
not today a good tool for concept design. To take
into analysis the fatigue issues, a concept design
method for fatigue needs to be established. The
bending response of secondary structures would
mean in cruise vessels that the vibration criteria
should be included into the optimization. The
items are left for future work.
2/18/2011 5:55:07 PM
 ACKNOWLEDGEMENTS
The research presented in the paper was carried
out in Finnish research project funded by Tekes,
STX Europe and Napa Ltd. The financial support
is gratefully appreciated.
REFERENCES
Andric, J. & Zanic, V., “The Global Structural Response
Model for Multi-Deck Ships in Concept Design Stage”,
Ocean Engineering, Vol. 37, 2010, pp. 688–704.
Bleich, H.H., “Nonlinear distribution of bending stresses
due to distortion of the cross section”, Journal of
Applied Mechanics, 1952;29, pp. 95–104.
Caprace, J.D., Bair, F. & Rigo, P., “Multi-Criteria Scant-
ling Optimization of Cruise Ships”, Ship Technology
Research, Schifftechnik, Vol. 57, No. 3, September
2010, pp. 210–220.
Carrera, E., “A Unified Formulation to Assess Theories
of Multilayered Plates for Various Bending Problems”,
Composite Structures, Vol. 69, 2005, pp. 271–293.
Carrera, E., “Historical Review of Zig-Zag Theories for
Multilayered Plates and Shells”, Applied Mechanics
Review Vol. 56, No. 3, May 2003, pp. 287–308.
Det Norske Veritas, “Direct Strength Analysis of Hull
Structures in Passenger Ships”, October 2007.
Fransman, J., “The influence of passenger ship
superstructures on the response of the hull girder”,
Transactions of RINA 1988, pp. 1–12.
Hughes, O., Mistree, F. & Zanic, V., 1980. A practical
method for the rational design of ship structures.
Journal of Ship Research 24(2): 101–13.
Hughes, O., 1983. Ship structural design: a rationally-based,
computer-aided optimization approach. New York:
Wiley, p. 566.
ISSC, Committee II.1—Quasi-Static Response, 13th
International Ship and Offshore Structures Congress,
18–22 August 1997, Trondheim, Norway, pp. 158–165.
Klanac, A., & Jelovica, J., “Vectorization and constraint
grouping to enhance optimization of marine struc-
tures”, Marine Structures 22(2), 2009, pp. 225–245.
Mantere, H., “Strength Analysis for the Design of Local
Ship Structures in Concept Stage”, Master’s Thesis.
Helsinki University of Technology, Department of
Mechanical Engineering, Ship Laboratory, 2007.
Espoo.
632
MARSTRUCT.indb 632
Naar, H., Varsta, P. & Kujala, P., “A theory of coupled
beams for strength assessment of passenger ships”,
Marine Structures Vol. 17, No. 8, 2004, pp. 590–611.
Niemeläinen, M., “Computational Method for Evalu-
ation of Hull Girder Strength in Concept Design
Stage”, Master’s Thesis. Helsinki University of Tech-
nology, Department of Mechanical Engineering, Ship
Laboratory, 2007, Espoo.
Reddy, J.N., “Mechanics of Laminated Composite Plates
and Shells – Theory and Analysis”, 2nd Edition, CRC
Press, pp. 726–738.
Remes, H., Klanac, A., Varsta, P. & Ehlers, S.,
“ConStruct—Platform for Conceptual Structural
Design”, Proceedings of the IMPROVE Confer-
ence, Dubrovnik, Croatia, 17–19 September 2009,
pp. 86–90, University of Zagreb, Croatia. pp. 86–90.
Richir, T., Caprace, J.D., Losseau, N., Bay, M.,
Parsons, M.G., Patay, S. & Rigo, P., “Multicriterion
Scantling Optimization of the Midship section of a
Passenger Vessel Considering IACS Requirements”,
10th International Symposium of Practical design of
Ships and Other Floating Structures, Houston, Texas,
2007, pp. 758–763. Ref. b.
Richir, T., Losseau, N., Pircalabu, E., Toderan, C. &
Rigo, P., “Least Cost Optimization of Large Pas-
senger Vessel”, Proceedings of the 1st International
Conference on Advancements in Marine Structures,
Glasgow, 2007, pp. 483–488. Ref. a.
Rigo, P., “Least-Cost Structural Optimization Oriented
Preliminary Design”, Journal of Ship Production
17(4), 2001, pp. 202–215.
Zanic, V., Andric, J. & Prebeg, P., “Decision Support
Methodology for Concept Design of Multi-deck Ship
Structures”, 10th International Symposium of Prac-
tival design of Ships and Other Floating Structures,
Houston, Texas, 2007, pp. 468–476.
Zanic, V., Andric, J., Prebeg, P., Stipcevic, M. & Piric, K.,
“RoPax Structural Design-Multi Level Decision Sup-
port Methodology”, 11th International Symposium
of Practival design of Ships and Other Floating
Structures, Rio de Janeiro, Brazil, 2010, pp. 490–501.
2/18/2011 5:55:09 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Optimization of monopile offshore wind structures

A. Thiry, F. Bair, L. Buldgen, G. Raboni & P. Rigo

University of Liège, ANAST, Liège, Belgium

ABSTRACT: As industrialization is essential for large-scale offshore wind deployment, a computerized

methodology to optimize the scantling of monopile steel structures has been developed using a genetic
algorithm. The objective functions of this tool are minimization of the structure weight and production
costs. The constraints implemented are mainly related to structural stability under extreme loads, reso-
nance check of the wind turbine support and fatigue strength at welded connections between shells and
stiffeners. This optimization should be invoked at the preliminary stage of the offshore project in order to
assess the impacts of the structure design on the fabrication and installation costs.

1 INTRODUCTION 2 GENERAL SCHEME OF THE

Since the very beginning of the century, wind
market has started to move offshore. Indeed oceans
locations offer a higher wind quality and the visual
impact of wind farms is significantly reduced.
Offshore wind power is now expected to represent
an important share of the future power supply
both in Europe and worldwide, reducing depend-
ency of industrial countries toward fossil fuels.
The main driver behind the methodology exposed
in this paper is the improvement of classical steel
solutions dedicated to monopile structures used in
offshore wind industry (Fig. 1). A structural optimi-
zation tool has thus been implemented to investigate
the gains in terms of weight and production costs
generated by the use of high tensile steel for shells, as
well as longitudinal and circumferential stiffeners.
OPTIMIZATION TOOL
The flowchart of the optimization tool described
in this paper is presented on Figure 2. The input
data implemented by the designer of the project
are the starting point of the computation. They
are related to the environment at the offshore loca-
tion, the characteristics of the wind turbine and
the scantling of the support structure (without any
consideration about the underground part of the
monopile).
The set of selected design variables and input
data are then used as a basis to assess the objec-
tive function (weight or production cost) and the
constraints (structural integrity) of the monopile
structure.

Figure 2. General scheme of the optimization

Figure 1. Steel monopile offshore wind turbines. algorithm.

633

MARSTRUCT.indb 633 2/18/2011 5:55:09 PM

 A genetic algorithm finally combines these
two elements—objective function and structural
integrity—over a certain number of iterations
in order to find the optimum scantling of the
support structure placed in the given external
conditions.

3 INPUT DATA

3.1 Environment
Environmental data summarizes the characteris-
tics of the site, data related to fatigue analysis and
load cases considered for the ultimate limit state
analysis.
Firstly, site data contains the values for water
depth d, the power law exponent α characterizing
the vertical distribution of wind speeds over the
tower height, the densities of the air ρa and of the
sea water ρw.
Secondly, for fatigue concern a distinction is
made between waves and wind actions. On one
hand, waves participation is presented under the
form of a list of sea states (or scatter diagram), each
one being characterized by a significant wave height Figure 3. Screen shot of the optimization tool.
HS, a mean zero up-crossing period TZ and a per-
centage of occurrence of the sea state PSS. On the
other hand, spectrums of punctual tower top loads monopile structure: the hub height Hhub, the
are used to describe fluctuating wind loadings on number of blades n, the rotor diameter drotor, the
the structure. weight of the rotor-nacelle assembly mTop, the tech-
Finally, data related to ultimate limit states are nical design lifetime of the turbine and its range of
listed under the form of a series of environmen- rotational speed.
tal situations and their associated wind and waves The interface of the optimization tool is pre-
conditions: average wind speed at hub height Vhub, sented on Figure 3.
water level elevation Δd compared to the mean still
water level MSL (elevation due to tide or storm for 4 ASSESSMENT OF CONSTRAINTS
example), wave height Hw and period Tw and a set
of punctual tower top loads. 4.1 Generals
The constraints implemented in the optimization
3.2 Support structure process are typically divided in two categories: geo-
metrical and structural constraints.
The monopile offshore wind tower considered in the
The first type of constraints refers to geomet-
study is an assembly of several conical or cylindri-
rical requirements such as equality of shell rings
cal tubular segments. The segments themselves are
diameters in one segment or decrease of shell
made of shell rings linked together with butt welds.
thicknesses and diameters while progressing to the
Each shell ring is characterized by a shell thickness,
top of the monopile structure.
upper and lower diameters, a height, a steel grade
The second type represents structural constraints
and a category of detail for the butt weld.
related to the verification of the structural integrity
Four framing systems are envisaged for the
of the offshore wind turbine towards fatigue, ultimate
scantling of shells belonging to the same tower
limit states and resonance phenomena. The verifica-
segment: unstiffened, longitudinally stiffened, ring
tion of those constraints requires the use of either
stiffened or orthogonally stiffened shells.
quasi-static or dynamic analysis of the structure.

3.3 Wind turbine 4.2 Building the 2D dynamic model

Some general data of the wind turbine are taken In order to perform dynamic analyses, a simple
into account for this preliminary design of the 2D dynamic model made of concentrated masses

634

MARSTRUCT.indb 634 2/18/2011 5:55:10 PM

 connected together with a translational spring is
built on the base of the scantling (Fig. 4). Nodes
are located at the intersection between shell rings
and are characterized by two degrees of freedom:
one horizontal translation xi and one in-plane
rotation θi (vertical translations are not taken into
account as they are supposed negligible compared
to horizontal translations). The model is perfectly
clamped at a distance from the sea bed level equal
to the height of the first shell ring. This assump-
tion allows the designer to take into account the
length required for the complete soil restraint to
develop around the monopile.
The aerodynamic damping generated by the
rotor is modeled with a single dashpot connected
to the tower top degree of freedom in translation
and the structural damping is expressed as a com-
bination of the generalized matrixes of masses
[M*] and rigidity [K*] (Rayleigh damping).
The value and the distribution of masses of the
wind turbine are not supposed to change. Thus,
cases where the offshore structure is covered with
a layer of ice or marine growth are excluded from
the dynamic analysis.
Basically, the simplicity of the chosen dynamic
model is justified by two considerations. On one
hand, the optimization process requires a great
number of iterations that should not take too
much computation time. And, on the other hand,
loss of accuracy in the results is not a major prob-
lem as the methodology is dedicated to early stage
design.
4.3 Resonance of the support structure
Excitations are likely to occur at frequencies that
are close to the natural frequencies of the offshore
wind turbine, leading to resonance phenomena.
Natural frequencies and corresponding modes
Figure 4. The 2D dynamic model of the offshore wind
turbine studied in the optimization process.
635
MARSTRUCT.indb 635
of the support structure are defined respectively
from the→generalized eigen values λ and eigen
vectors V of the matrixes of masses [M] and
rigidity [K]. The Campbell diagram related to
the excitation frequencies for the rotor motion
(1n) and the blades passing (3n for a 3-bladed
wind turbine) can then be used to check if reso-
nance of the support structure is avoided within
the rotational speed range of the wind turbine
(Fig. 5).
4.4 Fatigue check
4.4.1 Generals
Fatigue strength is verified at each structural
detail of the support structures (butt welds, ring-
stiffener connections, etc.) thanks to the Miner
rule (chech that cumulative fatigue damage
Dfat < 1).
In addition, the procedure developed for fatigue
analysis assumes that wind and waves effects are
completely uncoupled. Hence, the total cumula-
tive fatigue damage Dfat results from the addition
of the fatigue damages due to wind actions Dfat,wind
and waves actions Dfat,wave.
Concerning the characteristics of the wind
tower, structural members are supposed to be con-
tinuous and steel is considered as an isotropic and
homogeneous material.
4.4.2 Fatigue due to waves actions
Over its life, the offshore wind turbine will expe-
rience series of sea states, each one generating
cyclic loadings and being responsible for a certain
percentage of the total cumulative fatigue dam-
age due to waves Dfat,wave in the structural compo-
nents. Thus, for each sea state listed in the scatter
diagram, the equations governing the dynamic of
the structure and the stress range histogram at the
Figure 5. Campbell diagram for a 3-bladed offshore
wind turbine.
2/18/2011 5:55:11 PM
 nodes are solved using stochastic dynamic theory The systems of equations are solved in the
in the frequency domain. modal basis to get the power spectral densities of
In the first step of the computation, a fully bending moments PSDM,wave in the structure at each
developed sea at infinite fetch is assumed and frequency step. Power spectral densities of normal
the Pierson-Moskowitz wave spectrum PSDH stresses PSDσ,wave,i in the structural details are then
(equation 1) is generated at each frequency step for deduced using simple beam theory (see example on
the concerned sea state. Fig. 6).
1 In the next step of the computation, the Rayleigh
H s2 − (f z )−4 counting method is invoked to assess the number of
SDH f ) =
PSD e π (1)
4π Tz4 f 5 fatigue cycles nwa and their associated stress range
Swa,k characterizing the sea state (equation 5).
where HS = significant wave height of the sea state;
TZ = mean zero-up crossing period; f = frequency. m2
nwa ( wa , k ) Td ( PRayeigh
(P gh ( Swa , k + 1 ) PRayeig
gh ( Swa , k )) (5)
Secondly, the value of the power spectral den- w
m0
Rayeig Rayeigh

sity of the water particles acceleration PSDaw,i is

deduced from the Pierson-Moskowitz spectrum
PSDH according to equation 2.
⎛1 ( f
2
SDaw ,i f ) = PSD
PSD S H f )⎜ ⋅
⎝2 sin h( kwave d )
where kwave = wave number; zi* = vertical position
of the node i from sea bed level; d = water depth;
i = numbering of the node in the dynamic model.
Once the water particles acceleration spectrum
PSDaw,i is established for each underwater nodes
of the model, the distribution of hydrodynamic
inertia loads spectrum is computed in the matrix
[PSDFwave]. Each element PSDFwave,i,j of this matrix
results from the Morison formulation and the
coupling between nodal forces inherent to the sto-
chastic theory (equation 3).
⎧Ai PSDaaw ,ii ⋅ Aj PSD
S aw
aw
PSD
S Fwwave ,ii j =⎨
⎩ 0
where nuw = number of shell rings located below
the mean still water level; A = parameter allowing
calculation of wave loads and distribution of
these loads on the nodes of the dynamic model
(see equation 4).
where Td = duration of the concerned sea state;
m0 = variance of the power spectral density of nor-
kwave ( z*i )) ⎞
z
mal stresses PDSσ,wave,i at node i; m2 = third spectral
⎟⎠ (2) moment of the power spectral density of normal
stresses PDSσ,wave,i; PRayleigh = probability of occur-
rence of the stress range Swa,k.
Figure 7 shows the normal stress range histogram
(nwa, Swa,k) obtained from the power spectral density
of normal stresses at the clamping point presented
on Figure 6.
At the end, the normal stress range histograms
associated to each sea state are summed and con-
ventional S-N curves based on the detail category
are used to assess the cumulative fatigue damage
due to waves DFat,wave.
4.4.3 Fatigue due to wind action
for i j ≤ nuw For the preliminary design procedure devel-
j
(3) oped in this study, wind loads coming from the
for i j > nuw
rotor-nacelle assembly are completely uncou-
pled from the dynamic of the support structure.

⎛ πD 2 ⎞
Ai Cm ρw ⎜ i ⎟ LF ,i (4)
⎝ 4 ⎠

where Cm = hydrodynamic inertia coefficient (taken

equal to 2); ρw = sea water density; Di = diameter of
the structure at node i; LF,i = length of the part of
structure submitted to the waves load above and
below the node i.
Note that the hydrodynamic drag term is not
taken into account in the analysis of wave induced
fatigue. This is due to the fact that sea conditions
considered for the fatigue analyses are less severe
than in extreme wave conditions, making the Figure 6. Power spectral density of normal stresses at
hydrodynamic inertia load predominant in the the clamping point of a structure when HS = 0.25 m and
Morison formulation. TZ = 4 s.

636

MARSTRUCT.indb 636 2/18/2011 5:55:13 PM


Figure 7. Normal stress range histogram of nor-
mal stresses at the clamping point of an offshore wind
turbine.
Physically, this approach means that the “transfer”
between the 3D wind speeds field and the tower
top loads is performed considering that the turbine
is connected to an infinitely rigid support. As a
result, the value of the cumulative fatigue damage
due to wind DFat,wind is only based on the variations
of tower top loads and their associated number of
cycles nwi.
These loads are first extrapolated over the height
of the structure to obtain the range of bending
moment ΔMwind,i at each node i (see equation 6).
Δ Δ
ΔF
wind
i ,i o ⋅Δ
Fx ,top ΔM y,top
o (6)
where ΔFx,top = range of axial load at tower top;
ΔMy,top = range of bending moment at tower top;
Δzi = distance between the concerned structural
detail and the top of wind tower.
The histogram of normal stress ranges Swind,i is then
computed on the basis of the section characteristics
(equation 7) and the Miner rule is finally applied to
assess the cumulative fatigue damage DFat,wind.
– Mx: bending moment in Oyz-plane;
– Mz: torque in z-direction.
For each load case considered in the design proce-
dure, those distributions of internal loads result from
the superposition of the three following actions:
1. Action of wind on the rotor and nacelle;
2. Action of wind pressure over the height of the
emerged tubular structure;
3. Action of waves on the substructure.
The 3D configuration of these three actions is
presented on Figure 8.
Similarly to the procedure developed for the
assessment of fatigue strength, wind loads on
the rotor and nacelle are first replaced by a set
of equivalent punctual loads multiplied by safety
coefficients and directly applied at the top of the
offshore wind turbine.
Secondly, a pseudo-elastic calculation is per-
formed to assess wind loading on the tubular
structure. A power law profile is chosen for the dis-
tribution of wind speed vwind and the correspond-
ing wind pressures qsd,wind are evaluated according
to equation 8.
1
i ( z ) = ρa ⋅ Cw ν wind 2
qsd ,wind wind
ind press i (z) (8)
2
where ρa = air density; Cwind press = safety coefficient
on wind pressures; vwind= wind speed given by the
power law distribution; z = vertical position.
The resulting drag load WSd,wind drag per meter
applied at height z is found from expression 9.

ΔMwiind ,i
Swiind ,i = (7)
Wi

where Wi = section modulus at the level of the

structural detail i.

4.5 Ultimate limit states

4.5.1 Design loads
The ultimate limit state analysis implemented in
this early optimization design stage considers six
distributions of internal loads which are:
– Fx: shear force in x-direction;
– Fy: shear force in y-direction;
– Fz: vertical force; Figure 8. Distribution of wind and waves actions
– My: bending moment in Oxz-plane; considered for the ultimate limit state analysis.

637

MARSTRUCT.indb 637 2/18/2011 5:55:16 PM

 i drag ( z ) = cd ( z ) ⋅ qsd ,wind ( z ) ⋅ D( z )
Wsd ,wind
wind (9) where ρw = water density; Cd = hydrodynamic drag
coefficient; Cm = hydrodynamic inertia coefficient;
where cd = drag coefficient calculated from speci- Di = outer diameter of the monopile at node i;
fications prescribed by [5]; D = diameter of the Li = exposed height at node i; zi: height of the node i.
→
tubular structure at height z. The displacements vector x of the structure at
This drag load is to be multiplied by a gust the time step t result from the superposition of the
response factor G to account for the dynamic ampli- eigen modes weighted by their respective dynamic
fication of the structure response (equation 10). amplification factors qk (equation 15). Dynamic
amplification factors are assessed iteratively on the
Wsd ,wind
i presss ( z )
pres = GW
Wssdd ,wind ddrag ( z ) (10) basis of the structure and load vector thanks to the
Newmark method.
where WSd,windpress = design load due to wind pres-
→ nddt →
sure on the tower. x(t ) = ∑ qk (t ) ⋅V k (15)
In the third part, the computation of internal loads k =1
due to waves is performed through a dynamic analy-
sis of the support structure in the time domain. where qk(t): dynamic amplification factor of the
The water elevation characterizing a load case →
mode k; V k kth eigen vector of the structure.
is taken into account and a regular wave profile is The displacement vector is then used to assess
generated from the wave height Hwave and period the evolution of internal loads of the support struc-
Twave introduced in the environmental data. ture submitted to waves actions and the maximum
The values for horizontal velocity vw and acceler- values observed over the simulation are listed.
ation aw of water particles at each time step are then The distributions of internal loads obtained for
derived from the linear wave theory (Airy) corrected each load case are finally combined to find the
with the Wheeler stretching formulation in order to envelop diagrams of the support structure.
describe their kinematics in terms of instantaneous
surface elevation (see equations 11 and 12). 4.5.2 Structural check
The strength of each shell ring submitted to the
⎛ d z* ⎞ design loads defined in the previous paragraph is
cos h ⎜ kwave ⋅
1 ⎛ 2π ⎞ ⎝ d ξ (t ) ⎟⎠ checked according to the specifications given in
vw ( z, t ) Hwave ⎜ ⎟
2 ⎝ Twave ⎠ sin h ( kwave d ) DNV [4] or Germanischer Lloyd rules [3]. The fol-
lowing failures modes are considered:
⎛ 2π ⎞
× cos ⎜ − t⎟ (11) – Shell buckling of unstiffened shell rings;
⎝ T ⎠wave – Panel stiffener buckling of longitudinally stiff-
ened shell rings;
⎛ d z* ⎞ – Panel ring buckling of transversally stiffened
2 cos h ⎜ kwave
1 ⎛ 2π ⎞ ⎝ d ξ (t ) ⎟⎠ shell rings;
aw ( z, t ) Hwave ⎜ ⎟ – Overall buckling of orthogonally stiffened shell
2 ⎝ Twave ⎠ sin h ( kwave d )
rings.
⎛ 2π ⎞
× sin ⎜ − t⎟ (12) An additional constraint is also implemented
⎝ t ⎠
wave for the overall buckling of the column.
where Hwave = wave amplitude; Twave = wave period;
kwave = wave number; d = water depth including 5 DESCRIPTION OF THE OPTIMIZATION
water elevation; z* = z-position from sea bed level; PROCESS
ξ = instantaneous surface elevation; t = time.
The water particle kinematic is used to assess the 5.1 Design variables
hydrodynamic drag (equation 13) and inertia loads
(equation 14) that are finally summed together to The design variables selected for the optimization
assess each component Fwave,i (t) of the load vector process are divided in two categories:
[F
Fwave] at time t. – Shell ring variables: shell thickness, lower and
upper diameters of the ring, steel grade;
1
Fi , wave ddrag (t ) = ⋅ ρw ⋅ Cd Di ⋅ Li | vw ( zi t ) | vw ( zi , t ) (13) – Segment variables: number and profile of longitu-
2 dinal and ring stiffeners distributed respectively over
the circumference and the height of the segment.
Di2
Fi , wave iinertia (t ) = ρw ⋅ Cm ⋅ π ⋅ ⋅ Li aw ( zi ,t ) (14) The height of shell rings and segments are not
4 supposed to change during the optimization as

638

MARSTRUCT.indb 638 2/18/2011 5:55:22 PM

 they are considered as geometrical data fixed by the
designer and the manufacturer of the structure.
5.2 Description of the optimizer algorithm
The algorithm chosen for the problem presented in
this paper is a Genetic Algorithm (GA). Genetic
algorithms are search algorithms that work via
the process of natural selection. They begin with
a sample set of potential solutions which evolves
towards a set of more optimal solutions after
several iterations. Within the sample set, poor solu-
tions tend to die out while better solutions mate
and propagate their advantageous traits, introduc-
ing better solutions into the set (though the total set
size remains constant). A little random mutation
guarantees that a set won’t stagnate while filling up
with numerous copies of the same solution.
In general, genetic algorithms tend to work
better than traditional optimization algorithms
because they are less likely to be led astray by
local optima. This is because they don’t make use
of single-point transition rules to move from one
single instance to another in the solutions space.
Instead, GA’s take advantage of an entire set of
solutions spread throughout the solution space, all
of which are experimenting upon many potential
optima.
However, a few criteria must be met in order for
GA to work effectively:
– The assessment of “how good” a potential solu-
tion is compared to other potential solutions
must be relatively easy;
– The breaking of a potential solution into inde-
pendent discrete parts must be possible;
– Genetic algorithms are best suited for situations
where a “good” answer will suffice, even if it is
not the absolute best answer.
5.3 Evaluation of potential solutions
The “fitness function” is responsible for perform-
ing the evaluation of solutions compared to each
other. Basically, this module returns a positive inte-
ger number, or “fitness value”, that reflects how
optimal the solution is: the higher the number, the
better the solution.
The fitness values are then used in a process of
natural selection to choose the potential solutions
that will survive in the next generation and those
that will die out. However it should be noted that
natural selection process does not merely select the
top x number of solutions. Instead, solutions are
chosen statistically so that it is more likely for a
solution with a higher fitness value to be selected,
but it is not guaranteed. This tends to correspond
to the natural world.
639
MARSTRUCT.indb 639
In the methodology developed in this study,
the equation implemented to assess the fit-
ness value combines the criteria of weight/cost
minimization and the criteria related to the
constraints (equation 16). The relative impor-
tance of each term is set thanks to a weighing
coefficient fP.
nc
Pj k
Fitk = α ⋅ θ k − β ⋅ ∑ (16)
nc
j −1
where Fitk = fitness value of the solution k;
θk = criterion of objective function minimization
given by the ratio between the minimum value of
the objective function and the value of the objec-
tive function for solution k; Pj,k = penalty associ-
ated to the constraint j for the solution k (see
equation 17); nc = number of constraints; α = fP;
β = 1/fP; fP = weighing coefficient.
⎧ 0 if the constraint j is satisfie
f d
⎪
Pj k = ⎨ ⎛ C max, j k ⎞ (17)
⎪1 − ⎜ C ⎟ if not
⎩ ⎝ j k ⎠
where Cmax,j,k = maximum permissible value for the
constraint j in the solution k (for example the max-
imum permissible stress to avoid shell buckling of
unstiffened panels) and Cj,k = value found for con-
straint j in the solution k (for example the design
Von Mises stress computed in the ultimate limit
states analysis of the solution k).
The goal of the optimizer is to increase the fit-
ness: this is done while decreasing the weight (or
the cost) and satisfying as much structural con-
straints as possible.
6 OPTIMIZATION OF A 5 MW OFFSHORE
WIND TURBINE
6.1 Generals
The computerized tool has been tested on the
scantling of a 5 MW offshore wind turbine. The
characteristics of the support structure, environ-
mental conditions, optimisation parameters and
results are described hereafter.
6.2 Characteristics of the offshore wind turbine
The main characteristics of the wind turbine are
summarized in the Table 1.
The initial monopile support structure consid-
ered in the study is made of steel S235 and its gen-
eral dimensions are listed in Table 2.
Concerning the dynamic of the structure, the
logarithmic decrements on the first mode due
2/18/2011 5:55:25 PM
 Table 1. Characteristics of the 5 MW offshore wind Table 3. Wind parameters for dynamic pressure zone
turbine. IEC IA.

Description Value Unit Description Value Unit

Wind turbine power 5 [MW] Reference wind speed Vref 37.5 [m/s]
Number of blades 3 [−] Average wind speed Vave 10 [m/s]
Rotor diameter 118 [m] Intensity of turbulences I 0.18 [−]
Nacelle mass (incl. rotor 390 [tons]
blades)
Speed range 4−14.2 [min−1] Table 4. Values of cumulative fatigue damages for the
initial scantling S235.

z-position Dfat,wave Dfat,wind Dfat

Table 2. Characteristics of the initial monopile support Section [m] [−] [−] [−]
structure (steel grade S235).
23 110 (tower top) 9.154e-9 8.425e-4 8.425e-4
Description Value Unit
22 105 1.077e-5 0.007 0.007
Hub height (above MSL) 80 [m] 21 100 1.235e-4 0.023 0.023
Tower length (above MSL) 77 [m] 20 95 3.534e-4 0.030 0.030
Length between MSL and seaground 30 [m] 19 90 7.260e-4 0.040 0.041
Height of shell ring below seaground 5 [m] 18 85 0.002 0.068 0.070
Outer shell diameter at top 4 [m] 17 80 0.003 0.083 0.086
Outer shell diameter at seaground 5.7 [m] 16 75 0.004 0.085 0.089
Shell thickness at top 25 [mm] 15 70 0.005 0.101 0.106
Shell thickness at seaground 115 [mm] 14 65 0.006 0.106 0.112
Weight 902.1 [tons] 13 60 0.009 0.124 0.133
12 55 0.009 0.117 0.126
11 50 0.012 0.138 0.150
10 45 0.014 0.139 0.153
to structural and aerodynamical damping were 9 40 0.017 0.150 0.167
assumed equal to 0.012 and 0.01 respectively. 8 35 (MSL) 0.019 0.161 0.180
7 30 0.020 0.156 0.176
6 25 0.021 0.152 0.173
6.3 Environmental conditions 5 20 0.019 0.130 0.149
6.3.1 Data related to fatigue analysis 4 15 0.014 0.091 0.105
The scatter diagram taken into account to assess 3 10 0.012 0.082 0.094
2 5 (seaground) 0.012 0.070 0.082
wave induced fatigue is made of 15 typical sea
1 0 (clamping) 0.012 0.070 0.082
states observed in the North Sea [1].
Fluctuating wind loads were calculated by an
external bureau of study for the wind turbine placed
in a dynamic pressure zone IEC IA (see Table 3). m
Ve 1 25 Vreff 62, 5 (18)
The cumulative fatigue damages due to wind s
and wave action for each butt weld of the structure
The extreme wave profile considered in the
are listed in the Table 4. Note that the category of
optimization process is a regular wave profile char-
detail for butt welds is set to 112.
acterized by a wave height Hw and wave period Tw
equal to 10 m and 14 s respectively. No current or
6.3.2 Data related to ultimate limit states water elevation compared to MSL were assumed
Time domain simulations were performed by the in this load case.
bureau of study with the software NREL aerodyn Figure 9 shows the envelop diagrams of inter-
to establish the set of extreme tower top loads due nal loads generated by the software for the initial
to the wind action coupled with the rotor dynamics. scantling S235 placed in the environment described
These data are not presented in this paper for con- previously.
fidentiality reasons.
The distribution of extreme wind speeds over
6.4 Optimization parameters
the tower height is based on the power law coef-
ficient with an exponent α equal to 0.14 and the The optimization carried out was based on the
highest occurring wind speed Ve50 at hub height for minimization of the structural weight of the off-
a recurrence period of 50 years (equation 18). shore wind structure.

640

MARSTRUCT.indb 640 2/18/2011 5:55:26 PM

 Table 5. General characteristics of optimal scantlings
S355.

Description Value Unit

Diameter at top 4 [m]

Diameter at seaground 5.3 [m]
Shell thickness at top 22 [mm]
Shell thickness at seaground 94 [mm]
Structural weight 712.5 [tons]

Figure 9. Envelop diagrams of internal loads for the

initial scantling S235.

Figure 10. Evolution of weight and production cost

during the optimization process of the structure S355.
As the methodology aims to highlight the advan-
tages of using high tensile steel in offshore structures,
this optimization was performed on an unstiffened
structure made of conventional steel grade S355.
The design variables selected for the process were
the shell thickness (ranging from 8 to 150 mm),
lower and upper diameters of shell rings (ranging
from 4 to 6 m). Optimizations based on the varia-
tion of number and profiles of stiffeners were not
envisaged.
6.5 Optimization results
The evolution of the structural weight and produc-
tion cost during the optimization process for the
unstiffened structure made of steel S355 is pre-
sented on Figure 10. It can be seen that the con-
vergence to the optimum solution is ensured after
about 1000 iterations.
The general characteristics of the optimum
solutions made of steel S355 are summarized in
the Table 5.
Figure 11. Drawings of initial scantling S235 and
optimal scantlings S355.
In this optimum scantling, the values of shell
rings diameters are slightly lower than in the initial
scantling but only for the lower part of the struc-
ture (see Fig. 11).
On the other hand, the shell thicknesses are
significantly lower in the optimal solution S355.
This leads to weight and cost reductions equal to
21% and 20.4% respectively compared to the initial
scantling made of steel S235.
7 CONCLUSIONS
In this paper, an optimization tool dedicated to the
early design stage of steel monopile offshore wind
turbines has been described. Constraints related
to the structural integrity of the support structure
are assessed and an optimum solution in terms of

641

MARSTRUCT.indb 641 2/18/2011 5:55:27 PM

 weight or production cost is obtained thanks to a
genetic algorithm.
The dimensions of the support structure are
reduced while using high tensile steel instead of
normal steel grade. As an example, the optimiza-
tion tool showed that the saves in terms of weight
and production cost can reach about 20% when
steel grade S235 is replaced by steel S355.
Finally, a certain number of elements could be
added to the methodology in future developments.
First, the optimization process could be extended
to the scantling of the underground part of the
support structure. Second, if the coupling between
fluctuating wind loading and structure dynamic is
considered at each step of the iteration process, the
accuracy of the cumulative damage found from the
fatigue analysis would be improved. And finally,
some phenomena such as irregular wave profiles,
marine currents, ice and marine growth formation
could be taken into account in ultimate limit state
analyses.
642
MARSTRUCT.indb 642
REFERENCES
Eurocode 1: Bases de calcul et actions sur les structures,
May 1995.
Guideline for the Certification of Offshore Wind Turbines,
Germanischer Lloyd WindEnergie, edition 2005.
Kühn M. 2001. Dynamics and Design Optimisation of
Offshore Wind Energy Conversion System, PhD Thesis,
Delft University of Technology, Section Wind Energy,
Department of Civil Engineering.
Kühn M., Cockerill T.T., Harland L.A., Harrison R.,
Schöntag C., Van Bussel G.J.W. & Vugts J.H. 1998.
Opti-OWECS Final Report Vol. 2: Methods Assisting
the Design of Offshore Wind Energy Conversion Sys-
tems, Delft University of Technology, ISBN 90-76468-
03-6.
Recommended Practice DNV-RP-C202—Buckling Strength
of Shells, Det Norsk Veritas, October 2002.
Van der Tempel J. 2006. Design of Support Structures for
Offshore Wind Turbines, PhD Thesis, Delft University
of Technology, ISBN 90-76468-11-7.
2/18/2011 5:55:29 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Isogeometric shape optimization of design-dependent structures

Minho Yoon, Bon-Yong Koo & Seonho Cho

National Creative Research Initiatives Center for Isogeometric Optimal Design and Department of Naval
Architecture and Ocean Engineering, Seoul National University, Seoul, Korea

ABSTRACT: Using an isogeometric approach, a continuum-based shape optimization method is devel-

oped for elasticity problems. To obtain efficient and precise adjoint shape sensitivity, precise normal and
curvature information should be taken into account in shape sensitivity expressions, especially for design-
dependent problems. In this approach, the basis functions generated from NURBS are directly used to
construct a geometrically exact model in response and shape sensitivity analyses. Refinements and design
changes are easily implemented within the isogeometric framework. The isogeometric design sensitivity
analysis provides more accurate sensitivity of complex geometries including higher order terms. Also, it
vastly simplifies the design modification without communicating with the CAD geometry during optimi-
zation process. We demonstrate some numerical examples, where the accuracy and efficiency of the iso-
geometric sensitivity are verified by the comparison with finite difference one. Also, numerical examples
for design-dependent design problems are demonstrated to verify the applicability and effectiveness of
the proposed method.

1 INTRODUCTION and shape sensitivity analyses. Refinements and

Typically in Finite Element (FE) based engineering
analysis, designs are embedded in CAD systems
and the FE meshes are generated from the CAD
data. The geometric approximation inherent in the
mesh may lead to accuracy problems in response
analysis and more adversely in design sensitiv-
ity analysis. Piecewise linear approximation of
geometry is the root cause. Even though a mesh
is constructed, further refinement requires tedi-
ous communication with the CAD system during
design iterations. The objective of isogeomet-
ric analysis is to develop an analysis framework,
employing the same basis functions as used in
the CAD systems and thus embedding the exact
geometry.
A continuum-based shape optimization method
using the isogeometric approach is developed
for elasticity problems. To obtain efficient and
precise adjoint shape sensitivity, correct normal
and curvature information should be taken into
account in shape sensitivity expressions, espe-
cially for design-dependent problems. However, in
conventional finite element methods using linear
interpolation functions, the normal and curvature
information is generally inaccurate or missing due
to lack of inter-element continuity of design space.
In the isogeometric approach, the basis func-
tions generated from NURBS are directly used to
construct an exact geometric model in response
design changes are easily implemented within the
isogeometric framework, which maintains exact
geometry without subsequent communication
with a CAD description. In addition to the ben-
efits of isogeometric analysis, the isogeometric
design sensitivity analysis has the following advan-
tages: Firstly, it provides more accurate sensitiv-
ity of complex geometries including higher order
effects such as normal and curvature information.
Secondly, it vastly simplify the design modifica-
tion of complex geometries without communicate
with the CAD geometry during optimization proc-
ess. Since the NURBS basic functions are used in
the isogeometric response and sensitivity analyses,
design modifications are easily obtainable using
the adjustment of control points.
2 ISOGEOMETRIC ANALYSIS
2.1 NURBS basis function
Since the details about an isogeometric analysis
method are described in many former researches,
a brief overview will be given in this section. In the
one-dimensional case, a knot vector Ξ is given as
Ξ = {ξ1, ξ2, …, ξn+p+1}, (1)
which is the set of the knots ξi in a parametric
space. p is the order of the basis function and

643

MARSTRUCT.indb 643 2/18/2011 5:55:29 PM

 n is the number of control points. Then, the total Body force f and surface traction t are applied
number of knots in the knot vector Ξ is n + p + 1. in the domain Ω and on the traction boundary Γt
From this knot vector, B-spline basis functions are respectively, where c, z, and bar z represent material
defined using the recursive formula as tensor, displacement, and virtual displacement.
Using an isoparametric mapping, geometric
Ni ,0 (ξ ) = {1 if ξi ξ < ξi +1
0 otherwise
(2)
point and response are expressed, in terms of
NURBS baiss functions and control points, as

and x(Ξ) = ΣWI (Ξ)BI (10)

ξ ξi ξi ξ and
Ni p (ξ ) Ni p −1 (ξ ) + p +1
ξi p ξi ξi p +1 ξi +1 z(Ξ) = ΣWI (Ξ)zI. (11)
× Ni 1 p −1 (ξ ) , p = 1, 2, 3, … (3)
Note that the NURBS basis functions are not
interpolatory. The variational equation (7) can be
Next, a B-spline curve is obtained from the written as
linear combination of B-spline basis functions
and corresponding control points, which are the
∫Ω ∑ J ∇ΞWJ
T
coefficients for each B-spline basis. For the given a( , ) : C : z I ∇TΞWI Ω d ΩΞ (12)
Ξ
n pairs of p-th order B-spline basis functions Ni,p I ,J

defined by Eqs. (2) and (3) and corresponding

control point coordinates Bi , the B-spline curve is and
defined as
∫Ω ∑ zJ WJ b JΩ d ΩΞ
−1
( ) T
n Ξ

∑ Ni p (ξ )Bi .
J
C (ξ ) (4)
∫Γ ∑ zJ WJ JΓ d Γ Ξ.
^ −1
i =1 + N
T
(13)
Ξ
J
Similarly, the B-spline surfaces and solids are
defined as a tensor product of coordinates as Note that the domain and boundary are still in
a continuum form and constructed by the com-
n m
bination of NURBS basis functions and control
S (ξ η ) ∑∑ i p (ξ M j q η )Bi j , (5) points.
i =1 j =1

n m l
S (ξ η,ζ ) ∑ ∑ ∑ Ni p ξ )M j q (η Lk ,rr ζ )Bi j ,k, (6) 3 ISOGEOMETRIC SHAPE
i =1 j =1 k =1
OPTIMIZATION FOR
respectively, where Ni,p(ξ), Mj,q(η), and Lk,r(ζ) are PRESSURE LOADING
B-spline basis functions in the ξ, η, and ζ coordi-
nates, respectively, with given knot vectors Ξ = {ξ1, In the gradient-based shape design optimization
ξ2, …, ξn+p+1}, Η = {η1, η2, …, ηm+q+1} and Ζ = {ζ1, problem, the exact evaluation of shape design
ζ2, …, ζ l+r+1}. sensitivity is very important, since it determines
one-dimensional search direction during optimiza-
tion. Therefore, inaccurate sensitivity information
2.2 Elasticity problems causes slow convergence rate or even yields the
Using the principle of virtual work, an equilibrium completely wrong shape.
equation for elasticity problems is expressed as For the fixed loading case, the shape design sen-
sitivity equation is derived as
a( , ) ( )), ∀ Z, (7)

where the bilinear strain energy and linear load ∫Ω cijkl zi, j zk ,l d Ω
forms are defined, respectively, as
= ∫ cijkl zi ,mVm, j zk ,l d ∫Ω cijkl zi, j zk ,mVm,l d Ω
Ω
a( , z ) ∫Ω cijkl zi, j zk ,l d Ω (8) −∫
Ω
Ω ∫ fi ,m ziVm d Ω
Ω
and + ∫ fi zV V
i m,m d Ω + ∫Γ i,m i md Γ
t z
zVVt
Ω
( ) = ∫ + ∫ t ti zi d Γ. (9) + ∫ κ (ti zi )Vm nm d Γ, (14)
Ω Γ Γ t

644

MARSTRUCT.indb 644 2/18/2011 5:55:29 PM

 where ż, V, n, and κ are displacement sensitivity,
design velocity field, normal vector, and curvature,
respectively. On the other hand, if we consider the
pressure loading case, i.e. ti pni , the sensitivity
equation is modified as

∫Ω cijkl zi, j zk ,l d Ω
= ∫ cijkl zi ,mVm, j zk ,l d ∫Ω cijkl zi, j zk ,mVm,l d Ω
Ω

− ∫ cijkl zi , j zk ,lVm,m d Ω
Ω

+∫ Ω ∫ fi zzV iVm ,m d Ω
Ω Ω

+ ∫ ( p, jV j ni zi + pV
t
V j j ni zi − pVi j n j zi )d Ω . (15)
Γ

As we can see in the last integral term in Eq. (15), Figure 2. Non-normal design perturbation.
the exact evaluation of normal vector is very cru-
cial for the precise sensitivity calculation. In the
isogeometric-based shape sensitivity analysis, nor- Table 1. Shape design sensitivity for non-normal design
mal vectors can be precisely calculated from the perturbation.
geometrically exact geometry using NURBS basis DOF FDM (A) % (B) %
functions. The limitation of Eq. (14) is that it can
only be applied in normal perturbation case, which 1x 2.935E-5 7.839E-6 374.495 2.938E-5 99.895
preserves normal vectors along the design boundary. 2x 1.283E-5 1.721E-5 74.575 1.284E-5 99.971
However, in a real engineering case, loading direc- 3x 1.140E-5 8.634E-6 132.037 1.140E-5 99.969
tions can be changed according to the design modi- 4x 1.866E-5 5.778E-5 32.304 1.868E-5 99.924
fication. In this sense, Eq. (15) can be useful in the 5x 8.162E-6 1.681E-6 485.470 8.181E-6 99.769
design-dependent pressure loading problems. 6x 1.409E-5 1.995E-5 70.629 1.411E-5 99.870
In order to verify shape design sensitivity
expressions for pressure loading cases, the quarter
model of a circular pipe is introduced as shown in
In order to generate non-normal shape design
Figure 1. Internal pressure of P is applied inside
velocity field, only three control points on the
the inner hole, and symmetric boundary condition
oblique line are perturbed as shown in Figure 2.
is given. Due to the symmetry, only the displace-
When pressure loading is applied along the curved
ment sensitivities in x-direction for control points
boundary, the direction of loading is simultane-
1 to 6 are considered. For the sensitivity verifica-
ously changed under this non-normal perturbation.
tion, analytic sensitivities from Eqs. (14) and (15)
In Table 1. analytic sensitivities are also compared
are compared with finite difference sensitivities.
with finite difference sensitivity. In the third column,
analytic sensitivity for a fixed traction loading case
is obtained, which shows large disagreement with
finite difference as shown in the fourth column.
This is because Eq. (14) does not express the direc-
tion change of loading. However, analytic sensi-
tivity for a pressure loading case still shows very
good agreement compared with finite difference
sensitivity as shown in the last column.

4 NUMERICAL EXAMPLE

For the isogeometric analysis, a rectangular structure

which left boundary is subjected to non-uniform
pressure load is considered. The model consists of
231 control points (262 DOF) and the order of a
basis function is quadratic. The plane stress analy-
sis is utilized with a thickness of 1 mm. Material
Figure 1. Shape sensitivity models. properties are given as: E = 207.4 Gpa, ν = 0.3.

645

MARSTRUCT.indb 645 2/18/2011 5:55:34 PM

 The bottom of plate is fixed and the magnitude of
pressure load is decreased as y coordinate increases.
For this initial model, an isogeometric analysis is
performed, and von-Mises stress distribution is
shown Figure 3(b).
For the given rectangular model, a shape design
optimization is performed to minimized the com- (a) Initial model
pliance, satisfying the requirement that the volume
is less than or equal to the initial volume. The opti-
mization problem is stated as:

∫Ω d Ω,
T
minimize C (16)

subject to V ∫Ω d Ω ≤ Vinitial , (17)

(b) 2nd step (c) 6th step

uilower ≤ ui ≤ uiupper. (18) Figure 4. Optimization history based on FEM.

In Eq.(18), ui is the perturbation amount of each

shape design parameter form the initial shape. For a
comparison purpose, finite-element method based
optimization is performed. The shape optimization
that is based on the finite element method, an unre-
alistic irregular shape can be obtained (Braibant
and Fluery, 1984). After 8 gradient calls and 177 (a) Initial model
function calls, the optimal shape based on FEM is
obtained as shown in Figure 4. As expected, some
non-smooth and wiggly boundary at the 2nd step
of the optimization history is observed.
The result of shape optimization based on iso-
geometric method is presented in Figure 5. After
6 gradient calls and 122 function calls, the optimal
shape is obtained. The comparison between FEM
(b) 3rd step (c) 6th step
based case and isogeometric based case is pre-
sented in Figure 5. Figure 5. Optimization history based on isogeometric
The reduction of computation cost based on method.
isogeometric is 31% of the FEM based case. Fur-
thermore, any non-smooth boundary during opti-
mization process is not detected. From above the
result, the advantage when using the isogeomeric

(a) Optimization history (b) Computation

Figure 6. Comparison between FEM and Isogeometric

based optimization results.

optimization is assured, in terms of fast conver-

gence and exact representing of loading.

(a) Model description (b) Initial analysis 5 CONCLUSIONS

Figure 3. Rectangular plate subject to non-uniform We presented some demonstrative numerical exam-
pressure load. ples for shape optimization, where the accuracy

646

MARSTRUCT.indb 646 2/18/2011 5:55:37 PM

 and efficiency of the isogeometric sensitivity is
compared with finite differencing. Also, the shape
optimization of design–dependent structures is
demonstrated to verify the applicability and effec-
tiveness of the proposed method. It turns out that
the shape optimization method yields physically
meaningful results.
ACKNOWLEDGEMENTS
This research was supported by Basic Science
Research Program through the National Research
Foundation of Korea (NRF) funded by the Ministry
of Education, Science and Technology (Grant
Number 2010-18282). The support is gratefully
acknowledged.
647
MARSTRUCT.indb 647
REFERENCES
Braibant V & Fluery C, 1984, Shape optimal design
using B-splines. Comput Methods Appl Mech Eng,
44:247–267.
Choi KK & Kim NH, 2004, Structural sensitivity anal-
ysis and optimization: Volume 1, Linear systems &
Volume 2, Nonlinear systems and applications, Springer,
New York.
Cho S & Ha S, 2009, Isogeometric shape design
optimization: exact geometry and enhanced sensitivity,
Structural and Multidisciplinary Optimization, 38(1):
53–70.
Hughes TJR, Cottrell JA & Bazilevs Y, 2005, Isogeometric
analysis: CAD, finite elements, NURBS, exact geom-
etry and mesh refinement, Computer Methods in
Applied Mechanics and Engineering, 194:4135–4195.
2/18/2011 5:55:42 PM
 Structural reliability, safety and environmental protection

MARSTRUCT.indb 649 2/18/2011 5:55:42 PM

 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Combination of primary loading effects under various wave

scatter diagrams

Nian-Zhong Chen
ABS, Houston, USA

Ge Wang
ABS, Shanghai, China

C. Guedes Soares & A.P. Teixeira

CENTEC, Instituto Superior Técnico, Lisbon, Portugal

ABSTRACT: A case study was conducted to investigate the extreme value of the combination of
Still-Water Bending Moment (SWBM) and Vertical Wave-Induced Bending Moment (VWBM).
Three wave scatter diagrams, namely ABS Base 1, Base 2, and one from IACS Rec No. 34, were used for
the prediction of the extreme value of the VWBM based on the extreme value theory and Response
Amplitude Operators (RAOs). The SWBM is modeled as an alternating renewal pulse process. The point-
crossing method, load coincidence method, and Ferry Borges-Castanheta method are utilized for the
prediction of the extreme value of the combination of SWBM and VWBM. A sample tanker under sag-
ging condition is utilized for the case study and results show that for the predicted extreme values of the
combination of SWBM and VWBM: 1) the results predicted based on the wave scatter diagram of ABS
Base 1 are higher than those predicted based on ABS Base 2 or IACS; 2) the results based on the wave
scatter diagrams of ABS Base 2 and IACS are very close; 3) the point-crossing method and load coinci-
dence method yield almost identical results; 4) The Ferry Borges-Castanheta method yields higher values
than those predicted by the point-crossing method or the load coincidence method; 5) the highest value
with a design period less than 20 years is 93.3% of the total design bending moment, which is predicted
by the Ferry Borges-Castanheta method based on the wave scatter diagram of ABS Base 1.

1 INTRODUCTION Still-Water Bending Moment (SWBM) and Vertical

There is a continuing and pressing need to develop
practical solutions and engineering tools that rea-
sonably qualify the structural condition of a ship,
taking into account the structural degradation and
the cumulative loading effects over the life of the
vessel. Risk and reliability-based approaches hold
the promise of being practical methods for evalu-
ating hull structural integrity and on which to
base a Hull Integrity Management (HIM) program.
The Structural Reliability Approach (SRA) has
also demonstrated that it has the potential to take
into account many of the uncertainties associated
with structural degradation (Chen and Wang,
2009) and loading effects. Once established, such
risk and reliability-based tools could contribute to
the implementation of a more effective mainte-
nance strategy by a ship owner or operator.
A key input to SRA is the extreme value of the
combination of primary loading effects, such as
Wave-Induced Bending Moment (VWBM). Due to
the fact that SWBM and VWBM are two different
stochastic load processes that vary with time, it is
practically impossible for both maxima of SWBM
and VWBM to happen simultaneously.
To investigate the extreme value of the combina-
tion of SWBM and VWBM, Söding (1979) modeled
the SWBM as a normal variable and the VWBM
an exponential variable and then combined the
two variables. This approach is in general a simpli-
fied way to model SWBM and VWBM that are
physically two stochastic processes. Guedes Soares
(1992) investigated the combination of SWBM
and VWBM using the Ferry Borges-Castanheta
method (Ferry Borges and Castanheta, 1971) and
point-crossing method (Larrabee, 1981). In his
work, the SWBM is modeled as an alternating
renewal pulse process (Guedes Soares, 1990a, b) and
the long-term VWBM is assumed to follow a Weibull
distribution (Guedes Soares and Moan, 1991).

651

MARSTRUCT.indb 651 2/18/2011 5:55:43 PM

 Wang and Moan (1996) conducted a comparison
of different methods for predicting the load com-
bination between SWBM and VWBM using vari-
ous probabilistic models for SWBM and VWBM.
Recently, Huang and Moan (2005, 2008) proposed
an approach of load combination prediction for
SWBM and VWBM in which SWBM in different
voyages and the time variation of VWBM are mod-
eled as Poisson square wave models.
In this paper, a case study is conducted to inves-
tigate the extreme value of the combination of
SWBM and VWBM. In the work, the SWBM is
modeled as an alternating renewal pulse process.
The extreme value of VWBM is calculated based
on the extreme value theory, RAOs, and three
wave scatter diagrams, namely ABS Base 1, ABS
Base 2, and IACS Rec No. 34. The point-crossing
method, load coincidence method (Wen, 1977),
and Ferry Borges-Castanheta method are used for
prediction of the extreme value of the combination
of SWBM and VWBM.
3 VERTICAL WAVE-INDUCED
BENDING MOMENT
The VWBM is usually calculated by means of:
1) Rule value, which is normally specified by
IACS; or 2) direct calculation based on the wave
scatter diagrams and RAOs. In this paper, the
extreme value of VWBM is calculated according to
the extreme value theory, RAOs and wave scatter
diagrams.
3.1 Probability distribution function of maxima
with positive value
If a random process X(t) of dynamic responses
of a ship is stationary and Gaussian with zero
mean and the variable of the maxima with positive
values of the process is represented as Ε, the prob-
ability density function and the cumulative distri-
bution function of the non-dimensional variable
H E / m0 can be given by (Ochi, 1973):

2 STILL-WATER BENDING MOMENT 2 ⎧⎪ ε ⎛ η2 ⎞

f( )= ⎨ p⎜ − 2⎟
exp
1 + 1 − ε ⎪⎩ 2π
2 ⎝ 2ε ⎠
Normally, the SWBM can be calculated according
to the relevant loading cases. The maximum value ⎛ η ⎞⎡
2 ⎛ 1 − ε 2 ⎞ ⎤ ⎫⎪
of SWBM calculated from all the loading cases is + 1 − ε 2η exp ⎜ − ⎟ ⎢1 − Φ ⎜ − η⎟ ⎥ ⎬
⎝ 2 ⎠⎢ ⎝ ε ⎠ ⎥⎦ ⎪⎭
usually regarded as the extreme value of SWBM in ⎣
the lifetime of a ship. (3)
Previous studies (Guedes Soares, 1988) have
shown that the normal distribution might be
appropriate to represent the statistical variability
F (η ) =
1+ 1− ε
2⎧
⎩ 2 (
⎨− 1 − 1 − ε
2
)
2+Φ
⎛η⎞
⎝ε⎠
⎡ ⎫
1 − ε 2 ⎞ ⎤⎪
of the SWBM. Therefore, the characteristic value ⎛ η ⎞2 ⎛
or the probable extreme value of SWBM during − 1 − ε 2 exp ⎜ − ⎟ ⎢1 − Φ ⎜ − η⎟ ⎥ ⎬
N voyages in a reference period T might be defined ⎝ 2 ⎠⎢ ⎝ ε ⎠ ⎥⎦ ⎪⎭
⎣
by (Moan, 2006):
(4)
where ε is the bandwidth parameter of the spec-
M sw,T sw sw
1
( /N ) (1) trum defined by Cartwright and Longuet-Higgins
(1956), which is given by:
where μsw and σsw are the mean value and the stand-
ard deviation of SWBM. Φ(.) is the standard nor- m22
ε = 1− (5)
mal cumulative distribution function. m0 m4
To account for the time dependency of SWBM,
an alternating renewal process is used herein. where
In the process, the time a ship spent during voy-
ages and in port is modeled as two pulse processes, ∞
∫ 0 S ( | Hs Tz ,θ )Ld (ω | )dω
2
respectively. The rate of occurrence of load pulses m0
in a time unit, that is, the arrival rate of the process, ∞
m2 ∫ ω 2S ( | H s Tz ,θ )L )L2d (ω | )dω (6)
is given by: 0
∞
m4 = ∫ ω 4S (ω | H s Tz ,θ )L2d (ω | θ )dω
1 0
λ= (2)
μ1 + μ2 where
ω – frequency
where μ1 and μ2 are the mean values of the duration S – wave spectrum
of the voyages and the time in port, respectively. Ld – dynamic response of the ship

652

MARSTRUCT.indb 652 2/18/2011 5:55:43 PM

 Hs – significant height for the sea state 4 LOADING COMBINATION
Tz – zero up-crossing wave period for the sea state
θ – wave heading angle The point-crossing method, load coincidence
method, and Ferry Borges-Castanheta method are
used for predicting the combined extreme value of
3.2 Most probable extreme value SWBM and VWBM.
Let (η1, η2, …,ηn) be a random sample of size n of H.
Then, let (ζ1, ζ2, …, ζn) be an ordered sample of ηi 4.1 Point-crossing method
with ζ1 < ζ2 < ⋅⋅⋅ < ζn. The probability density func-
tion of ζn is thus given by (Ochi, 1981): For a scalar stationary process X(t), if the level a is
not time-dependent, the upcrossing rate can be
g (ζ n ) nnff (ζ n ) ⎡⎣ F (ζ n )⎤⎦
n 1
n− given by Rice’s formula:
(7)
+∞
v + (a ) = ∫ x fXX (a x ) dx (10)
The most probable extreme value can thus be 0
obtained by letting the derivative of g(ζn ) with
respect to ζn be zero: where fXX x, x ) is the joint probability density
function of X(t) and X (t ).
If X(t) = X1(t) + X2(t), and X1(t) and X2(t)
f ′ (ζ n ) F (ζ n ) + (nn ⎡⎣ f (ζ n )⎤⎦ = 0
2
(8) are mutually independent continuous processes,
it follows that:
+∞ +∞
3.3 Wave spectrum fXX (a,x
x) ∫−∞ ∫−∞ fX X ( x , x )
1 1
The Bretschneider spectrum is used herein for the
prediction of the extreme value of VWBM, that is: × fX
2X 2
(a − x1, x − x1 ) dx1dx1 (11)
Then,
5H s2ω p4 ⎡ 5 ⎛ω ⎞4⎤
exp ⎢ − ⎜ ⎟ ⎥
p
S( ) = (9) +∞ +∞ +∞
16ω 5 ⎢⎣ 4 ⎝ ω ⎠ ⎥⎦ v + (a ) = ∫
x1 =−∞ ∫x =−∞ ∫ x
1 2
x f 
x1 1 X1X1
(x x )
× fX
2X 2
(a − x1 x2 ) ddx2dx1dx1
where ωp is the modal (peak) frequency corre- +∞ +∞ +∞
sponding to the highest peak of the spectrum. +∫ ∫
x1 =−∞ x1 =−∞ x 2 ∫ x f 
x1 2 X1X1
( x1 x1 )
3.4 Wave scatter diagrams
× fX
2X 2
 ( − ) dx2dx1dx1 (12)

Three wave scatter diagrams are utilized herein for The integral shown in Eq.(12) is usually not easy
prediction of the VWBM. Two of the wave scatter to calculate analytically. However, the upper bound
diagrams are from ABS, namely ABS Base 1 and of v+(a) can be derived as (Larrabee, 1981):
ABS Base 2 (ABS, 2010). Another one is from IACS
Rec No. 34 (2010), which is called ‘IACS’ herein. ∞ +
The wave data from ABS Base 1 and Base 2 v + (a ) ≤ ∫ v ( y ) fX 2 a − y dy
d
−∞ 1
are determined based on the unrestricted service ∞ +
wave environment. ABS Base 1 is represented by +∫ v ( y ) fX1 a − y ddy (13)
−∞ 2
the North Atlantic wave data developed by Walden
(1964) for site-specific analysis. ABS Base 2 is rep- where vi+ is the upcrossing rate of the process
resented by the North Atlantic route wave data X i(t). fX i (.) is the probability density function
from ABS wave database for route-specific analysis. of Xi(t).
The wave data from IACS Rec No. 34 is represented If X1(t) or X2(t) has a discrete distribution, this
by the North Atlantic data derived from BMT’s upper bound formula provides an exact solution.
global wave statistics. In general, the solution of the upper bound formula
is exact, if:
3.5 Vertical wave-induced bending moment P [ X i (t ) 0 d X j (t ) < 0 ] = 0 (14)
Based on the sea states provided in the wave scatter
diagrams and the dynamic responses of the ship
4.2 Load coincidence method
from RAOs, the extreme value of VWBM is then
calculated according to the formulae of the extreme For two non-negative impulse type processes X1(t)
value given in Section 3.2. and X2(t) with the pluses returning to zero, Wen (1977)

653

MARSTRUCT.indb 653 2/18/2011 5:55:45 PM

 proposed an approximate formula for prediction
of the upcrossing rate of the X(t) = X1(t) + X2(t) as
follows:
v + (a ) ≈ τ m (a ) + τ m 2 χ 2 (a )
+ τ m m 2 ( μ1 + μ2 ) χ12 (a )
where Msw,T , Mw,T, Mc,T are the characteristic values
of SWBM, VWBM, and combined extreme bend-
ing moment with a time period T, respectively.
However, to compare the Mc,T and the total
design bending moment Md, a factor is introduced
(15) herein as:

where τmi is the mean pulse arrival rate of the proc-
ess Xi(t), and:
χ i (a ) 1 FX i (a )
a 5
χ12 ( ) χ 2 (a ) + ∫ χ1(a u ) fX 2 u )du
0
where fX i (.) and FX i (.) are the probability density
function and the cumulative distribution function
of Xi(t).
Studies of Wen (1977, 1990) show that the load
coincidence method usually provides conservative
results and the method can be applied for the com-
bination of Poisson processes, intermittent proc-
esses, and pulse and intermittent process, etc.
4.3 Ferry Borges–Castanheta model
Ferry Borges and Castanheta (1971) proposed a
simple but practically accurate load combination
model in which the combined stochastic process is
represented by a sequence of rectangular pulses of
fixed duration and the intensities of pulses are
modeled as random variables.
Therefore, for a combined process X(t) = X1(t) +
X2(t), if there are m occurrences of process X2(t)
during each occurrence of X1(t), the cumulative
distribution function of the maximum combined
value of X(t) when there are n occurrences of X1(t)
during the time period T is given by:
Mc,T
ψ= (19)
Md
CASE STUDY
(16)
A sample tanker is utilized for this case study.
In this study, only the sagging condition is consid-
ered and the SWBM is modeled as an alternating
renewal pulse process in which the average dura-
tion of one voyage and the average time in port of
the tanker are assumed to be 12 and 1.8 days,
respectively.
The VWBM is calculated based on wave scatter
diagrams, RAOs and the extreme value theory.
There are three wave scatter diagrams utilized for
the calculation. Two are from ABS (Base 1 and
Base 2) and the other is from IACS Rec No. 34.
The point-crossing method, load coincidence
method, and Ferry Borges-Castanheta method are
used for prediction of the extreme value of the
combination of SWBM and VWBM.
The principal dimensions of the tanker and
the stochastic properties of SWBM and VWBM
are shown in Tables 1–3, respectively. Note that the
mean value of the SWBM in Table 2 is normalized
by the total design bending moment Md, which is
defined as 100 units.

{ }
m n
⎡ FX ( x − x ) ⎤ fX ( x ) ddx
x Table 1. The principal dimensions of the tanker.
Fmax,T ( x ) = ⎣ ⎦ (17)
Nomenclature Value
The Ferry Borges–Castanheta model has been Length btw perpendiculars (m) 230.0
shown to be a sufficiently good estimate of the Beam (m) 32.3
cumulative distribution function of a long-duration Depth (m) 18.3
combined stochastic load process (Turkstra and
Madsen, 1980).
Table 2. Stochastic properties of SWBM.
4.4 Load combination factors
The load combination factors for SWBM and Distribution type Normal
VWBM are usually defined as follows:
Mean value 23.4
Mc,T Mw,T Coefficient of variation 0.3
ψ sw =
M ssw,T
Mc,T M sw s ,T
ψw = (18) Table 3. Stochastic properties of VWBM.
Mw,T
Mc,T Distribution type Weibull
ψc = Shape parameter 0.809
M sw,T Mw,T

654

MARSTRUCT.indb 654 2/18/2011 5:55:49 PM

 The calculated results are shown in Figures 1 (a)
and 2, and Table 4. Figure 1 shows the relationship Wave scatter diagram: ABS Base 1
between the ratio ψ and the design period T for 0.95
wave scatter diagrams, in which ψ = Mc,T /Md where
Mc,T is the combined extreme value of SWBM and
0.9
VWBM with a design period T and Md is the total
ψ

(a) 0.85
Point Crossing Method
0.95
0.8 Ferry Borges - Castanheta Model
Point Crossing Method
0.9 Load Coincidence Method

ψ 0.75
0 5 10 15 20
0.85
Design period (year)
(b)

0.8 Wave scatter diagram: ABS Base 2

Wave scatter diagram: ABS Base 1 0.9
Wave scatter diagram: ABS Base 2
Wave scatter diagram: IACS
0.75
0 5 10 15 20
Design period (year) 0.85
ψ
(b)
Load Coincidence Method
0.95
0.8
Ferry Borges - Castanheta Model
Point Crossing Method
0.9 Load Coincidence Method
ψ 0.75
0 5 10 15 20
0.85 Design period (year)
(c)
0.8 Wave scatter diagram: IACS
Wave scatter diagram: ABS Base 1 0.9
Wave scatter diagram: ABS Base 2
Wave scatter diagram: IACS
0.75
0 5 10 15 20
0.85
Design period (year)
ψ
(c)
Ferry Borges - Castanheta Model
0.95 0.8
Ferry Borges - Castanheta Model
Point Crossing Method
0.9 Load Coincidence Method
0.75
0 5 10 15 20
Ψ 0.85 Design period (year)

Figure 2. Relationship between the ratio ψ predicted

0.8
Wave scatter diagram: ABS Base 1 by different load combination methods as a function
Wave scatter diagram: ABS Base 2 of design period T (ψ = Mc,T /Md. Mc,T —the combined
Wave scatter diagram: IACS extreme value of SWBM and VWBM with a design
0.75
0 5 10 15 20
period T; Md —the total design bending moment).
Design period (year)

Figure 1. Relationship between the ratio ψ and the
design period T for wave scatter diagrams (ψ = Mc,T/Md.
Mc,T —the combined extreme value of SWBM and VWBM
with a design period T; Md —the total design bending
moment).
design bending moment. Figure 2 shows the rela-
tionship between the ratio ψ predicted using differ-
ent load combination methods as a function of
design period T. Table 4 shows the ratios ψ with a
design period of 20 years.

655

MARSTRUCT.indb 655 2/18/2011 5:55:52 PM

 Table 4. The ratio ψ with a design period of 20 years.
Wave scatter diagrams
ABS
Base 1 Base 2 IACS
Ferry Borges 0.933 0.894 0.888
and Castanheta
model
Point-crossing 0.909 0.874 0.869
method
Load coincidence 0.909 0.874 0.869
method
ψ = Mc,T /Md ⋅ Mc,T —the combined extreme value of
SWBM and VWBM with a design period T; Md —the
total design bending moment.
It can be seen from Figure 1 that: 1) the predicted
extreme values of the combination of SWBM and
VWBM based on the wave scatter diagram of ABS
Base 1 are higher than those predicted based on
the scatter diagram of ABS Base 2 or IACS; 2) the
results based on the wave scatter diagram of ABS
Base 2 and IACS are very close.
Figure 2 shows that: 1) the point-crossing
method and the load coincidence method yield
almost identical extreme values of the combination
of SWBM and VWBM; 2) the Ferry Borges-
Castanheta method yields higher extreme values of
the combination of SWBM and VWBM than
those predicted by the point-crossing method or
the load coincidence method.
Both Figure 1 and Figure 2 shows that the pre-
dicted extreme values of the combination of
SWBM and VWBM with a design period less than
20 years are all lower than the total design bending
moment. As seen in Table 4, the highest ψ is 0.933,
which is predicted based on the Ferry Borges-
Castanheta method and the wave scatter diagram
of ABS Base 1.
6 CONCLUSIONS
A case study to investigate the extreme value of the
combination of SWBM and VWBM was per-
formed. Three wave scatter diagrams, namely ABS
Base 1, ABS Base 2, and one from IACS Rec
No. 34, were used for the prediction of the extreme
value of the VWBM based on the extreme value
theory and RAOs. The point-crossing method,
load coincidence method, and Ferry Borges-
Castanheta method were utilized for the prediction
of the extreme value of the combination of SWBM
and VWBM.
656
MARSTRUCT.indb 656
For the sample tanker, the case study shows:
• The predicted extreme values of the combina-
tion of SWBM and VWBM based on the wave
scatter diagram of ABS Base 1 are higher than
those predicted based on the wave scatter dia-
gram of ABS Base 2 or IACS.
• The predictions based on the wave scatter
diagrams of ABS Base 2 and IACS are very close.
• The point-crossing method and the load coinci-
dence method yield almost identical extreme val-
ues of the combination of SWBM and VWBM.
• Ferry Borges-Castanheta method yields higher
extreme values of the combination of SWBM
and VWBM than those predicted by the point-
crossing method or the load coincidence method.
• The predicted extreme values of the combination
of SWBM and VWBM with a design period less
than 20 years are all lower than the total design
bending moment. The highest predicted value is
93.3% of the total design bending moment.
ACKNOWLEDGEMENTS
Many thanks to Dr. Lyuben D Ivanov for his valu-
able discussion on the load combination methods
and Dr. Roger Basu and Mr. Jim Speed for improv-
ing the manuscript. The views expressed in the
paper are those of the authors and do not neces-
sarily reflect those of ABS.
REFERENCES
Cartwright, D.E. & Longuet-Higgins, M.S. 1956. The sta-
tistical distribution of the maxima of a random func-
tion. Proceedings of the Royal Society of London.
Series A, Mathematical and Physical Sciences
237:212–232.
Chen, N.Z. & Wang, G. 2009. Fracture Mechanics
and Reliability Based Inspection Planning for Ship
Structures”, The 10th International Conference on
Structural Safety and Reliability, Osaka, Japan,
pp. 2017–2024.
Ferry-Borges, J. & Castanheta, M. 1971. Structural safety.
Laboratoria Nacional de Engenhera Civil, Lisbon,
Portugal.
Guedes Soares, C. 1990a. Stochastic Modelling of Maxi-
mum Still-Water Load Effects in Ship Structures.
Journal of Ship Research 34(3):199–205.
Guedes Soares, C. 1990b. Stochastic models of load
effects for the primary ship structure. Structural Safety
8:353–368.
Guedes Soares, C. 1992. Combination of primary load
effects in ship structures. Probabilistic Eng Mech 7:
103–111.
Guedes Soares, C. & Moan, T. 1988. Statistical analysis
of still water load effects in ship structures. SNAME
96:129–156.
2/18/2011 5:55:54 PM
 Guedes Soares, C., & Moan, T. 1991. Model Uncertainty
in the Long Term Distribution of Wave Induced
Bending Moments for Fatigue Design of Ship Struc-
tures. Marine Structures 4:295–315.
Huang, W., & Moan, T. 2005. Combination of global
still-water and wave load effects for reliability-based
design of floating production, storage and offloading
(FPSO) vessels. Appl Ocean Res 27:127–141.
Huang, W., & Moan, T. 2008. Analytical method of
combing global longitudinal loads for ocean-
going ships. Probabilistic Engineering Mechanics 23:
64–73.
IACS. 2010. Common Structural Rules. International
Association of Classification Societies, www.iacs.org.
Larrabee, R.D., & Cornell, C.A. 1981. Combination of
various load processes. J Struct Div ASCE 107:223–238.
Moan, T., Shu, Z., Drummen, I., & Amlashi, H. 2006.
Comparative reliability analysis of ships—considering dif-
ferent ship types and the effect of ship operations on
loads. In: SNAME Annual Meeting, Ft. Lauderdale, FL.
657
MARSTRUCT.indb 657
Ochi, M.K. 1973. On prediction of extreme values.
Journal of Ship Research 17:29–37.
Rules for Building and Classing Steel Vessels. 2010.
American Bureau of Shipping.
Soding, H. 1979. The prediction of still-water wave bending
moments in containerships. Schiffstechnik 26:24–41.
Turkstra, C.J., & Madsen, H.O. 1980. Load combination
in codified structural design. J Struct Div Proc ASCE
106(12):2527–2543.
Walden, H. 1964. Die Eigenschaften der Meerswellen in
Nordatlantischen Ozean. Deutscher Wetterdienst,
Einzelveröffentlichungn, No. 41 (in German).
Wang, X., & Moan, T. 1996. Stochastic and deterministic
combinations of still water and wave bending moments
in ships. Marine Structure 9:787–810.
Wen, Y.K. 1977. Statistical combination of extreme loads.
J Struct Div ASCE 103:1079–1093.
Wen, Y.K. 1990. Structural load modelling and combina-
tion for performance and safety evaluation. Amsterdam:
Elsevier.
2/18/2011 5:55:55 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Uncertainty assessment of the ultimate strength of a stiffened panel

Y. Garbatov, M. Tekgoz & C. Guedes Soares

Centre for Marine Technology and Engineering (CENTEC), Technical University of Lisbon,
Instituto Superior Técnico, Lisboa, Portugal

ABSTRACT: The objective of this work is to analyze various uncertainties related with the prediction of
the ultimate strength of a stiffened panel. The effect of different structural parameters on the uncertainty
is evaluated based on Monte Carlo simulation and an ANOVA methodology is used to determine the
most relevant parameters. The ultimate strength is predicted by the finite element method and the influ-
ence of plate thickness, Young modulus, yield and ultimate tensile strain of material, the shape of the
initial geometry imperfection and slenderness ratios are accounted for.

1 INTRODUCTION last decades. Analytical formulations have been

A ship structure is basically an assembly of stiff-
ened panels and an estimation of the maximum
load carrying capacity or the ultimate strength of
these elements is of high importance for the safety
assessment and economical design of the structure.
It is well recognized that the ultimate strength is a
base for structural design (Paik et al. 2006).
Stiffened panels behaviour under the effect
of compressive loads is particularly important
because the failure is generally in an unstable
mode, which has harmful consequences from the
point of view of safety.
Stiffened panels are fundamental structural com-
ponents in marine structures. Typical examples are
the hull girder and superstructure of a ship and the
deck of offshore platforms. Parameters of major
importance for the behaviour of stiffened plates are
length/width ratio of the panel, stiffener geometry
and spacing, aspect ratio of the plate between stiff-
eners, plate slenderness, residual stresses, initial dis-
tortions, boundary conditions and type of loading.
The possible failure modes of a stiffened panel
under longitudinal compression may be classified
as plate buckling and ultimate collapse, interframe
flexural buckling of the longitudinal stiffeners and
restrained torsional buckling of stiffeners (Guedes
Soares and Søreide 1983).
However, most structures are designed to pre-
vent overall grillage buckling. For short panels,
local plate buckling may be the critical mode and
for a long panel inter-frame flexural buckling of
the stiffener with associated plate flange becomes a
potential failure mode. Panels with heavy stiffeners
will follow a column mode of collapse.
A vast amount of research has been carried
out on the behaviour of stiffened plates in the
developed mainly for design purposes. The devel-
opment of non-linear computer programs has
rendered possible in-depth studies of the effect of
different imperfections.
The ultimate strength of ship plates is very
important from the design and safety viewpoint
because the collapse loads of plates can often act
as an indicator of the ultimate strength of the
whole stiffened panel in ship structures.
Guedes Soares (1988a), and Guedes Soares
and Gordo, (1997) studied several approaches and
proposed design equations for the compressive
strength of plate elements and stiffened panels
under uniaxial loading. Expressions resulting from
the same type of approach have been derived for
plates subjected to transverse and to biaxial load-
ing by Guedes Soares and Gordo, (1996 a,b).
That type of expressions were used for the reli-
ability assessment of plates by Guedes Soares and
Silva, (1992), who compared the results with the
ones of using other formulations.
Guedes Soares and Kmiecik, (1993) have shown
that assessing the plate strength with a more accu-
rate non-linear finite element code would lead to
the same general type of results although with
different numerical values. They have adopted a
Monte Carlo simulation to represent the variabil-
ity of the plate ultimate strength as a result of the
variability of several of its governing parameters,
which was an alternative to the procedure used in
Guedes Soares (1988b). However they reproduced
the variability of the initial imperfections that
existed in a database of plate measurements.
In design and during the service life, the struc-
ture is required to have an adequate margin of
safety accounting for various uncertainties due to
natural variability, inaccuracy in procedures used

659

MARSTRUCT.indb 659 2/18/2011 5:55:55 PM

 for the evaluation and control of load effects, in
structural response and also variation in building
procedure.
The load-carrying capacity of structures is nor-
mally evaluated using simplified design formula-
tions or by using more refined computations such
as nonlinear elastic-plastic large deformation finite
element analyses with special modeling related to
geometrical and material properties, initial imper-
fection, boundary conditions, load application and
finite element mesh size.
In the present work experimental design tech-
niques are used for Monte Carlo simulations that
generate samples of random values of the various
parameters, accounting for the balance among
them. For each set of values finite element calcu-
lations are made to determine the corresponding
plate ultimate strength. The factors used in the
experimental design are plate thickness, the Young
modulus, the yield stress, the ultimate tensile strain
of material and the shape of the initial geometry
imperfection, which are the most important fac-
tors as was demonstrated in earlier studies.
Secondly an analysis of variance is applied to
study the generated dataset of ultimate strength
and to determine the most significant factors. The
analysis of means is examined by the F-ratio, which
is the ratio of between-groups variance divided by
within-groups variance. The F-ratio effectively pro-
vides an estimate of the extent to which the distri-
butions from two (or more) groups or conditions
overlap. The more the distributions overlap the less
likely it is for the means to differ and vice versa.
When the means of each group are same, the
F- ratio will be very small. As the difference between
the groups means increases, so the distribution
overlap decreases. The between-groups variance
becomes larger about the within-groups vari-
ance as it is increased the difference in the means.
The F-ratio increases as the p-value decreases and
the difference in the means increase (Montgomery,
1997).
Finally, the uncertainties induced by the input
parameters in the ultimate strength assessment are
also evaluated.
The factors that have been identified as the most
important ones are the ones that need to be included
in the design equations that will be developed.
Furthermore if an uncertainty analysis of the type
of the one done by Guedes Soares (1988b) is to be
done, the present results can be used to identify
which variables to include in the analysis.
2 FINITE ELEMENT MODELLING
A typical longitudinally stiffened panel element,
originally bounded on each end by a transverse
660
MARSTRUCT.indb 660
frame and on the sides by girders, is modelled as a
unit panel representing the whole grillage structure.
The ultimate strength of a unit panel is analyzed
based on the finite element method performed
with the commercial software ANSYS (2009),
which enables modelling of elastic plastic mate-
rial properties and large deformations. Eight-node
quadrilateral shell elements are used to model the
plate and stiffener.
The kinematic assumption was large displace-
ment and rotation but small strain. The material
modelling is assumed to be bilinear elastic-perfectly-
plastic with hardening. Both a load-displacement
control method and an automatic-displacement
control method were used in the solution scheme.
The applied load is uniaxial compression.
The initial geometry imperfection is generated
with static deformed geometry. Linear buckling
analysis or linear static analysis with prescribed
displacements was performed before the incremen-
tal load-displacement analysis.
Stiffened plates with the plate mean slenderness
equal to β = 2.14 are analyzed. The aspect ratio l/b of
the plate is 5.55. The stiffener in the stiffened plate
has cross section of a standard type ‘L’ bar. The
area of the stiffener, A is approximately 33 percent
of the area of the plate, b ⋅ t. The non-dimensional
mean slenderness ratio, λ of the stiffened plate cal-
culated with a full plate width is λ = 0.44.
For a stiffened plate of a length l, width b and
thickness t, the plate and column slenderness is cal-
culated by (Faulkner, 1975):
b σy l σy
β= and λ = (1)
t E πρ E
where σy is the yield stress, and E is the Young’s
modulus and ρ = I A is the radius of gyration
for a plate with full width. The length of the stiff-
ened panel is 4480 mm and breadth is 807 mm
respectively. The flange thickness and breadth are
15 mm and 90 mm and for the web are 11 mm and
250 mm respectively.
The boundary conditions of the stiffened plate
were modelled as simply-supported avoiding rigid
body motion. V = W = Ux = 0 were imposed at all
the nodes located at the transverse edge where the
load was applied. V = W = Ux = 0 were imposed at
all the nodes located at the transverse edge oppo-
site the one where the load was applied. V = W = 0
is assumed for all the nodes located at longitudinal
edges.
The size of mesh used is relatively coarse and
the choice of its size considers not only the objec-
tive to get the best possible solution for the ulti-
mate strength, but also to achieve relatively quick
convergence in the solution. A part of the analysis
2/18/2011 5:55:55 PM
 is using Monte Carlo simulations that need to 3.500E+06
perform relatively high number of calculations, 3.000E+06
which are very much time consuming. The balance
between the mesh size and the number of simula- 2.500E+06

Load, [N]
tions was defined to achieve acceptable output. 2.000E+06
However, due to the fact that the objective is
to analyze to the contribution of different fac- 1.500E+06

tor uncertainties in the calculation of ultimate 1.000E+06

strength it is important that all the calculations are Plate thickness=0.014 m
5.000E+05 Plate thickness=0.015 m
performed under the same conditions. Plate thickness=0.016 m
The focus here is to analyze the effect of dif- 0.000E+00
ferent parameters contributing to the ultimate 0.000 0.003 0.006 0.009 0.012
Displacement, [m]
strength response. The analysis includes the effect
of the variation of the plate thickness A, the Young
Figure 1. Effect of plate thickness on load—
modulus B, the yield stress C, the ultimate tensile displacement (E = 210 GPa, σy = 315 MPa, σT = 500 MPa,
strain of material D, and the shape of the initial w0 = 4.9 mm, m = 1 and n = 1).
geometry imperfection E. These are the variable
identified by Guedes Soares, (1988b) as the most
important ones contribution to the uncertainty of 3.500E+06

predictions. 3.000E+06
Six sets of initial geometric imperfections are
2.500E+06
defined with distorted profile in the plate and out-
of-straightness in the stiffener. Faulkner, (1975) and
Load ,[N]

2.000E+06
Smith et al. (1988) have reported that the imperfec-
1.500E+06
tions in a plate can be assumed to be proportional
to β2. They suggest that the initial deformation 1.000E+06
Young Modulus=1.80E11 Pa
w0 for the average imperfection can be calculated as 5.000E+05 Young Modulus=2.10E11 Pa
w0 = 0.1β2t. The maximum permissible camber tol- Young Modulus=2.40E11 Pa
erance, wmax, for a standard shape is usually assumed 0.000E+00
0.000 0.002 0.004 0.006 0.008 0.010 0.012
to be 0.2% of the length of the length. The initial Displacement, [m]

geometry imperfections have been modelled as:

Figure 2. Effect of Young modulus on load—
displacement (t = 15 mm, σy = 315 MPa, σT = 500 MPa,
⎛ mx ⎞ ⎛ ny ⎞
w ( x, y ) w0 sin ⎜ π i π
sin (2) w0 = 4.9 mm, m = 1 and n = 1).
⎝ l ⎟⎠ ⎜⎝ b ⎟⎠

3.500E+06
where l is the length of the panel and b is the breath
of the panel. x and y are the Cartesian coordinates 3.000E+06
of the any location over the plate. m and n are wave 2.500E+06
numbers, and w0 represents the quantity of initial
Load, [N]

2.000E+06
deflection. The shape of the initial imperfections
is defined by the coefficients m and n, where n is 1.500E+06

a constant which has a value of one and m ranges 1.000E+06 Yield Stress=3.00E+08
from one to six. Yield Stress=3.15E+08
5.000E+05
Yield Stress=3.30E+08
0.000E+00
3 ULTIMATE STRENGTH ASSESSMENT 0.000 0.002 0.004 0.006 0.008 0.010 0.012
Displacement, [m]

The ultimate strength of a structure is defined as

the point beyond which an additional increment of
applied loading cannot be supported.
The case considered is a simply supported lon-
gitudinally stiffened plate subjected to axial loads
acting at its short edges, which can represent the
deck plating of a ship under bending moments.
Each of the five parameters studied is considered
at its reference value and a lower and higher value is
chosen to asses its influence on the ultimate strength.
Figures 1 to 5 present the load-displacement curves
Figure 3. Effect of yield stress on load—displacement
(t = 15 mm, E = 210 GPa, σT = 500 MPa, w0 = 4.9 mm,
m = 1 and n = 1).
of the ultimate strength response for different val-
ues of each of the input parameters. It can be seen
that while some variations produce changes in the
ultimate strength, others do not.
The variation of the plate thickness is signifi-
cant (Fig. 3) and the effect of the shape of the

661

MARSTRUCT.indb 661 2/18/2011 5:55:55 PM

 3.50E+06
0.80
1 2 3 4 5 6
3.00E+06
1 A A R R R R

Normalized ultimate strength

0.75
2.50E+06 2 A A R R R R
0.70 3 R R A R R R
Load,[N]

2.00E+06

4 R R R A A A
1.50E+06
0.65
5 R R R A A A
1.00E+06 Ultimate tensile strain=4.800E+08
0.60
6 R R R A A A
Ultimate tensile strain=5.00E+08
5.00E+05 1 2 3 4 5 6
Ultimate tensile strain=5.600E+08 Ranked yield stress

0.00E+00
,0.000 ,0.002 ,0.004 ,0.006 ,0.008 ,0.010 ,0.012
Dsiplacement,[m]
Figure 6. Box plot and null hypothesis acceptance/
rejection of the yield stress.
Figure 4. Effect of ultimate tensile strain on load—
displacement (t = 15 mm, E = 210 GPa, σy = 315 MPa,
w0 = 4.9 mm, m = 1 and n = 1). considered include waves longer than the funda-
mental buckling mode. The plate ultimate strength
may be larger than that of a perfect one, when the
3.500E+06
initial geometric imperfection shape is much differ-
3.000E+06 ent from the original collapse mode, having thus a
2.500E+06 strengthening effect. This may be explained with
the fact that such an initial imperfection shape will
Load, [N]

2.000E+06
disturb occurrence of the plate collapse mode.
1.500E+06 However, one of the parameters most strongly
1.000E+06 m=1, n=1
influencing ultimate strength, in addition to the
m=2, n=1 ones already discussed is the slenderness (Faulkner,
5.000E+05 m=3, n=1
m=4, n=1 1975, Guedes Soares, 1988 a,b). The effect of slen-
0.000E+00 derness is significantly important for the varia-
0.000 0.003 0.006 0.009 0.012
Displacement,[m] tion of ultimate strength and for the uncertainty
induced into the structural response and is included
Figure 5. Effect of initial imperfection shape on load— into the study presented here.
displacement (t = 15 mm, E = 210 GPa, σy = 315 MPa,
σT = 500 MPa and w0 = 4.9).
4 MONTE CARLO SIMULATIONS

initial geometric imperfections is insignificant, Monte Carlo simulation (Hammersley and

because the shapes chosen do not correspond to Handscomb, 1975, Rubinstein, 1981) is used for
the most weakening ones. It is known that in other generating a set of plate parameters before evalu-
cases its effect is significant (e.g. Guedes Soares ating deterministically the ultimate strength of
and Kmiecik, 1993). each of the stiffened panels defined by the sets of
The variation of Young modulus and ultimate random inputs. This method is used because the
tensile strain has been demonstrated small effect ultimate strength analysis is complex, nonlinear
(Figs. 4 and 6) while the yield stress has a moderate and depends on more than just a couple uncertain
effect on ultimate strength (Fig. 5). parameters.
However Figures 1 to 5 show the effect of any By using random inputs, the deterministic model
individual parameter conditional to the fixed value of the ultimate strength is essentially turned into a
of other ones involve in the study. To get complete stochastic one. Monte Carlo simulation is catego-
study of the effect of different parameters chosen rized as a sampling method, because the inputs are
here, all of them are kept in the present analysis. randomly generated from probability distributions
The initial geometric imperfection pattern in the to simulate the process of sampling from an actual
longitudinal direction is a multi half-wave shape population. The probability distributions for the
and that in the transverse direction is one half- input parameters are chosen in the way that most
wave shape. The ultimate strength becomes signifi- closely matches the existing data and best repre-
cant only when the stiffened panel is subjected to sents the current state of knowledge.
the initial geometry imperfection with a half wave The procedure of Monte Carlo simulation
number larger than 5 because the aspect ratio in this corresponding to the uncertainty propagation
example is 5.5. It is known that when the number are defined in several steps: create a parametric
of the half waves is decreasing, it does not reflect model, y = f(x1, x2, …, xq); generate a set of ran-
in the ultimate strength reduction because all cases dom inputs, xi1, xi2, …, xiq; evaluate the model and

662

MARSTRUCT.indb 662 2/18/2011 5:55:57 PM

 Table 1. Statistical descriptors. used here is the Fisher’s least significant difference
(LSD) method. The Fisher LSD method compares
A B C D E F all pairs of means with the null hypotheses H0:μi = τj
for all i ≠ j using the t-statistics (Montgomery,
mm GPa MPa MPa – –
1997).
Pdf N* N* N* N* N* U** Assuming a two-sided alternative hypothesis,
Mean 15 210 315 500 0.416 the pair of means μi and μj would be declared sig-
COV 0.08 0.12 0.1 0.1 0.12 nificantly different if:
Min 1
Max 6 yi − y j > LSD
SDij ,
* ⎛1 1⎞ (4)
Normal probability density function; **Uniform probability
density function. LSD
SDij tα / , DF
FW SE ⎜ + ⎟
MS
⎝ ni n j ⎠

store the results as yi; repeat the previous two steps
for i = 1 to n; analyze the results using histograms,
summary statistics, confidence intervals, etc.
In the study presented here, uniformly distrib-
uted random numbers are used as an input for
a normal distribution function to represent the
uncertainty in a ultimate strength model originat-
ing from the uncertainty in plate thickness A, the
Young modulus B, the yield stress C, the ultimate
tensile strain of material D, and column slender-
ness E. The shape of the initial geometric imper-
fection F is modelled as a uniform distribution
function. The statistical descriptors of the param-
eters are given in Table 1.
In order to obtain an error less than 2% in the
Monte Carlo simulations, 550 samples have been
produced for each parameter involved in the ulti-
mate strength calculation.
where DFW is the within group degree of freedom,
MSE is the within group mean square and ni, nj is the
number observation in groups i and j respectively.
The results for null test for pair wise compari-
sons of the means are equal or not are given in
Figures 6 to 12 (right), where “R” is for rejection
and “A” is for acceptance of the null hypothesis
and Figures 6 to 12 (left) shown the box plot of the
studied input parameters vs. normalized ultimate
strength.
It has to be pointed out that the plate thick-
ness and plate slenderness ratio ranks are fully
rejected by the null hypothesis for any par i ≠ j
demonstrating a significant effect on ultimate
strength. Figure 6 shows a mixed behaviour of
1 2 3 4 5 6

1 A A R R R R
5 FACTORIAL ANALYSIS 2 A A A R R A
3 R A A A A A
At this point it is interesting to make comparisons of
4 R R A A A A
the means (Bailey, 2008) of the plate ultimate strength
as a function of the input random parameters the 5 R R A A A A

plate thickness A, the Young modulus B, the yield 6 R A A A A A

stress C, the ultimate tensile strain D, of material,
column slenderness E, plate slenderness ratio F and
the shape of the initial geometry imperfection G. Figure 7. Box plot and null hypothesis acceptance/
The input parameters A to F are ranked into six rejection of the ultimate tensile strain.
levels from 1 to 6 where any level range, r = i + 2, is
defined by:
0.80
1 2 3 4 5 6
Mean + iStDev < r ≤ M + (i + 1) StDev, i ∈ [−3, 2] (3) 1 A R R R R R
Normalized ultimate strength

0.75

2 R A R R R R
when i = −3 then Mean = min and when i = 2 then
Mean = max. 0.70 3 R R A R R R
A null hypothesis, H0:τ1 = τ2, H0:τ1 = τ3 etc., 4 R R R A R R
where τ1 is the parameter effect, is tested by the use
0.65
5 R R R R A R
of the analysis of variance, knowing that some of 0.60
6 R R R R R A
the parameter levels means are different. When the 1 2 3 4 5
Ranked plate thickness
6

F-test p-value is less than the limit at 0.05 level,

the null hypothesis is rejected. To study this issue Figure 8. Box plot and null hypothesis acceptance/
multiple comparisons are analyzed. The method rejection of plate thickness.

663

MARSTRUCT.indb 663 2/18/2011 5:55:58 PM

 0.80
1 2 3 4 5 6
1 A A A A A A
Normalized ultimate strength

0.75

2 A A A A A A
0.70 3 A A A A R A
4 A A A A A A
0.65
5 A A R A A A
0.60
6 A A A A A A
1 2 3 4 5 6
Ranked Young modulus

Figure 9. Box plot and null hypothesis acceptance/

rejection of the Young modulus.

0.80
1 2 3 4 5 6 Figure 13. Interaction plot.
1 A A A A R A
Normalized ultimate strength

0.75

2 A A R A R A
0.70 3 A R A A R A The statistical importance of the ultimate
4 A A A A R A tensile strain and the shape of initial geometry
0.65
5 R R R R A R imperfection in shown Figures 7 and 10 revealed
0.60
6 A A A A R A similar behaviour. Figure 9 shows that the ranks
1 2 3 4 5
Shape of initial imperfection
6
3–5 demonstrate statistical importance for the var-
iation of Young modulus on ultimate strength.
Figure 10. Box plot and null hypothesis acceptance/ Figure 11 shows that the null hypothesis fully
rejection of the shape of initial geometry imperfection. accepted any par combination of column slen-
derness ratio ranks demonstrating that the varia-
0.80
tion of this parameter, in the range that has been
1 2 3 4 5 6 defined here, statistically is not producing different
1 A A A A A A results.
Normalized ultimate strength

0.75

2 A A A A A A The main effects of different parameters are

0.70 3 A A A A A A plotted in Figure 13 where the means of the nor-
4 A A A A A A malized ultimate strength for each level are shown.
0.65
5 A A A A A A The results plotted in Figure 13 confirm the ones
0.60
6 A A A A A A already discussed above with respect to the null
1 2 3 4 5 6
Ranked column slenderness ratio hypothesis acceptance/rejection. The effect of plate
thickness and plate slenderness ratio is large. Yield
Figure 11. Box plot and null hypothesis acceptance/ stress and ultimate tensile strain reveal significant
rejection of the column slenderness ratio.
effect and the Young modulus and the initial imper-
fection shape demonstrated almost no effect.
In this analysis, for all levels of the ranked
0.80
1 2 3 4 5 6
parameters, from 1 to 6, averaged over all levels of
1 A R R R R R
ultimate tensile strain show good linear interaction
Normalized ultimate strength

0.75

2 R A R R R R between most of the pairs, but not all plots show

0.70 3 R R A R R R parallel lines demonstrating a change in the impor-
4 R R R A R R tance trend of the par.
0.65
5 R R R R A R
0.60
6 R R R R R A
1 2 4 5
Ranked plate slenderness ratio
6 6 UNCERTAINTY ASSESSMENT

Figure 12. Box plot and null hypothesis acceptance/ Uncertainty is associated with the range of values
rejection of the plate slenderness ratio. that can reasonably be attributed to the analyzed
quantity. When uncertainty is evaluated, the level
of confidence has to be indicated. It is expected
the effect of the yield stress, demonstrating that that the values of the variable actually lie within
for any rank par i ≠ j where i ∈ [1, 3] ∩ j ∈ [3, 6] the range defined by the uncertainty interval.
(due to the symmetry of the acceptance/rejection The uncertainty is also a consequence of
matrix only the upper triangular one is discussed) the unknown sign of the random effects and
the null hypothesis is rejected revealing the statisti- is therefore expressed as an interval about the
cal importance of the specified ranks. result. It is evaluated by combining a number of

664

MARSTRUCT.indb 664 2/18/2011 5:55:59 PM

 uncertainty components. The components are
quantified either by evaluation of the results of
several repeated analyses or by estimation based on 0.80 0.80

Normalized ultimate strength

Normalized ultimate strength

data from records of previous measurements. 0.75 0.75

In most cases, repeated analyses are distrib- 0.70 0.70

uted about the average in the probability distri- 0.65 0.65

bution, in which there is a greater probability the 0.60 0.60

value lies closer to the mean than to the extremes. 1 2 3 4 5 6

Ranked column slenderness ratio
1 2 3 4 5 6
Ranked plate slenderness ratio

Uncertainty components quantified by means

other than repeated analyses are also expressed as Figure 16. Marginal plot of column (left) and plate
standard deviations, although they may not always (right) slenderness ratio vs. normalized ultimate
strength.
be characterized by the normal distribution.
The marginal plots and the scatter of the ranked
plate thickness, the Young modulus, the yield stress,
the ultimate tensile strain, column and plate slender-
ness ratio and the shape of the initial geometry imper- 0.80

Normalized ultimate strength

fection vs. normalized ultimate strength are given in
Figures 14 to 17. This is used to define a nominal or 0.75

characteristic measure of capacity that corresponds 0.70

to a specified percentage below the probability curve 0.65

for the corresponding random variable. 0.60

The method of combining the uncertainty 1 2 3 4

Shape initial imperfection
5 6

components is aimed at producing a realistic

rather than pessimistic combined uncertainty. Figure 17. Marginal plot of shape imperfection vs.
This usually means working out the square root normalized ultimate strength.
of the sum of the squares of the separate com-
ponents (the root sum square method). The com-
bined standard uncertainty may be reported as it correspondingly, the higher the level of confidence
stands (the one standard deviation level), or, usu- the value lies within that interval.
ally, an expanded uncertainty is reported. This is To present the 95% confidence intervals on each
the combined standard uncertainty multiplied by individual mean is defined as:
what is known as a coverage factor. The greater
this factor the larger the uncertainty interval and, μi = μ + τi, (5) i = 1,2, …a

A point estimator of μi is μi = Yi . It may be

assumed that errors are normally distributed; each
factor average is normally distributed with a mean
0.80 0.80
μi and a variance σ2/n. Thus, if σ2 were known,
Normalized ultimate strength

Normalized ultimate strength

0.75 0.75
the normal distribution may be used to construct
0.70 0.70
a confidence interval. Using MSE as an estima-
0.65 0.65
tor of σ2. The square root of MSE is the “Pooled
0.60
1 2 3 4 5 6
0.60
1 2 3 4 5 6
StDev”, based on the confidence interval and the
Ranked plate thickness Ranked Young modulus
t-distribution, since
Figure 14. Marginal plot of plate thickness (left) and
Young modulus (right) vs. normalized ultimate strength.
T (Yi − μi ) MS
SE
n (6)

where the t-distribution has a(n − 1) degrees of

freedom.
0.80
This leads to the 100(1 − α) percent confidence
Normalized ultimate strength

0.75
interval on the mean of the ith treatment μi, which
0.70
leads to:
0.65

0.60
1 2 3 4 5 6
MS
SE
yi − tα / ≤ μi ≤ yi
Ranked yield stress

,a ( n − ) n
Figure 15. Marginal plot of yield stress (left) and (7)
MS
SE
ultimate tensile strain (right) vs. normalized ultimate + tα / ,a ( n−
n )
strength. n

665

MARSTRUCT.indb 665 2/18/2011 5:56:02 PM

 Probability plots and the calculated 95% 0.6 0.7 0.8
1 2 3
percentile and 95% confidence intervals are shown 99.9

0.7462
99
in Figures 18 to 24, where the ultimate strength is 95 95 95
90
presented as a function of different parameters
and their ranks. 50

10

0.7377

0.7384
1

Percent
0.6 0.7 0.8 0.1
4 5 6
99.9 99.9
1 2 3

0.7506
99
0.7603
99
95 95 95 95 95 95
90 90

50
50
10
10
0.7946

0.7481
1
Percent

0.7493

0.7625
0.1 1
99.9
4 5 6 0.1
0.7300

99 0.6 0.7 0.8 0.6 0.7 0.8

95 95 95
90 Normalized ultimate strength
50

10
Figure 21. Probability plot of normalized ultimate
strength conditional to ranked Young modulus.
0.7348

0.7295

1
0.1
0.6 0.7 0.8 0.6 0.7 0.8

Normalized ultimate strength

0.6 0.7 0.8
1 2 3
99.9
Figure 18. Probability plot of normalized ultimate

0.7522
99
strength conditional to ranked yield stress. 95 95 95
90

50
0.6 0.7 0.8

1 2 3 10
99.9
0.7479

0.7400
Percent
0.7332

99 1
95 95 95 0.1
90 4 5 6
99.9

0.7331
50 99
95 95 95
10 90
0.7160

0.7437

1
Percent

50
0.1
4 5 6
99.9
10
0.7524

99
0.7519

0.7421
95 95 95 1
90
0.1
50 0.6 0.7 0.8 0.6 0.7 0.8
Normalized ultimate strength
10
0.7488

0.7536

1
0.1 Figure 22. Probability plot of normalized ulti-
0.6 0.7 0.8
Normalized ultimate strength
0.6 0.7 0.8
mate strength conditional to the shape of initial
imperfections.
Figure 19. Probability plot of normalized ultimate
strength conditional to ranked ultimate tensile strain.
0.50 0.75 1.00
1 2 3
99.9
0.7541

0.6 0.7 0.8 99

1 2 3 95 95 95
99.9 90
0.6845

99
95 95 95 50
90

10
50
0.8202

0.7458
Percent

1
10
0.1
4 5 6
0.6661

0.7186

1 99.9
Percent

0.7396

0.1 99
4 5 6
99.9 95 95 95
90
0.7558

99
95 95 95
90 50

50 10
0.7394

0.7506

1
10
0.1
0.7368

0.7758

1 0.50 0.75 1.00 0.50 0.75 1.00

0.1 Normalized ultimate strength
0.6 0.7 0.8 0.6 0.7 0.8
Normalized ultimate strength

Figure 23. Probability plot of normalized ultimate

Figure 20. Probability plot of normalized ultimate strength conditional to the ranked column slenderness
strength conditional to ranked plate thickness. ratio.

666

MARSTRUCT.indb 666 2/18/2011 5:56:07 PM

 0.6 0.7 0.8
mean value, but also tells how stable the esti-
1 2 3
99.
mate mean value is and what the degree of

0.7556
99
95 95 95 uncertainties is.
90
The analysis performed here may also be used
50
to define a nominal or characteristic measure of
10 capacity that corresponds to a specified percentage
0.7770

0.7358
below the probability curve for the corresponding
Percent

1
0.1
99.9
4 5
0.6992
6
random variable. For instance, one may base char-
99
95 95 95
acteristic strength or capacity on a lower bound or
90
95% exceedence value, while a characteristic load
50 or demands may be based on an upper bound or a
10 5% exceedence value.
0.7172

1
0.1
0.6 0.7 0.8
Normalized ultimate strength
Figure 24. Probability plot of normalized ultimate
strength conditional to the ranked plate slenderness ratio.
It can be observed form Figures 18 to 24 and
Figure 2 that the standard deviations of ultimate
strength as a function of studied parameters along
ranks show relatively small variation. If the stand-
ard deviations are averaged for all ranks, the small-
est one is calculated for plate thickness followed
by plate slenderness, shape of initial imperfection,
ultimate strain, column slenderness, the Young
modulus and yield stress.
However the standard deviation shows more
than just possible range around the estimated
0.6816
0.6 0.7 0.8
7 CONCLUSIONS
This work analyzed the influence of plate thick-
ness, Young modulus, yield and ultimate tensile
strain of material, column and plate slenderness
and the shape of the initial geometric imperfection
of stiffened plate.
The analysis pointed out that the plate thickness
and plate slenderness ratio ranks are fully rejected
by the null hypothesis for any par i ≠ j demon-
strating a significant effect on ultimate strength.
A mixed behaviour of the effect of the yield stress
on ultimate strength demonstrated that for some
ranks the null hypothesis is rejected revealing
the statistical importance of the specified ranks.
The statistical importance of the ultimate tensile

Table 2. Statistical descriptors of normalized ultimate strength conditional to the ranks of different parameters.

Rank 1 2 3 4 5 6

Yield stress Mean 0.731 0.714 0.705 0.699 0.691 0.688

StDev 0.038 0.028 0.026 0.026 0.023 0.051
N 11 69 209 179 66 16
Ultimate tensile strain Mean 0.674 0.695 0.700 0.706 0.708 0.707
StDev 0.026 0.023 0.026 0.026 0.027 0.029
N 7 89 181 186 72 15
Plate thickness Mean 0.644 0.667 0.693 0.715 0.732 0.750
StDev 0.014 0.011 0.016 0.014 0.014 0.015
N 11 76 186 197 65 15
Young modulus Mean 0.696 0.702 0.699 0.704 0.708 0.706
StDev 0.026 0.027 0.024 0.028 0.026 0.034
N 15 75 183 188 77 12
Shape of initial imperfection Mean 0.705 0.710 0.700 0.706 0.690 0.703
StDev 0.026 0.026 0.024 0.028 0.026 0.024
N 89 94 83 86 93 105
Column slenderness ratio Mean 0.724 0.701 0.703 0.701 0.704 0.702
StDev 0.058 0.032 0.026 0.023 0.022 0.029
N 5 84 204 172 67 18
Plate slenderness ratio Mean 0.748 0.725 0.710 0.691 0.672 0.654
StDev 0.017 0.018 0.015 0.016 0.016 0.016
N 27 87 180 175 64 17

667

MARSTRUCT.indb 667 2/18/2011 5:56:11 PM

 strain and the shape of initial geometry imperfec-
tion revealed similar behaviour. The null hypoth-
esis fully accepted any par combination of column
slenderness ratio ranks demonstrating that the
variation of this parameter, in the range that has
been defined here, statistically is not producing dif-
ferent results.
The main effects of different parameters have
confirmed the conclusions derived based on the
null hypothesis for acceptance/rejection.
The uncertainties introduced to ultimate
strength calculation by different parameters have
been demonstrated by probability plots. The small-
est standard deviations of ultimate strength, aver-
aged for all ranks, is calculated for plate thickness
followed by plate slenderness, shape of initial
imperfection, ultimate strain, column slenderness,
the Young modulus and yield stress.
The analysis performed may be used to define a
nominal or characteristic measure of capacity that
corresponds to a specified percentage below the
probability curve for the corresponding random
variable.
ACKNOWLEDGEMENTS
The work reported here is a contribution to
the activities of the MARSTRUCT VIRTUAL
INSTITUTE, (www.marstruct-vi.com), in partic-
ular, its Technical Subcommittee 2.3 on Ultimate
Strength.
REFERENCES
ANSYS, 2009, “Online Manual”, Release 11.
Bailey, R., 2008, “Design of Comparative Experiments”,
Cambridge University Press.
Faulkner, D., 1975, “A Review of Effective Plating for
use in the Analysis of Stiffened Plating in Bending and
Compression”, Journal of Ship Research, 19, pp. 1–17.
668
MARSTRUCT.indb 668
Guedes Soares, C. & Søreide T.H., 1983. “Behaviour of
Stiffened Plates under Predominantly Compressive
Loads”, Int. l Shipbuilding Progress, 30, pp. 13–27.
Guedes Soares, C. & Silva, A.G., 1992, “Reliability of
Unstiffened Plate Elements Under In-Plane Com-
bined Loading”, Proceedings of the 11th Interna-
tional Conference on Offshore Mechanics and Arctic
Engineering (OMAE´92), ASME, pp. 265–276.
Guedes Soares, C. & Kmiecik, M., 1993, “Simulation
of the Ultimate Compressive Strength of Unstiff-
ened Rectangular Plates”, Marine Structures, 6,
pp. 553–569.
Guedes Soares, C., 1988a, “Design Equation for the
Compressive Strength of Unstiffened Plate Elements
with Initial Imperfections”, Journal of Constructional
Steel Research. 9, pp. 287–310.
Guedes Soares, C., 1988b, “Uncertainty Modelling in
Plate Buckling”, Structural Safety, 5, pp. 17–34.
Guedes Soares, C. & Gordo, J.M., 1996a, “Compressive
Strength of Rectangular Plates Under Biaxial Load
and Lateral Pressure”, Thin-Walled Structures, 24,
pp. 231–259.
Guedes Soares, C. & Gordo, J.M., 1996b, “Compres-
sive Strength of Rectangular Plates Under Transverse
Load and Lateral Pressure”, Journal of Construc-
tional Steel Research, 36, pp. 215–234.
Guedes Soares, C. & Gordo, J.M., 1997, “Design
Methods for Stiffened Plates Under Predominately
Uniaxial Compression”, Marine Structures, 10(6),
pp. 465–497.
Hammersley, J. & Handscomb, D., 1975, “Monte Carlo
Methods”, London, Methuen.
Montgomery, D., 1997, “Design and Analysis of Experi-
ments”, John Wiley & Sons Inc.
Paik, J., Brennan, F., Carlsen, C., Daley, C., Garbatov, Y.,
Ivanov, L., Rizzo, C., Simonsen, B., Yamamoto, N. &
Zhuang, H., 2006, “Condition Assessment of Aged
Ships”, 16th International Ship and Offshore
Structures Congress, Committee V. 6, University of
Southampton, 2, pp. 255–306.
Rubinstein, R.Y., 1981, “Simulation and the Monte
Carlo Method”, J Wiley & Sons, New York.
Smith, C., Davidson, P., Chapman, J. & Dowling, P., 1988,
“Strength and Stiffness of Ships’ Plating under In-plane
Compression and Tension”, Trans. RINA, 130.
2/18/2011 5:56:13 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Safety analyses for bulk carriers using metamodels of still water loads

P. Georgiev
Technical University of Varna, Bulgaria

ABSTRACT: This paper presents an approach that combines metamodeling technique and Monte-
Carlo simulation for safety analyses of still water loads for bulk carriers. The approach is applied to study
the influence of deviations from cargo loading plan on net load of double bottom and still water bending
moments. As example a Handymax BC-A type ship is used. The metamodels approximate the work of
installed on board mandatory loading instrument and give the relation between the distribution of cargo
and the trim and still water bending moments in controlled sections. The Monte-Carlo simulation uses
the fitted metamodels to obtain a rich set of statistical data that permit preparing event tree analysis for
possible overloading of double bottom and the evaluation of likelihood for exceeding of permissible
bending moments.

1 INTRODUCTION bulk carriers that are applicable for ships con-

In the last two decades the attention of IMO and
IACS has been focused on various kind of legis-
lation leading to improvement of safety of bulk
carriers.
At the beginning of 1996 came into force amend-
ments called the Enhanced Survey Programme
(ESP) in order to enhance bulk carrier inspection.
In December 2002 the Maritime Safety Committee
(MSC) adopted amendments to SOLAS chapter
XII and the 1988 Load Lines Protocol. The revi-
sions of bulk carrier safety concern, bow height
and reserve buoyancy, fore deck fittings, water level
detectors and pimping arrangements, hatch covers
and securing mechanisms, harmonized notations
and design loading conditions, means of access,
immersion suits, free-fall lifeboats, double side skin,
using of loading instrument for ships with length
over 150 m capable of providing information on
hull girder shear forces and bending moments.
Following the 1998 publication of the report
into the sinking of the bulk carrier Derbyshire,
the Maritime Safety Committee initiated a further
review of bulk carrier safety, involving the use of
Formal Safety Assessment (FSA) studies to help
assess what further changes in regulations might
be needed. FSA is a structured and systematic
methodology, aimed at enhancing maritime safety,
including protection of life, health, the marine
environment and property, by using risk analyses
and cost benefit assessment (MSC 2007). The main
steps in FSA are: identification of hazards; risk
analysis; risk control options; cost benefit assess-
ment; and recommendations for decision making.
All revised and newly developed rules are imple-
mented in Common Structural Rules (CSR) for
tracted for construction on or after 1 April 2006.
For the ships are to be assigned one of the follow-
ing additional service notations: BC-A; BC-B or
BC-C. For BC-A ship design loading conditions
shall include at least one cargo loaded condition
with specified holds empty, with cargo density
3.0 t/m3, and the same filling ratio in all loaded
cargo holds at maximum draught with all ballast
tanks empty. Additionally, CSR define procedure
for determination of hold mass curves that spec-
ify the maximum and minimum mass of cargo in
each cargo hold and in any two adjacent holds as a
function of the draught at middle position.
In the operation of ships, the BLU (Bulk Load-
ing and Unloading) Code provides guidance to
masters of bulk carriers, terminal operators and
other parties concerned with the safe handling,
loading and unloading of solid bulk cargoes. In
resent years the ship operators have experienced
increased pressure from terminals to load cargo
as quickly as possible. INTERCARGO carried
out a confidential survey of ships’ masters and
the main conclusions (MSC 2008) are: there are
terminal instructions, requiring a 14-hour turn-
round and 16,000 t/hr loading rate, for capesize
bulk carriers; the maximum loading rate should
be approximately twice the ballast pump capacity,
i.e., typically ∼10,000 t/hr; the BLU Code is not
being universally applied.
The loading condition and loading/unloading
plans are prepared by on-board loading instrument
and the results to a great extent depend on the
input information for cargo distribution between
the holds. According to the conclusions of ISSC
2006 (ISSC 2006), based on (Rizzuto 2006) for dry
bulk cargoes, no accurate and direct measurement

669

MARSTRUCT.indb 669 2/18/2011 5:56:13 PM

 of cargo level in hold is available. The information
has always been derived from the final draft surveys.
This type of data, however, does not help in find-
ing out the cargo weight distribution in the various
compartments. Even in the event of a very accurate
monitoring of drafts, it is not possible to derive the
actual distribution of cargo in every single hold.
Discrepancies from the loading plan in content of
single holds are expected to be substantial but are
not quantifiable by the crew. At the present stage of
the investigation, no model has been developed to
quantify these uncertainties for bulk carriers.
Taking into account the above, the research
questions for the investigation is: “What is the
influence of deviation from loading plan on still
water loads and are there any hazards?” At the
same time, according to FSA methodology for
hazard identification and risk analyses where data
is unavailable, calculation or simulation and analy-
sis techniques could be used. These techniques are
fault tree and event tree analysis, failure mode and
effect analysis, what if analysis, etc.
The present study combines metamodeling
technique, on-board loading instrument and
Monte-Carlo simulation for safety analyses of bulk
carriers and to answer the question. This combine
approach is proposed by Georgiev (2010) for prob-
abilistic presentation of bending moments.
As example ship a 42700 tDW Handymax bulk
carrier type BC-A with main dimensions Lpp/B/D =
177/30/16.2 m and five cargo holds is used.
2 METAMODELING TECNHIQUE
2.1 Background
Statistical techniques and design of computer
experiments are used successfully in ship design
(Georgiev & Damyanliev 2005, Georgiev 2008). The
basic approach is to construct approximation of the
analyses code (software) i.e., to construct a “model of
the model” or metamodel (Kleijnen & Sargent 2000).
The main steps of the metamodeling process are as
follows: 1) Determine the goal of the metamodel;
2) Identify the inputs ant their characteristics;
3) Specify the domain of applicability; 4) Identify
the output variable and its characteristics; 5) Specify
the metamodel; 6) Specify a computer experimen-
tal design; 7) Fit the metamodel; 8) Determine the
validity of the fitted metamodel.
The left hand side of Figure 1 presents schemat-
ically the metamodeling technique.
2.2 Metamodels of still water bending
moments and trim
The Still Water Bending Moment (SWBM) depends
on cargo distribution. The uncertainties in this
distribution lead to uncertainties in calculated by
670
MARSTRUCT.indb 670
Figure 1. Connection between metamodeling tecnique
and Monte-Carlo simulation.
the loading instrument SWBM. The example ship is
equipped with ALCOS v. 2.5 (Auto Loading Compu-
ter On-board System) loading instrument, developed
by specialist from Technical University of Varna
and in the metamodeling process it is considered as
computer simulation. The goal of the metamodels
is to obtain the relationship between the amount of
cargo in the holds and SWBM and the trim of the
ship. The metamodels will eliminate the necessity
of calculation by loading instrument of thousands
loading conditions in Monte-Carlo simulation.
For present study an alternate loaded condition
(holds No 1, 3 and 5 are loaded) with cargo density
3.0 t/m3. However, variability in bending moments
due to different cargo distribution could occur only
when the cargo holds are partially filled. The cargo
loading plan is at 100% consumables and zero trim.
The input variables for the metamodel are devi-
ations dPi, i = 1,2,3 from planned amount of cargo
in the three loaded holds. For each deviation three
values are considered: −10%, 0 and +10%. The
coded values in range (−1,1) are calculated by the
following formula:
dP
Pi
xi = , i = 1, 2, 3 (1)
10
The amount of cargo in each hold is obtained by:
Pi ( xi / 10)P
)P0i i = , 2, 3 (2)
where P0 is the planned amount of cargo.
The output variables are SWBM at different
sections related to the permissible ones and the
trim. Fitting of metamodels is based on RSM
(Response Surface Methodology). Polynomial
regression is chosen due to its transparency and
simplicity and expected low order of non-linearity
of considered responses. The response surface
( ) is presented by:
y b0 + x T b x T Bx (3)
2/18/2011 5:56:13 PM
 Table 1. Polynomial regression coefficients for studied The Monte-Carlo simulation calculates the fitted
responses. metamodels for trim and SWBM each time using
different randomly-selected values for the amount
x1 x2 x3 b of cargo in holds.
x1 −0.018583 −0.021596 −0.003580 1.368173 In reality, the process of cargo loading is moni-
x2 −0.001311 0.016679 0.251147 tored by the crew and terminal representatives. To
take into account the human factor, the follow-
x3 0.016678 −0.979926
ing constraints for the total amount of cargo and
Trim, m b0 = 0.859706
obtained trim are accepted:
x1 −0.001536 0.000809 0.003450 0.096288
x2 0.000946 −0.001600 −0.091438 3
x3 −0.003963 −0.008829 Qmin ∑ Pi ≤ Qmax tmin ≤ trim
min t iim ≤ tm ax,
max
(4)
Fr.93, − b0 = 0.758067 i =1
x1 −0.001424 0.000241 0.002706 0.125465
x2 0.000567 −0.001191 −0.164884 where Qmin is the minimum amount of cargo taken
x3 −0.002969 0.023898 as 97% from the planned one, Qmax is the maximum
Fr.129, − b0 = 0.601142 amount limited by summer draught, tmin is the max-
imum aft trim accepted as 0.005 L = 0.88 m and tmax
x1 −0.001129 −0.000310 0.001735 0.109279
is the maximum trim to bow equal to 0.04 m.
x2 0.000131 −0.000675 −0.087475
For the input variables a double truncated nor-
x3 −0.001683 0.022291
mal distribution with lower and upper truncation
Fr.165, − b0 = 0.963968
point ZL = ZU = 2, mean zero and truncated stand-
ard deviation σT = 5% is accepted. Using the tables
In this model it is assumed linear effects in x (Khasawneh et al. 2005) this corresponds to normal
and, two-factor interaction and pure quadratic distribution with mean zero and σ = 5.684%. For
term in x (the term xTBx). samples generation RiskAMP MS Excel® Add-in
One of the important steps in computer experi- was used (http://www. riskamp.com/). To investigate
ment is the selecting a design with and appropri- how accurate are the particular estimates five sam-
ate number of runs and levels for each variable to ples with 10,000, 20,000, 30,000, 40,000 and 50,000
ensure sufficient design space coverage. The present runs are generated. Additionally, the investigated
study uses uniform design proposed by Fang range of trim is subdivided into smaller ranges
(2006). The designs are marked by Un(qs) where as follows: T1: [−0.19 ÷ 0.04); T2: [−0.42 ÷ −0.19);
n = number of experiments, q = number of levels T3: [−0.65 ÷ −0.42); T4: [−0.88 ÷ −0.65). To prepare
and s = maximum number of factors. Designs for the calculations by the metamodels and to sift
different number of factors, levels and experiments out the feasible points a small computer program,
could be taken from (www.math.hkbu.edu.hk/). written in Pascal was used.
For present study U15(33) design was used. The number of feasible points at different start-
Installed on-board loading instrument ALCOS ing number of runs is shown in Table 2. The total
was used to calculate the responses. It is necessary to number of feasible points is about 25% from start-
evaluate and to record the results for only 15 load- ing number of runs.
ing conditions that take less than 15 minutes time. The distribution of cargo is shown in Figure 2.
The least squares estimates for trim and SWBM for The graph includes the histogram for T1 trim range
eight frames are made by JMP® software. Table 1 and fitted normal probability distributions for others.
presents the polynomial coefficients for trim and One can see the correct relation between the cargo
SWBM at the muddle of holds No 2, 3 and 4. mass in the three holds in bubble plot. The mass
For deterministic computer experiment where of cargo in middle hold 3 is greatest (bigger circles)
a random error does not exist, the metamodel
accuracy is checked by the modelling error that
is the discrepancy between the true output Y’ Table 2. Number of feasible points for trim ranges.
from the simulation model and Y’’ from the meta-
model. The maximum modelling error for SWBM Trim range
No of
is less than 0.2% from the permissible values and for runs T1 T2 T3 T4 Total
trim the maximum error is 17 mm (Georgiev 2010).
10 × 10 3 696 676 567 501 2440
20 × 103 1430 1347 1158 1020 4955
3 MONTE-CARLO SIMULATION 30 × 103 2197 2034 1797 1465 7493
40 × 103 2869 2674 2416 1950 9909
The connection between the Monte-Carlo simula- 50 × 103 3565 3415 3012 2397 12389
tion and fitted metamodels is shown in Figure 1.

671

MARSTRUCT.indb 671 2/18/2011 5:56:16 PM

 1000 800
-T1 -T1
Hold No 1 Hold No 3
-T2 -T2
800
-T3 600 -T3
-T4 -T4
600
400
Frequency

Frequency
400

200
200

0 0
11200 11600 12000 12400 12800 13200 12400 12800 13200 13600 14000 14400 14800 15200
Cargo mass, t Cargo mass, t
a) b)
1200
-T1
Hold No 5
1000 -T2
-T3
800 -T4

600
Frequency

400

200

0
12000 12400 12800 13200 13600 14000 14400
Cargo mass, t
c)

Figure 2. a, b, c) Cargo distribution for different trim ranges; d) Bubble plot for cargo distribution in the holds.

when the mass in hold 1 and 5 are smallest (bottom 1.0

95%CI-Fitted function
left corner) and vice versa (top right corner). 99%CI-Fitted function
wr %

An approximate (1−α)100% confidence interval 0.8

-T1 -T3
for unknown population mean m is
RelativewidthofCI,

-T2 -T4
0.6
⎛ σ ⎞ wr=19.255N-0.4917
⎜⎝ μ ± z1−α / 2 ⎟ (5) R2=0.997
N⎠ 0.4
wr=14.686N-0.4921
where μ is sample mean,σ is standard deviation, 0.2
R2=0.997

N is number of runs, and z1−α/2 is the number that 0 1000 2000 3000 4000
Φ(zγ) = γ (Φ denotes the standard normal c.d.f). It Numberofruns, N
is common practice in simulation to use and report
the absolute (wa) and relative (wr) widths of the Figure 3. Relative width for 95% and 99% confidence
confidence interval (CI), defined as for cargo mass in Hold No 3.

σ w
wa 2 z1 α /2 and wr = μa (6) Table 3. Descriptive statistics for mass of cargo in holds
N at T1 trim range.

In Figure 3 the relative width in percent (wr) for Item Hold 1 Hold 3 Hold 5
α = 5% and = 1% significance levels is shown. The
results include all trim ranges and as one can see Plan P0, t 12536 13884 13186
from the trend lines, the accuracy of estimation MaxLoad, t 12536 13884 13.835
of the mean is proportional to N−0.5. The maxi- μ, t 12446.0 13818.4 13141.5
mum relative width of CI is less than 1% and goes 95% CI, t ±9.6 ±18.3 +11.2
down to less than 0.4% at 50,000 starting runs. 99% CI, t ±12.6 ±24.0 ±14.7
The descriptive statistics for mass of cargo dis- σ, t 291.9 556.8 341.7
tribution in the holds at T1 range for sample size CoV,% 2.3 4.0 2.6
N = 3565 are shown in Table 3. The table presents Skewness −0.0001 0.0314 −0.0299
the planned and maximum permissible (Max- Ex. Kurtosis −0.1283 −0.4916 −0.2440
Load) amount of cargo.

672

MARSTRUCT.indb 672 2/18/2011 5:56:17 PM

 4 SAFETY ANALYSES
4.1 Possible overloading of double bottom
According to CSR, the maximum allowable or
minimum required cargo mass in a cargo hold, or
in two adjacently loaded holds, is related to the net
load on the double bottom. This load is a func-
tion of draft, cargo mass in the cargo hold, as well
as the mass of fuel oil and ballast water contained
in double bottom tanks. The relation between the
draft and the load is given by hold mass curves pre-
pared for every single hold and for two adjacent
holds as well. In operation the maximum allowable
cargo mass shall be limited to MHD that is the maxi-
mum cargo mass allowed to be carried with speci-
fied holds empty at maximum draught.
The event tree analysis method is used to ana-
lyse the probability of occurrence (Kuo 2007). The
events are examined using a binary logic (Yes/No)
and the process goes in one direction to derive out-
comes. The tree consists of nodes, branches and
leaves. The probability of the outcome is obtained
multiplying along the branches (Figure 4).
At present study, we consider two events. Let
event B is the overloading of double bottom and
T is the trim range. The overloading exists when
the mass of cargo is greater than maximum per-
missible from corresponding hold mass curve. The
two events are considered as not independent and
the conditional probability of B, given that T has
occurred is given by
P(B T )
P(B T ) = , P (T ) ≠ 0.
P (T )
Let the trim ranges Ti i = 1,2, …,4 are events
that partition the whole trim range T. Following
the theorem of the total probability we obtain
4 rather at trim range T2 or T3 than T4;
( )
P ( B ) = ∑ P B Ti P (Ti ), i = 1, 2, ..., 4
i =1
Figure 4. Event tree for two events A and B.
MARSTRUCT.indb 673
For conditional probability of B, given that Ti
has occurred the following is valid:
P ( B Ti )
P ( B Ti ) = , P (Ti ) ≠ 0; i = 1, 2, ..., 4 (9)
P (Ti )
In case where it is necessary to reverse the order
in conditional probabilities the Bayes’ theorem is
used:
P (Ti B ) =
(
P (Ti )P B Ti ), i = 1, 2, ..., 4 (10)
P(B )
The masses of cargo in the holds are considered
as not independent and the event of overloading
is labelled B1, B3 and B5. The outcome for every
branch is marked with a string that includes the
number of trim range and combination of three
characters (‘1’ or ‘0’) that represent the overload-
ing in holds No 1, 3 and 5 respectively. The sample
with total 12389 points is used and the event tree
is shown in Table 4. The last column includes the
values of product rules.
Having the results from the event tree we can
draw different conclusions. The probability of
overloading of every separate hold and combi-
nation of holds for whole trim range is shown in
Figure 5. The probability that there is no overload-
ing is about 31% and probability for overloading
of hold No 3 (the middle hold) is about 39%.
Conditional probabilities for overloading of
every hold separately at given trim region can be
(7) seen in Figure 6. From that figure several conclu-
sions can be drown:
− The probability to have no overloading is great-
est (42%) at trim range T3: [−0.65 ÷ −0.42) m;
− An overloading of hold No 3 (that is with great-
est probability of about 39%) could be expected
(8) − The overloading of hold No 1 is most probable
(30.1%) at trim range T1: [−0.19 ÷ 0.04) m.
The information obtained from the event tree
analyses could be useful during the operation of
the ship to stress the crew attention to most critical
places during the cargo loading and how to inter-
pret the situation in case of considerable deviation
from cargo plan. The same information could be
used during the design process to evaluate the vul-
nerability of the ship to the deviations from cargo
loading plan.
4.2 Still water bending moments
During the Monte-Carlo simulation for fea-
sible points the SWBM is calculated by fitted
metamodels. Descriptive statistics and confidence
673
2/18/2011 5:56:21 PM
 Table 4. Event tree for overloading of double bottom.

Ti B1 B3 B5 Outcome

228/1369 Y 0/228 Y 1111 0.0

228/228 N 1110 0.01840
1369/3565 Y
1141/1369 N 67/1141 Y 1101 0.00541
1074/1141 N 1011 0.08669
3565/12389
1403/2196 Y 0/1403 Y 1011 0.0
1403/1403 N 1010 0.11325
2196/3565 N
793/2196 N 0/793 Y 1001 0.0
793/793 N 1000 0.06401
37/811 Y 0/37 Y 2111 0.0
811/3415 Y 37/37 N 2110 0.00299
774/811 N 154/774 Y 2101 0.01243
3415/12389 620/774 N 2011 0.05004
1488/2604 Y 0/1488 Y 2011 0.0
1488/1488 N 2010 0.12010
2604/3415 N
1116/2604 N 0/1116 Y 2001 0.0
1116/1116 N 2000 0.09008
1/406 Y 0/1 Y 3111 0.0
406/3012 Y 1/1 N 3110 0.00008
405/406 N 273/405 Y 3101 0.02204
3012/12389 132/405 N 3011 0.01065
1306/2606 Y 13/1306 Y 3011 0.00105
1293/1306 N 3010 0.10437
2606/3012 N
1300/2606 N 36/1300 Y 3001 0.00291
1264/1300 N 3000 0.10203
0/156 Y 4111 0.0
156/2397 Y 4110 0.0
156/156 N 155/156 Y 4101 0.07199
2397/12389 1/156 N 4011 0.00046
1036/2241 Y 56/1036 Y 4011 0.00302
980/1036 N 4010 0.05292
2241/2397 N
1205/2241 N 234/1205 Y 4001 0.01264
971/1205 N 4000 0.05244

100 No H1only H3only H5only

1.2

80
39.4 43.6
6.5
Probability, %

42.9
60

4.4
40 18.2 27.4
30.1

20 42.0
32.7 27.1
22.2
0
T1 T2 T3 T4

Figure 5. Probability (%) of overloading of every cargo Figure 6. Probability for overloading of separate holds
hold and combination of holds. at given trim range T1–T4.

interval for the mean at all controlled sections for
trim range T1 are shown in Table 5. The last two
rows include values for SWBM for the planned
loading condition and location of the section. The
relative width of 95% confidence interval for three
sections and corresponding trend line are shown
in Figure 7. The least accuracy of estimation of
the mean is obtained for frame 129. The frame is
located at the middle of the ship and the standard
deviation of SWBM for this frame is greatest.

674

MARSTRUCT.indb 674 2/18/2011 5:56:23 PM

 Table 5. Descriptive statistics for SWBM at trim range T1.

Frames

Statistic Fr. 75 Fr. 86 Fr. 93 Fr.111 Fr.129 Fr.139 Fr.147 Fr.165 Fr.183

μ, − 0.5259 0.7732 0.8483 0.7473 0.5907 0.6292 0.7302 0.9520 0.6449

95% CI ⋅ 102, − ±0.089 ±0.138 ±0.171 ±0.258 ±0.301 ±0.287 ±0.256 ±0.195 ±0.110
99% CI ⋅ 102, − ±0.117 ±0.182 ±0.225 ±0.339 ±0.395 ±0.377 ±0.336 ±0.256 ±0.145
σ, − 0.0271 0.0421 0.0522 0.0785 0.0916 0.0873 0.0780 0.0593 0.0335
CoV, % 5.2% 5.4% 6.2% 10.5% 15.5% 13.9% 10.7% 6.2% 5.2%
Skewness −0.0512 −0.0582 −0.0584 −0.0596 −0.0604 −0.0601 −0.0591 −0.0561 −0.0538
Ex. Kurtosis −0.4256 −0.4047 −0.3919 −0.3858 −0.3851 −0.3783 −0.3672 −0.3335 −0.3238
Plan, − 0.5372 0.7845 0.8595 0.7579 0.6010 0.6399 0.7413 0.9639 0.6515
X/L 0.3245 0.3721 0.4023 0.4801 0.5579 0.6011 0.6357 0.7135 0.7913

3.0 16
%
wr = 42.475N-0.5013 ; R2=0.9956
wr = 64.005N-0.5038 ; R2=0.9946 14
2.5
wr = 24.328N -0.4971 2
; R =0.9966
12
Relativewidth, wr,%

2.0
10
1.5
8

1.0 6

- CoV
0.5 4
-σ
x/L
Fr. 111 Fr. 129 Fr. 165 2
0.0 0.3 0.4 0.5 0.6 0.7 0.8
0 1000 2000 3000 4000
Number of runs, N Figure 8. Standard deviation and CoV along ship length.

Figure 7. Relative width (wr) for 95% confidence for

SWBM for three frames.
max permissible
1.0
The standard deviation and CoV are biggest
at the middle sections. From Figure 8 one can see
that the maximum standard deviation is about 9%
from the permissible SFBM and CoV is about 16%. 0.8
SWBM, -

The most important remark based on the

descriptive statistics of SWBM is illustrated in
Figure 9. The plot includes the SWBM related to the 0.6
permissible values for initial loading condition (the
plan), and obtained for controlled sections mean
and mean ±2 × σ. The shaded area corresponds to
95.45% of all possible values for SWBM that could 0.4
exist for this trim range. Due to the deviation from
μ−2σ μ
cargo plan and uncertainties in cargo distribution μ+2σ Plan
we will have a range of possible SWBM values. 0.2
For frame 165 (X/L = 0.71) there are values greater 0.4 0.6 x/L 0.8
than the permissible ones (rounded rectangle) and
overloading may occur. The relative frequency for Figure 9. Region around the mean value of SFBM.

675

MARSTRUCT.indb 675 2/18/2011 5:56:25 PM

 such overloading for considered trim ranges is as
follows: T1—780/3565 = 0.22; T2—741/3415 = 0.22;
T3—576/3012 = 0.19 and T4—362/2397 = 0.15.
The curve for mean values is very close to the
curve of the plan i.e., the values obtained by load-
ing instrument. For all frames the samples for
SFBM have negative excess kurtosis that shows
distribution with lower, wider peak around the
mean and thinner tails. A negative skew for all
frames indicates longer left tail in the distribution,
but the values are very small and one can conclude
that the data are not skewed.
5 CONCLUSIONS
It has been recognized nowadays that the safety
of the ship is permanent concern of IMO, IACS
and world maritime community, especially in case
of bulk carriers where much is done. The present
investigations worldwide are based on the new
FSA methodology whose goal is to discover haz-
ards and quantify the risk.
In this study an attempt has been made to com-
bine metamodeling technique and Monte-Carlo
simulation to investigate the influence of uncer-
tainties in cargo distribution on still water loads.
These uncertainties originate from possible devia-
tions from cargo plan during loading of cargo.
For bulk carriers with length over 150 m it is man-
datory to have installed loading software on board.
The loading computer should check compliance with
the limits not only for the global still water bending
moment and shear force but also in the local strength
diagram for the hold. The local strength diagram
stipulates the limits with respect to:
− The allowable cargo intakes in each individual
cargo hold as a function of the actual draught;
− The allowable cargo intake for two adjacent
cargo holds as a function of the actual draught.
In most cases the cargo mass and consumables are
given by the user as input to the loading computer.
The work of loading instrument is approximated by
metamodels that give very accurate results.
The Monte-Carlo simulation calculates the fitted
metamodels for trim and SWBM each time using
different randomly-selected values for the amount
of cargo in holds. To take into account the role of
the crew during the loading some constraints con-
cerning the trim range and total amount of cargo
are accepted. The constraints eliminate about 75%
from the number of starting runs. Notwithstanding,
the obtained relative width of 95% confidence
interval with reasonable number of starting runs
(50,000) is less than 0.3% for the mean of cargo
mass in the holds and less than 1.1% for SWBM.
676
MARSTRUCT.indb 676
The results obtained from the proposed
approach give new kind of information that could
be useful for the crew at preparing and monitoring
the loading plan. Based on the results from studied
ship it can be concluded that the probability that
there is no overloading is about 31% and prob-
ability for overloading of hold No 3 (the middle
hold) is about 39% in trim range from 0.04 m fore
to 0.88 m aft. The probability to have no overload-
ing is greatest (42%) at trim range −0.65 ÷ −0.42 m.
An overloading of hold No 3 (that is with greatest
probability of about 39%) could be expected rather
at trim range −0.65 ÷ −0.19 than for greater trim
aft. The overloading of hold No 1 is most probable
(30.1%) at trim that is close to even keel.
Considering SWBM there is a strong depend-
ency between ship length and standard deviation
and CoV in full load condition and the values are
greater for middle sections. Taking into account
the variation in cargo mass in loaded holds it is
possible to obtain region of variation of SWBM,
and study the probability for exceeding of some
critical value
The proposed procedure is applicable in the
design phase and during the operation as well. In
design the procedure can evaluate the vulnerabil-
ity of the ship to deviations in cargo plan during
the loading and in operation the results could be
useful to stress the crew attention to most critical
places during the cargo loading and how to inter-
pret the situation in case of considerable deviation
from cargo plan.
The procedure may be implemented as separate
module in loading software to study the effects of
uncertainties in cargo distribution on still water
loads and safety of bulk carriers.
Finally, it could be useful to present shortly the
actions taken from DNV to increase the loading
flexibility of bulk carriers (Vårheim 2008). DNV
has introduced a new class notation called Easy
Loading, EL. This is a voluntary notation which
will provide more automated control of the critical
parameters during the loading and allow for eas-
ier and more flexible loading sequences. In order
to obtain the EL notation, the following must be
complied with:
− The EL notation can only be assigned to bulk
carriers with a BC-A or BC-B class notation and
an additional GRAB [X] notation.
− Relevant loading sequences, with average load-
ing rates, must be stated.
− Each step in the loading sequence must be docu-
mented, from the commencement of cargo load-
ing until full deadweight is reached, step-wise
synchronized with the de-ballasting operation.
− There must be sufficient de-ballasting capacity
to meet the average loading rate requirements.
2/18/2011 5:56:28 PM
 − The ship must be designed such that at least 50%
of the maximum permissible cargo intake per
cargo hold can be loaded in one pour.
− An automatic draught-reading system must be
fitted.
− An on-line ballast tank level monitoring system,
linked to the loading computer, must be fitted.
ACKNOWLEDGEMENTS
This work has been performed in the frame of
EC 6th Framework Programme Specific Targeted
Project “Handling waves” TST5-CT-2006-031489,
www.mar.ist.utl.pt/handlingwaves/.
REFERENCES
Fang, Kai-Tai, Runze Li & Sudjianto, A. 2006. Design
and modeling for computer experiments. Taylor &
Francis Group.
ISSC. 2006. Special task committee VI.1, Reliability
Based Structural Design and Code Development,
16th International Ship and Offshore Structures
Congress, 20–25 August, 2006 Southampton, UK
Vol. 2, 317–389.
Georgiev, P. & Damyanliev, T. 2005. Metamodels in Ship
Design. Second International Congress on Mechani-
cal and Electrical Engineering and Marine Industry
(MEEMI), Varna, Bulgaria, 7–9 October 2005, Vol. 2,
pp. 268–278.
677
MARSTRUCT.indb 677
Georgiev, P. 2008. Implementation of metamodels in ship
design. Maritime Industry, Ocean Engineering and
Coastal Resources, eds. C. Guedes Soares & P. Kolev,
Taylor & Francis Group, London, 2008, pp. 419–428.
Georgiev, P. 2010. Probabilistic presentation of the bend-
ing moments of bulk carriers using metamodels.
Tenth International Conference on Marine Sciences
and Technologies—Black Sea 2010, October 7–9,
2010, Varna, pp. 82–89: ISSN 1314–0957.
Khasawneh, M.T., Bowling, S.R., Kaewkuekool, S. &
Cho B.R. Tables of a truncated standard nor-
mal distribution: A doubly truncated case. Quality
Engineering, 17:227–241, 205, Taylor & Francis Inc.
Kleijnen, J.P.C. & Sargent, R.G. A methodology for
fitting and validating metamodels in simulation.
European Journal of Operations Research, Vol. 120,
No. 1, 2000, 14–29.
Kuo, Ch. 2007. Safety management and its maritime
application. The Nautical Institute.
MSC. 83rd session. 2007. Formal Safety Assessment.
Consolidated text of the Guidelines for Formal Safety
Assessment (FSA) for use in the IMO rule-making
process (MSC/Circ.1023-MEPC/Circ.392). MSC
83/INF.2, 14 May 2007. IMO.
MSC. 84th session. 2008. Dangerous goods, solid car-
goes and containers. Bulk carrier loading rates. MSC
84/INF.8, 3 March 2008. IMO.
Rizzutto, E. 2006. Uncertainties in still water loads of
tankers and bulkers. Proceedings of the International
Conference on Ship and Shipping Research, NAV 2006.
Genoa, Italy, Centro per gli Studi di Technica Navale.
Vårheim, R. 2008. Bulk carrier update. 1, 2008, DNV,
pp. 6–7.
2/18/2011 5:56:28 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Reliability assessment of intact and damaged bulk carriers

A.W. Hussein
Department of Naval Architecture & Marine Engineering, Suez Canal University, Port Said, Egypt

C. Guedes Soares
Centre for Marine Technology and Engineering, Technical University of Lisbon, Instituto Superior Técnico,
Lisboa, Portugal

ABSTRACT: This paper studies the reliability of two single hull bulk carriers subjected to damage.
Two damage scenarios are assumed, collision at side and grounding at the bottom. The ultimate strength
is calculated using the Progressive Collapse Method. The reliability is assessed for both damage scenarios
and compared with the intact ship reliability. In the damaged condition, the ship should not operate with
a very high speed, and she will avoid very rough sea. Therefore, the related wave-induced bending moment
may be smaller than that for the normal design extreme condition. On the other hand, damage to the vessel’s
hull may result in the ingress of water and for fluid cargoes cargo out-flow may occur which will result in a
change of the loading condition and a variation to the still water bending moment. The reliability is assessed
and sensitivity analysis is performed to study the importance of the variables after and before damage.

1 INTRODUCTION the degradation of strength due to corrosion and

Bulk carriers represent an important type of ship in
the world fleet and thus the interest in studying its
track record of safety (Guedes Soares and Teixeira
2001). New requirements have been appearing for
these ships leading to upgrades in design that lead
to improved safety (Kolev et al. 2005).
The interest of safety connected with the
improved skills of modelling and structural anal-
ysis led the community to consider at the design
stage various accidental scenarios so as to design
the designs as robust as possible to the most fre-
quent accidents of small and moderate intensity.
Several authors have looked at the reliability of
bulk carriers. Guedes Soares and Teixeira (2000),
studied the reliability of two different structural
designs for a bulk carrier and one tanker. The load
effects were defined based on an extreme model
for both still-water and wave-induced bending
moments. It was shown that the failure in sagging
has a higher probability of failure for both ship
types. It was shown that the wave-induced bend-
ing moment is the most important variable for the
full-load condition. Besides, the still-water bending
moment has a large effect on the reliability results
for tankers in the ballast load condition. For bulk
carriers, the importance of the still-water bending
moment increases for both loading conditions.
Guedes Soares and Garbatov (1999a), have
analyzed the same ships, but used a time variant
reliability formulation that was able to account for
fatigue as well as the repair actions due to main-
tenance during ships lifetime. A more complete
analysis method is however presented in Guedes
Soares and Garbatov (1999b).
Fang and Das (2005), studied the relationship
between the risk evaluation and structural reliabil-
ity, and then reviewed the evolution of structural
reliability applied to ship structure. The authors
presented the limit state function of a damaged
ship with consideration of the special situations
after collision and grounding. The failure prob-
abilities of damaged ship were obtained based
on the Monte Carlo simulation technique, given
different damage scenarios and external loads
conditions. The authors assumed that the still
water bending moment given by Classification
Societies’ Rules is the maximum with a probability
of exceedance of 5% and the value fits to a nor-
mal distribution according to Guedes Soares and
Moan (1988). The mean value was taken as 60%
of the rule value.
Khan and Das (2008) made a reliability analysis
for a double skin tanker and a bulk carrier consid-
ering the combined effect of vertical and horizon-
tal loads. It was concluded that the wave-induced
vertical bending moment and vertical ultimate
strength are the most important parameters. The
ultimate strength in vertical bending is the most
important factor in the structural design, but in
case of the bulk carrier, it is noticed that the model
uncertainty factor for horizontal strength showed

679

MARSTRUCT.indb 679 2/18/2011 5:56:28 PM

 significant sensitivity. In that study the SWBM
model was calculated based on the rule value.
Studies of the reliability of damaged ships have
been done by Luis et al. (2009) and Hussein and
Guedes Soares (2009), although in both cases
addressing tankers.
The main aim of this paper is to calculate the
reliability of two bulk carriers in intact and dam-
age cases. The reliability calculations need to spec-
ify the stochastic model of the variables included
in the limit state function. Defining the ultimate
strength can always be done using many methods
such as Smith’s method or finite element methods.
The wave induced bending moment model can
also be defined if the ship lines and the load dis-
tribution are available. The stochastic model of
wave induced load effects is defined based on the Figure 2. Bulk 2 (183 m length).
evaluation of the wave induced load effects that
occur during long-term operation of the ships in
a seaway. In this paper, the reliability is calculated One can see from the figures that the two ships have
for intact and damaged bulk carriers considering longitudinally stiffened bottom and deck. Bulk 1
the SWBM model based of analyzing the data in has a transversely framed lower hoper tank while
the loading manual. Sensitivity analysis is made this area is longitudinally stiffened in Bulk 2.
to determine the importance of the variables and
compare it before and after damage.
3 ULTIMATE STRENGTH
OF INTACT SHIPS
2 MAIN PARTICULARS
OF THE STUDIED SHIPS The ultimate strength of the two ships is calcu-
lated using Smith’s method (Smith 1977). The
In this study two bulk carriers are considered. The ship cross section is divided into small elements
two ships are single side ships. The properties of composed of stiffeners and attached plating. The
the ships are presented in Table 1. Figures 1 and 2 stress–strain relationships of individual elements
show the cross sections of the two bulk carriers. are defined as given in the new IACS Common
Structural Rules (IACS 2006). Then a progres-
Table 1. Ship properties. sive collapse analysis is performed assuming that
a plane cross-section remains plane and each
Bulk 1 Bulk 2
element behaves according to its average stress–
LBP (m) 177 183 average strain relationships.
B (m) 30 32.26 The ultimate strength was calculated for the two
D (m) 16.2 17 ships and the results are presented in Table 2. One
CB 0.82 0.83 can see from the table that the sagging moment is
always lower than the hogging moment, which indi-
cates that the ship is more vulnerable to collapse in
sagging than in hogging.

4 ULTIMATE STRENGTH
OF DAMAGED SHIPS

Damaged structural elements are unable to carry

longitudinal stresses and should be excluded from
the calculation of the ultimate bending moment.
The inner hull is considered the final barrier sepa-
rating the oil from sea. It is assumed to withstand
longitudinal stresses. In this study the vertical dam-
age extent of the hull is considered to be less than
Figure 1. Bulk 1 (177 m length). the double hull height.

680

MARSTRUCT.indb 680 2/18/2011 5:56:28 PM

 Table 2. Ultimate bending moment in MN.m. where Mdamage is the ultimate moment of the dam-
aged section and MIntact is the ultimate moment of
Sag Hog the intact section.
The ultimate strength is calculated for both
Bulk 1 3142 4311
Bulk 2 2524 3895
Bulk carriers in sagging and hogging. Figures 4
and 5 show the RIF for the damages ships in the
five scenarios.
ABS (1995) considers that for grounding on
rocky sea bed there is considerable rupture of the
double bottom structure. The damaged bottom
structure is assumed to be in the most unfavour-
able location anywhere on the flat bottom within
the fore part of the hull between 0.5 L and 0.2 L
aft from F.P. Bottom structures are assumed to
be damaged over a considerable length and the
damaged members should be excluded from the
hull girder. The bottom shell plating for a width
4 m or B/6, whichever is greater, are assumed to
be damaged. The attached girders and the bottom
longitudinals within this area are to be damaged
and inactive.
Collision is considered with another ship on
one side which results in extensive rupture on the
side structure. It is assumed in the most favourable
location anywhere between 0.15 L aft from F.P.
and 0.2 L forward from the A.P. The collision is
assumed to be located at the upper part of the side
shell, down from the stringer plate of the strength
deck. The shell plating for the vertical extend of
4 m or D/4, whichever is greater, and the attached
girders and side longitudinals are to be damaged Figure 3. Damage locations.
and inactive.
Five damage scenarios are assumed; three in the bulk 1 bulk 2 Intact Sagging
1.00
bottom due to grounding and two in the side due 0.95
to collision. Damage due to grounding might be at 0.90

the keel area, the bottom below the hopper tank or 0.85

at the bilge. While damage due to collision might

RIF

0.80

be at the shear strake area or at the joint of the 0.75

hopper tank with the side. 0.70

0.65
Smith’s method (Smith 1977) has been exten- 0.60
sively used to determine the hull girder ultimate keel bottom below hopper
tank
bilge sheer strake and side at hopper tank
upper hopper tank joint
strength of the intact ship. Many authors calcu- Damage location side

lated the ultimate strength of the damaged ship

by Smith’s method too such as Gordo and Guedes Figure 4. Ultimate strength in sagging for damaged
ships.
Soares (1996,1997 and 2000), Fang and Das (2004),
Das and Fang (2005), Luis et al. (2007), Jia and
Moan (2008), Khan and Das (2007). The effect of Table 3. UBM after damage.
damage is simulated by removing the damaged ele-
ments from the midship section and re-calculating Bulk 1 Bulk 2
the ultimate strength of the section.
Sag Hog Sag Hog
The residual strength index is a way of comparing
the ultimate strength of the damaged hull with the Intact 3140 3895 2526 3895
intact one. The Residual Strength Factor (RIF) used Keel 3103 3601 2482 3601
in this study was defined by Fang and Das (2004) as: Bottom below hopper tank 3115 3655 2496 3655
Bilge 3140 3846 2524 3846
M Damage Sheer strake 2902 3714 2262 3714
I =
RIF (1) Side at hopper tank joint 3140 3838 2486 3838
M Intact

681

MARSTRUCT.indb 681 2/18/2011 5:56:29 PM

 1.00
bulk 1 bulk 2 IntactHogging
Guedes Soares and Dogliani (2000), found that the
0.95 still water moment vary significantly from depar-
0.90 ture to arrival. The authors assumed Gaussian
0.85
distribution for both the departure and arrival
RIF

0.80

0.75
conditions and a Gaussian distribution at a ran-
0.70 dom point in time. The mean and standard deri-
0.65 vations of this distribution are the average of the
0.60
keel bottom below hopper bilge sheer strake and side at hopper tank respective values at departure and arrival.
tank
Damage Location
upper hopper tank
side
joint
For the bulk carrier three different loading
conditions are defined, namely: alternate hold
Figure 5. Ultimate strength in hogging for damaged loading condition (LC1), homogeneous hold
ships. loading condition (LC2) and ballast condition
(LC3). In each loading condition, a percentage
Table 4. Stochastic modeling of UBM. of ship life can be identified according to an esti-
mate of the operational profile for the ship type.
Sagging Hogging A homogeneous hold loading condition refers
to the carriage of cargo, evenly distributed in all
μ σ μ σ cargo holds. This condition is usually adopted for
low density cargoes such as coal and grain, but may
Bulk 1 3142 251.36 4311 344.88 also be permitted for high-density cargoes under
Bulk 2 2524 202 3895 312 certain conditions. However, heavy cargo, such as
iron ore, is often carried in alternate holds. This
type of cargo distribution raises the ship’s centre
From the figures one can see that in sagging of gravity, which eases the ship’s rolling motion.
condition, the worst scenario which causes the In this paper the SWBM distribution is calcu-
worst ultimate strength after damage is when the lated based on studying the data of the loading
sheer strake is damaged. The loss in the ultimate manual. The mean value and the standard devia-
strength due to side damage at the hopper tank tion are calculated as a function of the maximum
joint in Bulk 1 is greater than that in Bulk 2. This allowed value defined in the loading manual.
can be attributed to the transverse stiffening of Table 5 shows the stochastic model of the still
Bulk 1. Although the loss in the ultimate strength water bending moment based on analyzing the
in hogging is more than that in sagging, never-
theless the sagging conditions still represents the Table 5. Stochastic model of SWBM (% max allowed).
critical condition.
Bulk 1 Bulk 2
4.1 Stochastic model of UBM μ σ μ σ
The ultimate bending moment is calculated using
the progressive collapse method. The ultimate LC1: Ballast 0.74 0.09 0.79 0.12
capacity distribution is lognormal with a mean LC2: Homogeneous 0.15 0.2 0.21 0.09
value as calculated by the code and the COV is 0.08 LC3: Alternate 0.85 0.04 0.85 0.09
the mean value. Table 4 shows the stochastic model
of all the ultimate capacity for the two ships.
Table 6. SWBM of Bulk 1.

Load condition Mean St. Dev.

5 STILL WATER BENDING MOMENT
LC1: Alternate 771 220
The still-water load effects result from the longi- LC2: Homogeneous 234 67
tudinal distribution of the cargo on-board. These LC3: Ballast 812 232
are likely to change at each departure and smaller
changes may occur during a voyage. Once the cargo
distribution is known, the still-water load effects Table 7. SWBM of Bulk 2.
can be calculated. However, they will vary with
time and consequently, at the design stage, they can Mean value St. Dev.
only be described by a probability distribution.
LC1: Alternate 1111 122
Guedes Soares and Moan (1982), identified that LC2: Homogeneous 195 14
the vertical still water bending moments amid- LC3: Ballast 979 258
ship can be described by a normal distribution.

682

MARSTRUCT.indb 682 2/18/2011 5:56:32 PM

 manual data. The values in the table are percent- observations follows an extreme distribution. Thus,
age of the maximum value defend in the loading the distribution of the extreme values of the wave
manual. One can see that the mean value of SWBM induced bending moment over the time period T is
in the ballast loading condition and the alternate obtained as a Gumbel law (Guedes Soares 1984):
loading condition is much higher than that in the
homogenous loading condition. ⎡ ⎛ x xn ⎞ ⎤
Fe ( xe ) = exp ⎢ exp ⎜ − e (3)
⎣ ⎝ σ ⎟⎠ ⎥⎦
6 WAVE INDUCED BENDING MOMENT
where xn and σ are parameters of the Gumbel
The stochastic model of wave induced load distribution.
effects is defined based on the evaluation of The Gumbel parameters xn and σ can be esti-
the wave induced load effects that occur during mated from the initial Weibull distribution using
long-term operation of the ships in a seaway. The the following equation:
short term structural response due to waves in
terms of mid ship bending moment is obtained 1
by linear hydrodynamic analysis and assum- xn w ⋅[ n ]k (4)
ing the Pierson-Moskowitz (PM) spectrum.
The long term response is then computed using 1− k
w
IACS (2000) North Atlantic scatter diagram σ =
k
[ ]k (5)
covering areas 8, 9, 15, 16. The resultant prob-
ability distribution is fitted by a Weibull model,
which describes the distribution of the peaks at a where w and k are the Weibull parameters and n is
random point in time: the mean number of load cycles expected during
the time period T.
⎡ ⎛ x⎞k⎤ Table 8 shows the stochastic model of extreme
FX ( x ) = 1 − exp ⎢ − ⎜ ⎟ ⎥ (2) wave induced bending moment for the adopted
⎢⎣ ⎝ w ⎠ ⎥⎦ operational profile. For each loading condition,
the distribution of the extreme wave load is calcu-
where w and k are respectively the scale and shape lated over the period of time that the ship spends
parameters of the Weibull distribution. in this condition TC. The average of wave period is
Figure 6 illustrates the long-term distribution assumed to be 7s in North Atlantic.
and the Weibull fit for the bulk carrier in alter- Table 9 shows the stochastic model of wave-
nate, homogeneous and ballast load conditions. induced loads based on the rule value with proba-
However one is normally interested in having the bility of exceedance 10−8. The Weibull distribution
probability distribution of the maximum ampli-
tude of wave induced effects in n cycles where n
corresponds to the mean number of load cycles Table 8. Stochastic model of WBM in MN. Bulk 1.
expected during the ship’s lifetime.
Weibull Gumbel
Gumbel (1958) has shown that whenever the parameters moments
initial distribution of a variable has an exponen- Load
tial tail, the distribution of the largest value in n condition w k Mean Std

LC1 84.2 0.950 1384.4 130.6

2500
LC2 97.4 0.993 1419.2 128.3
LC1 LC2 LC3
2000
LC3 90.1 0.933 1640.7 150.1
VBM (MN.m )

1500

1000
Table 9. Stochastic model of WBM in MN. Bulk 2.

500 Weibull Gumbel

parameters moments
0 Load
1.E-09 1.E-08 1.E-07 1.E-06 1.E-05 1.E-04 1.E-03 1.E-02 1.E-01 condition w k Mean Std

Q(x) LC1 79.17 0.95 1301.1 122.7

LC2 90.41 0.993 1316.3 118.9
Figure 6. Long term distributions for the three loading LC3 74.87 0.933 1362.3 124.7
condition.

683

MARSTRUCT.indb 683 2/18/2011 5:56:34 PM

 is assumed for long-term distribution and then 7.1 Sensitivity analysis
Gumbel parameter are calculated for one year
A sensitivity analysis is done to know the impor-
modelling. The same scale factors, which were
tance of all the variables include in the limit state
calculated for Bulk 1, are used for Bulk 4.
function and its influence on the ship reliability.
Positive sensitivity indicates that an increase in this
variable leads to an increase in the reliability and
7 RELIABILITY ASSESSMENT
vice versa. The sensitivity factor for a give limit
IN INTACT CONDITION
state function is given by:
In the present reliability assessment, a time inde-
pendent first order reliability formulation corre- αi =
1 . ∂g ( x ) (7)
sponding to one-year operation is considered. The n 2
⎛ ∂g ( x ) ⎞ ∂xi
limit state equation corresponds to the hull girder ∑ ⎜⎝ ∂x ⎟⎠
failure under vertical bending: i i

g(x) = Mu ⋅ XR − [Mwn ⋅ Xst ⋅ Xnl + Msw ⋅ Xsw] (6) One can notice from the sensitivity analyses
done that the ultimate capacity importance is the
where, MU is the ultimate capacity with a model highest among all the variables in LC1 and LC3.
uncertainty factor XR. MWV is the wave bending while the WBM is the most important variable in
moment with model uncertainty factors; Xst for LC2. The SWBM importance increases in the bal-
the linear response calculation and Xnl for nonlin- last condition while the WBM importance increases
ear effects. MSW is the random still water bending in the homogenous loading condition while the
moment with a model uncertainty factor XSW. The SWBM showed very low importance. The results
values of the above mentioned uncertainties are are presented in Tables 12 and 13. Figures 7 and 8
taken as presented in Table 10. show the results of Bulk 1.
The reliability calculations were carried out using
the computer program COMREL (Gollwitzer
et al. 1988), considering the distribution of the Table 12. Bulk 1.
extreme values of the load effects in the three load
conditions. The reliability indices for the different LC1 LC2 LC3
load conditions are presented in Table 11 for the
Sagging
intact condition.
SWBM −0.32 −0.28 −0.55
It is clear from the results that sagging is more
WBM −0.33 −0.61 −0.30
critical than hogging. The homogenous hold loading
UBM 0.53 0.41 0.45
condition has the highest reliability, while the alter-
nate and ballast loading give very low reliability. LC1 LC2 h LC3

Hogging
Table 10. Summary of uncertainties of the models.
SWBM −0.27 −0.17 −0.49
Variable Distribution μ δ WBM −0.44 −0.71 −0.36
UBM 0.51 0.37 0.47
XR Normal 0.85 0.1
Xst Normal 1 0.1
Xnl Normal 1 0.1 Table 13. Bulk 2.
XSW Normal 1 0.1
LC1 LC2 LC3

Table 11. Reliability indices in intact condition. Sagging

SWBM −0.31 −0.28 −0.40
Bulk 1 Bulk 2 WBM −0.34 −0.56 −0.33
UBM 0.55 0.43 0.52
LC1 sag 0.488 0.616
LC1 hog 2.571 3.508 LC1 LC2 h LC3
LC2 sag 2.829 2.72 Hogging
LC2 hog 4.365 4.904 SWBM −0.23 −0.17 −0.31
LC3 sag 0.414 0.616 WBM −0.53 −0.71 −0.50
LC3 hog 1.980 3.421 UBM 0.51 0.39 0.51

684

MARSTRUCT.indb 684 2/18/2011 5:56:36 PM

 LC1 s LC 2 s LC 3 s
8 RELIABILITY ASSESSEMNT IN
0.80

0.60
Sagging
DAMAGED CONDITION
0.40
Sensitivity variables

0.20

0.00
In the damaged condition, the ship should not
-0.20
SWBM WBM UB M
operate with a very high speed, and it will avoid
-0.40
very rough seas. Therefore, the related wave-
-0.60

-0.80
induced bending moment may be smaller than that
for the normal design extreme condition. On the
Figure 7. Sensitivity variables of intact ship in sagging. other hand, damage to the vessel’s hull may result
in the ingress of water and, for fluid cargoes cargo
LC 1 h LC 2 h LC 3 h out-flow can occur, which will result in a change
0.60
Hogging of the loading condition and a variation to the
0.40
still water bending moment (Santos and Guedes
Sensitivity variables

0.20

0.00
Soares 2008).
-0.20
SWBM WBM UBM Regarding the still water load coefficient, this is
-0.40
usually considered equal to 1.0 for the intact ship.
-0.60 The ABS (ABS 1995) guide for damaged ships rec-
-0.80 ommends that for hogging the value used should
be 1.1, while for sagging should be 0.9 as shown
Figure 8. Sensitivity variables of intact ship in hogging. in Table 14.
Thus, the bending moment according to the
7
lc1 s lc2 s lc3 s Bulk 1 Sagging ABS guide is equal to:
6
5
4
Mt = Kus ⋅ Msw + Kuw ⋅ Mw (8)
β

3
2
1
where Kus and Kuw are load combination coeffi-
0
intact keel bottom keel sheertrake hopper side
cients due to damage. The reliability is assessed
considering the ultimate bending moment after
Figure 9. Reliability indices Bulk 1 with damage. damage presented in Table 3 and the load combi-
nation factors presented in Table 14. Results are
lc1 h lc2 h lc3h Bulk 1 Hogging
presented in Tables 15 and 16.
8
7
6
5
4
Table 14. Bending moment combination factors
β

3
2
1
(ABS 1995).
0
intact keel bottom keel sheertrake hopper side
Intact Grounding Collision
Figure 10. Reliability indices Bulk 1 with damage.
Hogging Kus 1 1.1 1
Kuw 1 0.5 0.7
lc1s lc2 s lc3 s Sagging
6.00 Sagging Kus 1 0.9 1
5.00
Kuw 1 0.5 0.7
4.00
3.00
β

2.00
1.00
0.00
Intact keel bottom bilge sheertrake hopper Table 15. Reliability indices of damaged ship (Bulk 1).
side

Intact Keel Bottom Bilge Sheer Hopper

Figure 11. Reliability indices Bulk 2 with damage. strake side

lc1 h lc2 h lc3 h Hogging Sagging

8.00
7.00 LC1 0.488 3.04 3.07 3.12 1.18 1.71
6.00
5.00
LC2 2.829 3.10 5.82 5.84 4.05 4.42
4.00
LC3 0.414 2.43 2.45 2.50 0.84 1.28
β

3.00
2.00
1.00
Hogging
0.00
Intact keel bottom bilge sheertrake hopper
LC1 2.57 4.10 4.13 4.29 3.22 2.51
side LC2 4.36 7.00 7.02 7.13 5.45 4.97
LC3 1.98 3.33 3.35 3.49 2.58 1.96
Figure 12. Reliability indices Bulk 2 with damage.

685

MARSTRUCT.indb 685 2/18/2011 5:56:37 PM

 Table 16. Reliability indices of damaged ship (Bulk 2).

Intact Keel Bottom Bilge Sheer strake Hopper side

Sagging
LC1 0.62 3.159 3.199 3.279 1.055 1.712
LC2 2.72 5.574 5.603 5.66 3.634 4.127
LC3 0.62 3.103 3.142 3.218 1.078 1.688
Hogging
LC1 3.51 4.927 5.036 5.388 4.562 4.214
LC2 4.90 7.205 7.281 7.528 6.096 6.251
LC3 3.42 4.793 4.898 5.24 4.226 4.421

Table 17. Sensitivity factors of Bulk 1 in sagging.

Intact Keel Bottom Bilge Sheer strake Hopper side

LC1 s
SWBM −0.32 −0.39 −0.39 −0.39 −0.38 −0.37
WBM −0.33 −0.31 −0.31 −0.31 −0.30 −0.32
UBM 0.53 0.54 0.54 0.54 0.54 0.54
xsw −0.29 −0.36 −0.36 −0.36 −0.35 −0.34
xnl −0.36 −0.26 −0.26 −0.26 −0.30 −0.30
Xst −0.36 −0.26 −0.26 −0.26 −0.30 −0.30
xr 0.40 0.44 0.44 0.44 0.41 0.42
LC2 s
SWBM −0.28 −0.27 −0.27 −0.27 −0.32 −0.29
WBM −0.61 −0.74 −0.74 −0.74 −0.69 −0.70
UBM 0.41 0.36 0.36 0.36 0.39 0.38
xsw −0.04 −0.04 −0.04 −0.04 −0.05 −0.04
xnl −0.42 −0.34 −0.34 −0.34 −0.38 −0.37
Xst −0.42 −0.34 −0.34 −0.34 −0.38 −0.37
xr 0.32 0.30 0.30 0.30 0.31 0.30
LC3 s
SWBM −0.55 −0.63 −0.63 −0.63 −0.63 −0.61
WBM −0.30 −0.23 −0.23 −0.23 −0.25 −0.26
UBM 0.45 0.46 0.46 0.46 0.45 0.45
xsw −0.21 −0.31 −0.31 −0.31 −0.27 −0.27
xnl −0.35 −0.23 −0.23 −0.23 −0.28 −0.28
Xst −0.35 −0.23 −0.23 −0.23 −0.28 −0.28
xr 0.33 0.36 0.36 0.36 0.33 0.34

One can see from the figures that the homog-
enous loading conditions represents the safest
condition, this is because of the uniform cargo
distribution inside the compartments. While the
ballast loading condition and the alternative load-
ing condition which give high SWBM give low
reliability. The worst reliability occurred when
the side is damaged, whether at the sheer strake
or at the joint of the hopper tank with the side.
9 SENSITIVITY OF THE VARIABLES
AFTER DAMAGE
Sensitivity analysis is made for the damaged con-
ditions to study the importance of the variable
included in the limit state function and compare its
importance before and after damage. The results
are presented in Tables 17 to 20.
One can conclude from the calculations the
following:
• When the ship is damaged at her bottom, the
importance of the variables after damage does
not depend on the position of the damage.
• When the ship is damaged at her side, the impor-
tance of the variables after damage does not
depend on the location of the damage.
• The change in the importance of the ultimate
bending moment after damage is very small.
• The importance of the SWBM increased after
damage.

686

MARSTRUCT.indb 686 2/18/2011 5:56:42 PM

 Table 18. Sensitivity factors of Bulk 1 in hogging.

Intact Keel Bottom Bilge Sheer strake Hopper side

LC1 h
SWBM −0.27 −0.41 −0.41 −0.41 −0.33 −0.35
WBM −0.44 −0.28 −0.29 −0.29 −0.39 −0.35
UBM 0.51 0.55 0.55 0.55 0.53 0.54
xsw −0.25 −0.38 −0.38 −0.38 −0.31 −0.33
xnl −0.35 −0.22 −0.22 −0.22 −0.30 −0.30
Xst −0.35 −0.22 −0.22 −0.22 −0.30 −0.30
xr 0.40 0.45 0.45 0.46 0.43 0.42
LC2 h
SWBM −0.17 −0.25 −0.25 −0.25 −0.19 −0.19
WBM −0.71 −0.74 −0.74 −0.74 −0.73 −0.72
UBM 0.37 0.36 0.36 0.36 0.36 0.37
xsw −0.03 −0.04 −0.04 −0.04 −0.03 −0.04
xnl −0.37 −0.33 −0.33 −0.33 −0.35 −0.36
Xst −0.37 −0.33 −0.33 −0.33 −0.35 −0.36
xr 0.29 0.30 0.30 0.30 0.29 0.30
LC3 h
SWBM −0.49 −0.65 −0.65 −0.64 −0.57 −0.59
WBM −0.36 −0.20 −0.20 −0.20 −0.30 −0.28
UBM 0.47 0.47 0.47 0.47 0.47 0.46
xsw −0.22 −0.35 −0.35 −0.35 −0.28 −0.28
xnl −0.33 −0.19 −0.19 −0.19 −0.27 −0.27
Xst −0.33 −0.19 −0.19 −0.19 −0.27 −0.27
xr 0.36 0.37 0.37 0.37 0.37 0.36

Table 19. Sensitivity factors of Bulk 2 in sagging.

Intact Keel Bottom Bilge Sheer strake Hopper side

LC1 s
SWBM −0.31 −0.39 −0.38 −0.38 −0.38 −0.37
WBM −0.34 −0.33 −0.34 −0.34 −0.31 −0.33
UBM 0.55 0.56 0.56 0.56 0.56 0.56
xsw −0.30 −0.37 −0.37 −0.37 −0.36 −0.35
xnl −0.37 −0.27 −0.27 −0.27 −0.32 −0.31
Xst −0.37 −0.27 −0.27 −0.27 −0.32 −0.31
xr 0.34 0.37 0.37 0.38 0.36 0.36
LC2 s
SWBM −0.28 −0.27 −0.27 −0.27 −0.32 −0.29
WBM −0.56 −0.68 −0.68 −0.69 −0.59 −0.63
UBM 0.43 0.40 0.40 0.40 0.42 0.41
xsw −0.08 −0.10 −0.10 −0.10 −0.11 −0.10
xnl −0.41 −0.33 −0.33 −0.33 −0.37 −0.36
Xst −0.41 −0.33 −0.33 −0.33 −0.37 −0.36
xr 0.28 0.27 0.27 0.27 0.28 0.27
LC3 s
SWBM −0.40 −0.48 −0.48 −0.48 −0.46 −0.45
WBM −0.33 −0.30 −0.31 −0.31 −0.35 −0.37
UBM 0.52 0.54 0.54 0.54 0.52 0.52
xsw −0.27 −0.35 −0.35 −0.35 −0.32 −0.31
xnl −0.37 −0.26 −0.26 −0.26 −0.31 −0.30
Xst −0.37 −0.26 −0.26 −0.26 −0.31 −0.30
xr 0.33 0.36 0.36 0.36 0.33 0.33

687

MARSTRUCT.indb 687 2/18/2011 5:56:42 PM

 Table 20. Sensitivity factors of Bulk 2 in hogging.

Intact Keel Bottom Bilge Sheer strake Hopper side

LC1 h
SWBM −0.23 −0.39 −0.39 −0.37 −0.28 −0.27
WBM −0.53 −0.36 −0.37 −0.39 −0.53 −0.55
UBM 0.51 0.56 0.56 0.56 0.51 0.51
xsw −0.22 −0.37 −0.37 −0.36 −0.27 −0.26
xnl −0.35 −0.24 −0.24 −0.24 −0.30 −0.30
Xst −0.35 −0.24 −0.24 −0.24 −0.30 −0.30
xr 0.33 0.39 0.39 0.39 0.35 0.35
LC2 h
SWBM −0.17 −0.25 −0.25 −0.25 −0.19 −0.19
WBM −0.71 −0.70 −0.71 −0.71 −0.72 −0.73
UBM 0.39 0.39 0.39 0.39 0.38 0.38
xsw −0.05 −0.10 −0.10 −0.10 −0.07 −0.06
xnl −0.36 −0.32 −0.32 −0.32 −0.34 −0.34
Xst −0.36 −0.32 −0.32 −0.32 −0.34 −0.34
xr 0.25 0.27 0.27 0.27 0.26 0.26
LC3 h
SWBM −0.31 −0.49 −0.49 −0.49 −0.33 −0.32
WBM −0.50 −0.30 −0.30 −0.30 −0.57 −0.59
UBM 0.51 0.54 0.54 0.54 0.48 0.47
xsw −0.22 −0.37 −0.37 −0.37 −0.24 −0.23
xnl −0.35 −0.22 −0.22 −0.22 −0.30 −0.30
Xst −0.35 −0.22 −0.22 −0.22 −0.30 −0.30
xr 0.33 0.37 0.37 0.37 0.32 0.32

Table 21. General SWBM stochastic model. The damage is simulated by removing the dam-
aged elements from the mid-ship section and
μ σ re-calculating the ultimate strength of the section.
The reliability is calculated considering the
LC1: Ballast 0.77 0.11
ultimate limit state function; the SWBM statistic
LC2: Homogeneous 0.18 0.15
model is calculated based on studying the data
LC3: Alternative 0.85 0.07
of the loading manual. The mean value and the
standard deviation are calculated as a function of
the maximum allowed value defined in the loading
• The importance of the WBM decreased after manual.
damage. A final conclusion can be given from that study
• In sagging condition, there was no big difference telling that the mean value and the COV values
between the importance of the variables when of a bulk carrier of the same size can be taken as
the ship is damaged at her side or at her bottom. shown in Table 21.
• In hogging condition, the importance of the The stochastic model of wave induced load
WBM is higher when the side of the ship is dam- effects is defined based on the evaluation of the
aged than that when the bottom of the ship is wave induced load effects that occur during long-
damaged. term operation of the ships in a seaway.
• In hogging condition, the importance of the Calculation of the reliability of the intact ships
SWBM decreased when the ship is damaged at showed that the sagging is more critical than
her side that when she is damaged at her bottom. hogging. The homogenous loading condition has
the highest reliability, while the alternate and ballast
loading give very low reliability.
10 CONCLUSIONS Sensitivity analyses showed that the ultimate
capacity importance is the highest among all the
This paper studies the ultimate strength and reli- variables in LC1 and LC3, while the WBM is
ability of two single side bulk carriers. The ulti- the most important variable in LC2. The SWBM
mate strength is calculated using Smith’s method. importance increases in the ballast condition while

688

MARSTRUCT.indb 688 2/18/2011 5:56:42 PM

 the WBM importance increases in the homogenous
loading condition while the SWBM showed very
low importance.
Reliability assessment of the damaged ships
showed that the homogenous loading conditions
represent the safest condition, while the ballast load-
ing condition and the alternative loading condition
gave low reliability. The worst reliability occurred
when the side is damaged, whether at the sheer strake
or at the joint of the hopper tank with the side.
Sensitivity analysis for the limit state function
variables after damage showed that the importance
of the variables after damage does not depend on
the location of the damage. The importance of the
SWBM increased after damage, while the impor-
tance of the WBM decreased after damage.
In sagging condition, there was no big difference
between the importance of the variables when the
ship is damaged at her side or at her bottom.
In hogging condition, the importance of the
WBM is higher when the side of the ship is dam-
aged than that when the bottom of the ship is
damaged. Besides, the importance of the SWBM
decreased when the ship is damaged at her side that
when she is damaged at her bottom.
REFERENCES
ABS (1995). “Guide for Assessing Hull-Girder Residual
Strength for Bulk Carriers”, American Bureau of
Shipping.
Fang, C. & Das, P.K. (2004). “Hull Girder Ultimate
Strength of Damaged Ships”, Proceedings of the 9th
Symposium on Practical Design of Ships and Other
Floating Structures, Luebeck-Travemuende, Germany.
pp. 309–316.
Fang, C.L, & Das, P.K. (2005). “Survivability and
Reliability of Damaged Ships after Collision and
Grounding”, Ocean Engineering, 32:293–307.
Fang, C & Das, P.K. (2004). “Hull Girder Ultimate
Strength of Damaged Ships”, Proceedings of the 9th
Symposium on Practical Design of Ships and Other
Floating Structures, Luebeck-Travemuende, Germany.
pp. 309–316.
Gollwitzer, S. Abdo, T. & Rackwitz, R. (1988). FORM—
Program Manual, Munich.
Gordo, J.M. & Guedes Soares, C. (1996). “Approximate
Method to Evaluate the Hull Girder Collapse
Strength”, Marine Structures, 9(1):449–470.
Gordo, J.M. & Guedes Soares, C. (1997). “Interaction
Equation for the Collapse of Tankers and Container-
ships Under Combined Bending Moments”, Journal
of Ship Research, 41(3):230–340.
Gordo, J.M. & Guedes Soares, C. (2000). “Residual
Strength of Damaged Ship Hulls”, Proceedings of 9th
IMAM Conference; Napoli, Italy. pp. 79–86.
Guedes Soares, C. (1984). Probabilistic Models for Load
Effects in Ship Structures, Division of Marine Struc-
tures, The Norwegian Institute of Technology, Report
UR-84-38.
689
MARSTRUCT.indb 689
Guedes Soares, C. & Dogliani, M. (2000). “Probabilistic
modelling of time-varying still-water load effects in
tankers”, Marine Structures, Vol. 13, pp. 129–143.
Guedes Soares, C. & Garbatov, Y. (1999a). Reliability of
Maintained Hull Girders of Two Bulk Carrier Designs
Subjected to Fatigue and Corrosion. Journal of Ship
and Ocean Technology. 3(1):27–41.
Guedes Soares, C. & Garbatov, Y. (1999b). Reliability
of Maintained Ship Hulls Subjected to Corrosion
and Fatigue under Combined Loading. Journal of
Constructional Steel Research. 52(1):93–115.
Guedes Soares, C. & Moan, T. 1982. Statistical Analysis
of Still-Water Bending Moments and Shear Forces on
Tankers, ore and Bulk Carriers, Norwegian Maritime
Research, Vol. 10, N° 3, pp. 33–47.
Guedes Soares, C. & Moan, T. (1988). “Statistical Analy-
sis of Still Water Load Effects in Ship Structures”,
Transactions of the Society of Naval Architects and
Marine Engineers (SNAME), New York, Vol. 96, 1988,
pp. l29–156.
Guedes Soares, C. & Teixeira, A.P. (2000). Structural
Reliability of two bulk carriers designs. Marine Struc-
tures, 13(2):107–128.
Guedes Soares, C. & Teixeira, A.P. (2001). Risk Assess-
ment in Maritime Transportation. Reliability
Engineering and System Safety. 74:299–309.
Gumbel, E.J. 1958. Statistics of Extremes, Columbia
University Press, New York.
Hussein, A.W. & Guedes Soares, C. (2009). “Reliability and
Residual Strength of Double Hull Tankers Designed
According to the new IACS Common Structural
Rules”, Ocean Engineering, 36(17–18):1446–1459.
IACS (2001). Recommendation No. 34, “Standard Wave
Data”, 1992, Rev. 1 June 2000, Corr. Nov.2001.
IACS (2006). Common structural rules for Bulk Carriers,,
International Association of Classification Societies,
London, http://www.iacs.org.
Jia, H. & Moan, T. (2008) “Reliability Analysis of Oil
Tankers with Collision Damage”, Proceedings of the
27th International Conference on Offshore Mechanics
and Arctic Engineering (OMAE 2008), paper OMAE
2008-57102.
Khan I.A. & Das P.K. (2007). “Structural Safety Assess-
ment of Damaged Ships”, Proceedings of 26th Inter-
national Conference on Offshore Mechanics and
Arctic Engineering, San Diego, USA, 10–15 June
2007. Paper no. OMAE 2007-29121.
Khan, I.A. & Das, P.K. (2008). “Sensitivity analysis of
the random design variables applicable to ship struc-
tures”, Maritime Industry, Ocean Engineering and
Coastal Resources. Guedes Soares & Kolev (eds)
Taylor and Francis, pp. 185–192.
Kolev, P., Tsenkov, M., Georgiev, P. & Guedes Soares, C.
(2005). Upgrade of Existing Bulk Carriers to Meet the
New Requirements, Proceedings of the International
Conference “Design and Operation of Bulk Carriers”
(RINA); London, UK. 87–93.
Luís, R.M., Hussein, A.W. & Guedes Soares, C. (2007).
“On the Effect of Damage on the Longitudinal
Strength of Ships”, Proceeding of the 10th Interna-
tional Symposium on Practical Design of Ships and
Other Floating Structures (PRADS 2007), Houston,
Texas, United States of America, American Bureau
of Shipping.
2/18/2011 5:56:42 PM
 Luís, R.M., Teixeira, A.P. & Guedes Soares, C. (2009).
Longitudinal Strength Reliability of a Tanker
Hull Accidentally Grounded. Structural Safety.
31(3):224–233.
Santos, T.A. & Guedes Soares, C. (2008). Global Loads
due to Progressive Flooding in Passenger Ro-Ro
Ships and Tankers. Ships and Offshore Structures.;
3(4):289–302.
690
MARSTRUCT.indb 690
Smith, C.S. (1977). “Influence of Local Compression
Failure on Ultimate Longitudinal Strength of a Ship’s
Hull”, Proceedings of the International Symposium
on Practical Design in Shipbuilding, Tokyo, PRADS,
Japan, pp. 73–79.
2/18/2011 5:56:43 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Structural optimization of the hold frame of a bulk carrier

considering lifecycle risk

Y. Kawamura
Yokohama National University, Yokohama-shi, Kanagawa-ken, Japan

M. Miyazaki
Yokohama National University, Yokohama-shi, Kanagawa-ken, Japan (Now in Imabari Shipbuilding Co. Ltd.,
Marugame-shi, Kagawa-ken, Japan)

ABSTRACT: In this study, we propose a new strategy of lifecycle structural optimization by using
structural reliability analysis and risk evaluation. As a simple example, optimization of the hold frame of
a bulk carrier is carried out. In the first optimization, construction cost is used as an objective function
and the failure probability is used as a design constraint. In the second optimization, the lifecycle cost
including the risk of failure is regarded as an objective function, while lifecycle benefit is used as an objective
function in the last optimization. It is noted that not only the dimensions of the structural members but
also a number of repairs (re-coating) are used as design variables in the lifecycle optimization.

1 INTRODUCTION of failure is defined as an objective function,

Recently, FSA (Formal Safety Assessment) was
developed for rule-making process in IMO. Also,
the concept of SLA (Safety Level Approach) is
proposed in the development of GBS (Goal-based
Standard), where safety level is defined by the risk
as a quantitative index. Moreover, the authors
proposed a method of inspection planning based
on the Remaining Life Benefit (RLB) considering
the risk of failure (Kawamura et al., 2009). In this
way, risk-based approach now becomes a power-
ful tool for safety assessment of ship structures,
and can be an alternative approach for ship struc-
tural design instead of the traditional rule-based
approach. However at present, application of the
risk-based approach to structural design is very
few, so that it is necessary to establish a reasonable
strategy for risk-based structural design.
In this paper, we studied about a new strategy
of lifecycle structural optimization considering
the risk of failure. The target structure of the
optimization is the hold frame structure of a bulk
carrier. Firstly, annual failure probabilities of the
hold frame in the life of the bulk carrier is com-
puted by structural reliability analysis using a
simple beam model of the hold frame. Then, by
using the calculated failure probability and the
risk, three types of optimization are carried out.
In the first optimization, construction cost is used
as an objective function and the failure probabil-
ity is used as a design constraint. In the second
optimization, the lifecycle cost including the risk
while the lifecycle benefit is used as an objective
function in the last optimization. In the latter
two optimization, not only the dimensions of
structural members but also a number of repairs
(re-coating) are used as design variables, so that
we call these optimization methods as “lifecycle
structural optimization”. After the results of
three types of optimization are compared, the
availability of the proposed optimization strategy
is discussed in this paper.
2 STRUCTURAL RELIABILITY ANALYSIS
FOR HOLD FRAME STRUCTURE
2.1 Target structural model
It is reported by ClassNK (1992) that a series of
serious accidents occurred in about 40 merchant
ships in 1990’s, where the average ship-years of
these ships are about 18 (very high), and the most
of the accidents occurred in large bulk carriers. The
report by ClassNK also says that drop off of side
shell plates occurred in some cases, and significant
structural deteriorations such as corrosion wast-
age were observed at the hold frames of the bulk
carriers. It seems that there might exist weakness
in the structural design of the hold frame that is
suffered from structural deterioration. From these
background, we selected the design of the hold
frame shown in Figure 1(a) as a target structure of
optimization.

691

MARSTRUCT.indb 691 2/18/2011 5:56:43 PM


(a) Hold frame of bulk carrier
(b) Frame structure for analysis
ps , pw
(c) Simple both-ends-clamped beam model
subjected to hydro-static pressure (ps)
and hydrodynamic pressure (pw)
Figure 1. Target structure (Hold frame of bulk carrier).
It is also reported that the accident occurred
more frequent when the bulk carrier was loading
iron ore compared with when it was loading other
cargoes such as coals. Because of the high specific
gravity, the iron ore might be stacked just in the
bottom of the cargo hold with low height, so that
the water pressure may directly apply to the hold
frame without internal pressure by the cargoes.
For this reason, we consider the failure of the
frame by the applied distributed load caused by
the hydrostatic and the hydrodynamic pressure.
For the structural analysis of the hold frame, we
simplified the hold frame structure to a simple
both-ends-clamped beam model composed of one
frame with side shell plate and flange (Fig. 1 (b)(c)).
Though this model is very simple from the view-
point of structural analysis, easy formulation of
structural reliability analysis is possible. By using
this model, we focus on the discussion about the
difference of the results of optimization by differ-
ent strategy of optimization.
2.2 Structural reliability analysis
In this study, structural reliability analysis is car-
ried out to compute annual probability of failure
of the frame for each ship year, and to calculate
the risk of the failure. Limit state function for the
structural reliability analysis is defined as follows,
g(t) = MU(t) − ML(t), (1)
where MU(t) is the strength of the frame and ML(t)
is the load, which are evaluated as bending moment
692
MARSTRUCT.indb 692
of the beam model. Evaluation point of the limit
state function is selected at the lower clamped end
of the beam, because it is reported that the failure
of the frame usually occurred around that point.
2.2.1 Evaluation of strength
Usually, ultimate strength for complicated struc-
tures can be evaluated by using non-linear finite
element method. However, such method is very
costly so that it is difficult to use it from the iterative
optimization procedure. For this reason, we con-
sider the simple beam model and use the following
formula for the evaluation of the ultimate strength
of the frame.
MU(t) = xuϕZ(t)σy (2)
In this equation, σy is yield stress of the material
of the frame, and Z(t) is the section modulus, and xu
represents the modeling uncertainty. It is assumed
that ultimate strength of the frame can be calcu-
lated by multiplying ϕ (=1.43) to the maximum
elastic bending moment (Z(t)σy). The value of ϕ
is decided from the estimated ultimate strength
by 3-dimensional non-linear finite element shell
analysis of the frame structure.
Also in this study, effect of corrosion wastage
to the reduction of the strength is considered. For
prediction of the corrosion wastage, Paik et al.
(2004) proposed a probabilistic model to estimate
the thickness reduction by uniform corrosion
that is derived from the statistic data of thickness
measurement for each location of bulk carriers.
In this study, Paik’s corrosion wastage model is
used, in which thickness diminution (r(t)) can be
estimated as follows,
⎧0 t ≤ tC
r(t ) = ⎨
C
⎩ 1 (t − tC ) t tC
(3)
where tc is a probability variable which represents
the life of anticorrosion coating, and t is a ship-
year. And C1 is a corrosion rate defined as a
probability variable of Weibull distribution. By
using the above corrosion wastage model, it is
possible to predict thickness reduction, r(t), for a
ship-year t. We assume that thickness of structural
members uniformly reduces by r(t) as the ship ages
(Figure 2). By applying thickness reduction to
the thickness of the shell plate, the web plate and
the flange in the cross section of the frame, it is
possible to account for the reduction of strength
(section modulus) by the uniform corrosion, when
strength is evaluated by using Equation (2).
2.2.2 Evaluation of load
Two kinds of bending moment (MS and MW)
in the beam model (Fig. 1(c)) are considered
that are caused by hydrostatic pressure (ps) and
2/18/2011 5:56:43 PM
 bs shown in Table 1. Figure 3 shows the definition
of design variables. As shown in this figure, only
the geometry of the cross section of the frame is
considered as design variables, and other dimen-
sions shown in Figure 4 are assumed to be auto-
ts matically determined from the design variables,
tw h Thickness hw (web-height). In Figure 3, the frame-space, S,
Reductionr is not a design variable but can be computed from
tf (t) the design variable, N, that represents a number of
frames in one cargo hold, by dividing the length
of the hold (lh) by N. It is noted that N is not a
continuous design variable, but is a discrete design
bf variable.

Figure 2. Effect of uniform corrosion. 3.1.2 Design constraint

In this optimization, we consider two types of
hydrodynamic pressure (pw), respectively. Then the constraint conditions.
load in the Equation (1) is defined as follows, (1) Constraint based on rules
We consider the following constraint introduced
ML(t) = xSMS + xWMW, (4) from the rules of the classification society (Class
where xS and xW are probability variables which NK, 2008).
represent modeling uncertainty. It is reported that hw
the accident of the hold frame structure generally < 60 (Web height (hw) and thickness (tw)) (5)
occurred when the bulk carrier was fully loading tw
iron ore (ClassNK, 1992), so that we assume that
the target bulk carrier is always in full load, and the
bending moment caused by hydrostatic pressure in Table 1. Principle dimensions.
the full loading condition (MS), can be assumed
as a definite value. On the other hand, we assume Ship length L 188 [m]
that the bending moment caused by hydrodynamic Breadth B 32.26 [m]
pressure, MW, is a probability variable. To derive Molded depth D 18.0 [m]
the probability distribution of MW, we assume Draft d 15.0 [m]
that the maximum bending moment, calculated by Gross tonnage G/T 35,750 [t]
applying the design pressure in the rule of the clas- Dead weight DWT 55.000 [t]
sification society (ClassNK, 2008), can be regarded Complement p 50 [persons]
as the extreme bending moment in 20 ship-years.
Then probability distribution of the peak of the
bending moment can be estimated. And finally,
S = lh N
extreme distribution of the bending moment (MW)
in one year can be estimated from the probability
distribution (Kawamura and Sumi, 2010). ts tw Design Variables
ts : Thickness of Shell Plate
hw tw: Thickness of Web Plate
3 OPTIMIZATION 1 (MINIMIZATION tf : Thickness of Face Plate
tf hw: Height of Web Plate
OF CONSTRUCTION COST)
bf : Breadth of Face Plate
N : Number of Frames in One Hold
3.1 Formulation of optimization problem bf

Firstly, we carried out the optimization of the

Figure 3. Design variables for optimization.
hold frame by minimizing the construction cost of
the structure. In this optimization, probability of
failure computed by the method described in the
previous section is introduced as constraint of the bottom end topend
(connected (connected to
optimization problem. hw
the wing
to the hop- 2.00hw lw 1.43hw
per tank) tank)
3.1.1 Design variables
Optimization is carried out for the hold frame Figure 4. Parametric definition of the dimensions of
structure of a handy-max sized bulk carrier the frame.

693

MARSTRUCT.indb 693 2/18/2011 5:56:46 PM

 CShl 2 < Z (Section modulus (Z)) (6) where Pfi is annual probability of failure at i-th
ship-year.
L ts ( i (ts )) (7)
3.1.3 Objective function
In this first optimization, objective function is
C′(0.03L + 7.0) < tw (Web thickness (tw) (C′ = 1.0))
defined as construction cost (Ctotal). We assume
(8) that the construction cost is composed of the steel
(2) Constraint about failure probability cost (Csteel) and the welding cost (Cweld) as follows,
Next, we consider that the computed probability
of failure by structural reliability analysis can be Ctotal = Csteel + Cweld (13)
used as design constraint for this optimization. To compute these costs, it is assumed that the
To define the acceptable limit of the failure steel cost per ton is ps = 80,000 [yen/ton], and the
probability for the structure, CIRIA proposed the welding cost per length is pw = 15,000 [yen/m],
following equation (Melcher, 1999), where [yen] is the Japanese currency, and 1 US$ is
about 100 yen when this study is carried out.
PfCIRIA = 10−4 ⋅ μ ⋅ tL ⋅ n−1, (9)

where μ is a factor which value differs by the type 3.2 Optimization by genetic algorithm
of the structure, tL is the life of the structure, and In this study, we developed a computer program of
n is a number of people who uses the structure. Genetic Algorithm (GA) for optimization proce-
Also, Bhattacharya et al. (2001) compared the dure. Because not only continuous design variables
target reliability of marine structures obtained but also a few discrete variables are included in the
from the past researches with the CIRIA’s criteria objective function, GA seems to be more suitable
and etc. However, these criteria is generally related than the general optimization methods that require
to the failure (accident) of the whole ship, while computation of derivatives. In this GA, 4 or 5 bit
our structural reliability analysis is just for a local length in a gene is assigned to each design variable.
part of the ship structure, so that the acceptable And it is assumed that each variable can take dis-
limit for this structural reliability analysis may crete values between the lower limit value and the
be less than the criteria for the whole ship. It can upper limit value as shown in Table 2. The number
be concluded that, it is very difficult at present to of genes in one generation is set as 40. Also in this
define an reasonable acceptable limit of the failure GA, rate of mutation is set as 0.05, and multiple-
probability for this structural reliability analysis. point crossover method at two points is used.
For this reason, we just assumed in this paper It is noted that a lot of structural reliability anal-
that the acceptable limit (PfN) of the structural ysis should be carried out during the optimization
reliability analysis can be defined based on the process, so that structural reliability analysis should
CIRIA’s criteria as follows, be called directly from the optimization program.
For this purpose, we use a numerical library named
PfN = kPfCIRIA, (10) NESSUS-PA (Southwest Research Institute (SRI),
2009) which is the library of commercial reliability
where tL, μ and n in the Equation (9) are assumed
software, NESSUS (Thacker et al., 2006).
as 30, 5.0 and 25, respectively. Also k is assumed
as 0.5. Then, the acceptable limit of the failure
probability (PfN) is calculated as PfN = 3.0 × 10−4. 3.3 Optimization result
And the constraint condition of this optimization
can be represented by the following equation. Figure 5 shows the result of the optimization.
It can be observed that the thickness and height of
PfL < PFN (11)
Table 2. Gene coding of design variables.
In this equation, PfL is the failure probability
in the lifecycle of the bulk carrier, and can be Design Lower Upper Gene
computed from the annual probability of failure variable limit limit length Interval
obtained from the structural reliability analysis.
N 27 42 4 1
In this paper, we use the following simple formula ts (mm) 13.0 20.5 4 0.5
to calculate PfL; tw (mm) 12.0 27.0 5 0.5
tf (mm) 10.0 41.0 5 0.5
tL i
∑∏( (12) hw (mm) 200.0 820.0 5 10.0
PfL Pf i − ) )Pfi bf (ww) 10.0 320.0 5 5.0
i =1 j =1

694

MARSTRUCT.indb 694 2/18/2011 5:56:48 PM

 S=1087(mm)
ts=20.5(mm) tw=13(mm) Design Variables
N=27
hw=360(mm) ts=20.5mm
tw=13.0mm
tf=27(mm) tf=27.0mm
hw=360.0mm
bf=180(mm) bf=180.0mm
Figure 5. Optimization result (Minimization of const-
ruction cost).
the web plate (tw, hw) is relatively small. This might
mean that reducing the size of the web is more
effective than reducing the sizes of the shell plate or
the flange in order to reduce the construction cost.
It is also deduced that the cost of welding becomes
lower by making the frame-space (S) higher with
a small number of frames (N). The computed
failure probability by using the Equation (11) is
2.92 × 10−4, which is almost same as the acceptable
limit, PfN, shown in the Section 3.1.2.
Though the failure probability computed
by structural reliability analysis is included in
the constraint condition, this constraint can be
regarded as the constraint for the bending strength.
However, the constraint related to the bending
strength is already included in the Equation (6)
which is attributed to the rule of the classification
society. Therefore, the constraint related strength
is doubly imposed to this optimization. It is possi-
ble to say that the optimization with the constraint
of failure probability may not produce significant
effect if the constraints based on the rule is already
included in the optimization.
4 OPTIMIZATION 2 (MINIMIZATION
OF LIFECYCLE COST (LCC))
4.1 Formulation of optimization problem
Recently, the method to evaluate the structure and
to create a plan of maintenance based on the life-
cycle cost including the risk of failure is proposed
(Kawamura and Sumi, 2009). Generally in ship
structural design, the relationship between the prob-
ability of failure (pf) and each component of LCC
can be represented as shown in Figure 6. In this
figure, the initial cost (construction cost, CI) is high
if the probability of failure (pf) is low, because a
lot of steel is necessary to construct a ship with
high strength. And the initial cost becomes lower
as the probability of failure becomes higher. On
the other hand, the maintenance cost (CM) and the
risk of failure (CRISK) may be low if the probability
of failure is low, and may become higher if the
695
MARSTRUCT.indb 695
Figure 6. Concept of structural design considering LCC.
Table 3. Definition of the design variable, M.
Design variable Feasible value Gene length
M 0, 4, 12, 16, 20, 24, 28 3
probability of failure becomes higher. Then in the
structural design considering LCC, optimal design
can be obtained by deciding the values of design
variables by minimizing the LCC. Based on the
above concept, we carried out optimization of the
hold frame structure.
4.1.1 Design variables
In this minimization of the LCC, we introduce a
design variable M, frequency of maintenance, in
addition to the design variables used in the previ-
ous optimization. Because we only consider the
uniform corrosion as the structural deterioration
effect by ageing, we also only consider re-coating
for protection of the corrosion as the maintenance
action. It is assumed that, after the re-coating is
carried out at a certain ship-year, progress of cor-
rosion is stopped for tc years during the re-coating
is effective. Table 3 shows the definition of the
design variable, M.
M can take the 8 values. If M = 4, 8 or 12, this
means that re-coating is carried out every 4, 8, or
12 years in the life of the ship (30 years). If M = 16,
20, 24 or 28, re-coating is carried out only once in
her life at the year of M, while no re-coating is
carried out when M = 0.
4.1.2 Constraint
In this optimization, the rule based constraint con-
ditions as defined in the previous Section (3.1.2 (1))
are considered, while the constraint about failure
probability (3.1.2(2)) is not used.
4.1.3 Objective function
Objective function of this optimization is the
lifecycle cost represented by the construction
2/18/2011 5:56:49 PM
 cost (Ctotal) defined in the previous section, the
maintenance cost (CMi) and the risk of failure (the
cost of failure) (CRISKi) as follows,
tL
LCC Ctotal + ∑ (CMi CRISKi )
i =1
As the annual maintenance cost (CMi) at i-th
year, only the cost of re-coating is accounted
for where the price of coating per unit area is
assumed to be 30,000 [yen/m2]. Also the annual
risk of failure (CRISKi) is computed by the following
equation,
CRISKi = αCFpfi
where pfi is annual failure probability computed by
structural reliability analysis, CF is expected loss of
damage of the ship, and α is the factor introduced
to consider the total failure probability of the
ship.
4.2 Optimization result
Figure 7 shows the result of the optimization.
In this solution of optimization, the trend is same
as the previous optimization in that the number
of frames is small and the frame space is large.
However, thickness and width of the flange and the
height of the web are relatively higher compared
with the solution in the previous optimization
(Fig. 5), so that the section modulus and the bend-
ing strength of this solution is higher. It is noted
that M = 0 is obtained for this optimum solution.
This means that re-coating should not be carried
out in the life of the ship. It can be concluded that
the structure with high strength without mainte-
nance action, where the construction cost is high,
gives the most effective solution when minimiza-
(14) tion of the LCC is considered. The structure with
low construction cost with good maintenance plan
does not become an optimum solution by using
this optimization strategy.
4.3 Sub-optimal solution
Figure 8 shows sampling points near the optimal
solution which are generated during the optimiza-
(15) tion process by the genetic algorithm. The hori-
zontal axis of the graph is failure probability at
30 ship-year, and the vertical axis represents the
LifeCycle Cost (LCC).
It can be observed that the distribution of the
points is convex downward, and that the optimal
solution is located at the minimal point. However,
it is noted that the distribution is multimodal and
not only the optimal point but also the sub-optimal
point exist as shown in the figure. The values of
the design variables of this sub-optimal solution is
shown in Table 4, where the sizes of the flange (tf, bf)
are smaller than the optimal solution. Instead,
the sub-optimal solution says that the mainte-
nance should be carried out once at 16 ship-year
(M = 16). Here it can be concluded that, other than
the optimal solution, almost optimal (sub-optimal)
solution can be obtained where the strength of
the structure is lower and the maintenance action
(re-coating) is required when LCC minimization is
carried out.

S=1087(mm)
¥30
LCC(×1.0E+07[Japanese Yen])

ts=20.5(mm) tw=13(mm) ¥29

Design Variables ¥28

N=27
¥27
hw=520(mm) ts=20.5mm
tw=13.0mm ¥26

tf=30.0mm Sub-Optimal Optimal

¥25
tf=30(mm) hw=520.0mm Design Design
¥24
bf=300.0mm 1.E-09 1.E-08 1.E-07 1.E-06 1.E-05
M=0(No recoating) Failure Probability at 30 ship-year (pf)
bf=300(mm)
Figure 8. Lifecycle cost and failure probability for the
Figure 7. Optimization result (Minimization of LCC). designs near the optimal design (Minimization of LCC).

Table 4. Optimal design and suboptimal design.

N ts [mm] tw [mm] tf [mm] hw [mm] bf [mm] M

Optimal design 27 20.5 13.0 30.0 520.0 300.0 0

Sub-optimal design 27 20.5 13.0 26.0 520.0 200.0 16

696

MARSTRUCT.indb 696 2/18/2011 5:56:50 PM

 5 OPTIMIZATION 3 (MAXIMIZATION Table 5. Reference ship and operational parameters.
OF LIFECYCLE BENEFIT (LCB))
DWTref [ton] 55,000
5.1 Formulation of optimization problem
Reference ship parameters
In the previous chapter, optimization of structural S [mm] 890 ts [mm] 16
dimensions and maintenance planning is carried tw [mm] 12.5 tf [mm] 16
out by minimizing the LifeCycle Cost (LCC) of the Hw [mm] 400 bf [mm] 140
bulk carrier. As a result, the most optimal solution Opertional parameters
is to design high-strength and maintenance-free RC [yen/(ton ⋅ voyage)] 1,920
structure. However, because the high strength Nr 9
structure means high weight of the ship, ship
owner may think that this kind of structures is
not preferable. For this reason, we carried out the
optimization by maximizing the LifeCycle Benefit S=1087(mm)
(LCB) in which not only the lifecycle cost but also
the operational revenue from the viewpoint of ship ts=20.5(mm) tw=13(mm)
owner is considered.
Design Variables
The objective function in this optimization is as
follows, N=27
hw=520(mm) ts=20.5mm
tL ⎛ tL ⎞ tw=13.0mm
LCB
C ∑ ROPEi − ⎜⎝Ctotal ∑ (CMi + CRISKi )⎟⎠ (16) tf=21.0mm
i =1 i =1 tf=21(mm) hw=520.0mm
bf=160.0mm
In this equation, ROPEi in the first term is annual M=16
operational revenue at i-th ship year, while the sec- bf=160(mm)

ond term is the lifecycle cost which is defined in

Figure 9. Optimization result (Maximization of LCB).
the previous optimization. Therefore, the LCB is
defined as the difference between lifecycle revenue
and the LifeCycle Cost (LCC). The annual opera- Optimal
281.6
LCB(x1.0E+08[Japanese Yen])

tional revenue is estimated as follows, 281.4

Design

(DWTT )
281.2
ROPEi reef + Wdifff RC × NR , (17) 281.0
280.8
where RC is defined as the revenue for the unit- 280.6
weight cargo per one voyage, and NR is the number 280.4
of voyages in one ship-year. DWTref is the dead- 280.2
weight of a reference ship (Table 5), which is 280.0

introduced to compute the increase of deadweight 1.00E-08 1.00E-07 1.00E-06 1.00E-05

(Wdiff) of the optimized ship from the reference
ship by saving the weight of the structure.
Stated another way, we assume that the amount
of cargo which can be loaded on the ship can be
increased by designing the hold frame structure
lighter, and that operational revenue can also
becomes higher as a result of weight saving.
It is noted that the design variables and con-
straint condition in this optimization is completely
same as the optimization in the previous chapter.
5.2 Optimization result
Figure 9 shows the optimized solution by maximiz-
ing LCB. It can be observed that the thickness and
the width of the flange become smaller compared
with the solution by minimization of the LCC.
By this change, the strength of the frame becomes
lower and the effect of corrosion to the reduction
Failure Probability at 30 ship-year (pf)
Figure 10. Lifecycle benefit and failure probability for the
designs near the optimal design (Maximization of LCB).
of strength may become larger. However in this
solution, strength of the frame might be kept enough
high by carrying out the maintenance (see M = 16 in
this solution in Fig. 9). Figure 10 shows the sam-
pling points near the optimal solution which are
generated during the optimization process by the
genetic algorithm. Unlike the Figure 8 in the pre-
vious optimization, sub-optimal solution does not
exist in this optimization.
By carrying out the optimization considering
not only the lifecycle cost but also the lifecycle
revenue, the optimized structure may have proper
strength with proper maintenance plan, which is
different from the simple high-strength structure.

697

MARSTRUCT.indb 697 2/18/2011 5:56:54 PM

 6 CONCLUSIONS
In this study, we proposed a strategy of “lifecycle
structural optimization” by using the structural
reliability analysis and the risk evaluation. Based
on this concept, the optimization is carried out for
the hold frame structure of a bulk carrier. Though
the analysis model used in the optimization is too
simple and rough for practical purpose, the strat-
egy proposed in this paper seems to have large
potential for future development of risk-based
design. The conclusions of this study are summa-
rized below.
1. We firstly carried out the structural optimization
in which objective function is defined as the
construction cost. The failure probability com-
puted by structural reliability analysis is included
as a constraint condition. It is possible to say
that the constraint of failure probability might
have same meaning as the constraints of the
rule related to the strength of the frame, so that
the constraint of failure probability may not
produce a significant effect to the optimization
results.
2. In the second optimization where minimization
of LifeCycle Cost (LCC) is carried out, we
consider the risk of failure as a part of objec-
tive function (LCC). Moreover, an additional
design variable related to the maintenance plan
(frequency of re-coating) is introduced. The
solution of this optimization indicates that high
strength structure without maintenance (re-
coating) is the most effective from the viewpoint
of LCC.
3. In the minimization of LCC, almost optimal
(sub-optimal) solution can be obtained. In this
suboptimal solution, though the strength of the
structure is lower than the optimal structure, a
proper maintenance (re-coating) is planned as
the sub-optimal solution.
4. In the last optimization, we carried out the
maximization of the LifeCycle Benefit (LCB).
By considering not only the lifecycle cost
but also the lifecycle revenue, the optimized
structure may have proper strength with proper
maintenance plan, which is different from the
simple high-strength structure obtained by
the minimization of the construction cost or the
minimization of the LCC.
5. As there are many assumptions in the definition
of uncertainties in the structural reliability
698
MARSTRUCT.indb 698
analysis and in the objective functions of
the optimization, it might be necessary to
evaluate these uncertainties more accurately
for practical application of “lifecycle structural
optimization”. Also if the curve around the
optimal point as shown in Figure 10 is rather
flat, the effect of the assumptions in the uncer-
tainties to the result of optimal design should be
properly evaluated.
ACKNOWLEDGEMENTS
This work is supported by the Ministry of Educa-
tion, Science, Sports and Culture, Grant-in-Aid for
Scientific Research (B) 20360391. Also this work is
supported by Kawasaki Shipbuilding Corporation.
The authors are grateful for these supports.
Appreciation is also extended to Professor Sumi in
Yokohama National University who gave us a lot
of advises for this research.
REFERENCES
Bhattacharya, B. et al. 2001. Developing target reliability
for novel structures: the case of the Mobile Offshore
Base, Marine Structures, 14, pp. 37–58.
ClassNK. 1992. Investigative Report about the Accident
of Aged Large Bulk Carriers (in Japanese).
ClassNK. 2008. Rules and Guidance for the Survey and
Construction of Steel Ships.
Kawamura, Y., Nishimoto, M. & Sumi, Y. 2009. A study
on a method for maintenance of ship structures
considering remaining life benefit, In Guedes Soares &
Das (eds), Analysis and Design of Marine Structures
(Proc. Marstruct2009), 2009, pp. 279–289.
Kawamura, Y. & Sumi, Y. 2010. Comparison of Two
Different Hold Frame Structures of a Bulk Carrier
Based on Structural Reliability Analysis, Proc. 11th
International Symposium on Practical Design of
Ships and Other Floating Structures (PRADS 2010),
pp. 1421–1431.
Melchers, R.E. 1999. Structural Reliability Analysis and
Prediction, John Wiley & Sons.
Paik, J.K. et al. 2004. A time-dependent corrosion wast-
age model for seawater ballast tank structures of ships,
Corrosion Science, 46, pp. 471–486.
Southwest Research Institute (SRI) 2009. NESSUS-PA
Users’ Manual, Version 0.5, November, 2009.
Thacker, B.H. et al. 2006. Probabilistic engineering analy-
sis using the NESSUS software, Structural Safety, 28,
pp. 83–107.
2/18/2011 5:56:58 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

A study on long-term prediction of corrosion wastage

T. Matsukura & Y. Kawamura

Department of System Design for Ocean-Space, Yokohama National University, Yokohama, Kanagawa, Japan

Eeteng Khoo
American Bureau of Shipping, Houston, Texas, USA

ABSTRACT: Corrosion is one of the most commonly found degradation mechanisms on marine
vessels. It is inevitable and can compromise the structural integrity if not properly addressed and
monitored. In extreme cases, it may have an impact on the safety of the vessel. Hence, proper and
wise corrosion management is essential. Many corrosion wastage models have been proposed by
researchers. One of the most important challenges to develop wastage models is consideration of
uncertainties within corrosion data. These uncertainties are generally dependent upon the environ-
ment, effectiveness of corrosion protection, or accuracy of the thickness measurement. The objective
of this study is to investigate these uncertainties involved and develop a process of generating corro-
sion wastage model. Characteristics of existing approaches are also reviewed and assessed while
developing a new model.

1 INTRODUCTION 1.2 Scope of work

1.1 Background
In recent years, experts in the marine industry
have recognized the need for developing a method
to predict corrosion wastage to assist in vessel
maintenance planning. Prediction models have
been proposed to monitor and predict the wast-
age condition of a structural component (Paik
et al. 2003, Guo et al. 2008, Yamamoto & Ikegami
1998, etc). The models are based on different
assumptions and methodologies. Since the inte-
gration of the models has yet to be done, there is
still confusion in the application of the models.
The main factor that makes the difference of cor-
rosion prediction appears to be large variability
in corrosion data that related to both thickness
measurement data and corrosion mechanism
itself. Therefore the greatest challenge in devel-
oping wastage prediction models is the uncer-
tainties in the thickness measurement data.
In this paper the uncertainties involved in
thickness measurement data and the characteris-
tics of existing approaches are investigated.
The process of data management for the wastage
prediction will be developed which has not clearly
been presented in previous research. The find-
ings will be used in the development of a model
to predict corrosion wastage from tanker meas-
urement data.
This study involves the collection and screening of
thickness measurement data, the development of a
corrosion wastage model using the screened data,
and probabilistic representation of the corrosion
model. First thickness measurement data are col-
lected from gauging reports in order to identify the
uncertainties. Then a method for processing these
data is established with the objective to propose a
prediction model using less scattered data. The
process includes the screening of data that are identi-
fied as a factor skewing the statistics of corrosion
wastage. In the final phase, comparative analysis of
existing corrosion models and the newly developed
model are conducted and the key factors contribut-
ing to the model differences are investigated.
2 CORROSION DATA COLLECTION
AND DATA SCREENING FOR
PREDICTION OF WASTAGE
2.1 Problems with existing database
The primary requirement for predicting corrosion
wastage is to obtain not only a reliable set of cor-
rosion data but also a sizable amount of data. Due
to the difficulty in simulating long-term real seawa-
ter condition in the laboratory, corrosion data
obtained from gauging reports is a better option

699

MARSTRUCT.indb 699 2/18/2011 5:56:58 PM

 Table 1. Summary of existing database for tankers.

Organization ABS NK KR TSCF

Ship type Single hull tanker Single hull tanker Single hull tanker Single hull tanker
Vessels 140 197 230 54
Measurement records 110,082 >250,000 33,820 Not known
Ship size (m or DWT) 168–401 m 100–400 m Not known 150,000–300,000 DWT
Ship age (years) 12–26 −23 12–26 −25
Build years Mostly 1970’s Not known Not known 1960’s (16)
1970–1975 (38)
Measured 1992–2000 Not known Not known Not known
Data source SafeHull condition Gauging records Gauging records Owner, class
assessment
Gauging reports 157 346 Not known Not known

for verifying corrosion wastage models. Existing

corrosion database are summarized in Table 1.
The data can be used for the prediction of corro-
sion wastage through statistical analysis.
However, there are uncertainties in the thickness
measurement data collected from a fleet of vessels,
in addition to the uncertainty in the corrosion
mechanism itself. Suspicious uncertainties that
have been reported in previous research[1][2] are
listed below:
i. Various service or corrosive environments of
each vessel
ii. Structural members or coating type differences
iii. Including the data of a new plate due to repair
work
iv. Measurement error
v. Uncertainty of initial thickness (i.e., gap
between actual thickness of plates and its design
scantlings due to manufacturer’s margin).
2.2 Data collection and preparation
for prediction of wastage
In this study, thickness measurement data of a sin-
gle tanker taken from the database provided by
ABS are used for the development of the long-term
corrosion model. This is to eliminate the uncer-
tainty caused by service environments, as indicated
in Section 2.1.
The collected raw data have to be filtered before
being used for analysis purposes. For example, as
general corrosion is the focus of the analysis, pit-
ting and grooving corrosion data are not included.
This is due to their locally advanced and relatively
high wastage. The data also contains measurement
data taken on plates that were replaced by a new
plate during the vessel’s service life. Gauged data
that indicate negative or zero wastage are also
removed, as these would skew the statistics of cor-
rosion wastage of the vessel.
Measurement Data Collection
•Gauging reports
•Repair reports
Data Categorizing
(in accordance with corrosion environment)
•Compartment type •Location
•Structural type
Screening
•Remove repaired member data
•Remove negative & zero wastage
•Remove pitting/grooving
Data Analysis
•Verification with corrosion model
Figure 1. Process flow of thickness measurement data
management.
Figure 1 illustrates the flow of data manage-
ment proposed for this study. As stated above,
thickness measurement data are to be reviewed
appropriately before the analysis phase. The proc-
ess includes categorization and screening of thick-
ness measurement data using the gauging reports
and repair reports. First, thickness measurement
data are collected from the gauging reports. The
repair reports indicate structures that have been
repaired and the time when each repair took place.
These details are used to screen out those repaired
members’ records. Before the screening phase data
should be categorized in accordance with corro-
sion environment. For example, the data were cat-
egorized by structural member type such as upper
deck plate, side shell plate, bottom plate, stiffeners
and so forth. The data were also categorized by
tank location and compartment type.

700

MARSTRUCT.indb 700 2/18/2011 5:56:59 PM

 Figure 2 illustrates a scatter plot of corrosion

Corrosion wastage (mm)

Plate A
wastage on the deck plate in a sample cargo oil tank 3.0 Plate B
before data screening. Here, vertical axis shows the + Plate C
amount of wastage, while horizontal axis shows the 2.0 Plate D +

age of the vessel. The negative values observed in Mean +

Figure 2 indicate the remaining thickness of the
gauged plate is larger than its design thickness, which
may be caused by rust on the surface or greater initial
thickness. The pink and green lines describe the wast-
age of two deck platings. Repair reports stated that
the plates had been cropped and renewed at age 27,
causing the drop seen in the figure. In accordance
with the management process shown in Figure 1,
these negative values and the data of renewed plates
are to be screened out from the raw data.
Since there could not be found significant differ-
ence of wastage distribution by tank location the
data in all cargo oil tanks were compounded in order
to enrich the data size. Figure 3 shows the scatter
plot of corrosion wastage on the main deck plate of
all cargo oil tanks of the tanker after screening.
There are 3,387 measurement points in total. Mean
value and mean + standard deviation of wastage at
several gauged years are also shown in this chart.
From Figure 3, the data appear to have a large
variation and the wavy pattern progress of “Mean”
can be observed. Since the plates are located over a
wide area of the vessel, data for a single plate was
also collected by taking the average of all the gaug-
ing points in a plate.
+
1.0 + +
+
0.0
0 5 10 15 20 25 30
Age of vessel (year)
Figure 4. Corrosion wastage progress on a single plate
in No. 4 center cargo oil tank.
The average of the wastage on four deck plates
in No. 4 center cargo oil tank is shown in Figure 4.
In this figure, mean wastage is indicated as a red
line and a wavy pattern trend of wastage is still
observed.
From the above discussion, a large variability
and wavy pattern of mean wastage are still present
in post-screening phases, i.e., where repaired plates’
data, pitting and grooving corrosion data, negative
and zero values, have been screened out. Also, the
data is highly scattered even on a single plate.
Therefore, a large variability in thickness measure-
ment appears to exist. This is likely due to the une-
venness of the corroded surface since it is virtually
impossible to gauge the thickness of the exact same
area at each survey.

3 CORROSION WASTAGE MODEL

3.1 Consideration of existing formulas

Figure 2. Scatter plot of corrosion wastage on deck
plate in No. 3 cargo oil tank of the tanker.
By using the screened data obtained through the
process detailed in the previous section, a formu-
lated curve that represents corrosion progress in
terms of age of vessel was derived. The least square
method based on weighted regression was used to
determine the curve that best “fit” the data.
Typical formulas applied in existing corrosion
wastage models follow:
d (t ) = a (t − c ) t > τC
(1)
d (t ) = 0 t ≤ τC

d (t ) = a (t − t )b t > t0
(2)
d (t ) = 0 t ≤ t0

⎡ ⎛ t − τc ⎞ ⎤
d (t ) = d∞ ⎢1 − exp ⎜ − ⎥, t > τc
⎣ ⎝ τ t ⎟⎠ ⎦ (3a)
d (t ) = 0 t ≤ τC

d∞
Figure 3. Scatter plot of corrosion wastage on deck τt = , d∞ d0 (3b)
plate in cargo oil tanks of the tanker. tan α

701

MARSTRUCT.indb 701 2/18/2011 5:56:59 PM

 d (t ) = a (t − c − t )b , t > τc + τt
(4)
d (t ) = 0 t ≤ τc + τt
where d(t) = amount of wastage, t = age of vessel,
t0 = corrosion initiation time, τc = coating life,
τt = transition time between coating durability and
corrosion initiation, d∞ = maximum wastage, α is
defined by d∞ and τt, d0 = initial plate thickness and
a, b = coefficients determined by verification with
measurement data (a is known as corrosion rate).
Eq. (1) proposed by Paik et al. (2003) represents cor-
rosion progress as a linear formula by defining the
constant corrosion rate. Eq. (2) by Guo et al. (2008)
represents a nonlinear formula that defines the time-
variant corrosion rate. The exponential model pro-
posed by Garbatov et al. (2007) is represented by
Eq. (3). The main distinction of this model com-
pared with the others is that the corrosion wastage is
assumed to converge at maximum wastage, d∞, dur-
ing the service life it is defined as smaller value than
the initial plate thickness. Yamamoto & Ikegami
(1998) proposed Eq. (4), which defines coating life,
τc, and transition time,τt, as random variables. This
formula may be more theoretical as coating life, τc,
and transition time, τt, are taken into account.
Although not included above, Melchers (2009) has
proposed a tri-linear model that is based on the
theory of oxidizing surfaces in the processes govern-
ing oxidation in the first few years.
Figure 5 schematically illustrates these corro-
sion wastage models. Here, the model proposed by
Melchers is illustrated as curve (5). Since the coef-
ficients of the models are related to the start of
corrosion (such as coating life, τc, and transition
time, τt), the initiation of corrosion, t0, needs to be
determined from the thickness measurement data.
These values can be assumed from the trend of
corrosion progress.
The corrosion rate in the early phase is known to
be relatively high and decreases over time. The linear
model (curve (1) in Fig. 5) can represent corrosion
progress more simplistically since it depends prima-
rily on the value of τc. Because τc is difficult to deter-
mine based on thickness measurement data, this
Figure 5. Schematic of existing corrosion wastage models.
702
MARSTRUCT.indb 702
model is not to be applied in this study. The tri-linear
model (curve (5) in Fig. 5) may not be appropriate
for this study since its objective is short-term predic-
tion of corrosion progress. In this study, Eq. (2) was
chosen, considering both simplicity of the formula
and consistency with corrosion data obtained. This
formula has also been commonly applied in the
industrial field for long-term prediction (Guo et al.
2008, Yamamoto & Kobayashi 2005).
As such, Eq. (2) is applied for the curve fitting
with mean and mean + standard deviation of the
corrosion data.
3.2 Derivation of corrosion model
by means of curve fitting
Figure 6 shows the result of curve fitting to the cor-
rosion data presented in Figure 3 by applying the
least square method. In addition to mean and
mean + standard deviation of wastage, 50 percentile
and 75 percentile of the data are also shown in the
chart. Here, corrosion initiation time t0 of each
model is determined from the trend of the wastage.
Corrosion models are to be developed by taking
into account the data size at each gauging survey.
At some ages, the amount of gauging data may be
limited and can be ignored when curve fitting is
carried out. Furthermore, to make the corrosion
wastage model more conservative and reliable,
the weighted regression method was applied.
The trends are presented by the ship ages where the
data indicates more severe wastage.
At a particular age (for example, 15 years old
shown as dotted area (a) in Fig. 6), the gap between
mean (data mean) and 50 percentile (50%) of the
data is relatively large. It is attributed to the fact that
a few extremely large values had skewed the statis-
tics of the corrosion data. Thus at those ship ages
(here, ship age 15 in Fig. 6), it is preferred that the
corrosion progress model for mean wastage (curve
mean) passes between mean (data mean) and
50 percentile (50%), and the model for mean +
standard deviation (curve mean + sd) passes between
mean + standard deviation (data mean + sd) and
75 percentile (75%).
Figure 6. Curve fitting to the mean and mean + standard
deviation of corrosion wastage, together with 50 and
75 percentile of the data.
2/18/2011 5:57:02 PM
 4 PROBABILISTIC REPRESENTATION Table 2. Results of Chi-square goodness of fit test to
several probability distribution forms with the data.
4.1 Probability density function
Ship age Best fit 2nd best fit 3rd best fit
As mentioned, there are many uncertainties in cor-
rosion data, so probabilistic representation is to be a) Center cargo oil tanks
applied to the model. Yamamoto & Kobayashi 11 Weibull Lognormal Gamma
(2005) defined corrosion initiation time as a ran- 15 Lognormal Weibull Gamma
dom variable that follows lognormal distribution. 18 Weibull Lognormal Gamma
However, the corrosion initiation time of the model 21 Weibull Lognormal Gamma
proposed in this study was assumed to be a con- 26 Weibull Gamma Lognormal
stant value because of the lack of the data that spe- 29 Weibull Gamma Lognormal
cifically states corrosion initiation. Paik et al. 33 Lognormal Gamma Weibull
(2003) assumed a constant corrosion rate to follow b) Wing cargo oil tank
Weibull distribution. Since the equation compen- 11 Gamma Lognormal Exponential
sates for a limited number of data, the constant 15 Weibull Lognormal Gamma
corrosion rate has been regarded as a convenient 18 Gamma Lognormal Exponential
way to grasp the trend of corrosion progress. It is, 21 Lognormal Weibull Gamma
however, not appropriate to the nonlinear model 26 Gamma Lognormal Weibull
applied in this study, as it cannot represent time- 29 Gamma Normal Weibull
variant corrosion rate. Guo et al. (2008) catego- 33 Gamma Normal Lognormal
rized the data by ship age and investigated the
best-fitted probability distribution forms to the
histograms of the data obtained. As a result,
Weibull distribution was applied to represent the for all ship ages and compartments. However,
probability density function of corrosion wastage Weibull distribution appears to be a good candi-
at every vessel age. date for representing the corrosion wastage on
In the same way as with Guo et al. (2008), corro- center tanks over the vessel’s service life. Gamma
sion data was categorized by vessel age and then distribution appears to be another candidate for
probabilistic distributions such as Weibull and wing tanks. Since the data obtained are based on a
Lognormal were fitted to the data by applying single vessel and details of the tanks are limited, it
either method of moments, maximum likelihood is difficult to calculate the difference in corrosive
estimates and least squares estimates. From these environment that affects the distribution forms in
methods, the least computationally intensive meth- center tanks and wing tanks. Therefore, Weibull
ods were used. Several probabilistic distribution distribution was applied for probabilistic represen-
forms were investigated by applying the Chi-square tation in this study, i.e., corrosion wastage at each
test to determine the most suitable distribution vessel age was assumed to follow the Weibull
form. Figure 7 shows the histogram of corrosion distribution.
wastage of deck plate in center cargo oil tanks of
the tanker at age 21. Three best-fitted probability
density functions were ranked by the Chi-square 4.2 Probabilistic representation by means
test. Table 2 also lists the results of the Chi-square of Weibull distribution
test to other ages and compartment types.
From the results shown in Table 2, no probabil- The probability density function of Weibull distri-
ity density function can consistently fit the data bution can be expressed by the following equation:

⎛ k⎞ x
k −1 ⎡ ⎛ x⎞k⎤
f x ( x, k , λ ) = ⎜ ⎟ exp ⎢ ⎥ , x, k , λ > 0
⎝ λ⎠ λ ⎢⎣ ⎝ β ⎠ ⎥⎦
(5)

The following approximate expression (Halder &

Mahadevan 2000) can also be used to determine
the shape parameter k and scale parameter λ in
Eq. (5) by means of mean μ and standard
deviation σ.

Figure 7. Histogram of corrosion wastage on deck −1.08

⎛σ ⎞ μ
plate in center cargo oil tanks of the tanker at ship age 21 k=⎜ ⎟ , λ= (6)
with three best fitted probability distribution forms. ⎝ μ⎠ Γ( k+ )
703

MARSTRUCT.indb 703 2/18/2011 5:57:03 PM

 By using Eqs. (5) and (6), a probabilistic model
can be developed from statistical values such as
mean and standard deviation. As presented in
Figure 6, the corrosion progress models for “mean”
and “mean + standard deviation” predict the wast-
age value at vessel age where gauging was not per-
formed. Thus, the models derived in Section 3.2
allow probability density function representation
over the vessel’s life through Weibull distribution.
A three-dimensional corrosion wastage model is
generated and shown in Figure 8. This represents
the probability density function which follows the
Weibull distribution of corrosion wastage d(t) on
deck plate in cargo oil tanks of the tanker over the
vessel age t.
By using the procedure described above, it is
possible to derive a probabilistic formula by using
gauging reports. This allows engineers to perform
probabilistic prediction of corrosion wastage over
the vessel’s service life for the purpose of risk- and
reliability-based inspection planning.
5 COMPARATIVE STUDIES
The process of deriving a corrosion wastage model
in this study was developed through the evaluation
of existing approaches. The corrosion progress
model for mean wastage obtained in this study
(see Fig. 6) is compared to the existing models
based on the data taken from a fleet of vessels, as
illustrated in Figure 9. The details of the data used
for each model are listed in Table 3. Note that
unlike existing models, the data used in this study
is taken from a single vessel. For comparison pur-
poses, several existing models especially for upper
deck plate in the cargo oil tank of tankers are
featured. Here, the corrosion rate for the same
structural member and compartment type pro-
posed by TSCF (1997) is also shown. The upper

Figure 9. The comparison of existing corrosion wastage

Figure 8. 3D corrosion wastage model for upper deck
plate in cargo oil tank of the tanker.
models based on a fleet of tankers’ data and the model
obtained in this study based on a single tanker’s data.
*The upper and lower limit of corrosion rate are shown
as an alternative (TSCF 1997). *50% and 75% are also
shown (Yamamoto & Kobayashi 2005).

Table 3. Summary of existing corrosion models for upper deck plate in cargo oil tank of tankers.

Corrosion model Paik et al. (2005) Garvatov et al. (2007) Guo et al. (2008)

Vessels >100 tankers <140 <140 tankers

tankers
Member Upper deck plate in Upper deck plate in Upper deck plate in
cargo oil tanks cargo oil tanks cargo oil tanks
Total number of data used About 5000 4655 4655
Corrosion initiation time 7.5 11.49 6.5
Corrosion model Yamamoto & Kobayashi (2005) TSCF (1997) This study
Vessels 60 tankers >32 tankers 1 tanker
Member Upper deck plate in Upper deck plate in Upper deck plate in
cargo oil tanks cargo oil tanks cargo oil tanks
Total number of data used 38856 Unknown 3387
Corrosion initiation time Follows Weibull distribution Uncoated 6.5 years
*mean = 9.114

704

MARSTRUCT.indb 704 2/18/2011 5:57:05 PM

 and lower limits of the standard are indicated in
the chart. The existing corrosion models in Figure 9
were previously introduced in Section 3.1, except
for the model by Yamamoto & Kobayashi (2005).
This is a newly proposed model based on Eq. (2),
while Eq. (4) was used in previous research
(Yamamoto & Ikegami, 1998). The curve of
“50 percentile” and “mean” of their model are
both shown in this chart for reference purposes.
In Figure 9, the corrosion rate, 0.03∼0.10 mm/year,
stated by TSCF (1997) covers most of the pre-
dicted values of the models. There is, however, a
high level of difference between the estimated wast-
age of each model. The model proposed in this
study and Guo et al. (2008) show relatively rapid
corrosion progress compared to other models.
It may be caused by the fact that the weighted
regression carried out in these studies leads to a
more conservative estimation.
6 CONCLUSIONS
Thickness measurement data of a single vessel
were collected and managed through categorizing
and screening in order to take into account the
uncertainties involved. By using the filtered data,
corrosion progress was formulated along with the
consideration of methodology and assumptions of
existing approaches. A probabilistic formula of the
model was then derived. This model provides engi-
neers with a probabilistic prediction of wastage
and the ability to perform a more rational way of
inspection planning. The results and discussions in
this study are summarized below.
1. Large variability of wastage values remain even
when the data are based on a plate. This shows
the impact of uncertainty caused by gauging of
corroded surface, as it is almost impossible to
gauge the same point at each survey.
2. Since significant differences between existing
corrosion wastage models exist, further investi-
gation and development of a corrosion model
and database are required. As for the models
proposed by Guo et al. (2008) and this study,
they were derived based on the weighted regres-
sion method taking into account data size at
each survey and severity of data. As these mod-
els show a relatively conservative estimation as
opposed to the other models, such data analysis
may be needed in order to not underestimate
the amount of corrosion wastage.
705
MARSTRUCT.indb 705
3. The corrosion wastage model proposed in this
research requires a continual database update to
make it applicable to other tankers, as the model
in this research was obtained based on data col-
lected from a single vessel.
Producing a long-term prediction of corrosion
wastage requires proper analysis of corrosion data
which contains a variety of uncertainties and
unreliability, especially when the data size is limited.
The process may include data screening and appli-
cation of the weighted regression method.
ACKNOWLEDGEMENT
The first author was in residence at ABS Corpo-
rate Technology from October 2009 to February
2010. The reported work is a snapshot of the first
author’s work at ABS. We sincerely appreciate the
opportunity for the internship, the financial sup-
port and the guidance concerning the research
work that ABS provided.
REFERENCES
Garbatov, Y., Guedes Soares, C. & Wang, G. Non-linear
time dependent corrosion wastage of deck plates of
ballast and cargo tanks of tankers, Journal of off-
shore Mechanics and Arctic Engineering. (2007);
129(1):48–55.
Guo et al. Time-varying ultimate strength of aging tanker
deck plate considering corrosion effect, Marine Struc-
tures 21 (2008), 402–419.
Haldar, A. & Mahadevan, S. Probability, Reliability, and
Statistical Methods in Engineering Design, John Wiley
and Sons, Inc., New York (2000).
Melchers, R. Validity and quality of deterioration models
for structural reliability assessment, Structural Lon-
gevity Vol. 1, No.1 (2009) 17–36.
Paik, J.K., Wang, G., Thayamballi, A.K., Lee, J.M. &
Park, Y.I. Time-dependent risk assessment of aging
ships accounting for general/pit corrosion, fatigue
cracking and local denting damage, Transactions
SNAME (2003), Vol. 111, 159–197.
Tanker Structure Co-operative Forum (TSCF). Guidance
manual for tanker structures, Witherby & Co. Ltd.
(1997).
Yamamoto, N. & Ikegami. K. A study on the degradation
of coating and corrosion of ship’s hull based on the
probabilistic approach, Journal of Offshore Mechanics
and Arctic Engineering (1998).
Yamamoto, N. & Kobayashi, H. Corrosion progress predic-
tion system for ship structural members, Transactions of
Nippon Kaiji Kyokai (2005), vol. 274, 1–9.
2/18/2011 5:57:06 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Optimization of composite maritime structures—effects of uncertainties

on design criteria and limits

L. Sanchez & J.W. Ringsberg

Department of Shipping and Marine Technology, Chalmers University of Technology, Gothenburg, Sweden

E. Johnson
Department of Building Technology and Mechanics, SP Technical Research Institute of Sweden, Borås, Sweden

ABSTRACT: Composite materials have been used successfully in vessels since the 1960s. Changes
during the last decade in international regulations, fuel prices and environmental concerns have dramati-
cally increased the interest of the marine industry in lightweight composite materials. There are many
questions regarding the material’s properties and behaviour that still remain, leading, in some instances to
too robustly designed crafts. Therefore, the new generation of composite vessels requires a finer tuning of
safety and weight, which can be achieved by analyzing thoroughly the behaviour of the material and the
effects of uncertainties and approximations found in the design process. This paper presents the study of
two design methodologies with different degrees of complexity, through deterministic and probabilistic
analyses, in order to determine their benefits, drawbacks and challenges.

1 INTRODUCTION 1.2 Examples of composite shipbuilding projects

1.1 Background
The political discussion on climate changes and
environmental regulations, together with raised
fuel prices and shipping’s competition with other
means of transportation pressures the shipping
industry to change. A first step is to employ energy
efficiency studies, which are of outmost impor-
tance to ensure the viability of newly designed
vessels. The computation of an “energy efficiency
index” aims to raise awareness of the shipbuilding
efficiency of the proposed ship against the actual
ship’s performance so that ship owners can gauge
and set targets for improvement over time.
Composite materials in ship structures can
be used as a mean to increase energy efficiency:
a reduction on the structure weight renders reduced
fuel consumption or an increase of the payload
capacity. Additionally, the price of composite mate-
rials is going down, making them a competitive
option compared to steel and aluminium materials.
However, although knowledge has been developed
on how to design large parts of merchant vessels
in composite materials, there are challenges and
barriers in class rules that have to be dealt with.
The opening is SOLAS regulation 17, which gives
naval architects the possibility to design composite
marine structures for non-High Speed Light Craft
(HSLC).
The composite shipbuilding era has just begun
for the next generation of large merchant vessels.
This era was preceded by several successful exam-
ples of medium-sized vessels manufactured in
various types of composite materials. For several
decades, minesweeper hulls have been made of
Glass-Reinforced Plastics (GRP) to reduce its
chances of detonating mines; it must be sound-
proofed to reduce its acoustic signature and be
made of a non-ferrous material to reduce its mag-
netic signature. The Visby Corvette was designed
by the Swedish Defence Materiel Administration
(FMV) and built by Kockums AB in Sweden. The
first ship of the class was launched in 2000, and the
hull is constructed with a sandwich design consist-
ing of a PVC core with a carbon fibre and vinyl
ester laminate. The Swedish project LASS (Light-
weight Construction Applications at Sea) aimed at
improving the efficacy of marine transport and to
increase the competitiveness of the Swedish ship-
building industry. This was accomplished through
the development and demonstration of techniques
for using lightweight materials for ship construc-
tion (see www.lass.nu).
The large scaled integrating EU-project BESST
(Breakthrough in European Ship and Shipbuilding
Technologies) aims to achieve a breakthrough
in competitiveness, environmentally friendliness
and safety of European built ships. Focusing on

707

MARSTRUCT.indb 707 2/18/2011 5:57:06 PM

 passenger ships, ferries and mega-yachts, the results 2.1 Types of failure
will be, to a large extent, applicable also to other
FRP laminates are heterogeneous materials, which
ships. A holistic life cycle performance assessment
may exhibit a wide assortment of distinct types of
on ship level will guide the technical developments
damage. Damage in FRPs is usually observed in
on system level. The results will be integrated in
three levels or scales: micro (fiber/matrix), meso
virtual showcases (ship concepts) demonstrat-
(lamina) and macro (laminate). For simplification,
ing the technical solutions as well as the life cycle
the assessed methodologies in this investigation
impact compared to current designs.
distinguish only two types of damage as failure
and consider their consequences: Fiber Fracture
1.3 Current study (FF) and Inter Fiber Fracture (IFF).
Figure 1 illustrates both types of failure and
The maritime industry customarily works with
their modes. The FF is defined here as the breakage
methodologies accepted by classification societies
of a significant number of fibers. There are differ-
aimed to guarantee reliable designs. Nevertheless,
ent modes of fiber failure due to dominant com-
there is the possibility of developing and employ-
pression in the fibers direction: micro buckling,
ing new methodologies, provided that their validity
kinking and fiber fracture due to shear. However,
is demonstrated. The aim of this investigation is
in this investigation these modes are not recog-
to assess a current design practice for Fiber Rein-
nized and a generalized FF due to compression is
forced Plastic (FRP) laminates, and to determine
used. The IFF denotes the appearance of a frac-
the benefits, drawbacks and challenges of a more
ture on a plane parallel to the fibers in the lamina
sophisticated material model and methodology.
in three possible modes (A, B or C) depending on
The evaluation of both methodologies is done by
the loading conditions (Puck & Schürmann 1998).
means of deterministic and probabilistic analyses
It propagates across the lamina thickness and is
upon a finite element model of a quasi-isotropic
carbon fiber epoxy laminate subjected to a biaxial
in-plane loading.
The ideas and theoretical framework have been
taken from the literature. For example, Sutherland &
Guedes Soares (1997) present a review of proba-
bilistic models for composite material strength.
Nakayasu & Maekawa (1997) present a comparison
of failure criteria in probabilistic fields and stochas-
tic failure envelopes of composite materials. Jeong &
Shenoi (1998) present reliability analyses using a
direct simulation method, followed by probabilistic
strength analyses of FRP plates using the Monte
Carlo simulation method in Jeong & Shenoi (2000).
Murotsu et al. (1994), Frangopol & Recek (2003)
and Nader et al. (2008) are examples of other ref-
erences where assessment of the reliability of fiber
reinforced composites have been dealt with.

2 TYPES OF FAILURE AND STRUCTURAL

ANALYSES

This investigation examines two structural analyses

for FRP laminates: linear elastic and linear elastic
with progressive degradation. The former is the
most commonly used in the maritime industry
today due to its simplicity and execution time. The
latter is frequently used for sophisticated analy-
ses of components. From this point onward we
will refer to the structural analyses as models for
simplification. Before explaining the details of
both methodologies, it is necessary to describe the
types of failure in the laminate recognized by both
methodologies. Figure 1. Recognized types of failure.

708

MARSTRUCT.indb 708 2/18/2011 5:57:06 PM

 stopped by the fibers of the neighbouring layers of
different fiber directions.
IFF is treated as a type of failure that governs
the degradation of the laminate. Mode A occurs at
transverse tension, due to a combination of normal
tension and shear stresses and is directed at a right
angle to the lamina surface. It results in a degrada-
tion of the transverse modulus, the shear modulus
and the longitudinal compressive strength while the
longitudinal stiffness is unchanged. Modes B and C
occur at transverse compressive stresses combined
with shear stresses with B at lower normal com-
pressive stresses and C at higher normal compres-
sive stresses. Mode B occurs at a right angle to the
lamina while mode C occurs at an oblique angle.
Since the crack surfaces are pressed on towards
each other, the cracking according to mode B
results in a reduction of the shear modulus, but a
maintained transverse modulus. The wedge effect
caused by the oblique fracture present in Mode C
may cause a violent delamination followed by local
buckling of the outer layers. The degradation of
the Poisson’s ratio due to any of the possible modes
of IFF is debatable. In this investigation it is con-
sidered negligible and therefore disregarded.
2.2 Failure criteria
This investigation utilizes three failure criteria
relevant for the maritime industry, all of them
approved by the ship and offshore classification
society Det Norske Veritas (DNV): maximum
strain, FiReCo AS’ modified Tsai-Wu and Puck;
see Det Norske Veritas (2009).
The maximum strain criterion compares the
strain components separately against in total, for
plane stress analyses, five critical strength values
(longitudinal tension and compression, transverse
tension and compression and shear strength). It
arguably distinguishes between different types of
failure, FF for longitudinal failures and IFF for
transverse and shear ones. There is no coupling
between the different stress components. The
classification rules use this criterion only to check
for FF.
The FiReCo AS’ modified Tsai-Wu criterion
(Det Norske Veritas 2009) has the same formula-
tion as the well-known Tsai-Wu criterion, the only
Table 1. Summary of the failure criteria.
Criterion Type of failure
FiReCo AS’ modified Tsai-Wu Generalized lamina
failure
Maximum strain FF
Puck IFF
709
MARSTRUCT.indb 709
difference being that it uses modified in-plane
tensile and compressive transverse lamina strengths.
It takes into account the interaction between
stresses; however, it does not distinguish between
types of failure. The criterion only predicts general
lamina failure. Finally, the Puck criterion (Puck &
Schürmann 1998) distinguishes between all differ-
ent types of failure and their modes described in
subsection 2.1.
2.3 Structural analyses
Two structural analyses are performed in this
investigation: linear elastic and linear elastic with
progressive degradation. Both of the methodolo-
gies define the Ultimate Limit State (ULS) as FF or
IFF Mode C in any of the laminas of the laminate.
The residual hygro-thermal stresses are neglected
to simplify the calculations and to disregard their
effect on failure prediction.
In the linear elastic analysis, the representative
elastic modulus governs the stress-strain relation
until the ultimate limit state is reached. Yet, to
do so, it completely disregards IFF and its effect
on the redistribution of stresses in the laminate.
Because of this, the strength, and after IFF, the
stiffness of the laminate, are over predicted.
The linear elastic with progressive degradation
model uses degradation factors to decrease the
value of the transverse and shear elastic modules
(E2 and G12) of the laminas after IFF is predicted.
This is done to simulate the loss of stiffness and
the innate redistribution of stresses as a conse-
quence of the matrix damage. The degradation of
the longitudinal compressive strength is neglected.
The loss of stiffness in the laminas due to IFF
depends on the mode of fracture. Mode A indi-
cates a fracture in the matrix with a gap because
of the tensile stresses. Even though the transverse
and shear stiffness of the lamina are zero at the
crack, in the damaged lamina, the loss of stiffness
is smaller due to the redistribution of the stresses
owing to the adjacent laminas. A very conservative
approach would be to set the degradation factors
to zero when the inter fiber fracture is of mode A.
However, in this investigation, the degradation fac-
tors are adjusted to match the experimental data.
This methodology of representing degradation
does not portray accurately the phenomena, with
it, degradation is considered to be a sudden event
instead of a gradual one. In reality the IFF develop
throughout the laminas gradually as the load
increases until a state of ‘critical crack density’ is
reached. In the literature there are more advanced
methods for modelling degradation; see Garnich &
Akula (2009).
Puck’s failure criterion is used to predict IFF in
both structural analyses. The ULS is predicted with
2/18/2011 5:57:07 PM
 all three criteria: FF with the maximum strain, IFF
Mode C with Puck’s, and the generalized lamina
failure with FiReCo AS’ modified Tsai-Wu. Puck’s
criterion was not used to predict FF since its pre-
diction of the failure stress for plane stress sce-
narios differ only in a few percent from the ones
obtained with the maximum stress criterion, which
gives very similar results to the maximum strain
criterion.
3 ANALYSIS CASE
The behaviour of a composite laminate plate
subjected to a biaxial in-plane loading is used as
the mean of evaluating the methodologies. This
simple geometry can be considered as a representa-
tive part of a ship structure such as a plate in a free
sheet field. Figure 2 depicts the analysis case: a
(square) plate with side dimensions a × a, subjected
to distributed loads (PX and PY) around the plate’s
edges.
For the global coordinates, as depicted in
Figure 2, utilize the sub-indices x and y, while lam-
ina coordinates employ the sub-indices 1 and 2, 1
being the aligned with the orientation of the fibers.
The boundary conditions are set as freely supported.
The influences of other boundary conditions, such
as fixed supported edges, are left for a future work.
The material utilized for the modelling of the
laminate is AS4/3501-6 carbon/epoxy unidirectional
lamina. This material is selected because it is well
documented in the literature, and hence, there is
access to previous numerical and experimental
results to compare with; see Soden et al. (1998,
2002). The unidirectional lamina exhibits different
degrees of non-linearity, a slight convex one under
axial loading (σ1, ε1) and a strong concave one
under in-plane shear (τ21, γ21) and transverse
Figure 2. Geometry, loads and boundary conditions of
the analysis case.
710
MARSTRUCT.indb 710
Figure 3. Stress-strain relation of the AS4/3501-6
carbon epoxy unidirectional lamina properties.
compressive (σ2, ε2) loading. However, in this
investigation the behaviour of the material was
assumed to be always linear elastic, and there-
fore, to guarantee congruence in the stress-strain
calculations, the ultimate strains and stresses fol-
low the relation shown in Figure 3. The interac-
tion of σ2 and τ21 on their stress-strain diagrams
(Puck & Schürmann 1998) due to micro cracking is
neglected. The expectancy values of the mechani-
cal material properties of the unidirectional lamina
are presented in Table 2.
The analyzed laminate is a quasi-isotropic
one with a [90°/±45°/0°]S stacking sequence of
equally thick laminas. The overall thickness of the
modelled laminate is of 1.1 mm.
4 FINITE ELEMENT ANALYSIS
The response of the laminate is calculated with the
finite element software ANSYS (ANSYS 2009).
The maximum strain, FiReCo AS’ modified
Tsai-Wu and Puck’s failure criteria were imple-
mented in the software through a set of macros in
the post processing. This is done in order to start
developing tools and methodologies that can be
easily incorporated to the current design practice
of the maritime industry.
4.1 Description of the FE model
Due to the in-plane loading conditions and the
ratio between the laminate’s thickness and side
length, the laminate is considered to be in a plane
stress state. Shell281 layered elements are used
to represent the laminate, in it, each layer repre-
sents a lamina, which may have its own material
properties, thicknesses and principal directions
orientation; see ANSYS (2009). The mechani-
cal properties of the AS4/3501-6 carbon/epoxy
unidirectional lamina are assigned to each layer,
2/18/2011 5:57:07 PM
 Table 2. Mechanical properties of four AS4/3501-6 carbon/epoxy unidirectional laminas.

Property Expectancy COV Distribution

Longitudinal modulus, E1 (GPa) 126 5% Normal

Transverse modulus, E2 (GPa) 11 5% Normal
In-plane shear modulus, G12 (GPa) 6.6 5% Normal
Major Poisson’s ratio, ν12 0.28 10% Normal
Longitudinal tensile strength, X1T (MPa) 1739 10% Weibull
Longitudinal compressive strength, X1C (MPa) 1480 10% Weibull
Transverse tensile strength, Y2T (MPa) 48 10% Weibull
Transverse compressive strength, Y2C (MPa) 200 10% Weibull
In-plane shear strength, S12 (MPa) 79 10% Weibull
Longitudinal tensile failure strain, ε1T (%) 1.380 10% Weibull
Longitudinal compressive failure strain, ε1C (%) 1.175 10% Weibull
Transverse tensile failure strain, ε2T (%) 0.436 10% Weibull
Transverse tensile compressive strain, ε2C (%) 1.818 10% Weibull
In-plane shear failure strain, γ12U (%) 1.196 10% Weibull
*Characteristic tensile fracture strain, ε1Tch (%) 1.10 10% N/A

*Calculated assuming 10 samples, not used in the Monte Carlo simulations.

Figure 4. Comparison of experimental and calculated stress-strain curves.

4.2 Results from FEA

following the stacking sequence and thicknesses
described in Section 3. In the linear elastic model
these properties are kept constant, while for the
linear elastic with progressive degradation model,
the lamina properties are modified following the
methodology described in Section 2. The degrada-
tion factors are calibrated with the biaxial tensile
loading experimental data. The factors 0.4 and 0.5
are used for degrading the transverse tensile and
shear stiffness, respectively.
Figure 4 shows a comparison of the experimental
and calculated stress-strain curves of the laminate
under a biaxial tensile loading with a σY/σX = 2
ratio. The figure also shows the IFF and FF
stresses, predicted by our selection of failure
criteria, in addition to the lamina where the dam-
age was predicted. In this loading case, every IFF is
of Mode A. It is clear that the linear elastic model
over predicts the stiffness and strength of the

711

MARSTRUCT.indb 711 2/18/2011 5:57:08 PM

 laminate. As mentioned before, the linear elastic
model ignores the redistribution of stresses that
occurs after IFF. Thus, the lamina stress σ1 and
the strains are too small, while the other in-plane
stresses of the lamina (τ21 and σ2) are too high.
It is important to observe the difference of the
predicted failed layer between the maximum strain
and the FiReCo AS’ modified Tsai-Wu criterion.
The former predicts failure in the 90° laminas,
while the latter does in the ±45° ones. The differ-
ence is due to the formulation of the FiReCo AS’
modified Tsai-Wu criterion, which is a so-called
‘interactive criterion’ where all the in-plane stresses
affect the failure stress prediction.
The linear elastic with progressive degradation
model follows quite well the experimental curve,
however, there are some discrepancies worth men-
tioning. The inter fiber fracture on the ±45° laminas
is apparently well under predicted. The value of
the transverse ultimate stress of the lamina (σu2)
affects greatly the prediction of this point. This
difference could be attributed to the very question-
able assumption that the fracture stress found in an
isolated lamina is applicable on a lamina embed-
ded in a laminate, as well as incapability of our
model to simulate the gradual increase of the crack
density. The FiReCo modified AS’ Tsai-Wu crite-
rion greatly over predicts the fiber fracture stress.
Again, the interactivity of the criterion explains
the reason for the over prediction, the reduction
of the transverse and shear stresses in the lamina
after inter FF due to the degradation of the elastic
modules influences positively the prediction of the
ultimate state stress. Just after the maximum strain
criterion predicts fiber fracture a gradual soften-
ing of the laminate is observed in the experimental
curve. This softening can be considered to be due
to gradual fiber fracture. Overall, the linear elastic
with progressive degradation model is much more
accurate in comparison with the simple linear
elastic one.
5 SAFETY FACTORS
In order to guarantee a reliable design, classifica-
tion societies state partial safety and model factors
that determine, along with the material’s ultimate
strength or strain, the maximum allowable loading
conditions. Figure 4 depicts three maximum allow-
able strains, marked A, B and C, as recommended
by two different composite structure standards.
The maximum allowable strain A was calcu-
lated with a sole safety factor against FF stated in
DNV’s High Speed Light Craft and Naval Surface
Craft classification rules (2010). This safety factor
is applied to the mean measured failure strain, and
712
MARSTRUCT.indb 712
Table 3. Safety, load and model factors used to determine
the maximum allowable strains.
*Factor A B C
Safety factor against FF, R 3.3 – –
Load model factor, γSd – 1.0 1.0
Resistance model factor, γRd – 1.046 –
Combined load effect and – 1.7 1.7
resistance factor, γFM
*All factors are for ULS, brittle failure and high safety
class.
assumes that the failure strain of the component
has a coefficient of variation smaller than 7%. The
maximum allowable strains B and C were calculated
with the partial safety and model factors of DNV’s
offshore standard of composite components; see
Det Norske Veritas (2009). An important detail
is that the factors are applied to the characteristic
tensile failure strain (2.5% quantile); see Table 2.
Table 3 contains a summary of the employed fac-
tors in all three cases.
The maximum allowable strain A is just above
of the IFF strain for the 0° layers, while the maxi-
mum allowable strains B and C are well over the
IFF strains for the 0° and ±45° layers. Therefore,
the laminate will develop some degree of matrix
cracking.
6 VARIABILITY
A probabilistic analysis is performed to get a
deeper insight of the effects that the mechanical
properties variability has on the onset of IFF. This
is done through 1000 Monte Carlo simulations
of the linear elastic with degradation model. The
variables shown in Table 2 are considered to be
stochastic; they are given the statistical distribu-
tions and variations shown in the same table. The
statistical modelling of the distributions to the ran-
dom variables follows the principles presented in
among others Lekou & Philippidis (2008); see also
Sriramula & Chryssanthopoulos (2009) for good
examples on quantification of uncertainty mod-
elling in stochastic analysis of FRP composites.
Figure 5 shows the cumulative distribution func-
tions of the IFF strains for each layer orienta-
tion and the maximum allowable strains. One can
appreciate in the figure that the maximum allow-
able strains are well located in a region were IFF
will most certainly occur for the 0° and the ±45°
layers, and possibly also for the 90° layers. Also, it
is interesting to notice that the variation of the IFF
strain for the 90° layers is considerably larger than
the ones of the other IFF strains.
2/18/2011 5:57:08 PM
 Figure 5. Cumulative distribution functions of the IFF strains.
7 CONCLUSIONS
The linear elastic model with progressive degrada-
tion is a powerful tool for optimizing the employ-
ment of FRP laminates. It provides valuable insight
on the accumulation of damage in the laminate.
However, it requires input data that can only be
obtained through thorough experimental testing.
The finite element analysis and the probabilistic
analysis gave valuable input on the benefits, draw-
backs and challenges of utilizing a model with pro-
gressive degradation for analyzing FRP laminates.
Whether the benefits of the progressive degrada-
tion model out weights the inherent drawbacks,
such as the increased complexity and computing
time, depends greatly on two key factors: the onset
and progressive saturation of IFFs and the maxi-
mum allowable loading conditions defined by clas-
sification rules. Figure 5 clearly shows that IFF is
very likely to occur in a wide range of strains close
to the maximum allowable strains. However, it does
not provide information on the level of IFF satura-
tion and its effect on the laminate. Generally, IFF
can be considered to be harmless, but for example,
under cyclic loading small delamination zones in
the crack tips might grow and lead to delamination
as well as stress concentrations that promote FF.
It is clear that with this laminate, if the maximum
allowable loading conditions defined by class rules
are to be followed, the linear elastic with progres-
sive degradation model does not provide added
benefits to the structural analysis. The laminate’s
behavior can be simulated accurately enough with
the simple linear elastic model. However, if experi-
mental data containing the IFF onset strain for
embedded laminas and the IFF saturation curves
were available, the linear elastic with progressive
713
MARSTRUCT.indb 713
degradation model could provide valuable insight
for motivating the use of higher maximum
allowable loads.
ACKNOWLEDGMENTS
The authors acknowledge the support from
Lighthouse Maritime Competence Centre (www.
lighthouse.nu), Sweden. This work is part of a
deliverable by the authors to the EU-project BESST
(www.besst.it).
REFERENCES
ANSYS, Inc. 2009. ANSYS academic research, Release
12.1.
Det Norske Veritas. 2009. Composite components.
Offshore standard DNV-OS-C501, April 2009.
Det Norske Veritas. 2010. Rules for high speed, light craft
and naval surface craft. January 2010.
Frangopol, D.M. & Recek, S. 2003. Reliability of fiber-
reinforced composite laminate plates. Probabilistic
Engineering Mechanics 18: 119–137.
Garnich, M.R. & Akula, V.M.L. 2008. Review of degra-
dation models for progressive failure analysis of fiber
reinforced polymer composites. Applied Mechanics
Reviews 62, 010801.
Jeong, H.K. & Shenoi, R.A. 1998. Reliability analysis
of mid-plane symmetric laminated plates using direct
simulation method. Composite Structures 43: 1–13.
Jeong, H.K. & Shenoi, R.A. 2000. Probabilistic strength
analysis of rectangular FRP plates using Monte
Carlo simulation. Computers and Structures 76:
219–235.
Lekou, D.J. & Philippidis, T.P. 2008. Mechanical property
variability in FRP laminates and its effect on failure
prediction. Composites: Part B 39: 1247–1256.
2/18/2011 5:57:08 PM
 Murotsu, Y. et al. 1994. Reliability design of fiber rein-
forced composites. Structural Safety 15: 35–49.
Nader, J.W. et al. 2008. Probabilistic finite element analy-
sis of modified ASTM D3039 tension test for marine
grade polymer matrix composites. Journal of Rein-
forced Plastics and Composites 27: 583–597.
Nakayasu, H. & Maekawa, Z. 1997. A comparative study
of failure criteria in probabilistic fields and stochastic
failure envelopes of composite materials. Reliability
Engineering and System Safety 56: 209–220.
Puck, A. & Schürmann, H. 1998. Failure analysis of FRP
laminates by means of physically based phenomeno-
logical models. Composite Science and Technology
58: 1045–1067.
Soden, P.D. et al. 1998. Lamina properties, lay-up config-
urations and loading conditions for a range of fibre-
reinforced composite laminates. Composites Science
and Technology 58: 1011–1022.
714
MARSTRUCT.indb 714
Soden, P.D. et al. 2002. Biaxial test results for strength
and deformation of a range of E-glass and carbon
fibre reinforced composite laminates: failure exercise
benchmark data. Composites Science and Technology
62: 1489–1514.
Sriramula, S. & Chryssanthopoulos, M.K. 2009. Quan-
tification of uncertainty modelling in stochastic
analysis of FRP composites. Composites: Part A 40:
1673–1684.
Sutherland, L.S. & Guedes Soares. C. 1997. Review of
probabilistic models of the strength of composite
materials. Reliability Engineering and System Safety
56: 183–196.
2/18/2011 5:57:09 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Rule development for container stowage on deck

V. Wolf, I. Darie & H. Rathje

Germanischer Lloyd SE, Hamburg, Germany

ABSTRACT: Germanischer Lloyd currently update rules for stowage and lashing of containers to
ensure safe and reliable container loadings and adequate dimensioning of lashing equipment. Various
research and development works, focusing mainly on performance of fully automatic locks and reliable
calculation of container and lashing loads, were performed. Essentials of the corresponding findings and
their impact on classification rules are presented. Particular attention is given to the new test procedure
for fully automatic locks.

1 INTRODUCTION Further, rule values for design accelerations are to

As a classification society, Germanischer Lloyd are
committed to provide state-of-the-art expertise for
ship safety. Thus, safe transport of containers at sea
represents one topic of current rule development
activities at Germanischer Lloyd. In recent years,
particular attention was directed to the stowage
and lashing of containers on a ship’s weather deck.
Container stacks transported on weather deck
are exposed to dynamic forces caused by sea
induced ship motions, gravity, wind, and green
water. In severe seas as well as in case of improper
container stowing and container overweight, these
forces may become excessive, causing, for example,
failure of locks or collapse of lower stacked
containers. Consequently, whole container stacks
can collapse and may be lost overboard, which
is more than an economic issue. Containers lost
overboard and floating at the surface represent a
hazard for ships and, in particular, smaller craft.
On top of this, deck containers may be loaded with
dangerous goods. Thus, container loses also repre-
sent remarkable environmental implications.
Classification rules for stowage and lashing of
containers include stipulations ensuring reliable
container loadings and proper dimensioning of
lashing systems (Germanischer Lloyd 2007). The
standards are based on engineering expertise and
are, moreover, comprehensively calibrated and
proven by experience. Nevertheless, the ongo-
ing changes in container transport, in particular,
regarding the transport of containers stowed on
weather deck make the continual maintenance
and update of classification rules essential to
retain the container transport at sea being safe.
For instance, adequate prototype tests are to be
developed for reliable approval of new designs of
lashing equipment, such as fully automatic locks.
be determined for ships with distinct roll behavior,
such as wide-bodied container ships.
A number of container damages and losses in
2006 triggered a range of research activities at
Germanischer Lloyd concerning the causes of
such incidents. The investigations focused mainly
on fully automatic locks, which were involved in
the above-mentioned events. Moreover, complex
dynamics of container stacks on deck were consid-
ered in-depth to enable a sound prediction of the
prevailing dynamic container and lashing loads.
In the following, an overview of the recent and
ongoing rule development work at Germanischer
Lloyd regarding the container transport at sea
is presented. Interim development stages were
reported in former publications (Rathje et al. 2008,
Wolf & Rathje 2009). Here, a summary of the cor-
responding findings is given for the purpose of a
clear context. In this paper, the emphasis lies on
the most recent findings and their implication in
classification rules. Finally, an outline of the next
steps is presented as well.
2 CONCLUDED INVESTIGATIONS
2.1 Analyses of container loses
In February 2006, three container ships, including
the 8400TEU liner Mondriaan from the Blue Star
Shipping Company, lost about 180 containers in
heavy seas in the Bay of Biscay within two days.
Subsequently, the project “Seaborne Container
Losses and Damages”—a joint development
project of Germanischer Lloyd, Blue Star Ship-
ping Company and Technical University of
Hamburg-Harburg—was initiated to analyze the
causes of those damage series (Rathje et al. 2008).

715

MARSTRUCT.indb 715 2/18/2011 5:57:09 PM

 On the basis of detailed information for
Mondriaan regarding ship’s course, speed, loading
condition and sea condition, Germanischer Lloyd
determined container racking, lifting, and pres-
sure loads at locations of interest, taking into
account the wave-induced hull pressures, the ship’s
response including the flexible hull vibrations, the
resulting container accelerations and, finally, the
container weight information from the loading
computer. Further, load-bearing capacity of the
fully automatic locks was comprehensively inves-
tigated in finite element analyses. The numerical
results proved that the combined racking, lifting,
and pressure loads on containers exceeded the
locks’ bearing capacity in the investigated case. In
addition, full-scale test series with fully automatic
locks and a common 20 foot container were per-
formed and confirmed thoroughly the results of
the numerical calculations.
The investigations revealed that the container
loses in the Bay of Biscay in 2006 happened in
extreme sea conditions. According to the wave scat-
ter diagram given in the IACS Rec. 34, this partic-
ular seaway occurs about 23 hours during 25 years
exposure time in the North Atlantic Ocean. Such
extremely rare events can hardly be covered by
reasonably practicable design values for stowage
and lashing of containers laid down in the classi-
fication rules. Thus, the fully automatic locks were
likely to fail due to the exceptional dynamic load
combinations in those storm conditions. A basic
malfunction of the fully automatic locks was not
observed. However, numerical calculations and
experiments showed that the performance of the
fully automatic locks can be significantly affected
by various parameters, such as the condition of
container corner castings. Moreover, identified
failure mechanisms specific for fully automatic
locks revealed that the actual combinations of the
simultaneously acting dynamic racking, lifting, and
pressure forces are, besides their respective maxi-
mum values, essential for lock bearing capacity,
especially for that lock type. Consequently, the
increasing relevance of the reliable prediction of
the sea induced container and lashing loads and,
thus, the need of better understand the complex
container stack dynamics became apparent. The
development of an approach for accurate calcula-
tion of container and lashing loads, allowing for
the relevant aspects regarding stack dynamics, was
in focus of further investigation at Germanischer
Lloyd as summarized below.
2.2 Motion simulation of container stacks
The sea and wind induced dynamic response of
container stacks depends on various system char-
acteristics, such as container flexibility, container
716
MARSTRUCT.indb 716
cargo weights, lashing bridge type, clearances
in lashing elements, lashing rod pretension, etc.
The physics of the resulting vertical and lateral
forces on containers and lashing equipment are
fairly complex, e.g., due to sliding, uplifting and
bouncing of stacked containers which may occur
even in moderate sea conditions.
Although long-established approaches allowed
for a kind of straightforward prediction of
maximum container and lashing loads, knowledge
of the prevailing dynamic forces acting on flexible
container stacks was lacking. To gain insight into
the stack dynamics, particularly, into the actual
combinations of vertical and lateral container
forces, a new routine for transient motion simula-
tion of flexible container stacks carried on weather
deck was developed based on finite element method
(Wolf & Rathje 2009). To reduce the computa-
tional effort, appropriate modeling and simulation
techniques were used. For instance, a simplified
model was adopted for locks, which was nonethe-
less capable of reproducing sliding and uplifting
of stacked containers relative to each other within
the locks’ clearances, taking into account friction
effects. The parameters for the simplified model,
such as lock stiffness and clearances, were derived
from detailed finite element models of the respec-
tive lock type as illustrated in Figure 1.
Detailed finite element models of different
container types, including, e.g., container door
and bottom assemblies, were utilized combined
with appropriate substructuring techniques to
reproduce accurately container stiffness and mass
inertia properties. In the model, stacked contain-
ers were linked by the simplified locks and, where
applicable, lashed by pre-stressed truss elements
representing lashing rods (Figure 2). Container
cargo was added individually to each container in
stack by means of additional mass elements.
Figure 1. Modeling techniques for locks.
2/18/2011 5:57:09 PM

Figure 2. Container stack model.
Thus, the overall model for container stacks was
capable of simulating sea and wind induced stack
response, considering the relevant dynamic factors,
such as container flexibility, container cargo, clear-
ances and pretension in lashing, interaction of
adjacent stacks, etc.
Within the scope of the joint industry project
“Lashing@Sea”, full-scale and model-scale tests on
stacked containers were performed to investigate
the dynamic response of stacked containers to var-
ious excitation patterns. In addition, Germanischer
Lloyd conducted static full-scale test series on
ISO-standard containers to gain further infor-
mation regarding significant system parameters.
Test data obtained from both test series served as
validation basis for the finite element model and
confirmed the model’s capability to adequately
calculate sea induced container and lashing forces,
taking into account various dynamic effects.
The new method for stack motion simulation
enabled an important insight into container stack
dynamics. First findings regarding the dynamic
forces on containers and lashing equipment were
reported (Wolf & Rathje 2009).
8400 TEU container ship with following main
particulars.
− Length between perpendiculars 317.2 m
− Molded breadth 43.2 m
− Design draft 13.0 m
− Scantling draft 14.5 m
− Maximum speed 24.5 kn
In the course of previous investigations, sea
induced rigid body motions were calculated for that
ship by the use of strip method. On the basis of
the corresponding results, combinations of ship’s
heave, pitch, sway, yaw, and roll yielding the most
critical loads on deck containers were determined,
taking into account stochastic aspects. For the GM
value of 3.5 m, the following simultaneously acting
instantaneous values for roll angle and the ship’s
vertical and transverse accelerations as well as the
roll period were calculated.
− Roll angle 24,2°
− Roll period 17.0 s
− Vertical acceleration (astern) 1.3 m/s2
− Transverse acceleration (astern) 0.5 m/s2
Resulting dynamic loads on container stack
shown in Figure 3 were calculated. The considered
container stack was located astern, about 16 m off
the ship’s center line. For the ship’s loading condition
corresponding to the GM value of 3.5 m, the vertical
distance of the stack’s bottom to the ship’s roll axis
was about 9.3 m. The stack configuration, including
stack height, container type, container weights, and
the lashing system, was gathered from the Cargo
Securing Manual and accurately included in the
finite element simulation as shown in Figure 4.
Stack response to the ship’s roll motion and
accelerations listed above was simulated for a time
span of about three roll periods. Since the con-
sidered container stack was not in the outmost
position in the bay, wind induced container loads
were not included in the numerical simulations.

3 RECENT RULE DEVELOPMENT

ACTIVITIES

3.1 Sea induced container and lashing forces

Typical configurations of container stacks carried
on weather deck were systematically analyzed,
utilizing the new method for stack motion
simulation. In the following, results are given for
a representative case demonstrating some basic
aspects regarding the physics of dynamic forces
acting on flexible container stacks.
The example presented in this paper was based
on an existing Cargo Securing Manual for an Figure 3. Position of the investigated container stack.

717

MARSTRUCT.indb 717 2/18/2011 5:57:10 PM

 Figure 5. Transverse racking forces on the top of the
third container from stack bottom.

Figure 4. Configuration of the investigated container

stack.

The above input data for the calculation of

dynamic container and lashing forces comprise the
ship’s sea induced motions occurring once during
the ship’s life time. Thus, the calculated container
Figure 6. Vertical forces on the top corners of the third
and lashing forces presented below reflected the container from stack bottom.
probability of occurrence of about 10−8, which cor-
responds to the design safety level for the ship’s hull.
So, the calculated container and lashing forces cor-
responded to extreme sea conditions. In practice,
however, a reasonably lower safety level underlies
the approval of Cargo Securing Manuals.
First, we look at racking, lifting, and compres-
sive forces acting on the top of the container in
third tier from the stack’s bottom, which were
mainly caused by gravity and inertia loads from
the upper containers. Figures 5 and 6 show the
transverse and vertical forces acting on the top
of the third container for about one roll period.
In addition, Figure 7 illustrates the instantaneous
values of those forces for the moment of maximum
roll angle to port. Considerable dynamic amplifi-
cations caused by sliding, uplifting and bouncing
of stacked containers can be seen, particularly, for
the vertical forces. Moreover, Figures 6 and 7 also
demonstrate that the vertical forces acting on the
container’s front end were noticeably higher than
Figure 7. Maximum transverse and vertical forces on
those on the container’s door end. the top of the third container from stack bottom.
Racking forces on top of the third container
were equal on the front end and door end, as
expected for the unlashed upper part of the stack.
Maximum racking forces of about 130 kN were
observed on both ends of the container. On the
contrary, the vertical forces on the container’s front
end and door end differed considerably as shown in
Figures 6 and 7. Maximum lifting and compressive
forces of about 210 kN and 320 kN, respectively,
were observed on the container’s front end. The
corresponding values on the door end were sig-
nificantly lower. Merely 80 kN for lifting force and
250 kN for compressive force were observed there.
Figure 8 shows vertical forces on top corners
of the container in the second tier from stack Figure 8. Vertical forces on the top corners of the
bottom. In this tier, lifting forces of about 420 kN second container from stack bottom.

718

MARSTRUCT.indb 718 2/18/2011 5:57:12 PM

 and 0 kN were observed on the container’s front
end and door end. Compressive forces were about
900 kN and 450 kN, respectively. Thus, the differ-
ence between vertical forces on the container’s front
end and door end increased significantly compared
to the container in the above tier. This indicated
that although the transverse racking forces on top
of the third tier were equal on the container’s front
end and door end, the resulting vertical reaction
forces on the bottom of the third tier, i.e., on the
top of the second tier were considerably higher on
the container’s front end than on the door end.
Static finite element analyses as well as static full
scale tests with a single 20 foot container were per-
formed to analyze lifting and compressive forces on
the container’s bottom induced by uniform racking
forces on the container’s top as shown in Figure 9.
Assuming the container as a “simple box”, equal
compressive forces would be expected on both ends
below the bottom corner castings diagonally oppo-
site to the racking loaded corners. Further, equal
lifting forces would be present on both ends below
the other two bottom corners. In the tests as well
as in numerical simulations, a completely different
Figure 9. Static full scale tests and finite element analyses
on a 20 foot container.
719
MARSTRUCT.indb 719
picture was observed. The overturning moment
induced by the transverse racking forces on the
container’s top was mainly balanced by lifting and
compressive forces on the container’s front end,
which were about twice as high as would be expected
for a “simple box”. On the container’s door end,
even more counterintuitive results were obtained.
Below the bottom corner diagonally opposite to
the racking loaded corner, lifting force or, in case
of unrestrained corner, uplifting of that corner was
observed (Fig. 9). Accordingly, a compressive force
was present below the opposite bottom corner.
To understand the causes for vertical reaction
forces on the container’s bottom observed in tests
and simulations, the container’s flexibility had to
be taken into account. The investigated container
was characterized by a flexible door assembly on
the one hand and stiff front, roof, side, and bot-
tom assemblies on the other hand, which is typical
for the most ISO freight containers. Thus, the pro-
nounced racking deformation of the door frame
implied an in-plane rotation of the roof and, in
consequence, opposite in-plane rotations of the
container’s sides. As a result, secondary reaction
forces arose on the container’s bottom corners
increasing the lifting and compressive forces on the
container’s front end and compensating them on
the container’s door end.
Thus, the full-scale tests and the associated
finite element analyses of a single container veri-
fied the simulation results for stacked containers.
Container stack simulations revealed that load
combinations acting on container stacks carried
on weather deck are basically different on the con-
tainers’ front end and door end. While lifting and
compressive forces are clearly dominant on the
front end, combinations of low or moderate verti-
cal forces and high racking forces are typical for the
door end. This represented a basically new picture
of the prevailing container and lashing loads and
initiated a substantial review of the classification
rules for stowage and lashing of containers, includ-
ing the development of the new test procedure for
fully automatic locks.
3.2 New test procedure for fully automatic locks
Besides strength tests stipulated for all lock types,
additional functionality tests are required for fully
automatic locks to ensure the reliability of such
locks in operation. Recently, Germanischer Lloyd
developed a new procedure for testing the func-
tionality of fully automatic locks. In the following,
the main technical aspects regarding the new test
procedure are presented.
Two potential failure modes were identified for
the fully automatic locks on the basis of previous
analyses of container loses. The first failure mode,
2/18/2011 5:57:14 PM
 which is also known, e.g., from semi-automatic
locks, is characterized by violent vertical pull-
ing out of the locks from the corner castings due
to excessive lifting forces on containers, which is
accompanied by considerable damage of the corner
castings. Since lifting forces are dominant on the
container’s front end, as demonstrated above, this
failure mode is relevant primary for locks at that
location. The second failure mode characteris-
tic for fully automatic locks is represented by the
locks’ lateral sliding out of the corner castings,
which can occur in case of high transverse forces
combined with low vertical forces. For such load
combinations, which are likely for the container’s
door end, friction between the locks and corner
castings may be insufficient to arrest the locks’ lat-
eral sliding induced by the transverse force. Since
the fully automatic locks are typically not form-
fitted, locks can laterally slide out of the corner
castings without notably damaging them. Thus, for
the comprehensive testing of fully automatic locks,
separate functionality tests are required for both
failure modes, i.e., a kind of pull test for the first
failure mode and a shear test for the second failure
mode. Reasonable test loads and appropriate test
setups were to be defined for both tests.
Numerical simulations of container stacks with
representative cargo weights and lashing arrange-
ment were performed to find simultaneously act-
ing vertical and transversal forces representing the
most critical realistic load combinations for locks in
operation. In the course of this, permissible values
for container and lashing loads according to clas-
sification rules were to be borne in mind. Thus, the
lock’s tensile force in operation, which is decisive
for the first failure mode, was limited by the maxi-
mum permissible lifting force of 250 kN. Further,
the maximum transverse force of 190 kN acting
on locks on the container’s door end, representing
the most relevant load component for the second
failure mode, was determined by the permissible
values for container racking force and lashing rod
forces. Figure 10 shows load combinations on the
container’s front end and door end in operation,
which were assessed to be most critical regarding
Figure 10. The most critical operational loads on locks
on container’s front end (left) and door end (right).
720
MARSTRUCT.indb 720
the first failure mode and the second failure mode,
respectively.
Additional safety factors for the calculated loads
were required for tests to take into account scat-
tering of various parameters in operation, such
as type of containers and lashing equipment, pre-
tention of lashing rods, type and securing of con-
tainer cargo, etc. For this reason, the increase of
the tensile force in pull test by 10% to 275 kN was
assessed to be sufficient, following the common
praxis. Accordingly, the transverse force in shear
tests was increased from 190 kN to 210 kN.
The basic test setup for pull test and shear test
is illustrated in Figure 11. In both tests, a test jig
comprising two rigidly linked ISO bottom corner
castings was utilized. In addition, two ISO top cor-
ner castings were fixed on the test platform. Mint
condition was required for all corner castings. The
top and bottom apertures of all ISO corner fittings
were to be of maximum width according to ISO
standard, i.e., 65 mm wide, which represented the
most adverse condition regarding lock perform-
ance. Further, distances between corner castings on
test platform and test jig were adjusted within the
size tolerances for ISO containers to yield the most
unfavorable condition for locks. Thus, the distance
between center lines of corner casting apertures on
the test jig was 4 mm less than the distance between
center lines of the corner casting apertures on the
test platform (Fig. 11). In case of the pull test, an
additional offset of 1 mm was required to allow for
further aspects described below.
The test setup for pull test and shear test
implied necessary simplifications of real condi-
tions in operation. So, the containers’ flexibility
and the associated deformations were not explic-
itly accounted for by the test setup. Further opera-
tion parameters, such as corrosion and abrasion of
corner castings, were not reflected by the test setup
as well. These simplifications were required to
ensure the practicability of tests and, in particular,
Figure 11. General test setup for pull test and shear test.
2/18/2011 5:57:16 PM
 the reproducibility of test results. In consequence,
appropriate correction factors for test loads and
modification of the test setup were required to
compensate the gap between the simplified test
setup and the real operation conditions. For this
reason, benchmark tests and finite element analyses
were performed.
Figure 12 shows the benchmark for the shear
test, where the effect of the flexibility of lower con-
tainer on lock performance was explicitly included.
Moreover, the condition of container corner cast-
ings represented fairly real operation conditions,
as shown in Figure 13 for container top corner
castings. In the benchmark tests, the load combi-
nation for the shear test, including the additional
safety factor for the transverse force as described
Figure 12. Benchmark for shear test for fully automatic
locks.
Figure 13. Container top corner castings in benchmark
tests.
721
MARSTRUCT.indb 721
above, was applied. The utilized locks were able to
withstand the test loads. After the shear test loads
were applied, the dominant compressive force and,
thus, the friction between the locks and corner
castings were steadily reduced to determine the
critical compressive force triggering lock failure
due to sliding out of the corner castings. In addi-
tion, the critical compressive force was determined
for the simplified test setup in the same way. The
critical compressive force observed in the simpli-
fied test setup amounted to about 60% of the criti-
cal compressive force in the benchmark tests. This
indicated that the performance of fully automatic
locks in test conditions would be better than in
operation. To compensate this effect, a reduction
factor of 0.6 was specified for the dominant com-
pressive force in the simplified shear test, thus,
implicitly accounting for container flexibility and
realistic conditions of corner castings.
For the pull test, an analogous correction of test
loads to compensate mainly the effect of container
flexibility would basically be possible. However,
this would imply a further increase of the test ten-
sile force on locks and, thus, an exceedance of the
strength of corner castings in the tests. Thus, the
lock performance could hardly be assessed under
such test loads due to the failure of the corner
castings. Therefore, appropriate modification of
the test setup instead of the test load correction
was intended for the pull test. Finite element analy-
ses were carried out to compare the performance
of fully automatic locks in view of excessive ten-
sile forces in test conditions and in the more real-
istic case of stacked containers. Numerical results
revealed again that deformation of flexible con-
tainers schematically shown in Figure 14 worsens
the performance of fully automatic locks.
The deformation of flexible containers resulted
in an unfavorable tilting of adjacent corner cast-
ings and, thus, in a reduction of the lock’s nose
overhang in the lower corner casting, which
Figure 14. Effect of container flexibility (left) and the
equivalent test setup modification for the pull test (right).
2/18/2011 5:57:17 PM

Figure 15. Test procedure for functionality tests of fully
automatic locks.
affected significantly the performance of consid-
ered locks (Fig 14). To allow for this effect in test
conditions, the corner castings were transversely
offset as shown in Figure 14. Based on numeri-
cal simulations of the pull test for different fully
automatic locks, the effect of the corner castings’
transverse offset of 1 mm was found to be equiva-
lent to the effect of the flexibility of a typical ISO
container.
Figure 15 summarizes the basic information
regarding the test setup, test loads and test
procedure for the pull test and the shear test, which
were required to assess the performance of fully
automatic locks in operation, considering both
failure modes relevant for this lock type. Numbers
in parentheses in Figure 15 indicate the sequences
for force application in both tests. The test loads
included safety factors and correction factors gained
from benchmark tests and numerical calculations as
described above.
First functionality tests of fully automatic locks
were carried out to validate the viability of the
new test procedure. Sufficient bearing capacity
was observed for some lock designs. On the other
hand, considerable shortcomings became appar-
ent for other lock designs tested so far. The new
test procedure proved to be a practicable approach
for sound testing of the functionality of fully
automatic locks.
4 CONLUSIONS AND OUTLOOK
Since the remarkable container loses in the Bay
of Biscay in 2006, Germanischer Lloyd invested
722
MARSTRUCT.indb 722
substantial effort to better understand the cor-
responding safety-related technical aspects and,
thus, to ensure further on the safe and reliable sea
transport of deck containers in the face of the
ongoing changes in container shipping. Research
work carried out to date focused mainly on the
sound assessment of the functionality of fully
automatic locks and the accurate calculation
of dynamic loads on flexible container stacks.
Investigations showed that the performance of
fully automatic locks was susceptible to vari-
ous parameters, such as seemingly minor design
details, actual condition of container corner
castings in operation as well as the combina-
tions of vertical and transverse dynamic forces
simultaneously acting on locks. The new method
for numerical simulation of container stacks
transported on weather deck enabled essential
insight into the complex physics of the container
and lashing forces and was the key part for the
development of an adequate test procedure for
fully automatic locks. Besides realistic load com-
binations, the new test procedure allowed more
accurately for relevant aspects and operation con-
dition, such as container flexibility and abrasion
of corner castings. Insight into dynamic forces
acting on flexible container stacks revealed that
transverse wind, inertia, and gravity loads on
container stacks induce considerably magnified
vertical forces on the containers’ front end. This
fact has not yet been explicitly accounted for by
the standard approach underlying classification
rules. Its relevance and an appropriate way to
account for realistic container and lashing forces
during approval of Cargo Securing Manuals are
currently being checked. Moreover, the gained
sound knowledge of realistic container and lash-
ing forces serves as basis for the ongoing safety
level assessment of existing stowage and lashing
systems. Furthermore, the determination of rule
values for design accelerations for ships with dis-
tinct roll behavior, such as wide-bodied container
postpanmax containerships with comparatively
high GM values, is planned.
REFERENCES
Germanischer Lloyd 2007. Rules for Classification and
Construction, I—Ship Technology, Part 1—Seagoing
Ships, Chapter 20—Stowage and Lashing of Containers.
Hamburg, Germanischer Lloyd.
Rathje, H. et al. 2008. Seaborne Container Losses and
Damages. Schiffbautechnische Gesellschaft e.V. Yearbook
2008.
Wolf, V. & Rathje, H. 2009. Motion Simulation of Con-
tainer Stacks on Deck. 2009 SNAME Annual Meeting
and Ship Production Symposium Proceedings, 21–23
October 2009 1: 277–285.
2/18/2011 5:57:19 PM
 Advances in Marine Structures – Guedes Soares & Fricke (eds)
© 2011 Taylor & Francis Group, London, ISBN 978-0-415-67771-4

Reliability analysis of marine structural components using statistical

data of steel strength

B. Yu & D.G. Karr

Department of Naval Architecture and Marine Engineering, University of Michigan, Ann Arbor, Michigan, USA

ABSTRACT: The statistical data for ship steel strength suggest that the best-fit probability density
functions vary depending upon the type of steel and the type of failure mode. Various studies suggest
that lognormal, Weibull, or Gumbel distributions may offer the best descriptions for yield stress, ultimate
stress and failure strain. In this paper, we select several distribution functions for yield limit states and
particular steels and study the resulting reliability of steel beam components. The beams are subjected to
compressive loading conditions as commonly encountered for marine structures and the loading resultants
are thus also treated as random variables. The influence of the use of various probability density func-
tions is studied. The results of including uncertainty in the structural geometry are also discussed. It is
found that the probability of failure can vary by orders of magnitude for similar nominal safety factors
depending upon the type of steel or the yield function employed.

1 INTRODUCTION stresses to address the reliability of the component.

The reliability of a structural system is the prob-
ability that it will survive for a given period of time
and/or under specified loading conditions (White
et al. 1985). Generally, the reliability analysis of
marine structures needs to consider three com-
ponents: 1) loading, 2) structural strength and
3) method of calculating probability of failure
(Ayyub et al. 1995).
In this paper, we consider a commonly used
marine structure component, that is, a beam sub-
jected to compressive loading. We generate the
random variable of applied stress on the beam
from two sources of uncertainty, each separately.
One is from the uncertainty of geometry caused by
imperfections in constructing the beam. The other
is from the uncertainty of the external forces that
act on the beam. Our motivation is to address how
one might expect the probability of failure to be
affected by choices of yield stress probability dis-
tributions in addition to the effects of geometric
and loading uncertainty.
The statistical data of the several steels’ mate-
rial yield strength is taken from a previous study
(Yu et al. 2010) in which five different distributions
were applied to describe the random behavior of
material characteristics including failure strain,
yield stress and ultimate strength. Results from sev-
eral thousand tests are included in this data base.
In the previous study, chi-square tests were applied
to test the goodness of fit and to find the best fit-
ting distributions. In this paper, these results for
yield stresses are referred and compared to applied
Although recent research (Daley, 2007) suggests
that marine structures are commonly designed to
exceed the yield point, in this paper we focus on
the limit state in order to examine the variability of
calculated probabilities of failure. Because the most
appropriate limit state distribution functions are
generally not well established, we are particularly
interested in how sensitive the calculated struc-
tural reliability is to the choice of distribution
functions. The method utilized here to calculate
probability of failure is based on several different
probabilistic descriptions of the basic variables,
which in this case are the geometric imperfection
and the external force in addition to the yield stress
distributions. A numerical integration method is
used to calculate the probability of failure.
In previous research, one or several of the above
components have been addressed. The variability
of loading of structures may be caused by uncer-
tainties of geometric variables such as plate thick-
ness and stiffener height. In the papers by Hess
and colleagues (Hess et al. 1997 and Hess et al.
2002), the uncertainties of geometric variables of
common marine structures were quantified. Their
mean values, standard deviation and distributions
recommended to describe their random behavior
were listed. Another factor that influences the
variability of loading is the uncertainty of exter-
nal forces and moments. The statistical charac-
teristics of external bending moments on marine
structures were also analyzed in the papers by Hess
(Hess et al. 2002) and Soares (Soares et al. 1991).
Methods of combining several random variables

723

MARSTRUCT.indb 723 2/18/2011 5:57:20 PM

 of external moments were also described (Soares
et al. 1991).
The randomness of structural strength includes
the uncertainties of material, such as yield stress,
ultimate strength and Young’s moduli. In Hess et al.
(2002), the random behavior of the above three
characteristics of Ordinary Strength (OS) and High
Strength (HS) steel was investigated and the mean
values, standard deviations and recommended dis-
tributions to describe their random behavior were
also discussed.
Different methods of calculating probability
of failure were also studied in previous research.
In the paper of White et al. (1985), the first and
second moment methods were described. Monte
Carlo simulation has also been applied to combine
multiple geometric random variables and gener-
ate the random variable data of the applied stress.
Based on this approach, a structure’s reliability
can be calculated using the conditional expectation
variance reduction technique (Ayyub et al. 1997).
The axial stress that has the maximum absolute
value is the compressive stress:
P Pw0 ymax
σC = + (4)
A 1 − P/PE I y
where A is the area of the cross section of the beam;
ymax is the maximum y coordinate of extreme fibers
from the centroidal axis.
2.1 Geometric imperfection uncertainty
In this section, initial deformation at the middle
point of the beam w0 is treated as a random
variable, while other geometric parameters and the
external force are assumed to be deterministic.
Since the initial deformation is usually caused
by imperfections in the production of the beam,
we assume that w0 has a normal distribution with
its mean value equal to zero and standard devia-
tion equals 2 mm.

2 APPLIED STRESS OF THE MARINE w0 2

−
STRUCTURALCOMPONENT 1 2v12
pdf (w0 ) = e , < w0 < ∞ (5)
2π v12
We consider a single steel beam with length of 1 m,
cross section 90 mm × 20 mm. We assume the cen- where ν1 is the standard deviation of w0.
troidal axis is initially bent and in the shape of a At the same time, we note that when w0 becomes
half-sine wave, as in Figure 1. The initial deforma- negative, ymax also becomes negative. As a result,
tion is assumed to be: the maximum compressive stress has the value:
πx P P |w0 | t
w w0 sin (1) σC = + (6)
L A 1 − P/PE 2 I y
Then the total deflection at the mid-span of the
Since the distribution of w1 = |w0| will have prob-
beam will be (Boresi et al. 2003):
ability density function:
w0
δ= (2)
−
w0 2
1 − P/PE 2 2v12
pdf (w1 ) = pdf (|w0 |) = e , 0 < w1 < ∞
2π v12
where PE = π EIy/L , is the Euler column buck-
2 2

ling load, E is the steel Young’s Modulus and Iy is (7)

the moment of inertia of the cross section of the
beam. Let
Therefore the maximum bending moment along the
beam is at the middle point and has the value of P Pt
a= ,b= (8)
A 2II y ( P/PE )
Pw0
M= (3)
1 − P /PE Then

σC = a + bw1 (9)
Therefore, the pdf of σC becomes
( a )2
− C
2 2 2
pdf ( C) = e 2b v
, a ≤ σC < ∞ (10)
Figure 1. Initially bent beam. b π v12

724

MARSTRUCT.indb 724 2/18/2011 5:57:20 PM

 It can be observed that the compressive stress Table 1. Mean value and standard deviation of external
increases as the external force P increases. In this forces for different materials for examining external force
case, the initial deformation of the beam has a uncertainty.
mean value of zero. For our beam column without
Compressive force
initial bending, P will reach the critical buckling
force PE before the compressive stress reaches the μP (kN) νP (kN)
yield stress. As a result, in order to avoid buck-
ling, we set the mean external force equal to 2/3 A 83.0 3.20
of PE. In this case, PE = 118.4 kN, P = 78.93 kN, AH32 91.0 2.00
the nominal safety factor for buckling is thus AH36 94.0 1.70
FSb = PE/P = 1.5. AH40 96.4 1.39

2.2 Loading, external force uncertainty Note that in this model the maximum compres-
sive stress will reach the yield stress before the
Generally, the external forces and moments that
external force reaches the critical buckling force.
marine structures encounter should also be treated
Therefore the failure mode of this probabilistic
as random variables. In this section, we assume the
model of compressive stress generated from load-
external force P has a normal distribution while the
ing uncertainty is yielding. To avoid yielding for
geometric parameters are assumed deterministic.
the nominal conditions, we set the mean value of
The probability density function of P is assumed
the compressive stress of each material to be 2/3
of the form:
of the corresponding mean value of yield stress to
have a nominal safety factor of approximately 1.5.
(P μP )2 In order to generate such stochastic characteristics
−
1 2v 2
pdf ( P ) = e , 0 P<∞ (11) of compressive stress, the mean value and standard
2π v p2 deviation of the external forces P for the analysis
of each material are listed in Table 1.
where μp and νp are the mean value and standard
deviation of P. 3 YIELD STRESS OF THE MARINE
Since STRUCTURE COMPONENT

In order to determine the reliability of the marine

P Pw0 t
σC = + (12) structure component, the compressive stress
A 1 − P/PE 2 I y resulted from loading conditions are compared to
the yield stress of the particular material. As in our
Here we assume w0 equals 3 mm. former study (Yu et al. 2010), the yield stress ran-
In order to let the beam be stable, P has to be dom variable is described using five different types
smaller than PE. Therefore, of the two solutions for of distributions. The cumulative probability func-
P from equation 12, we choose tions of each kind are:
Lognormal:

b( C b2 ( C)− PE σ C
4 AP 1 ⎡ 1n(σ y ) μ ⎤
P= = g( C) (13) F1( y) f ⎢−
erfc ⎥ (16)
2 2 ⎣ 2V 2 ⎦

where where μ and ν are the mean value and variance of

histograms of yield stress.
tw0 Weibull:
b( C Aσ C + PE PE a2 , a2 =
AP (14)
2I y
⎡ ⎛σ ⎞ k⎤
F2 ( ⎢
y ) = 1 − exp − ⎜
y
⎥ (17)
Thus the probability density function of the ⎢ ⎝ u ⎟⎠ ⎥
⎣ ⎦
compressive stress is:
where
( g (σ C ) )2
− P
1 2vP2 ⎡ ⎛ 2⎞ 1⎞⎤
pdf ( C) = e | g ′(σ C ) |, (15) 1 ⎛
μ = uΓ( + ), v 2 = u 2 ⎢Γ 1 + ⎟ − Γ 2 1 + ⎟ ⎥ (18)
2π v p2 k ⎣ ⎝ k ⎠ ⎝ k ⎠⎦

725

MARSTRUCT.indb 725 2/18/2011 5:57:24 PM

 Table 2. Mean and standard deviation of compressive stress.

Compressive stress from Compressive stress

geometric uncertainty from loading uncertainty Yield stress

μ (σc) ν (σc) μ (σc) ν (σc) μ (σy) ν (σy)

(MPa) (MPa) SF (MPa) (MPa) SF (MPa) (MPa)

A 106.9 47.6 2.747 186.5 20.3 1.575 293.7 13.14

AH32 106.9 47.6 3.594 248.4 20.2 1.547 384.2 16.35
AH36 106.9 47.6 3.993 281.4 21.3 1.517 426.9 16.28
AH40 106.9 47.6 4.409 313.9 21.2 1.501 471.3 19.66

Gumbel maximum:

⎡ ⎛α σy ⎞⎤
F3 ( y) = exp ⎢ exp ⎜ ⎥ (19)
⎣ ⎝ β ⎟⎠ ⎦

where

v 6
a = μ − γβ β = , γ = 0.5772 (20)
π

Gumbel minimum:
Figure 2. Compressive and yield stress distributions of
⎡ ⎛ αy σ ⎞ ⎤ A part.
F4 ( y ) = 1 − exp ⎢ exp ⎜ ⎥ (21)
⎣ ⎝ β ⎟⎠ ⎦

where

v 6
α = μ + γβ , β = , γ = 0.5772 (22)
π

Gumbel mix:
1
F5 ( y)
⎡ F ( y )Gumbel max
2⎣ (23)
+ F ( y )Gumbel min ⎤⎦
Figure 3. Compressive and yield stress distributions of
A part 2.
In Table 2, the mean values and standard devia-
tions of the yield stresses and compressive stresses
generated from geometric uncertainty and force
uncertainty are compared. The nominal safety fac-
tors SF = μ(σy)/μ(σc), for both geometric and load-
ing uncertainty cases are also provided in the table.
Comparisons of the distribution of compressive
stress and yield stress are shown in Figures 2 to 9.
The chi-square test method was used in our
former study (Yu et al. 2010) to evaluate the good-
ness of fit of each distribution of yield stress.
Chi-square values of all, first half and first quar-
ter of the intervals in the histogram of yield stress
were calculated. Since the overlap of compressive
stress and the yield stress is mainly located at the Figure 4. Compressive and yield stress distributions of
lower quarter of the yield stress distribution, both AH32 part 1.

726

MARSTRUCT.indb 726 2/18/2011 5:57:30 PM

 Figure 5. Compressive and yield stress distributions of
AH32 part 2. Figure 8. Compressive and yield stress distributions of
AH40 part 1.

Figure 6. Compressive and yield stress distributions of

AH36 part 1.
Figure 9. Compressive and yield stress distributions of
AH40 part 2.

Table 3. Chi-square value for overall histogram of yield

stress.

A AH32 AH36 AH40

Lognormal 2093 277.2 124.8 8.030*

Weibull 3009 3937 1133 54.17
Gumbel max 255.1* 2625 229.1 25.72
Gumbel min 5256 1.168e4 2227 68.68
Gumbel mix 672.5 129.7* 95.10* 23.26

Figure 7. Compressive and yield stress distributions of Table 4. Chi-square value for first quarter histogram of
AH36 part 2. yield stress.

A AH32 AH36 AH40

the overall and first-quarter chi-square values are
Lognormal 252.7 4.850* 7.160* 1.190
of interest here. The data from Yu et al. (2010) are
Weibull 875.4 157.4 123.7 7.000
listed again here in Tables 5 and 6. The smallest Gumbel max 38.98* 2291 150.4 17.31
chi-square values indicate the best-fitting distribu- Gumbel min 967.3 179.8 143.1 8.560
tions of each material, which are indicated by an Gumbel mix 200.9 81.52 39.35 0.670*
asterisk in Tables 3 and 4.

727

MARSTRUCT.indb 727 2/18/2011 5:57:32 PM

 Table 5. Probability of failure from geometric the cumulative probability functions of the yield
uncertainty. stress. Five different cdfσ y functions are generated
based on equations 16 to 23. Each of the above
A AH32 AH36 AH40 five cdfσ y functions is used for each of the four
(10–3) (10–5) (10–6) (10–7) materials.
The numerically calculated probabilities of fail-
Lognormal 1.763 2.342 1.927 1.388 ure of each material using equation (26) and are
Weibull 1.826 2.734 2.434 2.489 listed in Tables 5 and 6.
Gumbel max 1.747 2.186 1.748 1.127
Gumbel min 1.840 2.869 2.628 3.283
Gumbel mix 1.793 2.528 2.188 2.205 5 DISCUSSION AND CONCLUSION

From Table 3, we note that the best overall fitting

Table 6. Probability of failure from loading uncertainty. distributions for each material are Gumbel maxi-
mum for material A, Gumbel mix for AH32 and
A AH32 AH36 AH40 AH36, lognormal for AH40. At the same time,
the best fitting distributions for the first quar-
(10−4) (10−6) (10−6) (10−7) ter are Gumbel max for A, Lognormal for AH32
and AH36 and Gumbel mix for AH40. Since the
Lognormal 1.607 6.675 3.988 12.39 overlap of compressive stress and the yield stress
Weibull 2.734 33.83 22.15 270.6 is mainly located at the lower quarter of the yield
Gumbel max 1.402 4.225 2.558 5.254 stress distribution, the probabilities of failure cor-
Gumbel min 3.275 51.49 39.94 635.5 responding to the best first quarter fitting distribu-
Gumbel mix 2.338 27.86 21.25 320.4
tions may give the most accurate estimation of the
resulting reliability.
The results in both Tables 5 and 6 also show that
the Gumbel maximum is the least conservative dis-
4 PROBABILITY OF FAILURE OF tribution which gives the smallest probability of
THE STRUCTURALCOMPONENT failure estimation for all four materials. On the
other hand, Gumbel minimum is the most conserv-
The yield limit state of the structural component ative distribution, it having the largest probability
occurs when the compressive stress reaches a value of failure estimation. Since the Gumbel mix is
x while the yield stress falls below x. We assume simply the average probability density function of
that the actual compressive stress value and yield Gumbel maximum and Gumbel minimum, it also
stress value are independent; the former is decided has the average probability of failure. The other
by geometric and loading conditions while the lat- two distributions make mid range estimations. The
ter is only related to characteristics of the material. Weibull distribution has similar shape in its lower
Therefore the probability of failure is: end as the Gumbel minimum, correspondingly, the
probability of failure calculated from Weibull is
f
failure )x0 = P ( c x0 andd y x0 ) close to but slightly smaller than the one calculated
= P( c x )P ( y x0 ) from the Gumbel minimum distribution. The log-
⎛ dx dx ⎞ normal estimation is close to the Gumbel mix but
= lim P ⎜ x0 − < σ C < x0 + ⎟ (24) less conservative.
dx → 0 ⎝ 2 2⎠
× P (σ y ≤ x0 ) As the safety factor becomes larger, the differ-
ence between the probabilities of failure calculated
from each distribution becomes smaller. However,
P failure
f )x0 = lim pdffσC (x d × cdf
( )dx cdff ( x0 )
(x (25)
dx → 0 y the order of conservativeness stays the same. As
a general conclusion, the order of distributions
∞ that give the least conservative estimation to the
P failure
f ) = ∫ pdffσC ( x )cdffσ y ( x )ddx (26) most is: Gumbel maximum, Lognormal, Gumbel
0 mix, Weibull and Gumbel minimum distribution.
All of the coefficients for the yield stress
where pdfσ c is the probability density function of (two-parameter) probability density functions we
the compressive stress. Equations 10 and 15 are the calculated based on the mean value and standard
general forms of pdfσ c calculated for the two dif- deviations from the test samples. The “best fit”
ferent conditions. For four different materials, four functions depend on whether considers the full
different pdfσc functions are generated from each sample data set or the first quarter sample set. One
of the equations 10 and 15. The cdfσy functions are would expect that the best fit functions indicated in

728

MARSTRUCT.indb 728 2/18/2011 5:57:34 PM

 Table 4 would result in the most accurate estimation
of the probabilities of failure.
It is interesting to note that for the geometric
uncertainty analyses (Table 5), there appears to be
relatively little difference between which probability
density function we choose for a particular mate-
rial. On the other hand we note there are consider-
able differences from one material to another; this
is at least in part due to the fact that the nominal
safety factors differ as per Table 2. Table 6 however
also shows considerable variation from material
to material, even though the nominal safety fac-
tors for the loading uncertainty case are nearly the
same. Evidently then, examining Tables 5 and 6,
some materials have inherently higher reliability
from a material yield strength stand point. Also
we note from Table 6 considerable variation in
the probability of failure from one distribution to
another for the same material. Structural reliabili-
ties may vary by more than an order of magnitude
from one distribution function to another although
each of the distribution functions coefficients we
determined from the same rather extensive data
set. From a practical stand point this highlights the
need for sound data bases for structural strengths
and loading conditions, together with careful defi-
nition of variability, in order to accurately establish
structural reliability.
REFERENCES
Ayyub, B.M., Beach, J.E. & Packard, W.T. 1995. Method-
ology for the development of reliability-based design
criteria for surface ship structures. Naval Engineers
Journal. Volume 107, Issue 1, 45–61, January 1995.
729
MARSTRUCT.indb 729
Ayyub, B.M., Muhuanna, R. & Bruchman, D.D. 1997.
Uncertainty in marine structural strength due to
variability in Geometry and material properties.
In Naval Surface Warfare Center Carderock Division.
Survivability, Structures, and Materials Directorate
Technical Report. W. Bethesda.
Boresi, A.P. & Schmidt, R.J. 2003. Advanced Mechanics
of Materials. New York: John Wiley & Sons.
Daley, C., Kendrick, A. & Pavic, M. 2007. New direc-
tions in ship structural regulations. 10th International
Symposium on Practical Design of Ships and Other
Floating Structures. Houston: ABS.
Hess P.E. III & Ayyub, B.M. 1997. Variability in geom-
etry and imperfections of surface ship structural
scantlings. In Naval Surface Warfare Center Carde-
rock Division. Survivability, Structures, and Materials
Directorate Technical Report. West Bethesda.
Hess, P.E., Bruchman D., Assakkaf, A. & Ayyub, B.M.
2002. Uncertainties in material and geometric strength
and load variables. Naval Engineers Journal. Volume
144, Issue 2, 139–166, April 2002.
Soares, C.G. 1991. Combination of primary load effects
in ship structures. Probabilistic Engineering Mechanics:
Issue 7, 103–111, September 1991. Lisboa, Portugal.
White, G.J. & Ayyub, B.M. 1985. Reliability methods for
ship structures. Naval Engineers Journal. Volume 97,
Issue 4, 86–96, May 1985.
Yu, B., Karr, D.G. & Wang, G. Statistical analysis of
tensile strength parameters of ABS steels. In review.
2/18/2011 5:57:36 PM