Whole

Stochastic analysis and robust design of stiffened composite
structures
Author:
Lee, Merrill Cheng Wei
Publication Date:
2009
DOI:
https://doi.org/10.26190/unsworks/22478
License:
https://creativecommons.org/licenses/by-nc-nd/3.0/au/
Link to license to see what you are allowed to do with this resource.
Downloaded from http://hdl.handle.net/1959.4/44217 in https://

unsworks.unsw.edu.au on 2023-07-14
Stochastic Analysis and Robust Design of
Stiffened Composite Structures
by
Merrill C. W. Lee
B.E. Aerospace Engineering (UNSW)
A thesis submitted in fulfilment of the requirements for the degree of

Doctor of Philosophy
School of Mechanical and Manufacturing Engineering

The University of New South Wales
August 2009
The work described in this thesis was conducted as part of a research program of the
Cooperative Research Centre for Advanced Composite Structures (CRC-ACS) Ltd
Dedicated to my wife, Janet.
Thank you for all the love and support you have provided
Originality Statement
I hereby declare that this submission is my own work and to the best of my knowledge
it contains no materials previously published or written by another person, or substantial
proportions of material which have been accepted for the award of any other degree or
diploma at UNSW or any other educational institution, except where due
acknowledgment is made in the thesis. Any contribution made to the research by others,
with whom I have worked at UNSW or elsewhere, is explicitly acknowledged in the
thesis. I also declare that the intellectual content of this thesis is the product of my own
work, except to the extent that assistance from others in the project’s design and
conception or in style, presentation and linguistic expression is acknowledged.
_______________________
Merrill Lee
_______________________
Date
i|Page
Copyright Statement
I hereby grant the University of New South Wales or its agents the right to archive and
to make available my thesis or dissertation in whole or part in the University libraries in
all forms of media, now or here after known, subject to the provisions of the Copyright
Act 1968. I retain all proprietary rights, such as patent rights. I also retain the right to
use in future works (such as articles or books) all or part of this thesis or dissertation. I
also authorise University Microfilms to use the 350 word abstract of my thesis in
Dissertation Abstract International. I have either used no substantial portions of
copyright material in my thesis or I have obtained permission to use copyright material;
where permission has not been granted I have applied/will apply for a partial restriction
of the digital copy of my thesis or dissertation.
_______________________
Merrill Lee
_______________________
Date
Authenticity Statement
I certify that the Library deposit digital copy is a direct equivalent of the final officially
approved version of my thesis. No emendation of content has occurred and if there are
any minor variations in formatting, they are the result of the conversion to digital
format.
_______________________
Merrill Lee
_______________________
Date
ii | P a g e
Summary
The European Commission 6th Framework Project COCOMAT (Improved MATerial

Exploitation at Safe Design of COmposite Airframe Structures by Accurate Simulation
of COllapse) was a four and a half year project (2004 to mid-2008) aimed at exploiting
the large reserve of strength in composite structures through more accurate prediction of
collapse. In the experimental work packages, significant statistical variation in buckling
behaviour and ultimate loading were encountered. The variations observed in the
experimental results were not predicted in the finite element analyses that were done in
the early stages of the project.
The work undertaken in this thesis to support the COCOMAT project was initiated
when it was recognised that there was a gap in knowledge about the effect of initial
defects and variations in the input variables of both the experimental and simulated
panels. The work involved the development of stochastic algorithms to relate variations
in boundary conditions, material properties and geometries to the variation in buckling
modes and loads up to first failure. It was proposed in this thesis that any future design
had to focus on the dominant parameters affecting the statistical scatter in the results to
achieve lower sensitivity to variation. A methodology was developed for designing
stiffened composite panels with improved robustness. Several panels tested in the
COCOMAT project were redesigned using this approach to demonstrate its
applicability.
The original contributions from this thesis are therefore the development of a stochastic
methodology to identify the impact of variation in input parameters on the response of
stiffened composite panels and the development of Robust Indices to support the design
of new panels. The stochastic analysis included the generation of metamodels that allow
quantification of the impact that the inputs have on the response using two first order
variables, Influence and Sensitivity. These variables are then used to derive the Robust
iii | P a g e
Indices. A significant outcome of this thesis was the recognition in the final report for
COCOMAT that the development of a validated robust index should be a focus of any
future design of postbuckling stiffened panels.
iv | P a g e
Acknowledgements
My heartfelt thanks go to my supervisors Prof. Don Kelly of UNSW and Dr. Rodney
Thomson of the Cooperative Research Centre for Advanced Composite Structures
(CRC-ACS) for the wonderful level of guidance I received throughout this Ph.D. They
have always provided tireless support and advice across all aspects of the project, and
have been an invaluable source of knowledge and inspiration both professionally and
personally. I would like to also thank Dr. Carl Reidsema for the insight he has provided
for this work.
The research described in this thesis was conducted as a part of a research program of
the CRC-ACS whose support is greatly appreciated. I sincerely thank Prof. Murray
Scott, Chief Executive Officer of the CRC-ACS, for giving me the opportunity to be
involved in such an interesting and challenging project. I would like to acknowledge all
the staff at the CRC-ACS, past and present, for the scientific and technical support and
the motivation I have received during my research.
I would like to express my gratitude to Prof. Richard Degenhardt and the staff of the
German Aerospace Center (DLR) in Braunschweig, Germany. Their expertise and
guidance have been critical to the success and timeliness of this project. In particular,
their dedicated support during my 6-month placement in the Institute of Composite
Structures and Adaptive Systems at DLR ensured that this time was both highly
productive and personally enriching, and I would like thank Richard again for such a
rewarding opportunity and being a fine host during my stay at the DLR as a Guest
Scientist.
This work was funded by the CRC-ACS with a Full Postgraduate Research Scholarship,
whose support I acknowledge as fundamental to this thesis. Additional financial support
was provided by the German Academic Exchange Service (DAAD) and UNSW. The
v|Page
contributions of the European Commission, Priority Aeronautics and Space, Contract
AST3-CT-2003-502723, and the Australian Government under the “Innovation Access
Programme – International Science and Technology” are also duly acknowledged.
I would like to extend my thanks to Dr. Adrian Orifici of RMIT for the many
discussions we have had over the years about postbuckling analyses and also to Dr.
Zoltan Mikulik of UNSW who has provided many insights on the modeling of
composite structures.
The publication of this thesis would not have been possible without the constant
encouragement and love from my wife, Janet. Thank you for sticking by me all this
time; it has been an interesting journey for the both of us thus far.
To my family and friends, I thank them for the wonderful support I have had not only
during this thesis but over my entire academic career. This work would not have been
possible without the thoughtfulness and understanding of the people close to me, and I
consider myself fortunate to have had such care and support throughout every stage of
my life. With respect to this I would like to mention my Aunt Rebecca who has always
been a constant supporter.
I would especially like to thank my parents and my brother for the nurturing and
supporting me throughout the key events of my life.
I would like to express my gratitude to my dear friend and colleague, Jonathan Vogt,
who has provided many an afternoon of friendship, often in the company of the
countless cups of coffee I have consumed these last three years. Thanks Jonno.
To all my fellow friends in research at UNSW, never give up.
To the one God who created me and his son, my Lord and Saviour, Jesus Christ.
vi | P a g e
Table of Contents
Summary .........................................................................................................................iii
Acknowledgements .......................................................................................................... v
Table of Contents .......................................................................................................... vii
List of Figures ...............................................................................................................xiii
List of Tables ..............................................................................................................xviii
Abbreviations and Acronyms ...................................................................................... xx
Nomenclature............................................................................................................... xxii
Chapter 1 Introduction ................................................................................................... 1
1.1 Background ........................................................................................................ 1

1.2 The COCOMAT Project .................................................................................... 3
1.3 Outline of Thesis .............................................................................................. 10
1.4 Research Outcomes .......................................................................................... 12
1.5 Participation in the COCOMAT Project .......................................................... 17
Chapter 2 Review of Literature ................................................................................... 19
2.1 Introduction to Stochastic Analysis .................................................................. 20
2.1.1 Stochastic modelling ................................................................................. 20

2.1.2 Applications of metamodelling ................................................................. 22
2.1.3 Measuring uncertainty in structures .......................................................... 24
2.1.4 Conclusions ............................................................................................... 25
2.2 Introduction to the Robust Design Concept ..................................................... 26
2.2.1 Definition of a robust design ..................................................................... 27

2.2.2 The Taguchi Method ................................................................................. 29
2.2.3 Types of Robust Design ............................................................................ 31
2.2.4 Mathematical formulations in Robust Design theory ............................... 33
2.2.5 Other developments in Robust Design ...................................................... 35
2.2.6 Conclusion ................................................................................................ 36
vii | P a g e
2.3 Robustness in the Design of Structural Components ...................................... 37
2.3.1 Progressive failure ..................................................................................... 37

2.3.2 Structural redundancy ............................................................................... 39
2.3.3 Case studies for progressive and disproportionate failure ........................ 40
2.3.4 Worked example for the effects of nonlinearity ....................................... 44
2.3.5 Conclusion ................................................................................................ 47
2.4 Variability in Fibre Reinforced Composites..................................................... 47
2.4.1 Analysis of composite laminates............................................................... 48

2.4.2 Effects of variations in fibre orientation ................................................... 50
2.4.3 Variations and defects in the matrix.......................................................... 51
2.4.4 Defects in bonding of structural elements ................................................. 52
2.4.5 Possible sources of damage in composites................................................ 53
2.4.6 Current research addressing variability in composites.............................. 54
2.4.7 Conclusion ................................................................................................ 55
2.5 Elastic Behaviour of Thin-walled Structures ................................................... 56
2.5.1 Bifurcation buckling and postbuckling of shells ....................................... 56

2.5.2 Experimental testing.................................................................................. 57
2.5.3 Recent developments in numerical methods for buckling and postbuckling
...................................................................................................................59
2.5.4 Mode shapes in buckling........................................................................... 61
2.5.5 Modelling imperfections ........................................................................... 62
2.5.6 Probabilistic methods and buckling .......................................................... 63
2.5.7 Conclusion ................................................................................................ 65
2.6 Placement of the Research in this Thesis ......................................................... 66
Chapter 3 Theory and Methodology ........................................................................... 70
3.1 Stochastic Analysis ........................................................................................... 71
3.1.1 Sensitivity Analysis (SA) procedures ....................................................... 71

3.1.2 The stochastic analysis procedure ............................................................. 72
3.1.3 Sample size................................................................................................ 76
3.2 Formulation for Indicators of Robustness ........................................................ 76
viii | P a g e
3.2.1 Derivation of Sensitivity ............................................................................ 77
3.2.2 Scaling of the Sensitivity parameter .......................................................... 78
3.2.3 Derivation of Influence ............................................................................. 78
3.3 Formulation for Robust Indices ........................................................................ 80
3.3.1 Derivation of the Robust Index R.I.I ......................................................... 80

3.3.2 Derivation of the Robust Index R.I.II ........................................................ 81
3.4 A Hand Calculation of the Stochastic Methodology ........................................ 82
3.4.1 Cantilevered beam with edge loading ....................................................... 82

3.4.2 Stochastic analysis of the cantilevered beams .......................................... 83
3.4.3 Results of stochastic analysis .................................................................... 83
3.5 Effect of Statistical Properties on Robust Indices ............................................ 87
3.5.1 Stochastic analysis of cantilevered beams with varying CoV .................. 88

3.5.2 Results of stochastic analysis .................................................................... 89
3.5.3 Discussion and conclusion ........................................................................ 92
3.6 Conclusion ........................................................................................................ 92
Chapter 4 Experimental Setup and Results ............................................................... 94
4.1 Experimental Setup .......................................................................................... 96
4.1.1 Buckling test facility ................................................................................. 96

4.1.2 Panel preparation ....................................................................................... 97
4.1.3 Measurement systems for the experiments ............................................... 98
4.2 Design of the COCOMAT Panels .................................................................. 101
4.2.1 COCOMAT Design 1 (D1) panel ........................................................... 101

4.2.2 COCOMAT Design 2 (D2) panel ........................................................... 103
4.3 Imperfections and Variations of the D1 Panel ............................................... 106
4.3.1 Variations in material properties ............................................................. 106

4.3.2 Variations in geometry of D1 panel ........................................................ 108
4.3.3 Other noted variations and imperfections in the stiffened panel............. 109
4.4 Results from Cyclic Loading of D1 Panel ...................................................... 111
4.4.1 Panels P28 and P29 – Asymmetric postbuckling mode .......................... 113
ix | P a g e
4.4.2 Panels P30 and P31 – Symmetrical postbuckling mode ......................... 116
4.4.3 Panels P35 and P37 – Pre-damaged Panels ............................................ 121
4.5 Discussion of Experimental Results for the D1 Panel.................................... 126
4.5.1 Degradation of panel from cyclic loading ............................................... 126

4.5.2 Numerical simulation of degradation ...................................................... 128
4.6 Results from Cyclic Loading of the D2 panel ................................................ 131
4.6.1 Impact damage on the P27 D2 panel....................................................... 131

4.6.2 Results from experiments........................................................................ 132
4.7 Conclusion ...................................................................................................... 135
Chapter 5 Stochastic Analysis of the COCOMAT D1 Panel .................................. 137
5.1 Investigation of Geometrical Imperfections in the D1 Panel ......................... 138
5.1.1 Residual stress in a composite laminate from the curing process ........... 138
5.1.2 Results from thermal finite element analyses ......................................... 140
5.1.3 Effect of geometrical imperfections on collapse of the D1 panel ........... 144
5.1.4 Conclusion .............................................................................................. 149
5.2 Stochastic Analysis of D1 panel ..................................................................... 150
5.2.1 Pre-processing of D1 panel for analysis.................................................. 150

5.2.2 Results of stochastic analysis of the D1 panel ........................................ 151
5.2.3 Coupling of bending and compression loads in D1 panels ..................... 158
5.2.4 Conclusion .............................................................................................. 160
5.3 Conclusion ...................................................................................................... 161
Chapter 6 Definition of Robust Indices for COCOMAT Panel Designs ............... 163
6.1 Validation of Initial Assumptions .................................................................. 164
6.1.1 Initial assumptions and data from material characterisation ................... 164
6.1.2 Finite element model for D1 panel.......................................................... 165
6.1.3 Results of analysis ................................................................................... 167
6.1.4 Discussion and Conclusion ..................................................................... 172
6.2 Study of Robustness for the D1 and D2 Panels .............................................. 173
6.2.1 Analysis of D1 panel ............................................................................... 174
x|Page
6.2.2 Finite element model for D2 panel.......................................................... 176
6.2.3 Analysis of D2 panel ............................................................................... 178
6.2.4 Discussion of results ............................................................................... 181
6.2.5 Conclusions ............................................................................................. 183
6.3 Conclusion ...................................................................................................... 183
Chapter 7 Redesign of COCOMAT Panels using Robust Indices .......................... 185
7.1 The Effect of Stacking Sequence on Laminate Robustness ........................... 186
7.1.1 Design of laminates for the buckling study............................................. 186

7.1.2 Geometry and boundary conditions of plates.......................................... 187
7.1.3 Finite element modelling of the laminate................................................ 188
7.1.4 Analysis of laminate configurations........................................................ 190
7.1.6 Conclusions ............................................................................................. 194
7.2 Parametric Study of Four D2 Panel Configurations ....................................... 194
7.2.1 Laminate design for hybrid D2 panels .................................................... 195

7.2.2 Stochastic analysis of hybrid D2 panels ................................................. 197
7.2.4 Conclusion .............................................................................................. 204
7.3 Conclusion ........................................................................................................... 205
Chapter 8 Conclusion ................................................................................................. 206
8.1 Summary of Key Findings and Achievements ............................................... 206
8.1.1 Review of literature ................................................................................. 206

8.1.2 Development of Theory and Formulation ............................................... 207
8.1.3 Results of DLR and CRC-ACS COCOMAT experiments ..................... 207
8.1.4 Stochastic Analysis of COCOMAT D1 Panel ........................................ 208
8.1.5 Definition of Robust Indices for COCOMAT panel designs .................. 208
8.1.6 Redesign of COCOMAT panels using Robust Indices ........................... 209
8.2 Future Work ................................................................................................... 210
8.2.1 Guidelines for the design of stiffened composite structures ................... 210
8.2.2 Redesigning the D1 panel for further testing .......................................... 211
xi | P a g e
8.2.3 Pseudo Particle Swarm Optimisation (PSO) technique .......................... 212
8.2.4 Extending Stochastic Analysis and Robust Design to other fields ......... 212
8.2.5 Robust Design and Damage Tolerance ................................................... 213
8.3 Closing Remarks ............................................................................................ 214
References .................................................................................................................... 215

Appendix A Markov Chain Methodology ................................................................ 233
A.1 Chan’s Markov Chain Methodology .............................................................. 233
A.1.1 Theory of Markov Chains ....................................................................... 234

A.1.2 Generation of Operational Flow Chart .................................................... 235
A.1.3 Obtaining Reliability and Robustness Indices ........................................ 236
A.2 Analysis of a Truss Structure using Markov Chains ...................................... 238
A.2.1 Geometry and loading conditions ........................................................... 238

A.2.2 Operational flow charts for Trusses ........................................................ 241
A.2.3 Stochastic boundaries for trusses ............................................................ 242
A.2.4 Results from Markov Chain analysis ...................................................... 243
A.2.5 Discussion and conclusions .................................................................... 244
Appendix B Assessment of Robustness for Two Composite Yachts ...................... 245
B.1 Design specifications for composite yachts ................................................... 246

B.2 Composite yacht with monocoque hull design ............................................... 248
B.3 Composite yacht with multiple frame hull design .......................................... 252
B.4 Discussion of results ....................................................................................... 257
B.5 Conclusion ...................................................................................................... 260
xii | P a g e
List of Figures
Figure 1.1: Comparison of design scenarios for stiffened CFRP structures. .................... 6
Figure 1.2: Benchmark results from experiment and finite elements at various levels of
axial compression (mm). ................................................................................................... 7
Figure 2.1: Comparison between robust and unstable maxima. ..................................... 27
Figure 2.2: Differences in output resulting from design selection. ................................. 28
Figure 2.3: Representation of the various Robust Design types in relation to a design. 32
Figure 2.4: Damage sustained by Ronan Point and schematic of failure joint. .............. 41
Figure 2.5: Damage sustained by WTC Towers One and Two. ..................................... 43
Figure 2.6: Cantilevered beam with axial loading. ......................................................... 44
Figure 2.7: Plot of normal density function of beam displacement against beam
displacement. ................................................................................................................... 46
Figure 2.8: Homogenisation of fibre-matrix properties in CLPT. .................................. 49
Figure 2.9: Defects in unidirectional fibres..................................................................... 51
Figure 3.1: Comparison between the results for a multi-variable and multi-variant
analysis. ........................................................................................................................... 74
Figure 3.2: Example of varying degrees of Sensitivity. .................................................. 77
Figure 3.3: Example of varying degrees of Influence. .................................................... 79
Figure 3.4: Cantilevered beam with end loading. ........................................................... 82
Figure 3.5: Metamodel of tip deflection against applied load. ....................................... 84
Figure 3.6: Metamodel of tip deflection against beam length. ....................................... 84
Figure 3.7: Metamodel of tip deflection against Young’s Modulus. .............................. 85
Figure 3.8: Metamodel of tip deflection against beam breadth. ..................................... 85
Figure 3.9: Metamodel of tip deflection against beam depth.......................................... 85
Figure 3.10: Plot of tip deflection against various beam depths. .................................... 87
Figure 3.11: Metamodel of tip deflection against applied load. ..................................... 89
Figure 3.12: Metamodel of tip deflection against beam length. ..................................... 89
Figure 3.13: Metamodel of tip deflection against Young’s Modulus. ............................ 90
Figure 3.14: Metamodel of tip deflection against beam breadth. ................................... 90
xiii | P a g e
Figure 3.15: Metamodel of tip deflection against beam depth........................................ 91
Figure 4.1: Photograph and schematics of DLR test facility. ......................................... 97
Figure 4.2: Casting of end blocks and finished end block. ............................................. 98
Figure 4.3: Location of strain gauges on the D1 panel and an enlarged cross-sectional
view. .............................................................................................................................. 100
Figure 4.4: Experimental setup viewed from above . ................................................... 101
Figure 4.5: Geometrical representation of the D1 test panel with stiffener design....... 102
Figure 4.6: Location of Teflon strips in pre-damaged panels P34 to P37. ................... 103
Figure 4.7: Geometrical representation of D2 panel. .................................................... 104
Figure 4.8: Location of impact damage for pre-damaged panels P26 and P27. ........... 106
Figure 4.9: Geometrical imperfections in the D1 panel due to residual stresses from the
curing process................................................................................................................ 109
Figure 4.10: Comparison between modeled and manufactured T-stiffener. ................. 110
Figure 4.11: Defects from the bonding process. ........................................................... 110
Figure 4.12: Defects from the manufacturing process showing (a) missing lamina and
(b) poor bond adhesion indicated by a lack of uniformity on centre stiffener. ............. 111
Figure 4.13: Out of plane ARAMIS displacement results for P29. .............................. 114
Figure 4.14: Load-shortening curves for cyclic loading of panel P29. ......................... 115
Figure 4.15: Thermography readings for panel P29. .................................................... 116
Figure 4.16: Out of plane ARAMIS displacement results for P30. .............................. 117
Figure 4.17: Load-shortening curves for cyclic loading of panel P30. ......................... 118
Figure 4.20: Out of plane ARAMIS displacement results for panel P35. .................... 122
Figure 4.21: Load-shortening curves for the cyclic loading of panel P35. ................... 122
Figure 4.23: Out of plane ARAMIS displacement results for panel P37. .................... 124
Figure 4.24: Load-shortening curves for the cyclic loading of panel P37. ................... 125
Figure 4.26: Plot of maximum applied load against number of applied cycles. ........... 128
Figure 4.27: Comparison of disbonds from experiment and initial disbonds applied in
FEM. ............................................................................................................................. 129
xiv | P a g e
Figure 4.28: Out of plane displacement results from finite elements for undamaged and
damaged D1 panels. ...................................................................................................... 130
Figure 4.29: Comparison of load-shortening plots for degradation between experiment
and FEM. ....................................................................................................................... 131
Figure 4.30: Impact damage for P27 showing; (a) damage on skin-side, (b) damage on
stiffener-side, (c) ultrasound image on skin-side and (d) ultrasound image on stiffener-
side. ............................................................................................................................... 132
Figure 4.31: Out of plane displacement results from ARAMIS for P24 and P27. ....... 133
Figure 4.32: Load-shortening curve for the collapse of panels P24 and P27. .............. 133
Figure 5.1: Finite element model for thermal analysis of the D1 panel. ....................... 140
Figure 5.2: Comparison of initial geometrical imperfections in panels due to residual
stresses from curing process.......................................................................................... 141
Figure 5.3: Metamodel of net skin deformation against Ply 1 Orientation. .................. 143
Figure 5.4: Metamodel of net skin deformation against Ply 2 Thickness. .................... 143
Figure 5.5: Boundary conditions for D1 finite element panel. ..................................... 145
Figure 5.6: Damage in skin-stiffener bond and the finite element model used to simulate
the failure. ..................................................................................................................... 147
Figure 5.7: Out of plane displacement plots from ARAMIS and FEA. ........................ 148
Figure 5.8: Load shortening plots for stiffened panels compressed to collapse. .......... 149
Figure 5.9: Finite element model with created node and MPC..................................... 151
Figure 5.10: Metamodel for maximum compression load against longitudinal Young’s
Modulus E11. ................................................................................................................. 153
Figure 5.11: Metamodel for maximum compression load against Blade extensional
stiffness A11. .................................................................................................................. 153
Figure 5.12: Sample load shortening curves for Samples 2, 8 and 32 obtained from
stochastic analysis using finite elements. ...................................................................... 154
Figure 5.13: Sample load shortening curves for Samples 9 and 20 obtained from
stochastic analysis using finite elements. ...................................................................... 155
Figure 5.14: Out of plane displacement plots corresponding to the load shortening
curves in Figures 5.12 and 5.13. ................................................................................... 156
Figure 5.15: Out of plane displacement plots of the D1 panel from the DLR benchmark
and COCOMAT experiments. ...................................................................................... 157
xv | P a g e
Figure 5.16: Plot of postbuckling mode shape at 3 mm compression against applied
rotation. ......................................................................................................................... 158
Figure 5.17: Anticlastic bending of a curved panel. ..................................................... 159
Figure 5.18: Rotational displacement of middle stiffener in finite element D1 panel. 160
Figure 6.1: Finite element model of D1 panel. ............................................................. 166
Figure 6.2: Load shortening plots comparing P30 and FEM D1 panels using measured
and assumed mean values. ............................................................................................ 168
Figure 6.3: Load shortening plots comparing upper and lower bounds of FEM D1
panels using measured and assumed values. ................................................................. 169
Figure 6.4: Metamodel of applied compression load against longitudinal Young’s
Modulus. ....................................................................................................................... 171
Figure 6.5: Metamodel of applied compression load against skin composite extensional
stiffness A11. .................................................................................................................. 171
Figure 6.6: Metamodel of applied compression load against blade composite extensional
stiffness A11. .................................................................................................................. 172
Figure 6.7: Comparison of results for D1 panel from experiment and finite element
analysis. ......................................................................................................................... 175
Figure 6.8: Load shortening curves for D1 FEM with upper and lower bounds. ......... 175
Figure 6.9: Finite element model of D2 panel. ............................................................. 177
Figure 6.10: Comparison of D2 buckling modes from experiment and finite element
analysis. ......................................................................................................................... 178
Figure 6.11: Load shortening plots from experimental panel P24 and finite element
analysis. ......................................................................................................................... 179
Figure 6.12: Load shortening curves for D2 FEM with upper and lower bounds. ....... 180
Figure 6.13: Metamodel of compression load against skin extensional stiffness. ........ 182
Figure 6.14: Metamodel of compression load against blade extensional stiffness. ...... 182
Figure 7.1: Laminate configuration and equivalent bending stiffness of plates used in
study. ............................................................................................................................. 187
Figure 7.2: Geometrical representation of flat plate. .................................................... 188
Figure 7.3: Finite element model of composite plate.................................................... 189
Figure 7.4: Eigenvector plot of the panel at bifurcation buckling. ............................... 190
Figure 7.5: Metamodel of Buckling Factor against laminate extensional stiffness. ..... 192
Figure 7.6: Metamodel of Buckling Factor against laminate coupling stiffness. ......... 192
xvi | P a g e
Figure 7.7: Metamodel of Buckling Factor against laminate bending stiffness. .......... 193
Figure 7.8: Plot of Young’s Modulus E11 and first derivative against global orientation.
....................................................................................................................................... 194
Figure 7.9: Laminate configurations and equivalent bending stiffness. ....................... 196
Figure 7.10: Load shortening curves for D2(b) FEM with upper and lower bounds. .. 199
Figure 7.11: Load shortening curves for D2(c) FEM with upper and lower bounds. ... 199
Figure 7.12: Load shortening curves for D2(d) FEM with upper and lower bounds. .. 200
Figure 7.13: Metamodel of compression load against longitudinal Young’s Modulus.
....................................................................................................................................... 201
Figure 7.14: Metamodel of compression load against skin extensional stiffness. ........ 201
Figure 7.15: Metamodel of compression load against blade extensional stiffness. ...... 202
Figure 7.16: Load shortening curves for experimental D2 and hybrid D2 FEM. ......... 203
Figure A.1: Operational Flow Chart for transitional matrix. ........................................ 235
Figure A.2: Pin-jointed truss structures subjected to a horizontal force. ...................... 239
Figure A.3: Operational Flow Chart for Truss A. ......................................................... 241
Figure A.4: Operational Flow Chart for Truss B. ......................................................... 242
Figure A.5: Plots of reliability and robustness of Truss A and Truss B. ...................... 244
Figure B.1: Region of structure that damage is simulated. ........................................... 247
Figure B.2: Isometric view of monocoque hull. ........................................................... 249
Figure B.3: Buckling occurring on the top deck of monocoque yacht. ........................ 249
Figure B.4: Metamodel of buckling factor against Shear Modulus G23 before and after
damage has been applied. .............................................................................................. 251
Figure B.5: Metamodel of buckling factor against skin extensional stiffness A11 before
and after damage has been applied................................................................................ 252
Figure B.6: Isometric view of multiple frame hull. ...................................................... 253
Figure B.7: Buckling occurring on the top deck of multiple frame yacht. ................... 254
Figure B.8: Metamodel of buckling factor against Poisson’s Ratio of the core material
before and after damage has been applied. ................................................................... 256
Figure B.9: Metamodel of buckling factor against skin extensional stiffness A11 before
and after damage has been applied................................................................................ 256
Figure B.10: Plots for change in buckling factor against damaged area between
monocoque and multiple frame hull designs. ............................................................... 259
xvii | P a g e
List of Tables
Table 1.1: List of COCOMAT participants. ..................................................................... 4

Table 1.2: Contributions of the COCOMAT project to previous design guidelines. ....... 8
Table 2.1: Application of Types I, II and III Robust Design in structural design. .......... 32
Table 2.2: Stochastic boundary for circular beam. ......................................................... 46
Table 2.3: Sources of damage that affect composite structures. ..................................... 53
Table 3.1: Stochastic boundary for Beam A and Beam B. ............................................. 83
Table 3.2: Results for stochastic analysis of Beam A and Beam B. ............................... 86
Table 3.3: Stochastic boundaries for Beam C, D and E. ................................................. 88
Table 3.4: Results for stochastic analysis of Beam C, D and E. .................................... 91
Table 4.1: Breakdown of panels tested in COCOMAT project. ..................................... 95
Table 4.2: Nominal geometry for the D1 panel. ........................................................... 102
Table 4.3: Nominal geometry for D2 panel. ................................................................. 104
Table 4.4: Nominal material properties for Hexcel IM7/8552. .................................... 107
Table 4.5: Nominal and measured geometry for the D1 panel. .................................... 108
Table 4.6: Summary of cyclic loading regime. ............................................................. 112
Table 4.7: Load carrying capability of P29 subjected to stiffener disbond. ................. 116
Table 4.10: Load carrying capability of P35 subjected to stiffener disbond. ............... 123
Table 4.11: Load carrying capability of P37 subjected to stiffener disbond. ............... 126
Table 4.12: Decrease in applied loads due to skin-stiffener disbonds. ......................... 128
Table 5.1: Stochastic boundary for the D1 FE stochastic analysis. .............................. 139
Table 5.2: Results of stochastic analysis for thermal analyses of the D1 panel............ 142
Table 5.3: Fracture properties for IM7/8552 carbon/epoxy unidirectional tape . ......... 147
Table 5.4: Results from stochastic analysis of D1 panel. ............................................. 152
Table 6.1: Boundaries for stochastic analyses for Hexcel IM7/8552 material. ............ 165
Table 6.2: Stochastic boundary for analysis of D1 panel. ............................................ 166
xviii | P a g e
Table 6.3: Settings for MSC.Nastran nonlinear solver. ................................................ 167
Table 6.4: Results of stochastic analyses. ..................................................................... 170
Table 6.5: Stochastic boundary for the D1 and D2 panels. ........................................... 174
Table 6.6: Results of stochastic analysis for the D1 panel. ........................................... 176
Table 6.7: Results of stochastic analysis for the D2 panel. ........................................... 180
Table 6.8: Summary of results from analysis of the D1 and D2 panels........................ 181
Table 7.1: Nominal geometry for composite plates. ..................................................... 188
Table 7.2: Stochastic boundary for the plates. .............................................................. 190
Table 7.3: Results of stochastic analysis for the plates. ................................................ 191
Table 7.4: Laminate configurations for D2 panel parametric study. ............................ 195
Table 7.5: Results of stochastic analysis for the hybrid D2 panels. .............................. 197
Table 7.6: Summary of results for stochastic analysis of the D2 panels....................... 203
Table A.1: Loading of members in Truss A and Truss B. ............................................ 239
Table A.2: Stochastic boundary for Truss A and Truss B. ........................................... 243
Table A.3: Results from Markov Chain analysis. ......................................................... 243
Table B.1: Principal dimensions for composite yachts. ................................................ 246
Table B.2: Stochastic boundary for the composite yachts. ........................................... 247
Table B.3: Results for the analysis of the monocoque yacht hull. ................................ 250
Table B.4: Results for the analysis of the multiple frame yacht hull. ........................... 255
Table B.5: Summary of results for analysis of monocoque and multiple frame hulls. . 257
xix | P a g e
Abbreviations and
Acronyms
Term Definition
2D, 3D Two Dimensional, Three Dimensional
Aernnova Aernnova Engineering Services
BVID Barely Visible Impact Damage
CAE Computer Aided Engineering
CFRP Carbon Fibre Reinforced Polymer
CLPT Classical Laminated Plate Theory
CoV Coefficient of Variation
CTE Coefficient of Thermal Expansion
COCOMAT Improved MATerial Exploitation at Safe Design of COmposite
Airframe Structures Through More Accurate Predictions of COllapse
CRC-ACS Cooperative Research Centre for Advanced Composite Structures
Limited, Australia
D1, D2 COCOMAT Design 1, COCOMAT Design 2
DLR German Aerospace Center (Deutsches Zentrum für Luft und
Raumfahrt)
EC European Commission
FE Finite Element
FEA Finite Element Analysis
FEM Finite Element Model
LVDT Linear Variable Displacement Transducer
MPC Multiple Point Constraint
NDT Non Destructive Testing
OLT Optically Lock-in Technology
POSICOSS Improved POst-buckling SImulation for Design of Fibre COmposite
Stiffened Fuselage Structures
xx | P a g e
Term Definition
PSO Particle Swarm Optimisation
RI Redundancy Index
SA Sensitivity Analysis
TM Taguchi Method
UNSW University of New South Wales
VCCT Virtual Crack Closure Technique
xxi | P a g e
Nomenclature
Term Definition
A Area/ Extensional component of laminate stiffness matrix
A Area
B Coupling component of laminate stiffness matrix
B Scaled Sensitivity
b Breadth/ Stiffener pitch
D Bending component of laminate stiffness matrix
d Depth
E Elastic modulus
f Function
G Shear modulus
h Height
I Moment of inertia
K Edge support length
L Length
m Sample of size
n Number of input variables
P Load
R Radius of curvature / Strength / Ordinal Rank
r Radius
r Number of output variables
R.I. Robust Index
S Stochastic analysis
W Arc length
w Width
X Input Variable
x Sampled Input
Y Output response
xxii | P a g e
Term Definition
β Sensitivity
Δ Change
μ Sample mean
ρ Influence
σ Standard deviation
υ Poisson’s Ratio
Subscripts
Term Definition
1,2,3 In ply coordinate system, fibre, matrix and through-thickness directions
11 Term in laminate stiffness matrix
c Compression
f Free length
L Longitudinal direction
s Symmetric
T Transverse direction
t Tension
Superscripts
Term Definition
I First Index
II Second Index
xxiii | P a g e
Chapter 1 Introduction
Chapter 1
Introduction
1.1 Background
The Boeing 787 and Airbus A350 aircraft represent the latest advances in design of
commercial transport aircraft. Both aircraft will have significant use of carbon fibre for
the airframe. This is due to the advantages of carbon fibre including the high strength to
weight ratio, resistance to fatigue and corrosion and the ability to form shapes that
conform to the aerodynamic requirements. These characteristics translate into light, fuel
efficient structures that should be easier to maintain than current aluminium airframes.
An enduring characteristic of the aerospace industry is the drive to produce aircraft that
operate more efficiently. This has largely been influenced by the reduction in profit
margins for operators due to increased prices for aviation fuel. The recent focus on the
efficiency of aircraft has also been driven by an environmental perspective with
demands by governments and the general public to reduce carbon emissions. In order to
achieve a lower carbon footprint, three areas in the design of aircraft have been
identified by industry as being the key to higher operational efficiency. These include
the propulsion systems, aerodynamics and structural design of the aircraft. This thesis
focuses on the last aspect, the design of stiffened composite fuselage structures.
Compared to traditional materials, such as aluminium, anisotropic composites are more

complex to manufacture due to the multiple processes required in design and
fabrication. Each step of the process will include variation in geometrical and material
properties that can affect the overall behaviour of the finished structure. Moreover the
1|Page
anisotropic properties cause the material to lose stiffness once the orientations of the
applied forces change.
This uncertainty is not new. Most data managed by stress engineers for designing
structures is not certain, existing within a range of values. Materials are not
homogenous, geometries are never perfect, loads fluctuate and boundary conditions are
never ideal. Since this is the case, using single deterministic values will lead to a
situation where engineers are unable to predict potential flaws and drawbacks when
designing structures. Therefore it may not be prudent to use deterministic values in
deign calculations but rather a stochastic approach, where the values within a plausible
range are considered. For example the nominal value for the tensile strength of Hexcel
IM7/8552 CFRP is considered 1740 MPa but, in reality, the tensile strength has been
found to vary about this mean value with a coefficient of variation of 12% (Lee et al.
2008).
With the advent of numerical methods such as finite element analysis (FEA), it is
possible to analyse and investigate the behaviour of structures in parallel with
experimental testing. While the results from physical experiments usually present a
range of scatter, this is not the case for numerical simulations as finite element models
(FEM) with perfect physical properties and boundary conditions are often used. In non-
linear failure modes such as postbuckling and collapse, the perfect models often exhibit
a different failure sequence or buckling pattern from the real-life experiments. This
presents an issue when attempting to benchmark the FEA results with the experiments
so that future tests can be simulated.
The current methodology for coping with design uncertainty is to apply safety factors
on to the structure so that premature failure does not occur. While safety factors do
provide a tolerance to variability and uncertainty in the analysis, they do not offer any
indication of the sensitivity of a design due to these uncertainty factors; the performance
of the structure can be highly unstable, thereby negating any gains through the factors of
safety. This is an important consideration as highly optimised structures, such as those
used in the aerospace industry, are required to survive unexpected loading and operating
conditions.
2|Page
This research has been driven by the desire to understand and account for the effects of
imperfections and variations in structural designs. It endeavours to create a
methodology so that the robustness of structures, when subjected to these uncertainties,
can be quantified and therefore assist the engineer in the selection of design
configurations. This is particularly important in stiffened curved shell structures where
the effects of boundary conditions, geometry, material and manufacturing quality can
drastically alter the failure mechanism when subjected to compressive loading.
It is an important requirement to have structures that perform within the designed

specifications. This is especially the case in non-linear failure modes such as buckling
and collapse which are highly sensitive to imperfections in geometry and variations in
boundary conditions. While it is acceptable for the actual structure to have an actual
designed critical load that is above failure, a situation where a load bearing member
fails before the predicted load is reached must be avoided. It is also important for the
failure to be contained within the bounds envisaged by the designer.
1.2 The COCOMAT Project
One approach for weight reduction in structural components is to design them for
postbuckling which allows the structure to continue carrying loads between the critical
buckling load and the load at which the structure collapses. Postbuckling structures have
been commonly used in metallic designs for decades but are not common in composite
designs due to concerns which relate to the ductility and durability of carbon fibre
reinforced polymer (CFRP) structures in operation and the accuracy of the available
design tools.
The European Commission (EC) 6th Framework project COCOMAT (Improved

MATerial Exploitation at Safe Design of COmposite Airframe Structures by Accurate
Simulation of COllapse) was a four and a half year project (2004 to mid-2008) aimed at
exploiting the large reserve of strength in composite structures through more accurate
prediction of collapse. There is a demand in the European aircraft industry for a
3|Page
reduction in developmental and operational cost, by 20% and 50% in the short and long
term, respectively. The COCOMAT project contributes to this aim by reducing
structural weight for safe design; it exploits considerable reserves in primary fibre
composite fuselage structures by accurate and reliable simulation of collapse.
COCOMAT aims to create a validated approach for including material degradation in
numerical analyses so that collapse can be more accurately predicted. This requires
knowledge about degradation as a result of static and low cycle loading in the
postbuckling range.
COCOMAT is based on the results of the prior EC 5th Framework project POSICOSS
(Improved POstbuckling SImulation for Design of Fibre COmposite Stiffened Fuselage
Structures) which developed faster and more reliable procedures for buckling and
postbuckling analysis of fibre composite stiffened panels of future fuselage structures,
created experimental databases and derived design guidelines. The COCOMAT project
extends the results from POSICOSS by accounting for material degradation, as current
design scenarios for composite structures require that the onset of degradation occurs
after the design ultimate load. It is well known that thin-walled structures made of
CFRP can tolerate repeated buckling without any change in their buckling behaviour.
However, the level at which a structure can go into the postbuckling regime without
severe damage must be determined.
Table 1.1 lists the 15 participants in the COCOMAT project.
Table 1.1: List of COCOMAT participants.
Partner Country Type
German Aerospace Center (DLR) Germany Research institution
Cooperative Research Centre for Advanced

Australia Research institution
Composite Structures (CRC-ACS)
The Swedish Defence Research Agency (FOI) Sweden Research institution
Aernnova Engineering Solutions (Aernnova) Spain Industrial partner
4|Page
Partner Country Type
Hellenic Aerospace Industries (HAI) Greece Industrial partner
Israel Aircraft Industries (IAI) Israel Industrial partner
PZL-Swidnik S.A. Poland Industrial partner
Agusta S.p.A. Italy Industrial partner
Israel Institute of Technology (TECHNION) Israel University
Karlsruhe University (UKA) Germany University
RWTH Aachen Germany University
Riga Technical University (RTU) Latvia University
Politecnico di Milano (POLIMI) Italy University
Samtech S.A. Belgium Software developer
SMR S.A. Switzerland Software developer
Figure 1.1 shows the current and future design scenarios for stringer stiffened composite
panels that COCOMAT aims to achieve. The left graph illustrates a simplified load-
shortening curve and highlights the current industrial design scenario. Three different
regions can be specified: Region I covers loads allowed under normal operating flight
conditions and is bounded above by the limit load; region II is the safety region and
extends up to the ultimate load; region III comprises the region from the ultimate load to
collapse which the current design codes do not allow for in CFRP stiffened structures.
There is a large unemployed structural reserve capacity between the current ultimate
load and collapse which can be exploited. The right graph of Figure 1.1 depicts the
future design scenario, where the ultimate load is shifted towards collapse as close as
possible. Another main difference to the current design scenario is that the onset of
degradation is moved from region III to region II due to reliable simulation of
degradation and collapse. This is comparable to metallic structures where plasticity is
already permitted in the safety region. However, it must be guaranteed that in any case
the onset of degradation must not occur below limit load. Moreover, the extension
requires an accurate and reliable simulation of collapse, which means to take into
5|Page
account degradation under static as well as under low cycle loading, in addition to
geometrical nonlinearity.
Figure 1.1: Comparison of design scenarios for stiffened CFRP structures.

(COCOMAT, 2004)
In the benchmark experiment for one of the panel designs tested in COCOMAT it was
found that an asymmetrical postbuckling mode shape was achieved at the point before
collapse occurred (Orifici et al. 2008). This differed from the finite element results
where there was a symmetrical postbuckling mode shape before collapse. It was thought
that the asymmetrical postbuckling mode shapes achieved through physical testing were
possibly due to three reasons, which were:
(i) Warping of the panels at curing and manufacture.

(ii) Imperfections in the stiffener blade width.
(iii) Asymmetric introduction of the load onto the panel.
The benchmark finite element models did not include any of the imperfections above
and hence no match with the benchmark experimental result was obtained. The
experimental result followed by the finite element results can be seen in Figure 1.2.
6|Page
Experiment
P23 Abaqus
Nastran
Imperfect
Nastran
Figure 1.2: Benchmark results from experiment and finite elements at various levels of
axial compression (mm).
In the design guidelines that were defined in the final report prepared for COCOMAT, it
was stated that:
‘Postbuckling stiffened structures undergoing damage up to collapse are sensitive to a

range of parameters, including: imperfections in the geometry, load or boundary
conditions; manufacturing variability such as ply misalignment or variations in bond
quality; and the unpredictable nature of buckling and postbuckling problems with
regards to mode shape development. In a design context, this variability must be
accounted for as part of the solution process. This means that for any design solution, it
is critical to understand whether variances in the design parameters within expected
tolerances affect the performance of the design. This is to ensure that the design solution
continues to provide the required performance in real manufacturing and loading
7|Page
scenarios.’ (COCOMAT Design and Analysis Guidelines, Orifici, Lee and Thomson
2008)
The final conclusions of the COCOMAT project are shown in Table 1.2. It can be seen
that statistical variability is an issue in bifurcation buckling and postbuckling and that
there is a need to address the large scatter that was observed in the experiments.
Table 1.2: Contributions of the COCOMAT project to previous design guidelines.

(COCOMAT Final Activity Report, Degenhardt 2009)
Current Industrial Practice Design Guidelines from COCOMAT
Local skin buckling generally Local skin buckling can be allowed at 40% of
not allowed below Ultimate Ultimate Load. Relatively large scatter to be
Load. expected for the 1st buckling load estimation.
Global buckling not allowed below Ultimate
Load. A safety factor (10%-20%) should be
Global buckling not allowed
maintained between global buckling (collapse)
below Ultimate Load
and ultimate load, to account for statistical
variability.
“No-Growth” approach. Use of May be changed, subject to further improvement
threshold values for Damage and reliability of computational tools developed
Tolerance. within COCOMAT.
Far-field strain level allowed:
Unchanged.
3000 -5000 με.
Size of acceptable damage: Size of acceptable damage: may be increased to
up to 0.5”. 2.0” at least.
This thesis was conducted in conjunction with the CRC-ACS as a contribution to the
COCOMAT project. The majority of the work was applied to the analysis of composite
postbuckling structures taking variations and imperfections into account. This was
integrated within collaborative work between CRC-ACS and the DLR Institute of
Composite Structures and Adaptive Systems in Braunschweig, Germany, which was
centred on experimental testing of fuselage-representative composite panels and
involved design, manufacture, testing, analysis and design optimisation.
8|Page
As a result of the considerations just discussed, the following aims were defined for the
thesis.
(i) To apply stochastic analysis to develop an understanding of the sources and

significance of the variability encountered in the buckling of stiffened shell
structures in the COCOMAT project.
(ii) To develop a design methodology to enable the design of more robust
configurations that exhibit a reduction in the significant variations.
To achieve these objectives the following work packages needed to be completed.
(i) To review literature pertaining to the current state of the art in design
philosophies relating to imperfection sensitivity in structural mechanics and
composite structures together with the areas relating to postbuckling failure.
(ii) The development of a stochastic simulation methodology to determine the
possible responses of complex structures in order to better understand
experimental results.
(iii) The determination of the variables that can be used to quantify the robustness
of composite structures.
(iv) The development of a quantitative measure for the robustness of structures to
provide a design methodology to reduce the variations encountered in the
experimental results.
(v) The validation of the underlying assumptions for variations used in the
simulations so that future analyses can be conducted with the use of realistic
scatter in the input variables.
(vi) The application of the developed methodology for the analysis of two stiffened
curved panel designs used in the COCOMAT project to demonstrate the
procedures developed.
9|Page
1.3 Outline of Thesis
A comprehensive literature review is presented in Chapter 2 where previous research

relevant to this thesis is summarised. The concept of stochastic analysis is first
introduced followed by the Robust Design philosophy which is presented together with
the advancement of the theory which originates from the manufacturing field. The
review of literature continues with a description of progress made in structural
engineering using theories such as progressive and disproportionate failure that are
comparable to Robust Design. As the focus of this work is on composite structures,
typical imperfections and damage sources in composite materials are described,
followed by the studies on stiffened postbuckling composite structures. Finally the
literature review is used as a basis to formulate the framework of the quantitative
methodology developed in this work.
In Chapter 3, the stochastic procedures developed in this thesis are defined. First a
methodology that defines a family of designs in which variations are randomly imposed
is described. This approach is possible due to the development of modern finite element
methods that automatically generate meshes and execute analyses with no user input.
The range of variation can therefore include geometry, curing conditions, material
properties and variations in loads and boundary conditions. The approach is also only
feasible because each analysis can be completed within a timeframe that permits the
analysis of the family of designs to be completed on a single computer or across
multiple platforms. The solutions can then be subject to statistical analysis to identify
trends and multiplicity is responses including different local buckling patterns.
A quantitative measure for the robustness of structures is then developed. This is firstly
motivated by the difficulty in accurately matching simulation results with experimental
testing as the analysed models often contain no discernable imperfections. It is also
motivated by the lack of any measure for the robustness of structures. Variability in
results should only be of concern if it affects the properties required for the design –
such as the growth of damage and the ultimate failure load. The stochastic methodology
is able to account for the scatter observed from manufacturing and operating conditions.
Robust Indices comprising of two measures R.I. I and R.I.II are derived using two
10 | P a g e
variables, Influence and Sensitivity. The approach is validated using a simple hand
calculation on the deflection of a cantilevered beam.
A description of the experiments performed in collaboration between the CRC-ACS and

the DLR for the COCOMAT project is provided in Chapter 4. The experimental set up,
as well as the specification of the panels tested, is shown followed by the results from
the experimental testing. These results provide information for material characterisation
as well as the imperfections and defects that can be found on manufactured panels. The
experimental results also include the degradation experienced by the panels subject to
cyclic loading. These results are used as a basis to validate the finite element analyses as
well as to provide plausible knockdown factors for the certification of future composite
panels. They therefore contribute to the outcomes for the COCOMAT project.
In Chapter 5 the stochastic methodology developed Chapter 3 is applied to the

COCOMAT panels where the mismatch between the experimental and FEA
postbuckling is investigated. The stochastic methodology is applied to identify the
causes of the curing deformations and the buckling modes that were obtained in the
experiments using the manufactured panels. The factors affecting the difference in
collapse loads in the experiments and numerical simulation are also investigated.
The aim of the COCOMAT program is to develop procedures that can be used in the
design of imperfection sensitive panels. Chapter 6 applies the Robust Indices
developed in Chapter 3 to design. An assessment of the initial input parameters is
conducted followed by a study on the robustness for two COCOMAT experimental
panel designs, D1 and D2, using the Robust Indices. The load carrying capability for the
panels is used as a basis for the comparison.
The Robust Indices are intended to be quantitative measures that can be easily applied
in future design studies where multiple designs can be effectively evaluated. In Chapter
7, an initial study into the effect of laminate configuration is conducted in order to
investigate the differences that can arise in buckling load and robustness. This is
followed by a design exercise where a redesign of the D2 panel is performed.
11 | P a g e
Parametric studies are conducted on both the skin and stiffener laminates to find a
design that is more robust than the experimental D1 and D2 panels.
Finally, a conclusion to this work is given in Chapter 8. A summary of the key findings
are presented, together with some thoughts for further work arising from the research. A
bibliography containing all references used in this thesis is also provided.
Appendix A contains the preliminary development of an alternate methodology using

Markov Chains to classify robustness for multiple-state systems. The research in this
thesis also supported another research programme involving composite yacht hull
design. Details of this work are presented in Appendix B.
1.4 Research Outcomes
The major outcome of this research has been the development of a methodology for
application of a stochastic analysis to the design of imperfection sensitive curved
postbuckling composite panels. To extend the work to design, a novel methodology has
also been developed that accounts for stability and robustness of structures with the
application of two quantitative measures of robustness.
This thesis has therefore produced valuable outcomes as CRC-ACS contributions

towards the COCOMAT project. These included portions from the literature review,
benchmarking analysis, the development of the stochastic methodology for the
measures of robustness as well as the analyses of the structures to which the
methodology has been applied. The research presented in this thesis is significant as it
provides a strong foundation by allowing critical structures to be designed for
robustness which will increase survivability, as uncertainties and imperfections can be
accounted for.
A comprehensive literature review has been conducted to review stochastic analysis and
the Robust Design philosophy and their relevance to structural design. This section also
summarises the areas in composite materials and postbuckling structures where Robust
12 | P a g e
Design is essential due to the existence of imperfections and variations in the structures
under operational environments. These outcomes were used to identify the current
shortcomings in the design of composite structures and the inability to reconcile the
differences in real-life operation and numerical simulations.
The stochastic methodology involving repeated sampling has been used in order to draw
out the possible responses that a structure will present under realistic manufacturing and
operating conditions. This has allowed the responses observed in the initial
benchmarking of COCOMAT to be replicated and understood. An unexpected bending
component in the structure, which was possibly due to exceedance of tolerances in the
lay-up of the stringer, and uneven loading along the radial axis during compression is
identified as the major causes of the observed asymmetry in the postbuckling mode
shape of the stiffened panel at collapse.
To extend the procedure to design the scatter has been characterised using two key
variables, Influence and Sensitivity. The Influence measures how much each input
variable affects the response while Sensitivity determines the sensitivity of the output
response with respect to the input variation. Two measures of robustness have been
derived in the form of the Robust Indices, R.I.I and R.I.II; both measures have been
derived from Influence and Sensitivity. The Robust Index has been developed so that it
can be a universal measure of robustness where the effects of imperfection can be
analysed and characterised.
The experimental results have revealed the worst-case scenarios for the degradation of
the skin-stiffener bond. The D1 panel design has 0o/90o laminae on the skin-stiffener
interface in order to have the worst-case fracture properties so as to achieve large
disbond growth. Under cyclic loading in the deep postbuckling region, the D1 panel
exhibited a maximum drop in load-carrying capability of 15% when loaded at 85% of
ultimate collapse. Hence a knockdown factor of 0.73 is proposed as this drop in
stiffness is used as a feasible knock-down factor when designing stiffened panels in the
future as it also contributes to the allowance of degradation before ultimate load is
reached.
13 | P a g e
The stochastic methodology has been successfully applied to the D1 and D2 panel
designs from the COCOMAT project and the results have revealed the parameters that
would most affect the buckling load of the panels. A redesign of the D2 panel was
undertaken with four separate laminate configurations and a parametric study based on
the R.I.I and R.I.II measures was carried out. This study was able to demonstrate the
effectiveness of the quantitative robustness measures and its valuable contribution in
design studies.
Quality control in the manufacturing of composite yachts has always been an issue as it
is a common occurrence in the industry to have designs that are not built to
specification. This includes issues with material quality and fibre misalignment. An
appendix has been included where the robust indices have been applied to the design of
a yacht hull. The work was done in support of another research thesis in yacht hull
design (Payne 2007). The measures from the Robust Indices were able to reveal which
parameters were affecting the design and also the robustness of the structures when
subjected to damage. This study also demonstrated that structural redundancy improves
robustness by allowing the loads to be transferred via alternative loading paths.
The key outcomes of the research presented in this thesis have resulted in the following
journal publications, international conferences, invited presentations and awards:
Journal Papers:
(i) Lee, M. C. W., Payne, R. M., Kelly, D. W. and Thomson, R. S., Determination
of Robustness for a Stiffened Composite Structure using Stochastic Analysis,
Composite Structures, (2008), Vol. 86(1-3), pp. 78-84.
(ii) Kelly, D., Lee, M. C. W., Orifici, A., Thomson, R. and Degenhardt, R.
Collapse Analysis, Defect Sensitivity and Load Paths in Stiffened Shell
Composite Structures, Computers Materials and Continua, (2009), Vol.10(2),
pp. 163-194.
14 | P a g e
(iii) Lee, M. C. W., Kelly, D. W., Degenhardt, R. and Thomson, R. S., A Study on
the Robustness of Two Stiffened Composite Fuselage Panels, Composite
Structures, (2010), Vol. 92(2), pp. 223-232.
(iv) Lee, M. C. W., Mikulik, Z, Kelly, D. W., Thomson, R. S. and Degenhardt, R.

Robust Design – A Concept for Imperfection Insensitive Stiffened Composite
Structures, Composite Structures (In Press).
(v) Lee, M. C. W., Mikulik, Z, Kelly, D. W. and Thomson, R. S. An Assessment of

Stiffened Composite Fuselage Panels based on the Robustness Parameter,
International Journal of Structural Stability and Dynamics (In Press).
Conference Proceedings:
(i) Lee, M., Kelly, D., Orifici, A. and Thomson, R., Postbuckling Mode Shapes of
Composite Stiffened Fuselage Panels Incorporating Stochastic Variables,
Proceedings of the 1st CEAS European Air and Space Conference – Century
Perspectives, Berlin Germany, 10 – 13 September 2007.
(ii) Lee, M. C. W., Payne, R. M., Kelly, D. W. and Thomson, R.S., Determination
of Robustness for a Stiffened Composite Structure using Stochastic Analysis,
Proceedings of the 14th International Conference on Composite Structures,
Melbourne Australia, 19 – 21 November 2007.
(iii) Lee, M. C. W., Kelly, D. W., Thomson, R. S. and Degenhardt, R., Imperfection
Investigation of Stiffened Composite Fuselage Panels for Postbuckling
Analyses, Proceedings of the 5th International Conference on Thin Walled
Structures, Brisbane, Australia, 18 – 20 June 2008.
(iv) Lee, M. C. W., Thomson, R. S., Degenhardt, R. and Kelly, D. W., A Stochastic
Study on the Robustness of a Stiffened Composite Structure, Proceedings of the
2nd International Conference on Buckling and Postbuckling Behaviour of
15 | P a g e
Composite Laminated Shell Structures, Braunschweig, Germany, 2 – 5

September 2008.
(v) Lee, M. C. W., Kelly, D. W. and Thomson, R. S., Is Robust Design Another
Name For Damage Tolerant Design?, Proceedings of the 13th Australian
International Aerospace Congress, Melbourne, Australia, 9 – 12 March 2009.
(vi) Kelly, D., Lee, M. C. W., Orifici, A., Thomson, R. and Degenhardt, R.
Collapse Analysis, Defect Sensitivity and Load Paths in Stiffened Shell
Composite Structures, Proceedings of the International Conference on
Computational and Experimental Engineering and Sciences 2009 Conference,
Phuket, Thailand, 8 – 13 April 2009.
(vii) Lee, M. C. W., Mikulik, Z., Kelly, D. W., Thomson, R. S. and Degenhardt, R.
Robust Design – An Evolution for Imperfection Insensitive Composite
Structures, Proceedings of the 15th International Conference on Composite
Structures, Porto, Portugal, 15 – 17 June 2009.
Invited Presentations:
(i) Robust Design of Composite Structures, Colloquium at the Institute of

Composite Structures and Adaptive Structures, German Aerospace Centre
(DLR), Braunschweig, Germany, 18th October 2007.
(ii) Development of a Methodology for the Robust Design of Composite Structures,

Colloquium at the Institute of Composite Structures and Adaptive Structures,
German Aerospace Centre (DLR), Braunschweig, Germany, 11th February
2008.
(iii) Determination of Robustness for Stiffened Composite Structures, Presented at

Airbus Deutschland, Hamburg, Germany, 18th February 2008.
16 | P a g e
(iv) Development of a Methodology for the Robust Design of Composite Structures,

Guest Lecture at Private University of Applied Science (PFH) Göttingen,
Stade, Germany, 22nd February 2008.
(v) Development of a Methodology for the Robust Design of Composite Structures,

Composites Australia/Composites CRC Annual Conference and Exhibition,
Melbourne, Australia, 13 March 2008.
Awards:
(i) Full Postgraduate Research Scholarship, Cooperative Research Centre for

Advanced Composite Structures, 2006.
(ii) Postgraduate Research Support Scheme (PRSS) Funding, University of New

South Wales, 2007.
(iii) The William McIlrath Postgraduate Travel Scholarship, University of New

South Wales, 2007.
(iv) German Academic Exchange Service (DAAD) Research Grant for Doctoral
Candidates and Young Scientists, 2007.
1.5 Participation in the COCOMAT Project
The author was the recipient of a German Academic Exchange Service (DAAD)
Research Grant for Doctoral Candidates and Young Scientists in 2007. This grant was
supplemented by a travel grant from UNSW that enabled the author to spend six months
at DLR Braunschweig from September 2007 to March 2008. The visit was supported
and supervised by Prof. Dr. -Ing Richard Degenhardt who was the principal coordinator
of the COCOMAT research program. At the time of the visit, testing for the
COCOMAT project had already commenced and four of the 12 planned experiments
had been completed. However the remaining panels were observed during testing by the
17 | P a g e
author. All the test panels were made available for inspection and access was provided
to the testing reports.
18 | P a g e
Chapter 2 Review of Literature
Chapter 2
Review of Literature
This chapter summarises the state of the art for Stochastic Analysis and Robust Design
and their application in the design and analysis of structural components with focus on
the design of composite stiffened structures in postbuckling. Stochastic analysis is a
discipline in mathematics that is applied when uncertainty in the input parameters is a
significant feature of the problem and has a significant effect on the solutions being
sought. Robust design is a methodology for design that attempts to create designs that
are insensitive to the variations and hence provide robust design solutions in the face of
uncertainty.
A comprehensive review of the relevant literature is provided, and summarises the key
research developments in Stochastic Analysis and Robust Design applied to engineering
structures. Application of stochastic analysis has led to procedures for solving directly
for the statistical properties for output parameters such as displacements and stresses for
linear solutions. The application in this thesis involves nonlinear response due to the
deep postbuckling behaviour of the COCOMAT panels prior to failure. Procedures that
can be applied to nonlinear analysis therefore are reviewed.
Key to the achievement of a design capability will be the development of a quantitative

measure for Robust Design. A section of the literature review is therefore focused on
the definition of robustness and the relevant mathematical formulations. Following this,
applications of robustness in structural mechanics are reviewed. Robust Design is a
relatively new field and so the capabilities of current published research are investigated
and critically examined.
19 | P a g e
As this thesis also contributes directly to the COCOMAT project, the postbuckling
behaviour of structures and experimental testing are also considered. Current finite
element codes can predict post-buckling response of thin stiffened skin panels. However
the reliable modelling of damage progression and failure is the goal of the COCOMAT
project and some of the world leading research is being undertaken within the CRC-
ACS (Orifici 2007, Orifici et al. 2008, 2009).
2.1 Introduction to Stochastic Analysis
This section introduces the concept of stochastic analysis and its applications in the
design of aerospace structural components. A stochastic analysis, as understood in the
context of this thesis, is essentially a methodology to account for the behaviour of
structures under non-deterministic conditions. Modern structural analysis currently
involves the use of finite element simulations to simulate the behaviour of components
experiencing loads. By using finite elements it is possible to understand the stress and
strain fields as well as the mechanisms that lead to failure. However one drawback is
that the simulation produces a response that is unique to the particular set of
deterministic input variables. An extension of this powerful simulation tool would be to
observe the response of structures subjected to imperfections and variations. The
foundations of the concept are discussed in this chapter followed by a section on
metamodelling. Metamodels are a result of the stochastic analyses and are
representative of structural response within a range of input variables. These models
allow information to be extracted about the response of the structure. Finally the review
on stochastic analyses shows the current developments on the determination of
uncertainty.
2.1.1 Stochastic modelling
Stochastic modelling involves modelling processes where the outcome is not

predetermined and may contain some indeterminacy. This is due to the inclusion of
random inputs leading to outputs which are not predetermined (Papoulis and Pillai
2001). The research in this thesis is a subset of stochastic processes as the focus is the
20 | P a g e
outcome or response of one discrete state in the process. The random inputs are
included into a structural analysis, such as the buckling of a panel, and the outputs, such
as buckling load and buckling mode, are considered.
According to Noguera and Watson (2006), stochastic simulations have been used in
many fields to understand the behaviour of various systems including, assembly line
balancing, production planning, system design and chemical batch processing. In the
last decade, this has been extended to structural analysis. With an increase in computing
power, designers are able to create computationally intensive simulations, however,
danger lies in trusting these simulations without validating the reality aspect required in
numerical modelling. The use of stochastics as a means to validate finite element
simulation models has been advocated by Marczyk (2000a). Stochastics if used with
Monte Carlo Sampling is able to introduce variability into simulations, reduce
prototyping sensibly as well as build user confidence in simulations.
Marczyk (2000b) states that optimality and uncertainty cannot exist together. He states
that there are three dogmas in Computer Aided Engineering (CAE) today, namely:
(i) determinism,
(ii) reductionism and
(iii) optimality.
Due to the nature of these beliefs, the designer has to remove uncertainty from CAE as
it reduces the scope to simplify the problem and seek an optimum that usually holds for
only a limited set of input variables. Instead of using optimality, it is believed that the
direction that designers should head towards is that of simulation. With the computing
power available today, the next logical step would be to use Monte Carlo based
simulations to study complex problems that originate from physics, and not the
simplified models that are based heavily on assumptions.
Marczyk, Hoffman and Krishnaswamy (2000) investigate crash scenarios for vehicles
and conclude that uncertainties in material properties and boundary and initial
conditions have to be taken into account as these variables contribute towards the paths
21 | P a g e
that the loads take. Bifurcations, or sudden changes in response, exist in crash scenarios
and they are caused by the angle of impact and impact velocity. Understanding the
effect that input variables have on the load paths is important so that the structures can
be designed for proper energy absorption. The key of this research is that only with
stochastic variables and multiple sampling is it possible to observe the true behaviour of
the crash. Validation of numerical methods via single deterministic analyses is not
possible.
MSC.Robust Design (MSC.Software Corporation 2004) was a product within

SimOffice which integrated the MSC.Patran pre-processor with the MSC.Nastran
processor to run stochastic simulations. Despite the name of the software, it was
essentially a software designed to conduct stochastic analysis. The software was
designed partly by Dr. J. Marczyk and was evaluated by Lee (2005) as part of a
Bachelor of Engineering thesis. It was found that although the software could identify
the variables affecting the output result, this alone did not provide a suitable measure of
robustness. An introduction to the Robust Design philosophy is presented in Section
2.2.
2.1.2 Applications of metamodelling
A metamodel is a family of models created by taking a single parent model and

selecting each of the variables using a random process so that a population of models is
obtained. The number of members in the population is a property of the metamodel, as
are the parameters for each random variable such as type of distribution and standard
deviation.
By definition, a simulation is representative of a real-world system while metamodels

are an approximation of a simulation model (Kleijnen and Sargent 2000). As such, the
metamodel can be used to determine the true relationship between outputs and their
related input variables (Reis dos Santos and Nova 2006). This is useful in understanding
real-world phenomena where the inputs are usually stochastic and especially with finite
element analysis where the output response is determined by the input variables. By
22 | P a g e
having multiple sets of random inputs, it is then possible to generate metamodels which
exist as a cloud of points, each one representing an analysis. This cloud of points thus
represents the approximate behaviour of the structure containing information of the
relationship that exists between the output and inputs.
The simplest procedure for developing a metamodel is the Monte Carlo method. The
form of the distribution for each input variable, the mean and the standard deviation are
defined. A random number generator is used to define each variable so that the
population conforms to the required distribution. For each member of the population the
variables are set independently so that all variables vary randomly within the defined
distributions. The result is a family of models that are analysed and the required output
responses are subject to statistical analysis to determine the stochastic properties for the
output responses. The output responses can include the mean and standard deviation of
a stress, deflection or buckling load.
Metamodels can be used for the ‘validation’ or ‘fitting’ of phenomena. Validation is the
substantiation that a model within its domain of applicability possesses a satisfactory
range of accuracy consistent with the intended application of the model (Sargent 1991)
while according to Kliejnen and Sargent (2000), fitting applies mathematical and
statistical techniques to a set of simulation data to estimate the parameter values and
evaluate the estimated parameter values with respect to the data set. This is done using
quantitative criteria. While validation requires knowledge about the problem entity and
the specified accuracy required of the metamodel, fitting is concerned with the
determination of a ‘good' set of parameter values from the data set.
Metamodels are also used to reduce computational time in numerical simulations.

Zadeh, Toropov and Wood (2009) have used metamodels for the design of composite
aerospace structures. Coarse mesh finite element models were used to generate
information on the structure followed by the use of fine mesh finite element models to
refine the final design. The metamodels were applied in order to reduce computational
time. Annicchiarico (2007) used metamodels in conjunction with evolutionary
algorithms to aid in the optimisation of structural shapes. For multiple input and output
systems, Li et al. (2006) propose the use of multiple dependent metamodels to
23 | P a g e
simultaneously solve a system of equations to individually estimate all the responses

simultaneously.
2.1.3 Measuring uncertainty in structures
The study of uncertainty falls within the area of research commonly known as
Sensitivity Analysis (SA) which is ‘the study of how the uncertainty in the output of a
model (numerical or otherwise) can be apportioned to different sources of uncertainty in
the model input’ (Saltelli 2002). According to Saltelli, the use of SA techniques in the
context of uncertainty analysis should satisfy the following requirements:
(i) global,
(ii) quantitative and
(iii) model free.
Being global means that the technique takes into consideration the entire input
distribution while model independence means that no assumptions on the model
functional relationship to its inputs is necessary in order for the SA method to produce
accurate results (Borgonovo 2007).
According to Borgonovo (2006), the current approaches to quantifying uncertainty fall

within three categories including:
(i) variance based methods which measure output variance,

(ii) nonparametric based methods which measure input-output correlations and
(iii) moment independent methods which consider the moment distribution of the
output.
A review of variance based methods found that Sobol Indices (Sobol 1993, 2003) are
one of the commonly applied methods for determining uncertainty. Sobol global
sensitivity indices provide information on the model structure in the absence of
correlations among the inputs. Also contributing to variance based methods that are
24 | P a g e
currently in use are Iman (1987), Andsten and Vaurio (1992), Helton (1993) and Saltelli
et al. (1999). These references apply variance-based methodologies to deal with
uncertainty along with sensitivity analyses over a broad range of applications and aid in
understanding the fundamental theory.
The nonparametric based methods mainly involve the application of correlation to

quantitatively measure uncertainty. Common methods include linear methods such as
the Pearson Correlation and nonlinear methods such as Spearman (1904). Regression is
often used as a tool for fitting within the class of nonparametric based methods (Helton
1993 and Frey and Patil 2002). However the results are often considered independently
with correlation. The moment independent method (Borgonovo 2006) is a higher order
method and is a relatively new development. It allows the influence of an uncertain
input in an uncertain system to be measured.
Applications of the methods discussed here can be found in Section 2.4.6 where the
review focuses on the variability that exists in fibre reinforced composites and in
Section 2.5.6 where imperfections affecting postbuckling are discussed.
2.1.4 Conclusions
From the review of stochastic modelling, it can be seen that the aim of the method is to
draw out all the possible responses from loading of structures. Monte Carlo simulations
are a good method to achieve this but require multiple sampling which can sometimes
make the analysis inefficient. For a large scale structure, the order of magnitude can be
up to 103 samples in order to derive a reliable mean value from the results. However the
concern of this thesis is with the factors driving the output response, which is a first
order derivative for a trend line. The input parameters that this thesis is concerned with
exist within a range of values that are dependent on the tolerances from materials
production and assembly which can be predetermined from inspection. The input
parameters and output responses in this thesis are also governed by the rules of
structural mechanics; hence a smaller sample size will suffice for reliable and realistic
quantities to be derived as this thesis is concerned with a measure of the relationship
25 | P a g e
between the output response and the input variables. Therefore a reduction in the
number of samples will have a lesser effect on the relationship between the output
response and input variables compared to the mean value of the output response.
Numerical simulations can be computationally intensive and, therefore, the minimum

required sample size should always be considered. Such a sample size has been
suggested by Will, Moeller and Bauer (2004):
(2.1)
where m is the sample size,

n is the number of input variables and
Y is the number of outputs.
The use of Monte Carlo methods has a natural progression into metamodels as the
repeated sampling process is able to generate a data set that is representative of the
response the structure has to a particular uncertain loading condition. As mentioned in
the previous section, fitting, in the context of metamodelling applies mathematical and
statistical techniques to a set of simulation data to estimate the parameter values and
evaluate the estimated parameter values with respect to the data set. This results in the
metamodel being useful for estimating the first order responses that are achieved as a
result of the sampling process.
Conclusions on the most ideal methodology for quantifying uncertainty will be made at
the end of this review in Section 2.6 when all the relevant areas of research pertaining
to this thesis have been collated and reviewed.
2.2 Introduction to the Robust Design Concept
The second aim of this thesis is to use the understanding of the stochastic properties of
the structural experiments executed in COCOMAT to define a measure that can be used
to guide a design process to a more robust design. This section therefore builds a
26 | P a g e
foundation for this development by establishing the aims and definition for Robust
Design. It begins with a qualitative definition of ‘robustness’, followed by a review of
the Taguchi Method created by Genichi Taguchi (1986), the principal contributor to this
concept. Robust Design can be classified into three distinct categories or types and this
is explained with examples encountered in engineering. The mathematical formulations
for Robust Design are then shown and the relevant developments pertaining to this
thesis are presented.
2.2.1 Definition of a robust design
Robust Design has been defined by Phadke (1989) as a concept ‘to make a product’s
performance insensitive to variations in material, geometry, manufacture and operating
environments’. The manufacture of products often involves a balance between the
controllable factors such as member sizing and uncontrollable noise such as material
and processing variations or the inability to accurately predict the boundary conditions
experienced by a structural component. Figure 2.1 shows a hypothetical comparison
between a robust maxima and an unstable maxima. Note how the performance of the
robust maxima remains constant away from the nominal input value while for the
unstable maxima a noticeable decrease in performance occurs when moving away from
the mean.
Output Response
Unstable Maxima
Robust Maxima
Input Variable
Figure 2.1: Comparison between robust and unstable maxima.
27 | P a g e
Robust Design can be a highly useful approach for analysing models of complex
systems for several reasons (Sanchez 2000). The theory is flexible as it can be applied to
a range of systems that can include analytical models, statistical modes or physical
prototypes. Robust Design is also multi-objective and non-deterministic (Zang, Friswell
and Mottershead 2005) and can thus be considered to fall within the sphere of multi-
disciplinary optimisation.
Decisions made in design can have a large impact on the scatter observed once it is in
operation. In Figure 2.2 it can be seen that both Design A and B have input variables
that share the same mean and standard deviation but the slope of the curve causes the
output to have a significant difference in mean and standard deviation. Therefore it is
important to ensure that the selected design variables have a minimal amount of scatter
in actual operation.
Output Response
Design B
Design A
Input Variable
Figure 2.2: Differences in output resulting from design selection.
It should also be noted that although variation and uncertainty are often used
interchangeably, they are by definition, not the same. Variations exist within a subset of
uncertainty and Zimmerman (2000) has provided a rough taxonomy of the general
properties of uncertainty. Causes of uncertainty include the following.
(i) Lack of information (most frequent cause of uncertainty).

28 | P a g e
(ii) Abundance of information (limited ability of humans to perceive and

simultaneously process large amounts of data).
(iii) Conflicting evidence (information supports different phenomena).
(iv) Ambiguity (linguistic information, which has context dependent different
meanings).
(v) Measurement (engineering measurement is never perfectly accurate).
(vi) Belief (a type of subjective information).
Of the seven causes of uncertainty, this thesis is concerned only with two; lack of
information and errors in measurement.
2.2.2 The Taguchi Method
The concept of Robust Design was originally devised by Genichi Taguchi (1986) in the
years following the Second World War. The work done by Taguchi was a reflection of
the state of industry in which Japan found itself; there was an acute shortage of quality
raw materials, high quality manufacturing equipment and skilled engineers. Despite
these overwhelming circumstances there was still a need for industry to manufacture
high quality products. Taguchi thought of robustness within the framework of the
quality of a product varying from its specification as opposed to being inside or outside
of the specification. To measure the quality, Taguchi defined a quadratic Quality Loss
Function as follows:
(2.2)
where L is the loss associated with parameter y,

m is the nominal value of the parameter specification and
k is a constant depending on the cost at the specification limits.
This function penalises the deviation of a parameter from the specification value that
contributes to deteriorating the performance of the product, resulting in a loss to the
customer. The key here is that a product engineer has a good understanding of what the
29 | P a g e
nominal size of the specification is. The usual lower and upper limits for the tolerance
of a given design parameter are changed to a continuous function that presents any
parameter value other than the nominal value as a loss.
The Taguchi Method advocates the use of orthogonal arrays in order to implement
Robust Design. This allows various combinations of inputs and variations to be
introduced and the resulting signal-to-noise ratios to be measured. The mean response
and standard deviation of two variables are considered for each combination. The
following define the mean response and standard deviation, respectively:
(2.3)
(2.4)
Using the mean response and standard deviation, the optimal input is then determined
using a Signal-to-Noise (SN) ratio. There are three optimal ratios that can be used
depending on the output response that is required. To achieve the smallest SN ratio
where the smallest response is desired, the following is used:
(2.5)
To achieve an SN ratio where variability about a mean is to be reduced, the following is

used:
(2.6)
To achieve an SN ratio where the largest SN ratio is required so that the largest response
can be obtained, the following is used:
30 | P a g e
(2.7)
2.2.3 Types of Robust Design
Another overview has been provided by Allen et al. (2006) where examples are
provided for the different levels at which Robust Design can be applied to the design
and manufacture of products. Three distinct types of Robust Design are identified.
These are commonly known as Type I, II (Chen et al. 1996) and III (Choi et al. 2005).
These three types of Robust Design allow the designer to accommodate all of the
possible variations in design, manufacture, operation and numerical simulation. They
are as follows:
(i) Type I Robust Design is used to identify design variable values that satisfy a set
of performance requirement targets despite variations in noise factors,
(ii) Type II Robust Design is used to identify design variable values that satisfy a
set of performance requirement targets despite variation in control factors
themselves and
(iii) Type III Robust Design is used to identify adjustable ranges for design
variables that satisfy a set of performance requirement targets and are
insensitive to variability within the system model.
Examples of Type I Robust Design include changes in ambient temperature, operating

environment, or other natural phenomena that are impossible or prohibitively costly to
control (Kalsi et al. 2001). Taguchi refers to this as Parameter Design. In Type II Robust
Design, the goal is to minimise variations caused by deviation in the control factors.
These are also factors that a designer has control over. These variations can possibly
result from manufacturing tolerance limitations, material quality variations, or even
evolving design preferences (Sunderasen, Ishii and Houser 1995). Taguchi refers to this
as tolerance design; it leads to a higher manufacturing cost and is a disadvantaged
solution compared to Type I Robust Design where there are no added manufacturing
31 | P a g e
costs involved. Type III Robust Design is a relatively recent development and involves
the determination of suitable tolerancing about a nominal design mean. A model may
incorporate simplifying assumptions or random factors. Examples include random
realisations of a microstructure in materials design that affect the accuracy and precision
of its predictions. Figure 2.3 shows the relationship between the types of Robust Design
and the response of the system.
Noise Factors
(Type I Robust Design)
Control Factors Design Output

(Type II Robust Design) (Type III Robust Design) Response
Figure 2.3: Representation of the various Robust Design types in relation to a design.
Table 2.1 presents a breakdown of the three types of Robust Design when used in the
context of designing structural components. It can be seen that all three types are
relevant to this thesis.
Table 2.1: Application of Types I, II and III Robust Design in structural design.
Type I Type II Type III
Temperature Structural layout/topology Modelling methodology
Boundary conditions Geometry Stochastic range
Imperfections Material Design tolerance
Fatigue/corrosion Designed failure modes Insensitivity to uncertainty
Unexpected events Manufacturing tolerance and variability in models
32 | P a g e
2.2.4 Mathematical formulations in Robust Design theory
Park et al. (2006) have conducted a review on Robust Design. Three theories including
the Taguchi Method, Robust Optimisation and Robust Design using the axiomatic
approach are presented. All current methodologies for Robust Design are a combination
or fall within one of the above methodologies. A Robust Design problem can be
mathematically represented within an optimisation framework. Find:
(2.8)
to minimise
(2.9)
subject to
(2.10)
where
(2.11)
where b is the design variable vector,

p is the design parameter vector,
z is the noise factor,
Gj is the nth constraint with noise,
bL is the lower bound of the design variable vector and
bU is the upper bound of the design variable vector.
The Taguchi Method (TM) has already been discussed above. The deficiency of TM is
that the variables in the matrix used for the method are discrete and this may not address
the entire design space. Therefore Robust Optimisation was developed using available
techniques in optimisation. Find:
(2.12)
to minimise
33 | P a g e
(2.13)
subject to
(2.14)
where
(2.15)
where μf is the mean of the function,

σf is the standard deviation of the function,
zb is the noise vector of b and
zp is the noise vector of p.
Robust Design using the axiomatic approach has been performed by Suh (2001). Suh
postulates the two following fundamental axioms in design.
(i) The independence axiom is to maintain the independence of the functional

requirements.
(ii) The information axiom is to minimise the information content of the design.
The independence axiom states that the independence of the functional requirements
(FRs) must always be maintained by an appropriate choice of design parameters (DPs).
Design equations may be expressed in one of three forms. When two functional
requirements are involved it can be expressed as one of the following equations:
(2.16)
34 | P a g e
The first equation is known as the uncoupled case where the FRs and the DPs are
independent. The second equation is known as the decoupled case where the
independence of the FRs can be maintained if DP1 is determined before DP2. The third
equation should be avoided, if possible, as no DP exists to satisfy the independence
axiom.
In Robust Design, the aim is to obtain the first equation in Equation (2.16) as the aim of
Robust Design is to have:
(i) a design that minimises the output response and

(ii) a minimum bias between the original design and modifications.
2.2.5 Other developments in Robust Design
In early 2005 a company, Ontonix (2008), was founded by J. Marczyk and extended the
applications of robustness from engineering into the areas of science and commerce. A
proprietary software OntoSpaceTM (2009) has been developed to measure a variable,
Complexity, within existing systems. Robustness and Uncertainty are both functions of
Complexity. Little detail is provided on the workings of the software or mathematical
formulations which are used to derive Robustness, Uncertainty or Complexity. However
a White Paper on OntoSpaceTM by Marczyk (2005) states that ‘low or high performance
scatter points to high or low quality, respectively, and has little to do with robustness.
Robustness is related to the character of the distribution, not to its spread’ and ‘the
commonly accepted definitions of robustness mistakenly state that a system with low
scatter in performance is robust. We state that a variable behaves in a robust manner if
its probability density function (PDF) is uni-modal. When the PDF is multi-modal the
system (in the case of a particular variable or channel) can spontaneously jump from
one peak (mode) to another.’ A uni-modal PDF has only one peak in the function while
a multi-modal PDF exhibits multiple peaks in the PDF, which points to a bifurcated
response where there is equal probability in various responses occurring within a
design.
35 | P a g e
It is true that having a multi-modal PDF reveals a lack of robustness in a structure. This
was shown in the variety of postbuckling mode shapes that were seen in Figure 1.2.
However the lack of robustness seen in the postbuckling of the stiffened panel was
attributed to the quality of manufacture and variations in boundary conditions.
Therefore it is acknowledged that some of Marczyk’s comments are applicable to this
thesis and the author seeks to reconcile the differences between insensitivity to
variations in the structure as well as uni-modal PDFs. While it is a concern that
structures may exhibit various responses, scatter in collapse loads are also factors that
determine the failure of a structure.
A search of the term, ‘measures of robustness’, revealed that this is a field that has been
previously explored in Control Engineering. Examples of work relating to measures of
robustness first appear in Patel and Toda (1980), Yadavalli (1985) as well as Woo and
Pujara (1988). However these papers refer to robustness for a particular steady state or
singular case with multiple input parameters and are therefore different from the
definition of robustness established in the previous sections where robustness is a
measure that is considered across multiple input parameters and load cases. In Section
2.3, the concept of robustness pertaining to structural mechanics is explored in further
detail.
2.2.6 Conclusion
From the study of the aims and methodology of Robust Design, it has been found that
the aim of Robust Design is to ensure that products are insensitive to variations and
uncertainties, which is positive for some aspects of experimental testing. In
experimental testing of structural components, test pieces are sometimes designed to be
highly sensitive so that empirical formulae and design codes can be created. However in
postbuckling experiments, having a sensitive design also means that the test pieces may
exhibit different buckling modes. This creates issues as benchmark experiments may
not represent the possible responses of all the test pieces. Due to the experimental noise
introduced through variations in geometry, boundary conditions and material properties,
the initial benchmark may not even match the numerical simulations used in designing
36 | P a g e
the experiment. The usefulness of Robust Design is in ensuring that the possible range
of responses is reduced. The formulations shown in Section 2.2.4 quantitatively
illustrate the aims of Robust Design. Even though they provide a means by which to
achieve insensitivity, there are no indications that a universally accepted measure of
robustness exists. This is especially the case for the analysis of aerospace structures.
This is also the case for the software developed by MSC.Software Corporation,
ANSYS® and Ontonix.
2.3 Robustness in the Design of Structural Components
This section considers the structural design concepts relating to robustness. The term
‘robustness’ has only been used in fairly recent times in structural design and analysis
but other key ideas such as progressive failure and redundancy have been in existence
for some time. Literature from civil engineering has been collated and two case studies
are presented in Section 2.3.3 to highlight the important features that robustness needs
to encompass. This is followed by a worked example of the effects of nonlinearity in the
design of subcomponent structures which ultimately contribute to disproportionate and
progressive failure.
2.3.1 Progressive failure
Progressive failure has become an area of interest most recently in light of the
September 11 2001 terrorist attack on the World Trade Center (WTC) in New York
City. Particular interest has been placed on ways to mitigate the damage caused by the
occurrence of extreme events such as terrorist attacks or environmental disaster. The
General Services Administration (GSA 2003) describes the phenomenon as follows:
‘Progressive collapse is a situation where local failure of a primary structural

component leads to the collapse of adjoining members which, in turn, leads to
additional collapse.’
37 | P a g e
The beginnings of this area of research can be traced back to the failure of Ronan Point
in East London on 16 May 1968 where a localised natural gas explosion resulted in the
disproportionate partial collapse of the tower structure (Pearson and Delatte 2005). Nair
(2006) provides four case studies linking disproportionate failure with progressive
failure: the aforementioned Ronan Point, the Murrah Building in Oklahoma City, the
WTC Towers One and Two and the WTC Tower Seven. The analysis of the failures are
concluded by mentioning that current approaches to designing structures against such
failures are handicapped by uncertainty of the event to be designed for and events not
anticipated in design guidelines and criteria. This provides a helpful starting point for
this section of the literature review as these concepts are tied in with the aims of
researching a novel method for quantifying the robustness of structural designs.
Although the focus has primarily been in the field of civil engineering, this is still
applicable to aerospace structures as both exhibit a similar level of complexity and is
especially useful for aerospace applications for which there are smaller factors of safety.
The concept of robustness in civil structures first appears in a publication by Canisius,
Sørensen and Baker (2007).
In terms of structural arrangements, a design is more robust, for example, if it has

multiple load paths so that it retains residual strength in the face of an unpredicted event
that compromises one of the paths. Mohamed (2006) says that multiple load paths also
ensure that forces can be easily transmitted and distributed to other load-bearing
members once an initial loss in one or more members has occurred. The re-distribution
of loads continues until a stable equilibrium has been reached. This may sometimes
occur when a considerable amount of structure has failed.
Doltsinis, Kang and Cheng (2005) analysed truss structures and found that the nature of
the response was path-dependent. They found that in some of their examples it was not
possible to concurrently maintain robustness and performance. It was also noted that
their methodology was suitable only in situations where the variations in the system
were small. Sandgren and Cameron (2002) analysed structural designs where the
robustness is built up topologically using a genetic optimisation algorithm. The
38 | P a g e
approach was computationally intensive as multiple sampling had to be undertaken but

produced results that were better than those produced through conventional analysis.
2.3.2 Structural redundancy
Another consideration in Robust Design includes the concept of cascading and

disproportionate failure. This consideration is related to structural redundancy.
Although redundancy is recognised as an important part of design, it is often dealt with
in a vague, non-quantified manner. For example many building codes specify ‘that a
structure should be robust in the sense that the consequences of structural failure should
not be disproportional to the effect causing the failure’ (Ellingwood 2005).
Fu and Frangopol (1990) considered probabilistic indices to measure structural

redundancy using the relationship between damage probability and system failure
probability. This Redundancy Index (RI) has been defined as:
(2.17)
where Pf (dmg) is the probability of damage occurring on the system and
P f (sys ) is the probability of the system failing.
The RI is a positive number as Pf (dmg) is always greater than P f (sys ) because not all
damage results in a global failure. In their work, Fu and Fragopol (1990) state that there
is no universally accepted definition of redundancy in structures.
Baker, Shubert and Fabert (2008) introduced an index, IRob, which is formulated by
dividing the risk associated with the direct consequence of localised damage and the
risk of subsequent damage. If a system is robust then no indirect failure should result. It
is as follows:
39 | P a g e
(2.18)
where RDir is the direct risk and

R Ind is the indirect risk.
Both of these approaches are probabilistic in nature and can be useful when attempting
to analyse for failure modes in structures. However they do not account for the
statistical scatter of input variables in terms of means and standard deviations. Neither
of the methods is able to account for the uncertainty of the input variables and provide
insight as to which ones are critical in the structural design or arrangement.
2.3.3 Case studies for progressive and disproportionate failure
Case Study 1: Ronan Point
Construction of Ronan Point started on 25 July 1966 and the building was handed over
to Newham Council on 11 March 1968, it cost approximately £500,000 to build. It was
80 ft by 60 ft in area and 210 ft high and consisted of 44 2-bedroom flats and 66 1-
bedroomed flats. On the morning of 16 May 1968 at 5:45 am, an explosion occurred in
Flat 90, on the 18th storey at the south east corner of the building. This resulted in the
collapse of one corner of the 23 storey block of apartments in east London as shown in
Figure 2.4 (a).
The Ronan Point Apartment Tower was constructed using the Danish Larsen-Neilsen
system which was developed in 1948. The Larsen-Nielson system was ‘composed of
factory-built, precast concrete components designed to minimise on-site construction
work. Walls, floors and stairways are all precast. All units, installed one-story high, are
load bearing’ (Systems 1968). This building technique encompassed all the panels and
joints including the method of panel assembly and the methods of production of the
panels. In the design, each level is supported by the load bearing walls directly beneath
it. Load transfer was through these load-bearing walls, and the walls and floor system
40 | P a g e
were fitted together using joints which were then bolted together and filled with dry
pack mortar to secure the connection. The joint can be seen in Figure 2.4 (b).
Load
bearing wall
Lifting
Washer rod
Nut
Open joint Load

finished with bearing wall
line motar
Joint between interior wall and floor slabs
(a) Ronan Point after partial collapse (b) Schematic of pre-fabricated joint
(Daily Telegraph, 1968) where failure occurred (Levy and
Salvadori, 1992)
Figure 2.4: Damage sustained by Ronan Point and schematic of failure joint.
The explosion that occurred in Flat 90 was caused by a leaking gas stove. The gas was
ignited when the resident of the apartment lit a match on the kitchen stove blowing out
sections of the outer and inner walls, which resulted in the collapse of wall and floor
sections from the top of the building to the ground.
The investigations found that the Ronan Point apartment tower was deeply flawed in
both design and construction. In the structural design, the tower was put together using
precast panels without the use of a structural frame. Therefore the apartment tower
lacked alternate load paths to redistribute forces in the event of a partial collapse. When
the walls on the 18th storey collapsed, this caused the levels above it to also fail as there
was no redundancy in the design. The 17th storey was not able to bear the combined
weight of the structures that had failed on the 18th storey and above and this lead to a
progressive failure where each level failed successively as equilibrium could not be
41 | P a g e
maintained. This failure was also disproportionate in the way that the failure of one
outer structural wall, led to the collapse of an entire section of the building block. Four
people died immediately as a result of the blast and the following collapse. A further 17
others were injured.
Case Study 2: WTC Towers One and Two
The twin towers of the WTC experienced a catastrophic failure on the morning of 11
September 2001 after a terrorist attack using a Boeing 767 aircraft impacted the upper
levels of each tower at high speed. The first attack occurred on Tower One at 8:46am
while Tower Two was attacked at 9:03am. Both aircraft were fully laden with fuel and
this resulted in an intense fire within each building. The initial impact resulted in the
loss of fireproofing from the structural steel columns which ultimately necessitated the
weight of the upper floors to be redistributed via other load paths. These fires, in
combination with the dislodged fireproofing, were responsible for a chain of events in
which the building core weakened and began losing its ability to carry loads. The floors
became weaker and began to deflect and sag, pulling inwards on the perimeter columns.
This caused the perimeter columns to bow inward and buckle—a process that spread
across the faces of the buildings resulting in the final collapse (NIST 2005). The NIST
report recommended that ‘the primary structural system should provide alternative paths
for carrying loads in case certain components fail’. This appears to be a
recommendation for structural redundancy in civil structures.
The WTC towers were constructed with a floor plan consisting of vast open floor areas
without intermediate walls or partitions and this design was repeated on every level.
This structural arrangement was accomplished through the use of the minimal number
of vertical supporting members. The configuration of the supporting members consisted
of a column line at the perimeter, a column line at the core and a connecting truss which
acted as a long slender bridge across almost one-half the width of the building (FEMA
2002). Regrettably the failure of one of its elements would not only result in an inability
to sustain the load of the remaining members, but would likely increase the extent of
failure due to interconnectivity of the few members which remained in a given spatial
42 | P a g e
plane. The failure of WTC Towers One and Two were from progressive collapse from a
lack of redundancy. Figure 2.5(a) shows the explosion at the time of the initial impact of
WTC Tower Two and Figure 2.5(b) is an enlarged view showing the impact damage to
the north face of WTC Tower One.
(a) Tower Two at the time of impact (b) Enlarged view of structural damage
to Tower One at point of impact
Figure 2.5: Damage sustained by WTC Towers One and Two.
(FEMA 2002)
It does not appear that simple changes in the structural design might have significantly
reduced the scale of the collapse, although, it does appear that fire protection material
with greater adhesion and strength could have allowed the fire to burn itself out without
failure of the steel structure. It is difficult to classify if the failure of both towers was
disproportionate as the initial damage from the impact was significant. However, the
final collapse that resulted can be described as progressive failure as each level
successively failed due to an inability to withstand the combined force of the levels
above. Tower Two collapsed at 9:59am and Tower One followed at 10:28am resulting
in a loss of 2750 lives.
The significance of the Ronan Point and WTC failures and the resulting work in civil
engineering design to this thesis is the similarity between the development of the
43 | P a g e
progressive failure in these structures and the possibility of progressive failure leading
to collapse in the buckling considered in this thesis for stiffened structures. In the
stiffened structures local buckling occurs in subcomponents of the structure early in the
load history. These local modes lead to redistribution of load and should not result in a
global collapse of the structure. Skin panels are usually designed to buckle first because
of the relatively high contribution the panels make to the structural weight, but the
combined effect of stiffeners and adjacent supported skin should carry existing and
higher applied loads. The designer should ensure that a cascade to collapse does not
occur and structural integrity is maintained in a robust design until the stiffeners fail or a
global buckle develops at a higher load.
2.3.4 Worked example for the effects of nonlinearity
The previous sections discussed the effects of disproportionate failure on a macro scale.
Disproportionate failures are nonlinear in the way that small localised failures lead to a
large global collapse. In multiple component structures the structure is composed of
numerous sub-components which can fail progressively if there is a lack of redundancy.
Unfortunately, these smaller subcomponents can also contribute towards the final
failure due to the existence of small variations and uncertainties.
Equally dire is the possibility that the expected response may not always be the same as
the actual response of the structure in operation once variations and uncertainties are
incorporated. Consider the clamped-free rod under the application of an axial load as
shown in Figure 2.6.
Figure 2.6: Cantilevered beam with axial loading.
44 | P a g e
Assuming that the rod area A has a nominal mean area of Ao and a standard deviation of
σA, the displacement δ of a rod with the corresponding mean area is:
(2.19)
where P is the applied force,

L is the length of rod,
E is Young’s Modulus and
Ao is the nominal cross-sectional area for the rod.
Using Equation (2.19) and the nominal values from Table 2.2 the expected deflection of
the beam is 1.389 mm.
^
From Kleiber and Hien (1992), the most probable displacement, is actually given by:
(2.20)
where σA is the standard deviation of the cross-sectional area

and the cross-sectional area A`, corresponding to the most probable displacement is:
(2.21)
Using Equation (2.20) and assuming a 0.1 coefficient of variation, the most probable
displacement is 1.403 mm which is a difference of about 1%. This is not a significant
shift until the upper and lower bounds for the beam are found. Refer to Table 2.2 for the
stochastic boundary for the analysis.
45 | P a g e
Table 2.2: Stochastic boundary for circular beam.

Input Variable Mean Possible Range
Load, P (N) 1000 NA
Length, L (mm) 1000 NA
Young’s Modulus, E (MPa) 72000 NA
Cross-sectional area, A mm2 10 7 - 13
Repeated sampling is performed 100 times with variation on the cross-sectional area.
The normal density function Z is plotted against the corresponding axial displacement.
The resulting plot can be seen in Figure 2.7. Note that the plot is not symmetrical about
the actual or expected mean displacement. Due to the nonlinearity of the displacement
function, the plot below is skewed towards the left. The outliers or inputs on the right
end of the plot causes the actual mean displacement to be higher than the expected
mean. The plot also shows that the displacement can exceed 1.7 mm which is beyond
three standard deviations in the output. It can also be argued that the skewing will be
more distinct if the Young’s Modulus were to be varied as well.
3
Stochastic Sample
Normal Density Function (Z)
2.5 Expected Mean

Actual Mean
2
1.5
0.5
0
1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9
Displacement (mm)
Figure 2.7: Plot of normal density function of beam displacement against beam
displacement.
46 | P a g e
The purpose of this section was to show that the disproportionate and progressive
failure that is observed in large structures can also be attributed to the behaviour of
subcomponent structures such as stiffeners and fasteners. This is parallel to Weibull’s
Extreme Value Distribution (Weibull 1951) which governs the time to occurrence of
failure for the ‘weakest link’ in failure processes.
An example where such nonlinear behaviour occurs is shown in Section 5.2 where the
initial geometrical imperfections of a stiffened panel are investigated. In the case of the
stiffened panel, it was found that there was a significant difference in geometrical
imperfection due to thermal stresses between the nominal panel and panels where there
were variations in material and laminate properties.
2.3.5 Conclusion
The concepts of disproportionate and progressive failure have been introduced and
discussed in the present section. It can be seen that both are related to robustness as they
are involved with investigating the effects of input variables on the output response. The
difference between the approaches in civil engineering and Robust Design is that critical
design features are not identified and actual scatter in properties are not considered.
From the case studies in Section 2.3.3, it can be concluded that structural redundancy
appears to contribute to robustness but it is important to examine if the alternate load
paths contribute to the prevention of collapse. The worked example in Section 2.3.4
demonstrates that it is possible for the actual mean values to deviate from the expected
mean values once variations are included and this can sometimes lead to premature
failure. Therefore any alternate load paths are useful only if the paths perform to the
designed specification.
2.4 Variability in Fibre Reinforced Composites
According to Daniel and Ishai (1994), ‘a structural composite is a material system

consisting of two or more phases on a macroscopic scale, whose mechanical
47 | P a g e
performance and properties are designed to be superior to those of the constituent

materials acting independently’. Although composite materials are inherently superior
to their constituent materials, these anisotropic materials also exhibit different modes of
failure not commonly found in metallic structures such as compression after impact,
fibre cracking through tensile failure and micro-buckling and matrix delamination, etc.
Composite laminates also exhibit variations in thickness and fibre volume fraction. This
is of particular relevance to postbuckling problems as buckling loads are very sensitive
to laminate stiffness and thickness. The source of these imperfections can be due to
variation in resin content in pre-preg materials, variation in consolidation pressure, and
variation in resin flow during the curing process, etc. There are also considerations for
local versus global imperfections and defects. Some defects can be expected over the
entire structure while others will be localised. The structural response may be more
affected by variations in certain locations. This definitely makes the numerical
investigation of defects/variations more difficult.
When the difficulty in predicting failure is coupled with complexity in design and
manufacture due to the multiple processes required in fabrication, it becomes apparent
that the possible uncertainties and variability that each composite structure experiences
must be accounted for in order to ensure that it does not fail prematurely. This section
seeks to address the imperfections and variations that will be found in fibre reinforced
composites. It covers variations and defects in the fibre and matrix, skin-stiffener
disbonding and possible sources of defects and damage.
2.4.1 Analysis of composite laminates
In the analysis of composites, the classical laminated plate theory (CLPT) is most
commonly used to assemble a finite number of orthotropic laminae into a laminate or
plate. The fibres and matrix for each lamina layer are homogenised by combining their
properties into an orthotropic layer. This can be seen in Figure 2.8.
48 | P a g e
Fibre 90o
0o
Matrix
(a) Fibre-matrix lamina (b) Orthotropic layer
Figure 2.8: Homogenisation of fibre-matrix properties in CLPT.
CLPT is based on two-dimensional (2D) plate theory that requires the plate thickness to
be at least two orders of magnitude less than the in-plane dimensions (Ochoa and Reddy
1992). It shares the same assumptions as the classical plate theory from the Kirchhoff
hypothesis which makes the following assumptions.
(i) Normals remain straight.

(ii) Normals remain the same length.
(iii) Normals remain right angle to the neutral plane.
In addition to the Kirchoff hypothesis, perfect bonding is assumed as follows.
(i) The bonding itself is infinitesimally small.

(ii) The bonding is non-shear-deformable.
(iii) The strength of bonding is as strong as it needs to be.
Daniel and Ishai (1994) and Baker, Dutton and Kelly (2004) give more detail on CLPT
and the constitutive equations for defining a laminate.
49 | P a g e
2.4.2 Effects of variations in fibre orientation
Fibre orientations have a direct effect on the behaviour of composite laminates as there
is a distinct loss of stiffness once the fibre or the applied load is off-axis. Sources of
variability in misalignment in fibre orientations include errors in layup and localised
waviness in fibres. Walker and Hamilton (2005) analyse the effect of variability in the
orientation of fibres for buckling panels. It was found that the resulting in-plane load
carrying capacity of the plate can be significantly lower than expected if the
manufacturing tolerances are neglected at the design stage.
The effects of misalignment in fibre orientations have been found to be significant in

affecting the longitudinal strength and stiffness of fibre-reinforced composites
(Yurgartis 1987). Yurgartis investigated a CFRP (APC-2) laminate produced by
industrial processes. A variation of ± 3 degrees about the mean fibre direction with a
standard deviation of 0.693 to 1.936 degrees was found. Piggott (1995) also analysed
the effects of fibre orientation variation and similarly found that there was a significant
effect on compressive strength and stiffness as well as shear strength and resistance to
delamination. Carbon reinforced PEEK specimens were tested and a standard deviation
ranging from about 1 to 6 degrees was found with corresponding compressive strengths
ranging from 1.9 to 1.5 GPa respectively.
A study of the imperfection sensitivity of fibres in the lamina, including micro-buckling

have been presented by Tomblin, Barbero and Godoy (1997). It was proven that the
laminae were sensitive to micro-buckling due to a symmetric bifurcation point and the
compressive strength of the laminate was dependant on the alignment of the fibres. The
cause of variability in the fibre orientation due to various manufacturing methods has
been investigated by Potter et al. (2007). It was found that unidirectional prepregs had a
significant amount of in-plane waviness when received from the manufacturer. The
effect of the waviness was investigated on laminates and components. It was found that
thin laminates are most susceptible to the effects of fibre misalignment. One of the
ways to reduce such defects in composite structures is to make corrections at the design
stage so that the structures are manufactured within acceptable tolerances (Potter et al.
50 | P a g e
2008). Figure 2.9 (a) and (b) are examples of fibre misalignments in unidirectional
prepregs.
(a) Misalignment of fibres in (b) Waviness of fibre tows across a flat

unidirectional prepreg surface
Figure 2.9: Defects in unidirectional fibres.

(Potter et al. 2008)
2.4.3 Variations and defects in the matrix
Matrix imperfections, such as voids or matrix cracks, usually occur at the fibre/matrix
interface while issues with porosity may occur from the interface. According to Zhang
and Taheri (2004), initial voids due to manufacturing may not be negligible as an
accumulation of imperfections at weak locations can cause major degradation of matrix
dominated properties. This has a negative effect on interlaminar shear or compression
stiffness and strengths.
Varna et al. (1995) examined the effect of voids on the failure mechanism of resin
transfer moulded laminates. Laminates with the highest average content of voids had a
transverse strain to failure as high as 2% whereas low void content laminates failed at
0.3%. Borg (2004) found that a high concentration of porosity could reduce
compression strength by up to 20%. On the other hand only a slight reduction of
properties was detected for laminates with a high fibre volume ratio. Matrix
51 | P a g e
imperfections can develop into delaminations under service loads which can
significantly affect the strength of the laminate.
2.4.4 Defects in bonding of structural elements
Stiffened composite structures are currently manufactured and assembled either by

using mechanical fasteners, co-curing the skin with the stiffeners, or secondary bonding
pre-cured stiffeners to the skins. Stiffeners that are mechanically fastened can
experience skin buckling between the fasteners when a compressive load is applied.
This is not the case for bonded stiffeners which will carry the shear load across the bond
until the growth of the skin-stiffener disbond reaches a critical size. The initiation and
growth of disbonds affect the collapse load of the structure and are therefore of
considerable interest to researchers in the composites field as the viability of bonded
joints is dependent on its response to a variety of operating conditions.
Skin-stiffener disbonding involves the separation of a stiffener from the skin in a

stiffened structure. The result of the disbond is a reduction in the buckling and collapse
load for the structure because they fail as two separate structures. In stiffened structures,
the initiation of disbonds usually occurs as a result of the design where the skin buckles
locally between the stiffeners. This has been observed under thermography and will be
found in the panels considered in this thesis in Chapter 4. Skin-stiffener disbonding can
be an explosive form of failure, which has typically occurred in a large number of
experimental investigations involving postbuckling stiffened structures. In addition to
affecting buckling and collapse, Krueger, Miguet and O’Brien (2000) found that the
disbond of stiffeners was also a crucial failure mode in pressurised fuselage structures.
Extensive research has been dedicated to analysing this important failure mode by
Krueger et al. (2000) who investigated skin-stiffener disbonding on a simple composite
plate. The same specimen was later employed by Krueger and Minguet (2007) who
predicted disbonding by a shell/3D finite element technique. More complex panels were
numerically considered by Yap et al. (2002, 2004) where an assessment of damage
criticality was presented for curved, multi-ribbed stiffened panels. Suemasu et al.
52 | P a g e
(2006) investigated the effect on the disbond growth of the buckling load of a stiffened
composite structure. More recent work has been undertaken in this area by Orifici et al.
(2007, 2008, and 2009) and Mikulik et al. (2008a, 2008b).
2.4.5 Possible sources of damage in composites
In addition to the imperfections in fibres and matrices, composite structures are affected
by sources of damage that do not usually concern metallic structures. In this thesis,
damage is considered as one of the variations that affect the response of a structure.
Some of the possible sources of damage in composites fall under the conditions for
Type I Robust Design. There are unpredictable and random events which occur during
manufacturing and in-service operation as well as long term damage caused by the
environment in which the components operate. Some of the causes are listed in Table
2.3.
Table 2.3: Sources of damage that affect composite structures.

Manufacturing In-service operation Environmental
(Cured laminates) (Random damage) (Long term damage)
Poor tooling Tool drop Moisture
Poor ply consolidation Hail stone Ultra-violet radiation
Low autoclave pressure Bird strike
Non-uniform autoclave Lightning
Temperature Runway debris
Service vehicle impact
It is recognised that some of the damage may remain undetectable, therefore composite
structures are often designed using the damage tolerant design philosophy. At present
there are a number of non-destructive testing (NDT) methods available that attempt to
find the damage. These include:
(i) testing using penetrant liquids,

(ii) ultrasonic testing,
53 | P a g e
(iii) eddy current testing,

(iv) radiographic testing and
(v) thermographic testing.
These techniques are employed depending on the size of the component, maintenance
requirements and after the occurrence of known damage from in-service operation.
2.4.6 Current research addressing variability in composites
Apart from the research undertaken in this thesis, there are few examples of Robust
Design being employed in the design and analysis of composite structures. This section
contains a summary of the procedures that have been identified.
Chiang (1996) used Robust Design techniques and finite elements to design
experiments for improving the accuracy of the Iosipescu shear test specimens. The
technique used Taguchi’s Method to identify the influential design variables.
Carrerè et al. (2009) investigated the effects of uncertainty in material databases to

model and design hat-like composite structures. Three methods including Monte Carlo
Simulation, Surface Response Modelling and the Stochastic Finite Elements method
were employed in the analyses. The amount of variability that was chosen was 10%
about the nominal means. It was found that Surface Response Modelling was the most
efficient in determining the approximate result. A parallel attempt was made by Rollet
et al. (2009) to improve the material database and to investigate the effect of variability
in the composite from a micro to meso scale. The findings seek to address the
propagation of constitutive uncertainties between scales occurring in the multi-scale
modelling of fibre-reinforced composites. The method is at present only suitable for
global changes in material properties across the structure.
The introduction of uncertainty into composites has also been examined by Shiao and
Chamis (1999) with the aim of increasing reliability in the assembled structure. With the
use of a software code – Integrated Probabilistic Assessment of Composite Structures
54 | P a g e
(IPACS) – developed in-house under NASA, they were able to identify the sensitivity
of input variables with respect to the output response in a probabilistic framework with
the aim of conducting reliability assessments. This work was further developed by
Chamis (2004) and it led towards the improvement of structural response with no
increase in weight by means of tailoring the behaviour of the composite material.
With regards to accounting for variation is real structures, attempts have been made in
recent years to introduce imperfections via stochastic modelling so as to achieve
plausible knock down factors. This can be seen in the work by Chryssanthopoulos and
Poggi (1995). Chryssanthopoulos, Giavotto and Poggi (1995) have also inspected
composite cylinders and characterised the defects found using statistical analysis.
António and Hofbauer (2008) have studied the effects of uncertainty in angle-ply
laminates using sensitivity analysis and perturbation methods. The aim of the study was
to compare the higher order methods which were more accurate but computationally
expensive with first order methods which were less accurate but less expensive. The
results showed that first order local methods provided a suitable level of accuracy in
determining the uncertainty in fibre composites.
2.4.7 Conclusion
A variety of variations and imperfections including the susceptibility to impact damage

in composite structures have been discussed in this section. It has been documented that
the fibre orientations may not always fall within the specification set out in the design.
Although the analysis undertaken was for local regions of a lamina, this misalignment
can also extend across the rest of the structure, causing the failure loads to differ from
the original design. Poor matrix cohesion and the existence of voids also lead to
premature failure due to the growth of delaminations and reduction of strength in the
lamina.
Both fibre and matrix variations can be accounted for using the CLPT theory. In the
CLPT method of analysis, the stiffness of the laminate is dependent on the stiffness,
55 | P a g e
orientation and thickness of each lamina. There has always been some difficulty in
matching results from simulations using CLPT and finite elements with experiments
due to the perfect and ideal parameters set up in the analyses. By introducing realistic
variations into the governing parameters, it should be possible to more closely match
the two.
The general response to manufacturing defects and damage, shown in Table 2.3, as well
as the defects in materials, is to design conservatively. High strength composites like
carbon fibre reinforced polymers (CFRP) can fail at high strain values without prior
warning due to the brittle nature of the material and the sensitivity to damage. This
added concern results in the need to understand and identify the key factors that lead to
failure as opposed to applying large margins of safety.
2.5 Elastic Behaviour of Thin-walled Structures
The research in this thesis involves the assessment of structural robustness in the
collapse of stiffened composite fuselage structures. Therefore the essential literature
involving these areas are summarised in this section. Although the field relating to the
influence of initial imperfections and the theory of imperfection sensitivity of elastic
and inelastic structures is well developed, it is essential that key factors affecting the
failures are understood in order to accurately simulate the collapse of stiffened
postbuckling structures. These include the issues pertaining to bifurcation buckling and
postbuckling, experimental testing and numerical simulations. This section mainly
contains literature involving the matching of experimentation and simulation results.
2.5.1 Bifurcation buckling and postbuckling of shells
In unstiffened structures, there is no distinct change in mode shape as the loading

progresses to structural collapse. However stiffened structures will undergo two stages
of buckling. As an axial compression is applied to well designed stiffened panels, such
as those used in the COCOMAT program, the unsupported skin between the stiffeners
experiences local buckling or bifurcation buckling. At this point the skin takes on a
56 | P a g e
sine-wave shape along the axial and transverse axis of the structure as it buckles locally
in the form of ‘squares’. The majority of the axial load is then transferred via the
stiffeners which remain unbuckled. As axial compression increases and peak or limit
loading is reached, the stiffeners will also begin to buckle. When this occurs, the
structure enters the deep postbuckling region where the local buckles on the skin,
influenced by the buckling of the stiffeners, start to form a large global buckle. The
stiffeners will buckle in sine-waves depending on the level of loading. Collapse
eventually occurs when local material failures across the transverse axis of the structure
form a line of failure which moves out of plane and results in material failure. This is
observed as a sharp drop in the load shortening plot. A detailed description of the failure
sequence of a COCOMAT panel is provided in Chapter 4.
Singer, Arbocz and Weller (1998, 2002) provide extensive information and a useful
reference on the buckling and postbuckling of thin-walled structures. A wide range of
experimental methods for testing and assessing buckling behaviour is provided together
with a summary of the state of the art of buckling theory and computations.
2.5.2 Experimental testing
Curved elastic structures in compression are highly imperfection sensitive and a wide
scatter can usually be observed from experimental testing. Initial attempts in buckling
experiments involved the generation of empirical data for characterising the buckling
behaviour of flat elastic structures. Examples of these early attempts include Lundquist
(1933) and Donnell (1934).
Recent attempts, however, have been focussed on matching experimental testing with
numerical simulations. Therefore attempts have been made in the experiments to ensure
that the known imperfections are meticulously measured and recorded in order to better
understand the failure mechanisms. This is especially significant for composite
structures where buckling is also dependant on the laminate properties. At the present
time, there is still limited published data with regard to experimental and theoretical
investigations into the buckling and postbuckling of CFRP aerospace structures. The
57 | P a g e
articles reviewed in this section present a short summary of recent attempts at large
scale testing of fuselage-representative structures.
The European BRITE – EURAM research project DEVILS (Design and Validation of
Imperfection-Tolerant Laminated Shells) was a project from 1994 – 1999 comprising of
11 European partners aimed at finding the exact conditions that would cause buckling
and failure in thin-walled structures. One of the goals was to correlate the results from
the experiments with those modelled using finite elements.
One component of the DEVILS project involved investigating if it was realistic to

consider imperfection sensitivity as a quantifiable property and whether any measures
could be used for the various load cases. Geier (1998) proposed three sensitivity indices
that can be used to determine imperfection sensitivity of axially compressed thin-walled
cylindrical shells subjected to axial compression. The indices took into consideration
the radius/thickness ratio, the buckling load factors, membrane and bending resistance.
The effectiveness of each sensitivity index was compared with the experimental results
from the DEVILS project.
Meyer-Piening et al. (2001) provide benchmark results for CFRP cylinders tested under
axial and torsion loading. The investigation showed that the buckling loads of cylinders,
which were imperfection-sensitive, under axial loading were not so sensitive when
subjected to combined loads. Furthermore, it was found that the stiffness eccentricity of
the laminate played a significant role on the magnitude of the axial buckling load, while
for combined loads this effect was somewhat reduced. Wullshleger and Meyer-Piening
(2002) tested geometrically imperfect unstiffened CFRP cylinders and found that the
FEM results differed depending on the code that was used, and good prediction was
dependant on the nodal resolution used. Geier, Meyer-Piening and Zimmermann (2002)
found that the stacking sequence of the laminate affected the influence that
imperfections had on the cylinders.
Unstiffened cylinders were also tested and analysed by Bisagni (2000) to study the
effects of imperfections on buckling cylinders. Bisagni and Cordisco (2003, 2006)
conducted similar tests with the aim of defining knock-down factors and weight
58 | P a g e
reduction for structures for collapse under the POSICOSS project. Zimmermann, Klein
and Kling (2006) performed, as part of the POSICOSS project, axial buckling tests on
stringer stiffened fibre composite panels and found that the construction of the stiffeners
affect the buckling capability of the composite panels. The way in which the stiffened
panels buckle, in the postbuckling region, is also dependent on whether an odd or even
number of stiffeners are employed. Abramovich, Weller and Bisagni (2008) tested an
assembly of stiffened panels in the form of a closed box to examine the response of the
structure under combined torsion and axial loading. It was found that the finite element
analyses tended to over predict the collapse loads for the structures, and that the
stiffener configuration is critical for torsion loading.
2.5.3 Recent developments in numerical methods for buckling and

postbuckling
One of the main issues that are of concern with the non-linear finite element analyses
for buckling and postbuckling analyses is the large amount of computational time
required for convergence to be reached. With regards to this, many researchers have
focussed their efforts on the development of more computationally efficient buckling-
specific solvers and element types and also on the development of other improved
methods of calculating postbuckling and collapse loads.
Numerical simulations of stiffened structures subject to collapse in finite elements are

usually computationally expensive due to the model experiencing large strains and
deformations and thus a requirement to have small time steps in order to achieve
convergence. SMR S.A., a COCOMAT partner, have successfully used the B2000++
finite element software (Doreille and Merazzi 2008) using a transient solver to achieve
significant reduction in computational run-times for the collapse analysis of stiffened
shell structures. The developments comprise a static non-linear finite element
continuation solver, an artificial damping procedure, a line search technique to improve
the convergence of the Newton algorithm, as well as an implicit dynamic non-linear
solver with linear multi-step time integration and variable time step size, controlled by a
59 | P a g e
local time integration error estimation. New shell elements with Assumed Natural Strain
and Enhanced Assumed Strain were created in order to reduce computational time.
Rittweger et al. (1995) developed a stiffness matrix for an element of a shell of

revolution considering arbitrary load distributions and initial geometric imperfections.
From the analysis using the new element, it was demonstrated that the imperfection
sensitivity of a shell can depend significantly upon the load distribution and true
boundary conditions, load distribution and arbitrary initial imperfections need to be
considered in order for failure to be accurately analysed.
Hong and Teng (2002) present a new finite element formulation where coupling
between different harmonics is dealt with rather than by the use of pseudo-loads in the
analysis of elastic doubly curved segmented and branched shells of revolution, subject
to arbitrary loads. This coupled approach allows easy implementation with the arc-
length method resulting in load-deflection paths which can be traced accurately. This
has been further improved by Teng and Hong (2006) where the finite element
formulation found that strong interaction amid various harmonic modes may exist and
the transition of deformation mode from one to another is gradual.
Apart from finite element analyses, solving schemes using semi-empirical methods have
been developed in order to reduce computational time. An issue with such methods is
the need to maintain a high level of accuracy despite the reduction in computational
time. The Generalised Beam Theory (GBT) has been used by Goçalves and Camotim
(2004) and Camotim et al. (2006) and more recently in postbuckling stiffened shell
structures by Silva et al. (2008). GBT achieves a high level of accuracy by discretising
the cross-section of the structure into various deformation modes such as bending,
distortion, shear and transverse extensional deformation modes. This allows the analysis
to be conducted with a reduction in the degrees of freedom by preselecting a set of
deformation modes.
60 | P a g e
2.5.4 Mode shapes in buckling
Under compression loading, long slender structures will usually adopt the minimum
energy state which can be seen in the buckling mode shape that they exhibit. The mode
shapes of the structure are highly affected by the geometry, boundary conditions and
loading of the structure. As the mode shapes reflect the buckling load being carried by
the structure, being able to accurately simulate the mode shapes ensures that the finite
element results are comparable with those obtained from experimentation. In addition,
the mode shapes can change as the load increases. Crisfield (1996) notes that accurately
capturing mode shape changes is a challenging mathematical exercise complicated by
the multiplicity of equilibrium states at buckling and the nonlinear and dynamic nature
of mode shape changes. Stiffened structures generally undergo two stages of buckling
from bifurcation or local buckling, where buckles form between the stiffeners, to deep
postbuckling, where the stiffeners and the skin buckle globally as a unit, before
collapse.
In the deep postbuckling region, panels have a tendency to ‘snap’ to higher order
buckling modes due to instability. The issues relating to numerical simulation of this
‘snapping’ have been investigated by Riks (1979), who devised a computational
procedure for tracking the critical bifurcation points and tracing the paths connected to
these instabilities, and by Crisfield (1981) who improved on the solution using the
Newton-Raphson Method for better convergence in the finite element analysis. Riks,
Rankin and Brogan (1996) simulated the collapse of a thin-walled composite cylinder
using a combination of the path tracing method (Riks 1979) in conjunction with a
transient integration method to further improve the convergence of the finite element
analysis.
Falzon and Hitchings (2003) investigated a secondary instability which was coupled
with a change in the mode shape for blade-stiffened panels loaded beyond initial
buckling. A modified explicit dynamic analysis was successfully devised to address this
issue and Falzon and Cerini (2006) have made progress in the numerical simulation of
mode shape changes by using a bracketing procedure to locate critical points, followed
by eigenmode injection to ‘point’ the displacement increment in the right direction.
61 | P a g e
2.5.5 Modelling imperfections
In order to have an accurate prediction for buckling loads and collapse, it is essential
that imperfections are included in the numerical simulations. The use of the term,
‘imperfection’ in buckling analyses typically relates to imperfections in geometry but
this can also been extended to localised structural stiffness and boundary conditions.
The pioneering work of Koiter (1945) led to an understanding of the influence that
imperfections have on the postbuckling and collapse of elastic structures through the
use of a sensitivity factor – the b-factor – to allow for the classification of imperfect
cylindrical shells. Koiter’s theory is exact at the bifurcation point and a close
approximation for postbuckling configurations near the bifurcation point (Singer,
Arbocz and Weller, 1998).
The application of Koiter’s theory has since been applied to a computer-aided

environment (CAE) package at Delft University of Technology by Arbocz and Hol
(1990). Tsouvalis et al. (2001) used CAE to reconstruct the actual geometry of thin-
walled cylinders for finite element analysis. The accuracy of the simulations was
increased by including the outer and inner wall geometry of the experimental panels.
Numerous researchers have since extended the study of imperfections for elastic
structures beyond localised geometric imperfections. Hilburger and Starnes (2002)
extended the study of imperfections to include shell-wall thickness variations, local
shell-wall ply-gaps associated with the fabrication process, shell-end geometric
imperfections, non-uniform applied end loads, and variations in the boundary conditions
including the effects of elastic boundary conditions.
Raj, Iyengar and Yadav (1998) acknowledge that it is impossible to control all the
variables in a manufacturing process and hence, for better understanding of structural
behaviour, all material properties should be considered stochastic. It was found that thin
plates with an aspect ratio less than 100 were found to be sensitive to deflection loading,
and the longitudinal modulus and in-plane shear modulus were critical when plates were
subjected to bending loads. Singh, Iyengar and Yadav (2002) note that although there is
a wealth of understanding generated from deterministic analyses, structural analyses
62 | P a g e
where the material properties are stochastic is still underdeveloped. Cylindrical panels
were analysed and found to be sensitive to boundary conditions and aspect ratio as well
as material properties.
Yadev and Verma (1997) varied the aspect ratios of cylindrical composite panels and
found that panels with low aspect ratios and high thickness ratios were more buckling
sensitive. Noor, Starnes and Peters (1997) have concluded that cylindrical panels are
most affected by edge displacement. A11, the extensional component of the stiffness
matrix, was also identified as being a critical buckling factor in a cylindrical panel.
Their methodology involved the use of numerical sensitivity response techniques. The
results do not contradict the results obtained by Singh et al. as a different methodology
was used to conduct the stochastic experiments.
Various FEA codes such MSC.Nastran, MSC.Marc and Abaqus currently allow for
imperfection modelling which admits variation of material and geometry parameters.
Initial benchmark FEA of Project COCOMAT panels by Orifici et al. (2008) have
included this imperfection feature in order to match load-shortening curves obtained
through experimentation.
2.5.6 Probabilistic methods and buckling
The issue of imperfections in buckling structures should be treated as probabilistic

considering that all actual imperfections, defects and loading occur randomly. This
problem has previously been examined by researchers such as Bolotin (1962), Fraser
and Budiansky (1969), Amazigo (1969) and Roorda and Hansen (1972), among many
others. Since the time of the foundational publications by the above researchers, the
study of imperfections has been seriously investigated as it is a key factor affecting
buckling loads of thin-walled structures. A collection of imperfection data has now been
collated under the International Imperfection Data Bank located at two locations in TU
Delft, Netherlands and at the Technion, Israel for correlation studies between the
manufacturing method of the shell and their geometric imperfections.
63 | P a g e
Arbocz and Hol (1995) have been involved in characterising imperfections in thin-
walled structures based on the manufacturing technique used. The measurements of the
initial geometric imperfections were represented using a Fourier series and the Fourier
coefficients were used to construct the second-order statistical properties that were then
included in the numerical simulations. The results were a good match with those
obtained from experimentation. Cederbaum and Arbocz (1996) have considered both
the effect of random initial geometry and loading using Koiter’s formulations. It was
concluded that the loading of the shell was also a critical factor in determining the
behaviour under compression loading.
Elishakoff (1978) was the first to conduct systematic studies of random geometric
imperfections affecting the reliability of shells using the Monte Carlo method
(Metropolis and Ulam 1949). However, the accuracy of the Monte Carlo method is
highly dependent on the number of samples used and therefore it can be inefficient if a
high level of accuracy is required. Nonetheless Elishakoff’s results using the method in
a later study (Elishakoff 1979) closely matched those obtained analytically proving its
applicability to the buckling problem. Elishakoff et al. (1987) proposed a first-order
second-moment analysis to determine the reliability of imperfect shells in compression.
It was found that the first order second-moment method was efficient as it produced
results that were in good agreement with those obtained using the Monte Carlo method
despite using a smaller sample size. Elishakoff has since published several articles on
probabilistic methods. Publications that relate to this thesis include Elishakoff, Cai and
Starnes (1994), where the Monte Carlo method is used as a stochastic tool to determine
probability density of the buckling load and the reliability of a column with initial
imperfections, and Elishakoff, Li and Starnes (1995), where the buckling of an elastic
plate is highly influenced by the misalignment of a stiffener location.
Adali et al. (2003) studied the optimal design for laminated composites under uncertain
buckling load. The ply angles were varied in the analyses and it was found that under
deterministic design conditions, the buckling load of the laminate was lower compared
to the laminates which had been designed using probabilistic robust design when the
loading was uncertain.
64 | P a g e
The generation of realistic knockdown factors ensures that the lower bounds of buckling
loads are effectively accounted for and also assists in providing less conservative design
criteria. Arbocz and Hilburger (2005) have successfully derived a knockdown factor
using a semi-analytical reliability approach. In the study, several composite cylinders
are experimentally tested and numerically modelled. Initial geometrical imperfections
from a database were applied on to the numerical models with the aim of generating
realistic knockdown factors. The results showed that the reliability-based knockdown
factor derived by the authors consistently provided a safe estimate of the buckling load
of the cylinder specimens. The knockdown factor was also found to be less conservative
compared to empirical methods which have previously been used. Huehne et al. (2008)
have shown that a common knockdown factor can be applied for all unstiffened
cylinders using a perturbation method. In this methodology, a perturbation load was
applied on to the mid-span of a cylinder between the edges resulting in a consistent
reduction of 30% in buckling load.
2.5.7 Conclusion
The buckling phenomena, comprising bifurcation buckling and postbuckling, has been
investigated by various researchers since the 1930s. Critical buckling and limit loads for
elastic structures were initially found through experimentation but this has now been
extended to the use of analytical and numerical methods. The prohibitive cost of
experimentation has been the driver for faster and more accurate numerical methods,
such as finite elements, over the last four decades. This has mainly been due to the non-
repeatable nature of the experiment as well as the need to obtain reliable results from a
suitable sample size.
Buckling is highly sensitive to both material and geometric imperfections and yet,
despite the large amount of research dedicated to examining these imperfections, there
is still no widely accepted method that is able to identify which imperfections should be
of concern. In addition to this, there is still difficulty in explaining the wide variety of
scatter that is observed in experiments and matching these with results obtained from
finite element analysis.
65 | P a g e
As mentioned in the previous section, accurately capturing the buckling mode shapes is
imperative as it is a reflection of the minimum energy state that the structure adopts.
While numerical schemes currently allow these mode shapes to be analysed in finite
elements, these are still dependent on the actual imperfections that are introduced. This
is especially the case where composite materials are used and the complexity of
imperfections increases, as shown in Section 2.4.
Finally it can be seen that there is still scope for research to be conducted such that
finite element analyses are completed before experimentation in order to reveal the
variety and source of responses in the buckling experiments so that the parameters of
the experimentation can be effectively tailored to answer the research questions.
2.6 Placement of the Research in this Thesis
A comprehensive review of literature in research areas pertaining to this thesis has been
presented. The review was conducted in order to support the aims of applying stochastic
analysis to develop an understanding of the sources and significance of the variability
encountered in the buckling of stiffened shell structures in the COCOMAT project as
well as the development of a design methodology which would enable the design of
more robust configurations that exhibit a reduction in the significant variations.
Stochastic analysis and the Robust Design concept were introduced. This was followed
by the parallel research and findings that have been applied in Civil Engineering,
namely Disproportionate and Progressive Failure theories. The various defects and
variations in composite structures were then summarised followed by the buckling and
postbuckling behaviour of stiffened structures under compressive loading. Stochastic
analyses are able to capture sufficient information on the structure so that real-world
phenomena that are observed can be simulated in finite elements. Metamodels which
are then generated from the stochastic analysis can then be used for fitting the behaviour
of the output response. As the metamodel is for fitting purposes and not for validation,
sensitivity analysis techniques using statistical methods can be used to classify the
behaviour of the structure. Nonparametric uncertainty methods including correlation
66 | P a g e
and regression should be applied as these measures are able to provide insight into
which inputs most affect the response as well as the overall response of the structure. A
methodology using the concepts found in the review of literature will be developed in
Chapter 3.
Robust Design was found to be a useful process of ensuring that the output response
falls within a desired range of values. However the current state of the art for use in
structural mechanics came in the form of the, now defunct, MSC.Robust Design
software. While the algorithm used in the software was able to identify which of the
input parameters had the closest relationship with the output response, no indication was
given about how sensitive that response was to the input variable. Therefore it is not
possible to achieve a complete representation for the robustness of a structure. Building
on the qualification of what robustness is, Chapter 3 addresses this issue by presenting
a new methodology to quantify the robustness of a structure. This is important as this
allows any improvements in robustness to be measured as design changes or studies are
conducted.
Disproportionate and Progressive Failure are two key concepts that can be transferable
to Robust Design in ensuring that initial local failures do not propagate into a global
failure mode. The hand calculation of a simple beam subjected to an axial load showed
how non-linearity could be a source of disproportionate failure as the propagation of
damage can be further extended due to the inability of individual loading paths to
transfer loads across. Robustness, therefore, needs to consider how each input variable
can adversely affect the response of a structure. The measures of robustness that are
formulated must incorporate a determination of how disproportionate the response is
with respect to each input variable as well as to all of the input variables.
In Section 2.4, it was shown that composite structures are susceptible to many defects
such as delamination, fibre misalignment, matrix voids, skin-stiffener disbond, etc.
These defects and variations ultimately affect the stiffness and strength of the structure
and must therefore be included in the finite element analyses. The variations that were
determined from materials characterisation in the COCOMAT program will be used as
67 | P a g e
initial assumptions and a comparison study of these variables will be conducted using
data obtained from material characterisation for the buckling analyses in Chapter 5.
One of the aims of this thesis is to apply stochastic analysis to develop an understanding
of the sources and significance of the variability encountered in the buckling of
stiffened shell structures in the COCOMAT project. This leads towards an attempt to
match experimental results with numerical simulations and will involve the applications
of all three types of Robust Design that are listed in Table 2.1. Buckling analyses are
highly nonlinear and the energy states represented by the stiffened shells and panels in
the deep postbuckling region are dependent on the initial imperfections and loading.
Chapter 5 will show the application of the stochastic methodology developed in
Chapter 3 with attempts to match the results from the initial experimental benchmark
discussed in Chapter 1. The analyses will use a multi-variant method in the stochastic
analyses to produce a small sample size in order to predict the responses that are
observed from the experiments.
From reviewing the literature on postbuckling analyses, it can be seen that postbuckling
analyses for stiffened shells in finite elements is computationally intensive due to the
large strains and displacements that the models experience. Many of the models were
solved using parallel computing and super computers. The computational time is
especially longer in the case for analyses running progressive failure algorithms. The
aim is, therefore, to use the data obtained from experimentation to determine how far
into the postbuckling region a finite element analysis should solve in order to obtain the
most computationally accurate and efficient solution. In the case of this thesis,
computational accuracy is defined as how closely the experimental postbuckling modes
match those of the finite elements; this is a function of the energy level of the panel and
would thus provide a good match in terms of the load-shortening plots. Computational
efficiency is defined in terms of time required to achieve an accurate result. Therefore
using these two criteria as a gauge, it is not a requirement to run the postbuckling finite
element analyses to collapse. Instead robustness can be ascertained using a finite
element run only to the point before a major drop in the load-shortening plot, not related
to a change in geometry from buckling modes snapping through various states.
68 | P a g e
From Section 2.5, it was found that many attempts have been made over the years to
investigate the effect of imperfections of shell structures. Initial geometric imperfections
have been thoroughly studied together with the effect of loading conditions. The
imperfections in composite structures also need to be investigated in order to have a
clear picture of the imperfections and defects that are of concern in a buckling analysis.
This will be shown in Chapter 6, 7 and Appendix B of the thesis where composite
structures with material defects are investigated. These will reveal which variables
highly affect the behaviour and response, as well as allowing the overall response of the
structure to be quantified in terms of robustness.
It should be noted that the aim of all design philosophies that are currently in use have a
similar goal of ensuring that premature failure does not occur. Similar to concepts such
as damage tolerance, Robust Design also ultimately aims to reduce the cost of
ownership for operators of aerospace vehicles. Therefore each chapter of this thesis
aims to build confidence that Robust Design is a viable option for increasing safety and
improving structural design in the aerospace industry.
69 | P a g e
Chapter 3 Theory and Methodology
Chapter 3
Theory and Methodology
This chapter presents the formulations for the stochastic analysis and quantification of
the robustness of a structure. The stochastic analysis is based on the well known Monte
Carlo sampling procedure. The only modification required is to minimise the size of the
family of samples due to the resources required to analyse each member. On the other
hand, no universal quantitative measure of robustness currently exists for structural
design even though the concept of robust design has existed for more than 40 years.
There is also no objective way to assess the effects of design changes on the robustness
of the structure. The sections dealing with robust design will therefore dominate the
work presented in this chapter.
The formulation for the stochastic analysis methodology is shown in Section 3.2 where
a Monte Carlo sampling procedure is defined. This section also includes the definition
of Influence and Sensitivity. Influence refers to the influence an input variable has on an
output response. Sensitivity refers to the sensitivity of an output response to an input
variable. Both variables will contribute to the definition of robustness of the structure.
In Section 3.3, the Robust Indices are derived using the Influence and Sensitivity
variables. The first index, R.I.I, is defined for each input variable and allows the
robustness of the structure to be defined with respect to that input variable. Using the
standard deviations, the Influence and Sensitivity variables can be quantitatively
measured for each variable. This list of indices forms the basis for quantifying a single
robustness index R.I.II for the entire structure, subject to variations and imperfections.
A cantilevered beam is again used as a worked example in Section 3.4 to illustrate the
70 | P a g e
methodology introduced for quantifying robustness. This procedure establishes the

framework that will be utilised to analyse the structures in the following chapters. In
Section 3.5, the effects of statistical properties on the metamodels as well as the
resulting measures of robustness are evaluated.
3.1 Stochastic Analysis
The stochastic analysis procedure defined in this section is applied in Chapter 5 when
the curing deformations and postbuckling mode shapes of the COCOMAT are
investigated. The stochastic analysis defines a family of designs that are analysed using
the finite element program MSC.Nastran. Features of the pre-processor MSC.Patran that
enable multiple models with randomly varying geometry and materials models to be
generated and solved will be used.
3.1.1 Sensitivity Analysis (SA) procedures
In a traditional multi-variable sensitivity analysis, the analysis is conducted by varying

only one input variable and observing the change in response. This can be seen in the
equation below where partial derivatives of complex functions are found with the other
variables in the equation considered as constants.
Y
Yi X i , i 1, n (3.1)
X i
where Y is the output response,

X is an input variable and
n is the number of input variables.
When more than one variable is changed the variation of each sample point from the
nominal design mean can be determined using:
71 | P a g e
Y Y Y
Y X 1 X 2 X n (3.2)
X 1 X 2 Xn
However this may not reveal the input parameters affecting the behaviour of the system
under actual operating conditions, as the method only allows the variation between the
output responses to be calculated. Hence a stochastic analysis procedure has been
devised in order to identify the variability through the determination of the key input
variables in an analysis.
3.1.2 The stochastic analysis procedure
With the introduction of variation, the input variables can now be considered stochastic
in nature, where each input variable is random and exists within a range of values
determined by its mean and standard deviation. The sampled input is defined as:
x Rand f ( , ) (3.3)
where x is the sampled input

μ is the mean
σ is the standard deviation
In most cases, the Gaussian normal distribution can be applied for the sampling
procedure. The normal distribution is suitable for describing many variables that tend to
cluster about a mean value. This is the case for many of the input parameters that are
used in this thesis where the input variations and defects include lamina orientations and
thicknesses, material stiffness and boundary conditions. Other probability distributions
can be applied depending on the situation. For example, in operational cases where the
‘infant mortality’ of components is an issue, the Weibull distribution can be applied. In
cases where the effects of random damage are to be applied, the Uniform distribution
can be used as there is an equal chance that the damage can occur at various locations
with varying severities.
72 | P a g e
The analysis can be described as multi-variant and each output result should consist of a
combination of input variables. Each input variable is stochastic, as shown in Equation
(3.3), and the resulting output response is as follows:
Y1 f1 X 1 , X 2 , X n
(3.4)
Ym f m X 1 , X 2 , X n

X is the input variable,
m is the sample size and
n is the number of input variables.
The plots in Figure 3.1 show the difference between a multi-variable and multi-variant
analysis for the definition that will be used in this thesis. In a multi-variable analysis,
the response is dependent on the variation of an input variable, while all the other input
variables are kept constant. In a multi-variant analysis, all the inputs are varied
simultaneously and the response is then seen in each metamodel plot.
73 | P a g e
Multi-variable Analysis
1.8
1.6
Output Response
1.4
1.2
0.8
0.6
85 90 95 100 105 110 115
Input Variable
Multi-variant Analysis
1.8
1.6
Output Response
1.4
1.2
0.8
0.6
85 90 95 100 105 110 115
Input Variable
Figure 3.1: Comparison between the results for a multi-variable and multi-variant
analysis.
It is important to note that the independence axiom (Suh 2001) has to be maintained
when selecting the input variables where variation is to be introduced. This is to ensure
that a correct solution is obtained. If variations are introduced to dependent input
variables, basic physical and mathematical laws will be violated. An example of this is
simultaneously varying both the volume and edge dimensions of a cube; the volume is
dependent on the edge dimensions and so, varying them independently would violate
both mathematical and physical laws.
74 | P a g e
Therefore, a stochastic analysis with a sample size of m and n input variables, where m
values of the output response Y are obtained through repeated analyses, can be described
as follows:
Snm fn Y1 ,Y2 ,Ym (3.5)
From the stochastic analyses, n metamodels can be obtained for each output response.
In the ith metamodel for the output response Y, the m values Y1,2,m are plotted against the
input variable Xi .
The process by which the stochastic plots are generated is akin to the Monte Carlo
method (Metropolis and Ulam 1949). The Monte Carlo method is used when it is not
possible to generate a solution using a deterministic formulation. A comprehensive
description of the method and applications can be found in (Bauer 1958). In the case of
this thesis, the aim is to generate a reasonable sample using repeated sampling in order
to predict the causes of scatter and measure the overall behaviour of a structure, in terms
of robustness.
First the set of variables to be considered for the structure is defined. Then the statistical
properties are defined for each variable. For example, the variable may have a normal
distribution with defined mean and standard variation. An algorithm is then required to
define values for the variable that conform to this distribution. If it is decided to create a
family of m=100 designs, for example, the random number generator is used to define
100 values of the variable that conform to a normal distribution with the defined mean
and standard deviation. This process is repeated for each variable in the model. The first
member of the family then uses the first value for each of the variables, the second
member the second value for each of the variables etc.
Each member of the family is then analysed using the finite element method. The result
of the analysis is to define the output response Y that is under investigation. The output
response could, for example, be the maximum deflection, the maximum stress etc. A set
of m values of Y are obtained that can be subject to a stochastic analysis.
75 | P a g e
3.1.3 Sample size
The Strong Law of Large Numbers states that the sample mean will converge ultimately
to the expected value as the sample size increases. That is :-
P lim X n 1 (3.6)
x
It is reasonable to assume that a large sample size is required if the value of interest is
the mean of the result. In robust design the interest is in the change of the output with
respect to the input. Numerical simulations can be computationally intensive and,
therefore, the minimum required sample size should always be considered. A review of
literature found that this is an area that has currently not been fully explored. However a
minimal sample size has been suggested by Will, Moeller and Bauer (2004) and this
will be applied for the analyses in this thesis. The recommended minimal sample size
can be calculated as follows:
m 2( n r ) (3.7)
where m is the sample size,

n is the number of input variables and
r is the number of outputs.
The remaining sections of this chapter use Equations (3.1) to (3.7) to introduce
variations and imperfections to the structures.
3.2 Formulation for Indicators of Robustness
In this section formulations are derived for the quantification of the metamodels which
have been generated. The methodology shown is applicable for all defects, variations
and imperfections in geometry, boundary conditions and material properties. The two
76 | P a g e
variables, Influence and Sensitivity, which are introduced in this section, can be used to
provide insight about the effect of each input parameter in the metamodels.
3.2.1 Derivation of Sensitivity
Sensitivity can be found using the least-squares method (Legrende 1805). This is a
measure of the gradient of the output result, with respect to the input variable. The
formulation for Sensitivity β is as follows:
(3.8)

X is the input variable and
the over-bar indicates the mean value.
The Sensitivity points to the amount of change that the output result experiences with
respect to a change in the input variable. Thus a large Sensitivity translates to a large
change in output, as the input is varied. This can be seen Figure 3.2 where plots
showing various levels of sensitivity are shown. Note that the Sensitivity is the gradient
of the line.
Low Sensitivity High Sensitivity

1 1
Output Response
Output Response
0 0
0 Input Parameter 1 0 Input Parameter 1
Figure 3.2: Example of varying degrees of Sensitivity.
77 | P a g e
3.2.2 Scaling of the Sensitivity parameter
It is a requirement that the sensitivities with respect to the input variables are scaled so
that any skew effects in the final result due to the scale of each input can be removed.
The scaling problem is frequently encountered when analysing composite structures
where, for example, the longitudinal Young’s modulus is a factor of 105 while Poisson’s
ratio exists at a factor of 10-1. The scaling used in this thesis is to multiply the sensitivity
of the input variable by the mean value for that variable. This allows the Sensitivity
parameters within each sample set to be compared. Scaled Sensitivity, B can be obtained
as follows:
(3.9)
where β is the Sensitivity of the input variable and

μ is the sampled mean of the input variable.
3.2.3 Derivation of Influence
The Influence is a measure of the correlation of the output result with respect to the
input variable. The Spearman Correlation (Spearman 1904) has been selected as it is
neither biased nor is there a requirement for the variables to be linearly related. The
formulation for the Influence ρ is:
(3.10)
where R is the ordinal rank,

Yi is the output response,
Xi is the input variable and
m is the number of samples in the metamodel.
78 | P a g e
An Influence factor of unity can be interpreted as an input variable being proportional to

the output. Conversely, if the Influence factor is -1, it can be concluded that the input
variable is inversely proportional to the output response. A high influence means that
the input variable lies on the sensitivity curve and has dominance over the response of
the output. Figure 3.3 shows Influence changing from 1.0 to -1.0. The distinct curves at
1.0 and -1.0 represent the Sensitivity. The Influence plot for zero influence is for points
randomly placed about a mean.
Figure 3.3: Example of varying degrees of Influence.
The stochastic analysis defined in this thesis is initially directed at defining all the
output responses encountered in the experiments conducted in the COCOMAT
program. This analysis first conducted a survey of the panels manufactured and tested to
determine the variables that would be included in the analysis of the panels, and the
type of distribution, mean and variance that would be used. In all cases normal
distributions were selected.
The second step is to define the size of the metamodels that would be used. Initially the
number of samples was set to 100 – roughly conforming to the sample size defined in
Equation (3.7). In practice the analyses were computationally intensive and so the
output responses were continuously monitored until the full set of buckling mode
shapes encountered in the COCOMAT experiments had been defined. This occurred at
a sample size of approximately 30 samples and the process was terminated since the
goals of this phase of the research had been achieved. In practice, if the goal was to
predict the outputs that could be expected prior to conducting any experiments a full
sample set would be analysed.
79 | P a g e
In the case of a-priori prediction the final step would be to analyses to determine the
best way to present the results. For the results presented in the thesis basing in
presentation on the buckling mode just prior to development of the global buckle that
preceded collapse would have had sub-branches since the path by which that final mode
was achieved was not unique.
The stochastic analysis that has been described in this section will be presented in
Chapter 5.
3.3 Formulation for Robust Indices
With Influence and Sensitivity, the robustness of structures can be quantified. The
Robust Indices that are described in this section are applied in Chapter 6 where two of
the experimental COCOMAT panel designs are analysed and in Chapter 7 when a
redesign is conducted for one of the panels in order to further improve its robustness.
3.3.1 Derivation of the Robust Index R.I.I
The Robust Index R.I.I with respect to an input variable is defined using the Influence
and Sensitivity of the output response for that input variable. The index is therefore
evaluated for each output response and for each variable.
(3.11)
where Xi is the input variable.
Clearly, a design is robust, with respect to changes in a design variable, if the sensitivity
BXi is small. However, the product B is used here because each sample point in the
metamodel shows the effect of simultaneous variation of all the variables. A low
correlation between the output and input variable in the metamodel indicates other
variables are causing the change and so the effect of a higher sensitivity can be reduced.
80 | P a g e
The measure of robustness for the design is then the minimum of the absolute value of
the indices recorded across all the variables and output responses.
(3.12)
where ρ is the Influence,

B is the scaled Sensitivity and
Xi is the input variable.
The family of indices [R.I.]Xi allows the individual parameters affecting the behaviour
of the structure to be identified and so possible design changes can thus be
recommended.
3.3.2 Derivation of the Robust Index R.I.II
A second measure can also be derived to indicate how representative the minimum
Robust Index in Equation (3.12) is for the design of the structure. To achieve this, the
standard deviation for Influence and scaled Sensitivity has to be found:

1 n

i
n 1 i 1
2
and (3.13)
B
1 n

Bi B
n 1 i 1
2
Combining the standard deviations for Influence and scaled Sensitivity, the formulation
for the second measure is as follows:
(3.14)
81 | P a g e
This measure indicates the reliability of the Robust Index from Equation (3.12). If the
values of Influence and scaled Sensitivity have a small scatter, the structure will have a
higher value for R.I.II. Given the nature of the Equation (3.14), there is a possibility that
this value will tend to infinity, should either standard deviation return a zero value.
3.4 A Hand Calculation of the Stochastic Methodology
3.4.1 Cantilevered beam with edge loading
The methodology shown in the previous sections is demonstrated in the example below.
Consider a simple cantilevered beam with an end loading as shown in Figure 3.4.
P
b
d
L
Figure 3.4: Cantilevered beam with end loading.
From Solid Mechanics, the deflection of the cantilevered beam is:
(3.28)
with second moment of area,
(3.29)
82 | P a g e
3.4.2 Stochastic analysis of the cantilevered beams
Table 3.1 shows the stochastic boundary for the stochastic analysis of two beams, Beam
A and Beam B. A 0.05 coefficient of variation has been used, and the samples follow a
normal distribution function. Both beams have the same nominal volume and similar
properties and loadings with the exception of the cross-sectional dimensions.
Table 3.1: Stochastic boundary for Beam A and Beam B.

Beam Input Variable Mean Possible Range
Load, P (N) 1000 850 - 1150
Length, L (mm) 1000 850 - 1150
Young’s Modulus, E (MPa) 72000 61200 - 82800
Breadth, bA (mm) 50 42.5 - 57.5
Beam A
Depth, dA (mm) 100 85 - 115
Breadth, bB (mm) 40 34 - 46
Beam B
Depth, dB (mm) 125 106.25 - 143.75
The following assumptions were used in the analyses.
(i) The stress experienced by each beam does not exceed the Yield Stress.
(ii) Transverse shear is not considered.
(iii) The dimensions used are only an example and not a representation of actual
tolerances existing in real life.
A sample size of 100 was used for the stochastic analysis.
3.4.3 Results of stochastic analysis
The tip deflection is plotted for both beams against the input variables in Figures 3.5 to
3.9. The sensitivity for each variable can be seen as the slope of the mean line in the
plots. Table 3.2 shows the level of influence the input variables have on the deflection
of the cantilevered beam. It can be seen that in this instance, the height of beam has the
83 | P a g e
greatest influence over the deflection. This is due to the inverse cube effect of the height
in the deflection equation. It also has the minimum [R.I.]Xi value and therefore
contributes towards the R.I.I value. It should therefore be the input variable of concern.
2.5
2 Beam A Beam B
Deflection (mm)
1.5
0.5
0
850 900 950 1000 1050 1100 1150
Force (N)
Figure 3.5: Metamodel of tip deflection against applied load.
2.5
2 Beam A Beam B
Deflection (mm)
1.5
0.5
0
850 900 950 1000 1050 1100 1150
Length (mm)
Figure 3.6: Metamodel of tip deflection against beam length.
84 | P a g e
2.5
Beam A
2
Beam B
Deflection (mm)
1.5
0.5
0
60000 62500 65000 67500 70000 72500 75000 77500 80000
Young's Modulus (MPa)
Figure 3.7: Metamodel of tip deflection against Young’s Modulus.
2.5
2 Beam A Beam B
Deflection (mm)
1.5
0.5
0
30 35 40 45 50 55 60
Breadth (mm)
Figure 3.8: Metamodel of tip deflection against beam breadth.
2.5
Beam A
2
Beam B
Deflection (mm)
1.5
0.5
0
80 100 120 140 160
Depth (mm)
Figure 3.9: Metamodel of tip deflection against beam depth.
85 | P a g e
Table 3.2: Results for stochastic analysis of Beam A and Beam B.

Relationship of Scaled
Influence Sensitivity [R.I.I]x R.I.II
Displacement and Sensitivity
Beam A
Load 0.26 1.67 × 10-3 1.68 × 100 2.30

Length 0.57 3.40 × 10-3 3.42 × 100 0.51
Young's Modulus -0.11 -8.42 × 10-6 -6.03 × 10-1 15.65 0.66
Breadth -0.27 -3.84 × 10-2 -1.91 × 100 1.96
Depth -0.77 -4.17 × 10-2 -4.60 × 100 0.31
Beam B
Load 0.11 3.18 × 10-4 3.20 × 10-1 27.67

Length 0.61 2.45 × 10-3 2.47 × 100 0.67
Young's Modulus -0.01 -1.09 × 10-6 -7.81 × 10-2 1123.89 1.18
Breadth -0.03 -8.80 × 10-3 -3.52 × 10-1 86.10
Depth -0.77 -1.88 × 10-2 -2.34 × 100 0.56
In Table 3.2, it can be seen that there are notable changes in influence and scaled
sensitivity as the cross-sectional dimensions are varied. In both cases, the beam height
remains the variable with the highest influence. For Beam B, there is a notable decrease
in sensitivity. Using the procedure it can be seen that overall Beam B is significantly
more robust than Beam A when the input variables are subjected to variations.
This result can be understood if the tip deflection of the beam is determined for three
families of beams with depth and width varied so the cross-sectional area remains
constant. In Figure 3.10 the nominal height for each family is different. If the deflection
is plotted against height, the plot in Figure 3.10 is obtained. Clearly the sensitivity of the
deflection to variations in depth (the slope of the mean line plotted separately for each
family) reduces as the depth increases. As a result, the robustness of Beam B should be
higher than the robustness of Beam A if the deflection of the tip is the output response
that is significant for the design.
86 | P a g e
8
7
6
Deflection (mm)
5
4
3
2
1
0
40 60 80 100 120 140 160
Depth (mm)
Figure 3.10: Plot of tip deflection against various beam depths.
3.5 Effect of Statistical Properties on Robust Indices
The Gaussian Normal distribution has been found to be applicable for representing the
type of scatter observed in geometry, boundary conditions and material properties.
However the distribution itself is dependent on two properties, which are the mean and
standard deviation. While it is easy to ascertain the mean values in design and analysis,
the standard deviation may sometimes be a property that is difficult to establish due to a
lack of statistical information about a structure, or the need to conduct stochastic
analyses before materials characterisation has been completed. In the previous section,
it was shown how the mean values affected the robustness of a cantilevered beam. In
this section the effect that the standard deviation has on the robustness of a structure is
investigated. The cantilevered beam with the loading condition shown in Figure 3.4 is
again used in this section. The standard deviations selected for the input variables are
arbitrary if no extra information is available on the variations and defects arising from
the input parameters.
87 | P a g e
3.5.1 Stochastic analysis of cantilevered beams with varying CoV
Three coefficients of variation, 0.025, 0.05 and 0.10 are applied on to the mean values
of the input parameters. Again, this is a five variable problem using the formulas shown
in Equations (3.28) and (3.29) to ensure that the problem is better understood. Table 3.3
shows the stochastic boundary for the stochastic analysis of three beams, Beam C, D
and E with the possible range of values that the input parameters can have due to the
corresponding CoV. The beams have a similar nominal mean and volume.
Table 3.3: Stochastic boundaries for Beam C, D and E.

Beam Input Variable Mean Possible Range
Load, P (N) 1000 925 - 1075
Length, L (mm) 1000 925 - 1075
Beam C
(0.025 CoV)
Breadth, b (mm) 50 46.25 - 53.75
Depth, d (mm) 100 92.5 - 107.5
Load, P (N) 1000 850 - 1150
Length, L (mm) 1000 850 - 1150
Beam D
(0.05 CoV)
Breadth, b (mm) 50 42.5 - 57.5
Depth, d (mm) 100 85 - 115
Load, P (N) 1000 700 - 1300
Length, L (mm) 1000 700 - 1300
Beam E
(0.1 CoV)
Breadth, b (mm) 50 35 - 65
Depth, d (mm) 100 70 - 130
A sample size of 100 was used for each beam.
88 | P a g e
3.5.2 Results of stochastic analysis
The tip deflection is plotted for the three beams against the input variables in Figures
3.11 to 3.15. The sensitivity for each variable can be seen as the slope of the mean line
in the plots. It can be seen that the samples with a smaller CoV are more tightly packed
compared to the samples with a higher CoV. Table 3.4 shows the quantitative results for
the analysis which were derived from the metamodels.
2.5
CoV 0.1
Deflection (mm)
2 CoV 0.05
CoV 0.025
1.5
(CoV 0.1)
1
(CoV 0.05)
0.5 (CoV 0.025)
0
600 800 1000 1200 1400
Load (N)
Figure 3.11: Metamodel of tip deflection against applied load.
2.5
CoV 0.1
Deflection (mm)
2
CoV 0.05
1.5 CoV 0.025
(CoV 0.1)
1
(CoV 0.05)
0.5
(CoV 0.025)
0
600 800 1000 1200 1400
Length (mm)
Figure 3.12: Metamodel of tip deflection against beam length.
89 | P a g e
2.5 CoV 0.1
CoV 0.05
Deflection (mm)
2
CoV 0.025
1.5
(CoV 0.1)
1 (CoV 0.05)
0.5 (CoV 0.025)
0
60000 65000 70000 75000 80000 85000 90000
Young's Modulus (MPa)
Figure 3.13: Metamodel of tip deflection against Young’s Modulus.
2.5 CoV 0.1
CoV 0.05
2
Deflection (mm)
CoV 0.025
1.5
(CoV 0.1)
1 (CoV 0.05)
0.5 (CoV 0.025)
0
35 40 45 50 55 60 65
Breadth (mm)
Figure 3.14: Metamodel of tip deflection against beam breadth.
90 | P a g e
2.5 CoV 0.1
CoV 0.05
Deflection (mm)
2
CoV 0.025
1.5
(CoV 0.1)
1 (CoV 0.05)
0.5 (CoV 0.025)
0
70 80 90 100 110 120 130
Depth (mm)
Figure 3.15: Metamodel of tip deflection against beam depth.
Table 3.4: Results for stochastic analysis of Beam C, D and E.

Beam C
Load 0.31 1.77 × 10-3 1.77 1.84

Length 0.67 3.69 × 10-3 3.70 0.40
Young's Modulus -0.25 -1.74× 10-5 -1.25 3.25 0.69
Breadth -0.15 -1.40× 10-3 -0.70 9.69
Depth -0.63 -3.79 × 10-3 -3.79 0.42
Beam D
Load 0.26 1.67 × 10-3 1.68 2.30

Length 0.57 3.40 × 10-3 3.42 0.51
Young's Modulus -0.11 -8.42 × 10-6 -0.603 15.65 0.66
Breadth -0.27 -3.84 × 10-2 -1.91 1.96
Depth -0.77 -4.17 × 10-2 -4.60 0.31
91 | P a g e
Beam E
Load 0.28 1.47 × 10-3 1.46 2.44

Length 0.54 2.98 × 10-3 2.97 0.63
Young's Modulus -0.23 -1.93 × 10-5 -1.41 3.15 0.84
Breadth -0.23 -2.48 × 10-3 -1.22 3.64
Depth -0.64 -3.56 × 10-3 -3.53 0.44
3.5.3 Discussion and conclusion
In Table 3.4, the R.I.I and R.I.II measures showed no discernable trend. However from
the influence values, it can be seen that the beam length and depth remain the input
parameters affecting the response and robustness of the structure. The analysis found
that the amount of scatter, in the case of this cantilevered beam, had little effect on the
robustness. This lends credence to Marczyk’s assertion that robustness is not a function
of quality. However it must be stated that this may be true only in this particular case
and not for all structures and their respective loading cases. Therefore this will be
investigated further when the stiffened composite fuselage structures are analysed for
robustness in Chapter 6.
3.6 Conclusion
This chapter has presented the formulations for the stochastic methodology and
determination of robustness that will be used in this thesis. The stochastic methodology
allows the possible responses of a structure to be found. This is especially important in
structures having highly nonlinear responses, such as the stiffened panels in the
COCOMAT project. Being able to also use the results to determine robustness is a
significant step forward in the Robust Design concept as it allows the robustness of
structural designs to be effectively quantified and compared. The stochastic
methodology and measures of robustness discussed here are employed in Chapter 5 to
92 | P a g e
validate the initial assumptions used in the postbuckling analyses where values
measured from material characterisation are compared with arbitrarily assumed values.
The methodology is also used to match the experimental results obtained in Chapter 4,
together with solutions obtained from finite elements. The measures of robustness are
also applied in Chapters 6, 7 and Appendix B to demonstrate the practicality of using
the methodology for structural analysis and design.
It may appear that the method used here is inefficient due to the repeated sampling that
is carried out. However, to determine the robustness of a structure, the aim of the
sampling is to establish the cause of the observed response, not the actual response
itself. Therefore the number of samples required is at the user’s discretion but there is a
need to satisfy Equation (3.7).
93 | P a g e
Chapter 4 Experimental Setup and Results
Chapter 4
Experimental Setup and Results
This chapter provides details for the experiments conducted by the German Aerospace
Center (DLR) for the COCOMAT project. Two stiffened panel designs were tested at
the DLR; the first panel, known as the Design 1 (D1) Panel, was designed by the DLR
and CRC-ACS while the second panel, known as the Design 2 (D2) Panel, was
designed by a COCOMAT partner, Aernnova Aerospace, which also manufactured both
panels used for the experimental testing. The experimental setup is presented along with
the measurement devices. This is followed by a section showing the material and
geometrical imperfections.
The following is a summary of the test program. Within the COCOMAT project, a total
of six panel designs were created for analysis and experimentation, so that parametric
studies could be conducted once experimentation was complete. A total of 43 panels
were manufactured and tested using the six designs. The breakdown is shown in Table
4.1. The work in this Chapter and thesis relate to the Design 1 (D1) and Design 2 (D2)
panels. CRC-ACS was involved with DLR and Aernnova Aerospace in designing and
analysing these panel designs.
94 | P a g e
Table 4.1: Breakdown of panels tested in COCOMAT project.

Design Design 1 Design 2 Design 3* Design 4 Design 6* Design 5*
Manufacturer Aernnova PZL IAI Agusta
Tester DLR PZL TECHNION POLIMI
Loading C C C C S C and S
Undamaged panels
Static
buckling test 2 1 1 - - 2
to collapse
Cyclic
buckling test 2 1 2 2 4 2
with collapse
Pre-damaged panels
Static
to collapse
Cyclic
with collapse
Panels (Total) 8 4 6 5 10 10
* model consisting of two panels
C = Compression, S = Shear
One of the aims of the COCOMAT project was to generate standards for the design of
postbuckling composite structures. Summary results for the experimental testing are
presented here, followed by a discussion on the growth of damage in the panels.
As mentioned in the previous sections, it is important to have a minimal sample size

during the sampling process in order to reduce computational time. Numerical
simulations of panels in the postbuckling state can also be computationally intensive
due to the large deformations and strains that are experienced by the model. Therefore
analysing the models to first failure in the deep postbuckling region without progressive
failure criteria will help to improve computational efficiency. Majority of the finite
95 | P a g e
element analyses in this thesis did not include damage growth and progressive failure
even though tools allowing progressive damage and failure had been developed in the
COCOMAT project. During the initial stages of the research in this thesis, it was found
that including progressive damage and failure could take up to a week to solve. The
results of the undamaged panels subjected to cyclic loading were therefore used to
investigate level of compression the panels can experience before the occurrence of any
failure in the fibre, matrix or bonding. This level of axial compression was then applied
in the finite element models for the stochastic analyses.
4.1 Experimental Setup
The buckling experiments for the COCOMAT project were conducted at the German
Aerospace Center (DLR) Institute for Composite Structures and Adaptive Systems in
Braunschweig, Germany. This chapter briefly details the experimental set up and
measuring equipment. A more complete and detailed summary of the facilities available
at the institute can be found in Degenhardt et al. (2007).
4.1.1 Buckling test facility
Figure 4.1 shows a photograph of the buckling test facility as well as a schematic plan.
The facility is capable of an axial compressive force of 1000 kN, and is able to test
cylindrical specimens with a maximum length and diameter of 1600 mm and 1200 mm,
respectively. The force applied to the panel is achieved via the drive plate on the bottom
of the specimen. This causes the panel to react against the cylindrical structure above,
which distributes the applied force into three load cells, thereby providing a smooth
force distribution. The test facility is displacement controlled using two linear variable
displacement transducers (LVDT) which measure the axial shortening of the panel. The
signals from these transducers are recorded and used for the control of the facility as
well as recorded as the applied displacement on the panels.
96 | P a g e
Axial support
Adaptation of shell length
Introduction of torsion
4.7 m 3 Load cells

Load distributor
Test shell
Drive plate
Torsion support
Facility frame
Hydraulic cylinder
Figure 4.1: Photograph and schematics of DLR test facility.
4.1.2 Panel preparation
Before loading the panel onto the testing machine, both ends of each panel are encased
in blocks made from a potting resin. The blocks have the dimensions shown in the
geometrical specifications in Figures 4.5 and 4.7. The potting serves to ensure that
loading onto the panel is linear on both edges, and to provide lateral support from the
edges to the free length of the panel. The first step involves casting the ends of the panel
into the blocks. Once the resin has cured, the blocks are removed and the ends are
milled to obtain a uniform loading surface. Figure 4.2 shows the casting of the end
blocks and the panel.
97 | P a g e
Figure 4.2: Casting of end blocks and finished end block.
4.1.3 Measurement systems for the experiments
This section describes the measurement systems used in the experimental testing of the
COCOMAT panels. Buckling experiments are relatively expensive, as the panels can be
collapsed only once. Hence, it is imperative that the maximum possible amount of
information is extracted from each experiment.
Pre-experimental non-destructive testing (NDT)
ATOS System
In the literature review in Chapter 2 it was found that geometrical imperfections are a
key factor affecting the buckling and postbuckling of thin-walled structures. Therefore it
is essential that the surface geometries are accurately measured. For the panels tested at
the DLR, the geometrical imperfections were measured using the ATOS system (GOM
2008). ATOS is a full-field digitising system that uses a laser scanner to conduct surface
measurements. Actual coordinates for the surface are generated by triangulating points
on the surface with multiple lasers. This is useful as both surface imperfections and
nominal radii can be determined.
The panel deforms during the curing process due to the effect of cooling of the
laminates (spring-back) and variations in the material properties and lay-up process. The
98 | P a g e
deformations from the spring-back affect the local buckling of the panel, while the
resulting panel radii are significant factors affecting the final collapse load. The
deformations are not corrected by the potting process since the geometry is preserved by
curing the potting mix with the panel in a relaxed condition. The plots from the ATOS
measurements are therefore useful when attempting to make comparisons between
experimental results and finite element simulations.
Ultrasonic scanning
Ultrasonic scans were carried out on the panels before and after experimentation. Prior
to the experiments, the panels were scanned in order to ensure that any structural
anomalies were detected before the panel is encased in potting. During the scanning, the
quality of the secondary bonds between the skin and the stiffener were also checked to
ensure that a uniform bond was achieved between the skin and stiffeners. After the
experiments, the panels were again scanned to check for interlaminar damage. This
included the panels subjected to impact testing to measure the extent of the damage
imposed.
Experimental measurement systems
Apart from the load-shortening data provided from the test facility, three other
measurement systems were employed to record additional information during the
experiments. These include the use of traditional strain gauges and imaging
technologies, such as the ARAMIS system and thermography. The readings from
ARAMIS and thermography can be found in Section 4.4.
Strain Gauges
Strain gauges were applied onto the panel before a final casting is done on the edges.
Once this is completed, wires are attached onto the strain gauges, and the panel is
installed onto the test facility. Strain gauges are attached onto both the skin and
stiffeners of the panels, as shown in Figure 4.3. A total of 25 strain gauges were
attached to the D1 panel, measuring the strain in the axial direction. Due to the design
99 | P a g e
of the panel, the panel skins are symmetric about the centre stiffener, which allows
useful comparisons to be made between the strains experienced in each bay.
S S2 S1 S0
F9 B8/7 B6/5 F4
B6 B5
F15 B14/13 B12/11 F10
7.75 mm F4
F21 B20/19 B18/17 F16
S25 S24 S23 S22

9 mm
Stringer side Enlarged cross-sectional view
Figure 4.3: Location of strain gauges on the D1 panel and an enlarged cross-sectional
view.
ARAMIS system
The ARAMIS system has been jointly developed by the Institute of Composite
Structures and Adaptive Systems and GOM GmbH (GOM 2008). It is a full-field
displacement system that allows the displacement of the panel skin to be measured. This
is especially crucial if the out of plane displacements for the panel are to be observed.
The ARAMIS plots can be seen in Section 4.4.
Thermography system
A thermographic system from the Institute of Polymer Technology of the University of

Stuttgart was employed in order to capture the disbonding of the skin-stiffener interface
during the cyclic loading. The measurements are made with Optically Lock-in
Technology (OLT) which is able to show the level of disbonding by measuring the
amount of heat radiating off the skin surface. For more information on thermography,
refer to (Busse et al. 1992). Experimental images of the thermographic measurements
can be found in Section 4.4.
100 | P a g e
Figure 4.4 shows the setup of the measurement systems, when viewed from above, with
ARAMIS on the stiffener side and the thermographic system on the skin side.
System 1
Camera
System 2
Camera
ARAMIS Thermography
System 3
Camera
Figure 4.4: Experimental setup viewed from above .
4.2 Design of the COCOMAT Panels
4.2.1 COCOMAT Design 1 (D1) panel
Undamaged specimens
This panel was one of two designed by CRC-ACS and DLR (Degenhardt et al. 2008).
The panel is curved, with five equally spaced T–stiffeners in the axial direction and was
designed to have a large postbuckling region before collapse, so that the geometrical
and material failures leading to collapse could be better observed. Particular attention
was paid to skin-stiffener disbond growth, and this was observed using cyclic loading
for some of the panels. Four of the D1 panels tested were undamaged panels and the
geometrical specifications can be found in Figure 4.5 and Table 4.2. These panels were
known internally as panels P28 to P31.
101 | P a g e
clamped end 3.0 mm

potting
12.5 mm
L Lf
1.5 mm
potting
loaded end Stringer Design
W
w
h skin
stiffener
b R
Figure 4.5: Geometrical representation of the D1 test panel with stiffener design.
Table 4.2: Nominal geometry for the D1 panel.

Parameter Value
Panel length, L (mm) 760
Panel free length, Lf (mm) 660
Panel radius, R (mm) 1000
Stiffener pitch, b (mm) 128
Number of stiffeners 5
Panel arc length, W (mm) 560
Stiffener width, w (mm) 32
Stiffener height, h (mm) 14
Skin-stiffener joint Bonded
Adhesive FM 300
Skin lay-up [90, ±45, 0]S
Stiffener web lay-up [±453, 06]S
Stiffener flange lay-up [06, ±453]
Lamina thickness (mm) 0.125
Material Hexcel IM7/8552
102 | P a g e
Pre-damaged specimens
Four panels, internally designated P34 to P37, were pre-damaged designs with a Teflon
strip in the skin-stiffener interface acting as an initial disbond and providing idealised
crack-tips. This form of artificial damage was chosen as it was the best way to
investigate the skin-stiffener separation within the project. The size of the Teflon strip
was determined to be 100 mm, as previous experimental experience revealed that the
size of the damage had to be approximately 15% of the free buckling length of the
panel. This magnitude was necessary to obtain crack propagation during a buckling test,
under axial compression. An investigation determined that two locations would be
suitable for the pre-damage and the locations are shown in Figure 4.6. With the
exception of P34 and P36, all the panels were tested under cyclic loading conditions.
240 mm
340 mm
100 mm
100 mm
Panels P34 and P35 Panels P36 and P37
Figure 4.6: Location of Teflon strips in pre-damaged panels P34 to P37.
4.2.2 COCOMAT Design 2 (D2) panel
Undamaged specimens
The D2 panel was designed by Aernnova Aerospace and CRC-ACS. It is a scaled down
representation of the type of fuselage panel that would be used in a 70-100 passenger
aircraft. Geometrical representations for the panels can be found in Figure 4.7 and the
geometrical properties can be found in Table 4.3. The D2 panel has longitudinal edge
103 | P a g e
supports in order to apply reproducible boundary conditions, up to the deep

postbuckling region during the experiment, as the panel skin extends beyond the
stiffener flange. Previous experiments of this panel design with skin run-off by
Zimmerman, Klein and Kling (2006), found that the postbuckling behaviour of the
panel was highly sensitive to the behaviour of the skin edge and hence, longitudinal
edge stiffeners were required to stabilise the behaviour. There were a total of four D2
panels (known internally as P24 to P27) manufactured, of which two (P26 and P27)
were impact damaged. One undamaged (P25) and one damaged (P27) panel were tested
under cyclic loading.
clamped end
potting
longitudinal
L Lf K
edge
potting
loaded end
W
w
h skin
stiffener
b R
Figure 4.7: Geometrical representation of D2 panel.
Table 4.3: Nominal geometry for D2 panel.

Parameter Value
Panel radius, R (mm) 1000
Stiffener pitch, b (mm) 156
Number of stiffeners 4
Panel arc length, W (mm) 624
104 | P a g e
Parameter Value
Edge support length, K (mm) 380
Stiffener width, w (mm) 56
Stiffener height, h (mm) 28.9
Skin-stiffener joint Bonded
Adhesive FM 300
Skin lay-up [±45, 0, 90]s
Stiffener web lay-up [±45, 02, 902]s
Stiffener flange lay-up [902, 02, ±45]
Lamina thickness* (mm) 0.152
*Note: Lamina thickness of 0.152 mm has been checked against
drawings and agrees well with actual panel thickness.
Pre-damaged specimens
Two D2 panels (P26 and P27) were pre-damaged using impact damage at critical
locations on the skin side of the panel. These locations were determined to be at the
skin-stiffener interface, which would have resulted in skin-stiffener disbond,
interlaminar delamination and fibre damage, when damaged with 20 J impacts. Cyclic
loading in the postbuckling region of the panels with damage sustained from the
impacts was intended to promote interlaminar damage growth and skin-stiffener
disbonding. These impacts were meant to simulate tool drop events and damage
sustained during operation, such as hail and foreign object damage, and also to show the
damage sustained under barely visible impact damage (BVID). Figure 4.8 shows the
locations where impact damage was applied to the D2 panels.
105 | P a g e
7 mm 7 mm
clamped edge
190 mm
320 mm
loaded edge
Figure 4.8: Location of impact damage for pre-damaged panels P26 and P27.
4.3 Imperfections and Variations of the D1 Panel
This section includes results from material characterisation, geometrical measurements

against the nominal designs and some noted defects found in the D1 panel. The material
characterisation was undertaken at the DLR and is applicable for both panel designs.
Focus is placed on the D1 panel in this section, as the benchmark panel shown in
Chapter 1 was of this design.
4.3.1 Variations in material properties
The data presented in this section are results obtained from the material characterisation
of Hexcel IM7/8552 unidirectional carbon fibre epoxy, used in COCOMAT. This
material was also used in the forerunner project, POSICOSS (Improved POstbuckling
SImulation for Design of Fibre Composite Stiffened Fuselage Structures) and hence,
the material characterisation includes data obtained from that project. This enables a
larger population size to be considered. Table 4.4 shows the results of the material
characterisation. The manufacturer’s data was obtained from Hexcel (2005). Note that
there is a significant difference between the mean values obtained from COCOMAT
and POSICOSS. Nevertheless, it is important to detail the full range of possible values
and standard deviations for future use, so that realistic values can be input into the
106 | P a g e
stochastic analyses. The range of possible values, as shown below, shows the minima
and maxima for each material property.
Table 4.4: Nominal material properties for Hexcel IM7/8552.

Mean value / Standard
Range of Hexcel
deviation
Possible Values Data
POSICOSS COCOMAT
Stiffness (GPa) (%) (GPa) (%) (GPa) (GPa)
Et L 192.3 1.17 164.1 3.01 155.8 - 197.4 164
Ec L 146.5 1.84 142.5 1.69 138.6 - 150.8 150
Et T 10.6 2.36 8.7 3.91 8.3 - 10.9 12
Ec T 9.7 6.77 9.7 4.85 9.0 - 10.4 -
GL T 6.1 2.28 5.1 13.73 3.7 - 6.3 -
Poisson’s Ratio - (%) - (%) - -
υL T ( t ) 0.31 5.55 0.28 14.44 0.22 - 0.33 -
Strength (MPa) (%) (MPa) (%) (MPa) (MPa)
Rt L 2715 3.42 1741 11.92 1523 - 2836 2724
Rc L 1400 4.93 854.7 9.04 472 - 1530 1690
Rt T 56 18.56 28.8 5.23 19.3 - 69.3 111
Rc T 250 6.6 282.5 18.16 229.9 - 310.1 -
t = tension, c = compression, L = longitudinal direction, T = transverse direction
With the exception of the tensile stiffness in the longitudinal direction, the other
material properties for the composite were found to be lower than those stated by the
manufacturer. The property that will directly affect the collapse load is the longitudinal
compressive strength, Rc L; the mean value measured is about 50% lower than the
manufacturer’s stated value. Judging from the range of values that were measured, it is
concluded that the quality of the material, in terms of strength, is highly questionable
and will be a source of variations when collapse loads are to be considered.
107 | P a g e
4.3.2 Variations in geometry of D1 panel
This section details the geometrical specifications for the D1 panel. Attention has been
paid to this panel design as it exhibits various buckling and postbuckling modes, due to
the high level of instability and variability in the design. One of the obvious disparities
between the manufactured panel and the panel design is the variation from the initial
geometry. The manufactured panels were subjected to residual stresses, due to the
curing process, causing the panels to no longer be uniformly curved. Another crucial
issue that arises from the curing process is that of the panels taking on varying nominal
radii or curvature, thereby affecting the buckling behaviour and final collapse load.
Therefore, a high amount of care needs to be taken when constructing composite
materials due to the non-isotropic nature of the individual laminae. Table 4.5 shows the
values that were measured from the panels, compared with the nominal geometry in the
design.
Table 4.5: Nominal and measured geometry for the D1 panel.

Measured
Standard Nominal
Mean Range of Possible
Deviation Design
Value Values
(%)
Panel length, L (mm) - - - 780
Panel free length, Lf (mm) 658.63 0.067 657.5 - 659.0 660
Panel radius, R (mm) 937.25 11.87 864 - 1034 1000
Stiffener pitch, b (mm) 132.65 0.49 132 - 133 132
Panel arc length, W (mm) 560.4 0.24 558 - 561 560
Stiffener width, w (mm) 32.37 1.40 31.5 - 33.0 32
Stiffener height, h (mm) 14.36 0.82 14.1 - 14.5 14
108 | P a g e
(a) Panel P28 (b) Panel P29 (c) Panel P30
(d) Panel P31 (e) Panel P35 (f) Panel P37
Figure 4.9: Geometrical imperfections in the D1 panel due to residual stresses from the
curing process.
ATOS measurements were taken for all the panels tested by the DLR and Figure 4.9
shows the results measured for each of the undamaged D1 panels. The colour palette
shows the deformation of the panels with respect to the nominal curvature. It can be
seen that there are significant differences in the nominally identical panels. The source
of these geometrical imperfections can possibly be traced to defects in the laminates.
This will be verified in Chapter 5.
4.3.3 Other noted variations and imperfections in the stiffened panel
Figure 4.10 presents the differences between the nominal stiffener design, the idealised
finite element representation modelled using shell elements, and the actual
manufactured stiffener. The forming procedure also resulted in a resin-rich area at the
middle of the stiffener flange, where it is bonded onto the panel skin. This region is
possibly a location from which failure in the bond may initiate. The effect of this local
resin-rich area cannot be captured by the shell element representation used in the finite
element analyses.
109 | P a g e
12.5 mm
3 mm
32 mm
1.5 mm
Nominal Design Finite Element Manufactured
Figure 4.10: Comparison between modeled and manufactured T-stiffener.
Another critical imperfection that affects the collapse load is the skin-stiffener bonding.
Slight imperfections will ultimately affect how the disbonds grow. Symmetric
postbuckling mode shapes provide higher collapse loads, compared to asymmetric ones,
as the failure involves the symmetrical disbond of the stiffeners from the skin. The
delamination between the skin and the stiffener is highly dependent on the final quality
of the bonding process. It was noted by Aernnova, that assessment of the bond quality,
from the C-scans that were conducted after manufacture, was a difficult task due to the
variations in adhesive thickness along the bonded joints. These variations in the bonding
will eventually result in different loads for separation due to changes in the strength and
fracture toughness of the joint. Some of the imperfections from manufacturing are
Figure 4.11: Defects from the bonding process.
Other defects that can possibly occur in the manufacture of composite structures include
the incorrect lay-up sequence being employed or even certain plies not being laid up at
110 | P a g e
all. This was found in one of the panels where one of the skin bays had a ply missing, as
shown in Figure 4.12. The missing layer of material, seen as a lighter shade on the third
skin bay, is highly visible through thermography. In Figure 4.12(b), poor bond adhesion
between the skin and stiffener are shown for one of the panels under thermography.
Poor adhesion in bond
Teflon
strip
missing skin lamina
stiffener
(a) Missing lamina (b) Poor bond adhesion
Figure 4.12: Defects from the manufacturing process showing (a) missing lamina and (b)
poor bond adhesion indicated by a lack of uniformity on centre stiffener.
4.4 Results from Cyclic Loading of D1 Panel
As part of the experimental program in COCOMAT, the panel was placed in the testing
facility and underwent cyclic loading. The quasi-static cyclic loading allowed the
degradation process to be observed more closely as thermography could be used to
show the level of disbond in the skin-stiffener interface. The loading regime also
revealed the decrease in load carrying capability, as the panels were subjected to
degradation via the cyclic loading. Table 4.6 shows a breakdown of the applied loads
and displacements on the D1 panels subjected to cyclic loading.
111 | P a g e
Table 4.6: Summary of cyclic loading regime.

Loading Cycles Load (kN) Displacement (mm) Comments
Panel P28
0 – 2000 65 0.68
2001 – 4000 75 0.80
4001 – 8000 93 1.09
8001 114.1 2.29 Collapse Cycle
Panel P29
0 – 2000 89.5 1.09
2001 – 3800 103 – 107 1.90
Panel P30
0 – 2000 86.0 1.00
2001 – 4000 97 – 108 2.00
4001 – 4100 100 – 103 2.30
4101 – 4200 102 2.50
Panel P31
0 – 2000 83 0.96
2001 – 4000 99 – 109 1.98
4001 – 4200 106 – 109 2.49
Panel P35
0 – 402 47 0.48
403 – 500 55 0.58
501 – 800 64 0.69
801 – 1200 72 0.80
1201 – 1600 76 0.85
1601 – 2000 83 1.01
2001 - 2400 91.2 1.49
2401 - 3200 95 - 96 1.80
3201 – 4001 95 - 98 2.00
112 | P a g e
Loading Cycles Load (kN) Displacement (mm) Comments

Panel P37
0 – 400 53 0.58
401 – 800 62 0.70
801 – 1600 70 0.82
1601 - 2000 76 0.93
2001 82.7 1.20
2002 – 2400 88 1.48
2401 - 3200 92.5 – 93 1.80
3201 - 4001 93 – 96 2.00
The test results are presented in three sets, due to the nature of the postbuckling and
panel design. Panels P28 and P29 exhibited asymmetric postbuckling modes while
panels P30 and P31 showed symmetry in postbuckling. The remaining two panels, P35
and P37, were pre-damaged and should, therefore, be considered separately. The
degradation of each set of panels is accompanied by a description of the collapse of one
panel in each set of results. The descriptions provided are for the final collapse cycle.
4.4.1 Panels P28 and P29 – Asymmetric postbuckling mode
Panel P28 was used as a benchmark panel for the cyclic loading. The panel was
subjected to the displacements as shown in Table 4.6 for 8000 cycles before collapse.
This allowed the panel to be axially compressed beyond local buckling and into the
postbuckling region. It was ascertained that there was no noticeable degradation in the
panel in the form of fibre failure or stiffener disbonds. Thermography failed to reveal
any skin-stiffener disbonds.
The axial compression of P29 resulted in an asymmetric postbuckling mode shape at

collapse, as shown in Figure 4.13. The figure shows the panel on the stiffener side and
the displacement below each ARAMIS plot is the axial displacement applied. The
corresponding load-shortening curve can be found in Figure 4.14. At the local skin
buckling level, see Figure 4.13(a), 13 half-sine waves appear on the skin bays. It can be
113 | P a g e
seen that an asymmetrical buckling pattern is emerging about the third skin bay from
the left. The skin displacement is 1.38 mm inwards (into the page) and 0.76 mm
outwards. The difference seen in the third skin bay is probably due to the disbonds
created by the cyclic loading during the experiment, as well as variations in the skin
material. At 0.92 mm applied displacement, Figure 4.13(b), the local buckles combine
into a larger skin buckle on the outer skin bays, while the inward buckling observed in
the inner bays continue to grow. The change in stiffness probably means that some of
the stiffeners have started to buckle.
The gradient change in the load-shortening curve, at about 1.1 mm of applied

displacement, signals that all of the stiffeners have started to buckle. This point is taken
to be the first global buckle and occurs with a corresponding load of 88 kN. The kink
seen in the curve is due to an enlargement of the skin buckling in the third skin bay, as
seen in Figure 4.13(c). The onset of collapse began at an applied displacement of 2.32
mm with a corresponding load of 110 kN. This was followed by a series of sharp aural
events, signalling a growth of the disbonds in the centre and centre-right stiffeners, as
seen in Figure 4.13(e). The maximum collapse load of P29 was 110 kN at 2.51 mm
applied displacement. Upon visual inspection of the panel, post collapse, it could be
seen that interlamina delamination had occurred at the 0o/45o interface in all the
stiffener blades, with the exception of the second stiffener from the left. The failures
were also coupled with matrix failures and fibre cracking of the 45 o plies. Fibre
fractures were observed on the skin bays located left and right of the centre stiffener.
(a) 0.69 mm (b) 0.92 mm (c) 1.32 mm (d) 1.99 mm (e) 2.52 mm (f) 2.91 mm
Figure 4.13: Out of plane ARAMIS displacement results for P29.
The load-shortening curves for P29 are shown in Figure 4.14 and these include the
readings taken during regular intervals of cyclic loading. The panel is subjected to 2000
cycles of displacement at 1.09 mm, which is just past local buckling with no noticeable
114 | P a g e
degradation occurring which means that the load-shortening curves can be accurately
reproduced up to 1.9 mm of applied displacement. The decrease in stiffness seen in the
plots up to 1.9 mm is a function of changes in mode shapes. However the decreases in
stiffness as the panel is cyclically loaded are caused by degradation in the skin-stiffener
bond. It can also be seen that the effect of degradation has no effect on the load-
shortening curve up to a displacement of 0.7 mm, which is when local buckling occurs.
The onset of degradation occurs after Cycle 2001 when the panel is subjected to an
applied displacement of 1.9 mm, which is within the postbuckling region. The progress
of degradation from cyclic loading can be traced in the imaging results produced by
thermography, as shown in Figure 4.15. As per the set up configuration of the
experiment, the images observed are from the skin-side of the panel. Disbonds begin to
appear on the centre and the centre-left (as observed from the figure) stiffeners. The
occurrence of the disbonds is due to local buckling of the skin, and the location is in
good agreement with the ARAMIS plots shown in Figure 4.13. By Cycle 3201, the
loading causes a full disbond to occur on both stiffeners, thereby removing any shear
carrying capability. The result is a less noticeable kink as the postbuckling mode shape
shifts after Cycle 3201 as seen in the plots of Figure 4.14.
120
Cycle 2001
100
Cycle 2401
Compression Load (kN)
Cycle 2601
80
Cycle 2801
60 Cycle 3001
Cycle 3201
40
Cycle 3401
20 Cycle 3601
Collapse
0
0 0.5 1 1.5 2 2.5 3 3.5
Axial Shortening (mm)
Figure 4.14: Load-shortening curves for cyclic loading of panel P29.
115 | P a g e
(a) 2000 (b) 2600 (c) 2800 (d) 3200 (e) 3600
(f) Collapse
Cycles Cycles Cycles Cycles Cycles
Figure 4.15: Thermography readings for panel P29.
Table 4.7 shows the decrease in load carrying capability of the panel as the cyclic
loading progresses, and as more disbonds appear. At an applied displacement of 1.9
mm, there is no significant loss of load carrying capability as the maximum drop in
capability when compared to Cycle 2001 is 3.9%, which was recorded at the final
collapse cycle.
Table 4.7: Load carrying capability of P29 subjected to stiffener disbond.

Displacement Change in Load
Loading Cycle Max Load (kN)
(mm) (%)
2001 106.95 1.90 N/A
2401 106.08 1.90 -0.81
2601 105.49 1.90 -1.37
2801 105.20 1.90 -1.64
3001 104.94 1.90 -1.88
3201 103.98 1.90 -2.78
3401 103.69 1.90 -3.05
3601 103.11 1.90 -3.59
3701 102.78 1.90 -3.90
Collapse 109.67 2.51 N/A
4.4.2 Panels P30 and P31 – Symmetrical postbuckling mode
When P30 was axially compressed, it resulted in a symmetric postbuckling mode shape
at collapse, as shown in Figure 4.16. The figure shows the panel on the stiffener side,
116 | P a g e
and the displacement below each ARAMIS plot is the axial displacement applied. The
corresponding load-shortening curve can be found in Figure 4.17. Local buckling occurs
at about 0.40 mm applied displacement. The right most skin bay has a maximum
outward displacement of 3.0 mm. The first global buckling occurs at 0.80 mm
displacement, at a load of 63 kN. The plots in Figure 4.16(b) to (e) show the progress of
the skin buckling, as the panel proceeds through the postbuckling region. The extreme
right skin bay appears to have a larger outward displacement, compared to the extreme
left, and this is due to the high level of disbonding caused by the cyclic loading. The
onset of collapse and the maximum collapse load coincide at an applied displacement of
2.93 mm, with a corresponding load of 107 kN. The sharp drop in stiffness at 2.93 mm
was probably caused by the eventual merging of the separate disbonded regions in each
stiffener, and a line of failure occurring across the breadth of the skin.
When the panel was inspected after the experiment, it was observed that ply
delamination of the 0o/45o interface only occurred on the centre stiffener. This was
accompanied with the matrix failure and fibre cracking on the 45o plies. This was not
found on the other stiffeners. Matrix and fibre cracking was found along the middle of
the skin-side of the panel, corresponding to where the middle stiffener had cracked.
Most of the damage on the panel at collapse was from skin-stiffener disbonding. The
load-shortening curves for P30 are shown in Figure 4.17 and these include the readings
taken during regular intervals of cyclic loading.
(a) 0.40 mm (b) 1.01 mm (c) 1.50 mm (d) 2.00 mm (e) 2.85 mm
Figure 4.16: Out of plane ARAMIS displacement results for P30.
The cyclic loading for P30 was conducted slightly differently to P29, as the sequence
for the stiffener disbonding was different. No degradation was observed during the first
2000 cycles when the panel had been loaded up to 1.0 mm displacement, which is just
beyond the first global buckling.
117 | P a g e
Cycles 2001 to 2401 appear to have minimal change in load carrying capability. At
Cycle 2601, there is a significant drop and this is caused by the disbonding across the
stiffener (refer to the results from thermography in Figure 4.18 to see the progress of
degradation). From the thermographic results, it can be seen that disbonds caused by the
local buckling appear on the centre-right stiffener (as viewed from the skin-side) and
later, on the centre-left stiffener. The degradation progresses to affect the outer
stiffeners (see Figure 4.18). The centre stiffener remains largely unaffected with the
exception of the disbond that occurs just below the resin potting. This is probably due to
poor adhesive quality in the bond. The data from the cyclic loading has been collated
and presented in Table 4.8. It can be seen that at 2.0 mm applied displacement, the
maximum loss in load carrying capability is 13.1%. The italicised values show the full
compression and displacement that was applied to the panels.
120
Cycle 2001
Cycle 2101
100 Cycle 2301
Cycle 2401
80 Cycle 2701
Cycle 2901
Load (kN)
60 Cycle 3101
Cycle 3301
Cycle 3501
40
Cycle 3701
Cycle 3901
20 Cycle 4001
Cycle 4101
0 Cycle 4201
0 0.5 1 1.5 2 2.5 3 3.5 Collapse
Figure 4.17: Load-shortening curves for cyclic loading of panel P30.
118 | P a g e
(a) 2100 (b) 2200 (c) 2400 (d) 3700 (e) 4100
(f) Collapse
cycles cycles cycles cycles cycles

(mm) (%)
2001 107.59 2.00 N/A
2101 107.37 2.00 -0.20
2301 106.36 2.00 -1.14
2401 106.04 2.00 -1.44
2601 99.64 2.00 -7.39
2701 99.24 2.00 -7.76
2901 99.01 2.00 -7.97
3101 98.56 2.00 -8.39
3301 98.38 2.00 -8.56
3501 98.00 2.00 -8.91
3701 97.90 2.00 -9.01
3901 97.59 2.00 -9.29
4001 97.42 2.00 -9.45
4002 103.18 2.30 N/A
4101 95.20 2.00 -11.5
4101 100.90 2.30 N/A
4102 95.04 2.00 -11.7
4102 103.97 2.50 N/A
4201 93.94 2.00 -12.7
4201 102.70 2.50 N/A
4202 93.45 2.00 -13.1
4202 102.15 2.50 N/A
119 | P a g e
Panel P31 had a similar loading profile to panel P30, with an applied axial compression
of 1.97 mm. The difference between the two panels can be seen in the way the stiffeners
disbond from the skin. Both panels produced symmetrical postbuckling modes up to
collapse and should, therefore, also have a similar stiffener disbond sequence that
occurs symmetrically about the centre stiffener. This can be seen in the thermographic
plots in Figure 4.19. At Cycle 2401, there is a 6.2% decrease in load carrying capability
and this is due to the large disbond which occurs on the centre-right stiffener. From
Cycle 2401 onward, the load-carrying capability gradually decreased due to other
disbonds appearing on the panel. The next drop occurs at Cycle 4002 when the applied
displacement is increased to 2.47 mm. The decrease is caused by the large disbond that
appears on the centre stiffener. An observation from the thermography results suggests
that the stiffeners that are on the immediate sides of the centre stiffener disbond first,
followed by the outer stiffeners. This is due to the bond having to carry higher shear
loads across from the skin to the stiffener. Table 4.9 shows the data collated from the
cyclic loading with the italicised numbers showing the full compression and
displacement that was applied.

(mm) (%)
2001 108.33 1.97 N/A
2201 102.69 1.96 -5.20
2401 101.62 1.97 -6.19
2601 101.44 1.97 -6.36
2801 101.28 1.98 -6.50
3001 101.18 1.97 -6.60
3201 100.97 1.97 -6.79
3401 100.70 1.97 -7.05
3601 99.77 1.97 -7.90
3801 99.58 1.98 -8.08
4001 99.23 1.97 -8.40
4002 99.16 1.97 -8.47
4002 108.44 2.48 N/A
120 | P a g e
4201 95.82 1.97 -11.5

4201 105.55 2.48 N/A
4202 95.70 1.97 -11.7
(a) 2400 (b) 2800 (c) 3600 (d) 4000 (e) 4200
(f) Collapse
4.4.3 Panels P35 and P37 – Pre-damaged Panels
Results for Panel P35
When P35 was axially compressed, it resulted in a symmetric postbuckling mode shape
at collapse, as shown in Figure 4.20. The figure shows the panel on the stiffener side
and the displacement below each ARAMIS plot is the axial displacement applied. The
corresponding load-shortening curves are presented in Figure 4.21, which include
readings taken at regular intervals during the cyclic loading. Local buckling in the
central skin bays occurs early on the loading due to the existence of the Teflon strip in
the middle stiffener. The first global buckling occurs at 0.81 mm displacement, at a load
of 69 kN. The plots in Figure 4.20(b) to (e) show the progress of the skin buckling, as
the panel proceeds through the postbuckling region. The onset of collapse and the
maximum collapse load coincide at an applied displacement of 2.28 mm, with a
corresponding load of 99 kN. The sharp drops in stiffness after that point were caused
by the merging of the disbonded regions on each stiffener, and the collapse of the panel
was in the form of the skin forming a plastic hinge across the transverse axis, where the
stiffeners had disbonded. When the panel was inspected after the experiment, it was
observed that ply delamination of the 0o/45o interface only occurred on the centre
121 | P a g e
stiffener. This was accompanied with the matrix failure and fibre cracking on the 45 o
plies. This was not found on the other stiffeners. Matrix and fibre cracking was found
along the middle the skin-side of the panel corresponding to where the middle stiffener
had cracked. Most of the damage on the panel at collapse was from skin-stiffener
disbonding.
Figure 4.20: Out of plane ARAMIS displacement results for panel P35.
120
Cycle 2001
100
Cycle 2401
80 Cycle 2801
60 Cycle 3201
40 Cycle 3601
Cycle 4001
20
Collapse
0
0 0.5 1 1.5 2 2.5 3
Figure 4.21: Load-shortening curves for the cyclic loading of panel P35.
No degradation was observed in P35 during the first 2000 cycles when the panel had
been loaded up to 1.0 mm displacement, which is just beyond the first global buckling.
The readings taken from thermography can be found in Figure 4.22. The load-carrying
capability of the panel decreased as the applied displacement was increased after Cycle
122 | P a g e
2001. The centre-right stiffener disbonded due to the cyclic loading at Cycle 2401 and it
was symmetrically repeated on the centre-left stiffener by Cycle 2801. The growth of
the disbonds was mainly concentrated about the centre-left and right stiffeners, with
some growth on the outer stiffeners as the cyclic loading progressed.
The centre stiffener remained largely undamaged despite the placement of the Teflon
strip along the middle length. It was initially thought that the pre-damage would cause
disbonds to initiate from the crack fronts of the damage, but this did not occur. It is
possible that the initial simulations had different properties from the experiment in the
form of fracture properties at the crack front. The data from the cyclic loading has been
collated and presented in Table 4.10. It can be seen that if reference is made to the 2.0
mm applied displacement, the maximum loss in load carrying is 5.04%. The italicised
values show the full compression and displacement that was applied on to the panels.
This is not a significant change compared to the decrease experienced by the
undamaged panels.

(mm) (%)
2401 96.19 1.80 N/A
2801 95.13 1.80 -1.11
3201 94.71 1.80 -1.54
3201 98.11 2.00 N/A
3601 92.72 1.80 -3.61
3601 95.88 2.00 N/A
4001 91.67 1.80 -4.70
4001 94.91 1.99 N/A
4002 91.34 1.80 -5.04
123 | P a g e
(a) 2000 (b) 2400 (c) 2800 (d) 3200 (e) 4000
(f) Collapse
Results for Panel P37
The effect of the Teflon strip in P37 is evident in Figure 4.23(a) as the centre-right
stiffener begins to separate, as the panel is compressed. The buckling in the panel was
slightly asymmetric due to the initial disbond and this can be seen in Figures 4.23(b) to
(e). The local buckling first appears on the far right skin bay at a compressive
displacement of 0.45 mm and a loading of 37 kN followed by the far left skin bay at
0.70 mm compression and a loading of 51 kN. The far left skin bay of the panel
transitions to the postbuckling state but this is not the case on the far right skin bay,
which remains in the local buckling mode, through to collapse. The corresponding load-
shortening curves in Figure 4.24 reveal that the panel transitions into postbuckling at
about 0.90 mm compression and 60 kN. The onset of collapse occurs at 2.53 mm
compression with a drop in the load-shortening curve, due to growth of the skin-
stiffener disbonds in the centre-right stiffener, as shown in Figure 4.25. Collapse occurs
suddenly at a displacement of 2.81 mm and at a loading of 89.1 kN.
Figure 4.23: Out of plane ARAMIS displacement results for panel P37.
124 | P a g e
100
90 Cycle 2001
80
Cycle 2401
70
60 Cycle 2801
50
Cycle 3201
40
30 Cycle 3601
20
Cycle 4001
10
0 Collapse
0 0.5 1 1.5 2 2.5 3 3.5
Figure 4.24: Load-shortening curves for the cyclic loading of panel P37.
(a) 2000 (b) 2800 (c) 3200 (d) 3600 (e) 4000
(f) Collapse
No degradation was observed in P37 during the first 2000 cycles, when the panel had
been loaded up to 1.0 mm displacement, which is just beyond first global buckling. The
load-carrying capability of the panel decreases as the applied displacement is increased
after Cycle 2001. The centre-right stiffener disbonds due to the cyclic loading at Cycle
2801. All the stiffeners, except the centre and pre-damaged stiffener, experience a
growth of disbonds caused by the cyclic loading.
It was initially thought that the pre-damage would cause disbonds to initiate from the
crack fronts of the damage, but this did not occur. This was similar to the behaviour
seen on P35 where no disbond growth was detected at the crack fronts on either end of
125 | P a g e
the Teflon strip. The data from the cyclic loading has been collated and presented in
Table 4.11. It can be seen that if reference is taken to the 1.8 mm applied displacement,
the maximum loss in load carrying capability is 16.14%. The italicised values show the
full compression and displacement that was applied on to the panels. There is a
significant drop in load-carrying capability of 11.5% when the collapse cycle (Cycle
4017) is compared with Cycle 4001.

(mm) (%)
2401 93.16 1.80 N/A
2801 92.50 1.80 -0.71
3201 92.05 1.81 -1.19
3201 95.73 2.00 N/A
3601 89.89 1.80 -3.51
3601 93.87 2.00 N/A
4001 88.85 1.80 -4.63
4001 92.95 2.00 N/A
4017 78.12 1.80 -16.14
4.5 Discussion of Experimental Results for the D1 Panel
4.5.1 Degradation of panel from cyclic loading
While it is apparent that the postbuckling modes have a direct impact on the level of
energy absorption for each panel, this appears to have no significant effect on the
ultimate collapse load of each panel. At a loading of about 2.0 mm for the undamaged
panels, it can be seen that panels P28 to P31 have a similar loading despite the
difference in load-shortening curves. The results can only be effectively analysed at this
axial displacement level as none of these undamaged panels were statically loaded to
collapse.
126 | P a g e
From comparisons of the ARAMIS and thermographic plots, it can be seen that the
skin-stiffener disbonds correspond with the local skin buckling pattern. The disbonds
occur due to differences in axial stiffness between the skin and the stiffener flange. As
the panel is axially compressed, the skin and stiffener buckle and failure of the bond
occurs where a tensile stress is present. The local skin buckling has a direct effect on the
disbonding and a loss of skin-stiffener bond leads to a larger region of the skin buckling
earlier then predicted, as the stiffener ceases to act as a simple support for local
buckling.
Figure 4.26 shows the loading-cycle plots for the panels subjected to cyclic loading. The
vertical axis shows the load being exerted on the panels while the horizontal axis shows
the number of cycles to which that the panel has been subjected. Data for these points
have been taken from Tables 4.6 to 4.10. It was assumed that the increase in loading
served to accelerate the disbond growth and this can be seen in the sudden drops in load
after 4000 cycles.
The discontinuities seen in each plot are caused by a failure of the bond across the
shoulder of the stiffeners. Under cyclic loading the disbond growth was caused by the
local skin buckling and the discontinuities were accompanied by a sudden failure of the
bond, as observed using thermography. It can be seen that with exceptions to the
discontinuities, there is a general trend where the loads decrease as the number of cycles
increase. This decreasing trend appears to be independent of the postbuckling mode
shape experienced by the panel. The pre-damaged panels appear to also follow this
trend.
Table 4.12 summarises the change in panel loadings, due to the disbond growth from
cyclic loading. The last column in the table shows the loads that the panels experienced
at the shown displacements during the last loading cycle, as a percentage of the collapse
load. It can be seen that for the undamaged panels, P30 experiences the largest drop in
loading with a 13.4% decrease, while for the damaged panels P37 experiences the
largest drop in loading with a 16.1%.
127 | P a g e
Maximum Load at Respective Shortening (kN) 110
105
P29 at 1.90 mm
100
P30 at 2.00 mm
95
P31 at 1.97 mm
90
P35 at 1.80 mm
85
80 P37 at 1.80 mm
75
2000 2500 3000 3500 4000 4500
Number of Cycles
Figure 4.26: Plot of maximum applied load against number of applied cycles.
Table 4.12: Decrease in applied loads due to skin-stiffener disbonds.

Panel Displacement (mm) Decrease in Load (%) % of Collapse Load
P28 1.09 0 81.5
P29 1.90 3.9 93.7
P30 2.0 13.4 91.5
P31 1.98 11.7 87.5
P35 1.80 5.0 92.5
P37 1.80 16.1 87.6
4.5.2 Numerical simulation of degradation
A finite element model for the damaged D1 panel was created using 7020 QUAD4 shell
elements using MSC.Patran, where the initial disbonds were created using separate
elements for the skin and stiffener flanges. The model was solved with MSC.Marc using
a progressive failure algorithm developed by Orifici (2007). This algorithm uses an in-
house Virtual Crack Closure Technique (VCCT) subroutine to generate the initial
disbond. This can be seen in Figure 4.27(b) with the disbonded region being
128 | P a g e
highlighted. The FEM uses user-defined Multiple Point Constraints (MPCs) to model
the bonding between the skin and the stiffener and the MPCs fail depending on the
loads set by the user. For more details on the algorithm refer to Orifici (2007). The Tsai-
Wu fibre failure criterion was also used for fibre failure during simulation. For details of
the undamaged finite element D1 model, please refer to Chapter 5, where the initial
benchmarks for the stochastic analysis and robustness measures are carried out.
Figure 4.27(a) shows the level of disbond growth observed using thermography. The
majority of the damage has occurred in the second and fourth stiffener. The regions of
damage were measured and it was found that the damage extended 220 mm along the
length of the stiffener bond and was located 280 mm below the top edge of the panel.
280 mm
220 mm
(a) Results from thermography (b) FEM of panel with initial disbond
Figure 4.27: Comparison of disbonds from experiment and initial disbonds applied in
FEM.
Figure 4.28 shows the finite element displacement plots, as seen from the stiffener-side.
It can be seen that bifurcation buckling occurs earlier in the damaged panel. The
buckling occurs locally between the skin bays where there are disbonds, as can be seen
in Figures 4.28(f) and (g).
129 | P a g e
Undamaged panel FE result
(f) 0.40 mm (g) 0.50 mm (h) 1.00 mm (i) 1.50 mm (j) 2.00 mm
Damaged Panel FE result

Figure 4.28: Out of plane displacement results from finite elements for undamaged and
damaged D1 panels.
Figure 4.29 shows the corresponding load shortening plots for both the experiment and
the simulation using finite elements. From the experiments it can be seen that panel P30
experienced a 13.4% decrease in load-carrying capability due to the growth of disbonds.
In the FEM a 10.1% decrease was experienced and the load shortening plots are in good
agreement with the experiment. In the experiment, no fibre failure was detected at the
cyclic loading displacement of 2.0 mm and this was similar to the results from finite
elements. In the FEM, no further disbond growth was detected as the model was
compressed to 2.0 mm.
130 | P a g e
120
No Degradation
(P30)
100
With
80 Degradation
(P30)
60 No Degradation
(FE)
40
With
20 Degradation (FE)
0
0 0.5 1 1.5 2
Figure 4.29: Comparison of load-shortening plots for degradation between experiment

and FEM.
4.6 Results from Cyclic Loading of the D2 panel
In this section, the results for two of the D2 panels, P24 and P27, are presented. P24 was
the benchmark panel for the series of experiments involving the D2 design, where an
undamaged panel was statically collapsed. On the other hand, P27 was pre-damaged
with 20 J at the skin-side locations shown in Figure 4.8 and cyclically loaded to
collapse. The D2 design is relatively stable in the postbuckling region hence, the
benchmark result from P24 can be directly compared with the results from the final
collapse load of P27. The results of this section provide a basis for the discussion in
Chapter 6 when the robust design and damage tolerance philosophies are compared
using the D2 panel design.
4.6.1 Impact damage on the P27 D2 panel
Figure 4.30 contains images taken after the panel was impact damaged. This includes
photographs taken of the surface damage, from both the skin and stiffener side of the
panel, as well as ultrasonic scans from both the skin and stiffener side of the panel,
131 | P a g e
showing the extent of damage caused by the 20 J impacts. It can be seen in Figure
4.30(b) that the impact damage has resulted in a delamination of the outer plies around
the stiffener shoulder. The damage extended into the skin-stiffener interface, which
caused a large disbond as seen in the ultrasound images.
(a) Damage seen from skin-side (b) Damage seen from stiffener-side
(c) Ultrasound image from skin-side (d) Ultrasound image from stiffener-side
Figure 4.30: Impact damage for P27 showing; (a) damage on skin-side, (b) damage on
stiffener-side, (c) ultrasound image on skin-side and (d) ultrasound image on stiffener-
side.
4.6.2 Results from experiments
Figure 4.31 shows the results of the out of plane displacements for P24 and P27 that
were observed using the ARAMIS system and Figure 4.32 shows the corresponding
load-shortening plot for the panel, from rest to collapse, for P24 and the final collapse
cycle for P27.
132 | P a g e
Panel P24
(g) 0.29 mm (h) 0.35 mm (i) 0.59 mm (j) 1.35 mm (k) 1.69 mm (l) 1.80 mm
Panel P27
Figure 4.31: Out of plane displacement results from ARAMIS for P24 and P27.
200
160
120
Panel P24
Panel P27
80
40
0
0 0.5 1 1.5 2 2.5
Figure 4.32: Load-shortening curve for the collapse of panels P24 and P27.
As the P24 was axially compressed, no damage in the fibre, matrix or bond was
detected up to the point of collapse. Bifurcation buckling occurs at 69 kN and 0.48 mm
compression. This was observed as a half sine wave forming on the outer skin bays of
the panel, and a noticeable kink on the load shortening plot. This progresses into one
and a half sine waves on the outer right stiffener. At 0.65 mm, the centre skin bay
begins to buckle and two buckles form at 0.94 mm axial compression. The panel
133 | P a g e
continues to buckle in ‘square’ patterns up to collapse with no large global buckle

developing. Collapse occurred at 1.82 mm compressive displacement but, due to the
dynamic nature of the failure, it was not determined whether the damage was caused by
failure in the fibre or the bond. Comparing the two load-shortening plots in Figure 4.32,
it was found that there was a drop in ultimate collapse load from 175 kN in P24 to 159
kN in P27, a decrease of 9%.
The effect of the impact damage can be seen from the ARAMIS plot in Figure 4.30(g)
as the regions in blue on the left side of the two stiffeners. The cyclic loading of the
panel introduced damage on to the panel, causing the panel to require less energy to
buckle when compared to P24. This can be seen in the ARAMIS plots in Figure 4.30
and as a noticeable drop in the load shortening curves in Figure 4.31. Bifurcation
buckling for P27 occurs at 0.37 mm and 51 kN axial compression, while the ultimate
collapse load is 159 kN with 1.71 mm axial shortening.
Panel P27 was cyclically loaded for 4202 cycles before it was subjected to collapse. The
cyclic loading caused a drop in the load carrying capability and this was identified as
failure in the material. The thermography readings for P27 can be found in Figure 4.33.
Despite the cyclic loading and the damage in the bonding, no disbond growth was
detected using thermography. At 3000 cycles, a failure initiated in the resin rich region
on the top of the right blade stiffener and this resulted in the failure progressing outward
towards the skin by 3400 cycles, as seen in Figure 4.33(d). A similar failure also
initiates in the lower left stiffener.
(a) 2000 (b) 2400 (c) 3000 (d) 3400 (e) 4000
(f) Collapse
134 | P a g e
4.7 Conclusion
This chapter has provided details on the experimental setup as well as the results from
the testing of the D1 and D2 panels. Extensive investigations were also conducted on
the D1 panels, showing the manufacturing defects and imperfections that emerged from
the manufacturing process. This includes the geometric imperfections measured using
the ATOS system and physical measurements taken of the panel geometry. Variations
were found in the Hexcel IM7/8552 unidirectional CFRP during material
characterisation and critical parameters, such as the longitudinal compressive Young’s
Modulus and Strength, were established to have mean values lower than the values
quoted by Hexcel. Manufacturing defects such as the poor bonding found in the FM300
and the missing ply in the P37 panel affect the experimental results to a significant
degree and need to be considered as factors that will result in a lower stiffness and
shorter fatigue life. The thermographic readings were found to be highly effective in
showing the level of damage that was progressing in the secondary bond between the
skin and stiffeners, for both the D1 and D2 panels.
The D1 panels were designed to have a large postbuckling region, so that the
postbuckling degradation and growth of skin-stiffener disbonds in the CFRP panels
could be better observed and understood. From the experiments it could be seen that the
undamaged panels will, in a worst case scenario, experience a 13 % decrease in load
carrying with no fibre failure, with the damage solely in the form of skin-stiffener
disbonding, if loaded at 92 % of the ultimate collapse load. In the pre-damaged P37
panel, it can be seen that the worst case scenario is a 16 % decrease in load, when
loaded at 88 % of ultimate collapse load. There was no degradation or disbond growth
observed when the panel was cyclically loaded at the bifurcation buckling load.
Therefore it can be concluded that the D1 panel, when loaded at 85 % of ultimate
collapse load, should experience a decrease of no more than 15 % in load carrying
capability due to damage growth in the panel. This translates to a knockdown factor of
0.73 from the ultimate collapse load.
The D2 design is much stiffer than the D1 design, given that the free length is shorter
and the stiffener blade is taller. Given the fact that the material stacking sequence is also
135 | P a g e
different, with the D2 design having higher extensional and bending stiffness values in
the composite stiffness matrix, it was expected that collapse would occur at a higher
load, compared to the D1 design. Unlike the D1 design, the D2 design panels did not
experience large global buckles in the postbuckling region before collapse. This resulted
in a panel design that is highly stable and predictable under compressive loading. Using
the statically loaded undamaged P24 panel as a benchmark result and the impact
damaged cyclically loaded P27 panel as a worst-case scenario, there was a 9 % drop in
the ultimate collapse load. In the cyclic loading of P27, no disbond growth was detected
using thermography. However, a decrease in load carried in each cycle was observed
during each progressive cycle and this can be attributed to damage growth in the fibre
and matrix of the composite laminate.
The results presented in this chapter are used as a basis for comparison and validation in
Chapters 5 and 6. Using the stochastic methodology shown in Chapter 4, finite
element models will be created in order to understand the variations seen in
postbuckling modes, as well as the initial geometrical imperfections that were found
using ATOS. The finite element simulations are different from the experiments from the
aspect that the simulated panels are solved statically loaded to collapse as it is not
possible to simulate the cyclic loading. From the experiments, the D1 and D2 panel can
be accurately modelled without failure up to 2.0 mm and 1.8 mm axial compression,
respectively. Application of the axial compression as determined from the experiments
will allow for maximum computational efficiency and allow the various postbuckling
modes to be observed.
In Chapter 6, the stochastic methodology is used to determine the robustness of both

the D1 and D2 panels and various configurations of the D2 panel. The analyses
conducted will also compare the D2 design for both robustness and damage tolerance.
136 | P a g e
Chapter 5 Stochastic Analyses of the COCOMAT D1 Panel
Chapter 5
Stochastic Analysis of the

COCOMAT D1 Panel
This chapter presents the application of the stochastic analysis procedure developed in
Chapter 3 to predict the buckling patterns encountered in the experiments. Two
experimental patterns were presented in Chapter 4, one symmetric and one
asymmetric. One of the factors that has a considerable effect on the buckling and
collapse loads is the nominal radii of each stiffened panel. This was measured using the
ATOS system and an attempt is made to correlate the initial geometry in the
experimental D1 panels with thermal analyses conducted using finite elements. Residual
stresses in the panel are the cause of the deformations and the stochastic methodology is
used to explain which parameters are affecting the final geometry, during the curing
process.
As highlighted in Chapter 1, none of the initial finite element results for the D1 panel
matched the experimental results for the benchmark panel, P23. This is an area of
concern in design as finite element models require the ability to accurately simulate the
possible responses, so that any unexpected behaviour can be anticipated ahead of time.
Therefore, imperfections in material properties and boundary conditions are introduced
to the D1 finite element models in this chapter to demonstrate that the postbuckling
modes observed in the experiments can be accurately captured using the stochastic
methodology and numerical simulations. The factors causing the differences in the
postbuckling modes are discussed and recommendations are made, so that the lack of
137 | P a g e
stability in compressive loading as observed in the various postbuckling mode shapes

can be avoided.
5.1 Investigation of Geometrical Imperfections in the D1 Panel
One of the obvious disparities between the manufactured panel and the panels analysed
using finite elements, is the difference in the initial geometry. The panels which are
manufactured have been subjected to residual stresses, caused by the curing process,
while those in the finite element environment are perfect, with the exception of minor
geometrical variation caused by numerical rounding in the pre-processor. The curing
process has resulted in the panels taking on varying nominal radii of curvature, thereby
affecting the buckling behaviour and final collapse load.
The curing process will be modelled in this section to determine the variation in the
geometry that is predicted to occur. The numerical rounding in the MSC.Patran pre-
processor will not be investigated further. It led to differences in the position of the
nodes in the third or fourth decimal place for nodes that should have been
symmetrically placed about a symmetry plane. This was sufficient to trigger asymmetric
buckling modes in the geometric nonlinear algorithms in MSC.Nastran – but the
imperfections are negligible compared to the geometry variations that occurred due to
the cure process.
5.1.1 Residual stress in a composite laminate from the curing process
Initial geometry is a significant factor affecting the buckling and postbuckling of

stiffened composite panels. It is not possible to manufacture a perfect structure as there
will always be geometric imperfections. However the order of magnitude of these
imperfections can be influenced by design and manufacturing processes. In the
COCOMAT project the panels have curved geometry and unsymmetrical laminates for
the stiffener flange which add to the complexity for manufacture. As such, the aim from
a manufacturing point of view would be to investigate the effect of the imperfections to
see if they significantly alter the loading characteristics of the panels.
138 | P a g e
In order to understand the effect of residual stress, finite element models of the stiffened
panels, with geometry and meshing similar to the model shown in the previous section,
were created and were subject to a curing simulation. The data for the coefficient of
thermal expansion (CTE) was taken from Kulkarni and Ochoa (2006). The longitudinal
CTE was set to -0.4×10-6/oC and the transverse CTE was 5.6×10-6/oC. The finite
element panel was subjected to an initial temperature of 177oC, as per the Hexcel (2005)
data sheet, and cooled to room temperature (25oC) using the MSC.Nastran nonlinear
solver. The panels were numerically modelled using the variability and imperfections
found during material characterisation. A total of 40 panels were created in finite
elements with variations introduced into the laminate, using the methodology as
described in Chapter 3. As this was a thermal analysis, care had to be taken to ensure
that the fixity of the model would not affect the deformations resulting from the residual
stresses. The material properties and stochastic range used for this analysis can be seen
in Table 5.1.
Table 5.1: Stochastic boundary for the D1 FE stochastic analysis.

Young’s Modulus, E11 (MPa) 142000 135000 - 150000
Poisson’s Ratio, υ 0.277 0.237 - 0.317
Shear Modulus, G12 (MPa) 5130 3014 - 7 240
Shear Modulus, G23 (MPa) 4000 3400 - 4600
0o Lamina orientation (deg) 0 -3.38 - 3.38
45o Lamina orientation (deg) 45 41.6 - 48.4
-45o Lamina orientation (deg) - 45 - 41.6 - - 48.4
o
90 Lamina orientation (deg) 90 86.6 - 93.4
Lamina thickness (mm) 0.125 0.116 - 0.134
The panels were modelled using 5460 QUAD4 shell elements using the MSC.Patran
pre-processor. In order to achieve a minimum computational time for solving each
model, the skins and stiffener flanges were modelled as single laminates. Additionally,
139 | P a g e
offsets were applied in order to obtain panels that were geometrically accurate, having
stiffness coefficients that were representative of the physical panel. Geometrical
properties for the panel can be found in Table 4.2. The boundary conditions for the
panel can be seen in Figure 5.1.
Translations: <x, y, z > Clamped condition middle of

Rotations: < θx, θy, θz > skin shell : <0, 0, 0 >
< 0, 0, 0 >
Figure 5.1: Finite element model for thermal analysis of the D1 panel.
5.1.2 Results from thermal finite element analyses
The results of the stochastic analyses can be seen in Figure 5.2 where the actual
displacements from the manufactured panels are compared with those obtained from
stochastic analysis and finite elements. The measurements for the manufactured panels
were achieved using the ATOS system described in Chapter 3.
140 | P a g e
0.5mm 0.5mm 0.5mm
- 0.5mm - 0.5mm
- 0.5mm
Sample displacements from ATOS System
0.775mm 0.535mm 0.852mm
-0.151mm 0mm
-1. 58mm
Sample displacements from thermal finite element analysis

Figure 5.2: Comparison of initial geometrical imperfections in panels due to residual
stresses from curing process.
The stochastic analysis indicates that the amount of deformation was significantly
larger, once variation was applied. The nominal panel had a net deformation of 0.57
mm, while the net mean deformation obtained from the analyses was 1.95 mm. Table
5.2 shows the results of the stochastic analysis. Included in the table are results showing
the effect of the skin input parameters. Input parameters relating to the blade were not
included as these were insignificant with regard to affecting the curing deformation. It
was found that the first two skin plies were significant in affecting the curing
deformation; the Influence was 0.516 and 0.458, respectively. Due to the CLPT
formulation the orientation and thickness for the laminas are used to calculate the
laminate stiffness. The basis for these two plies highly influencing the curing
deformation is due to their positioning as the plies furthest from the neutral axis. The
R.I.I for the panel (highlighted as bold in the table), 0.162, was taken from the
Orientation of Ply 1.
141 | P a g e
Table 5.2: Results of stochastic analysis for thermal analyses of the D1 panel.
Input Variables Influence Sensitivity Scaled Sensitivity
Material Properties
Young’s Modulus, E11 0.0735 0.00000656 0.974
Young’s Modulus, E22 0.191 0.000160 1.87
Poisson’s Ratio, υ -0.106 -7.51 -2.54
Shear Modulus, G12 -0.0850 -0.000418 -2.50
Shear Modulus, G23 0.0994 -0.000280 -1.14
Shear Modulus, G31 -0.251 -0.00000874 -0.0524
Skin lamina
Thickness -0.150 -2.90 -0.360
Ply1
Orientation 0.519 0.132 11.9
Thickness 0.486 58.7 7.33
Ply 2
Orientation -0.0913 -0.0199 0.891
Thickness -0.0650 58.7 7.33
Ply3
Orientation -0.0605 -0.0664 -3.01
Thickness -0.145 -19.2 -2.40
Ply4
Orientation 0.105 0.0300 0.00861
Thickness -0.0813 -23.1 -2.88
Ply5
Orientation -0.143 -0.0587 0.00903
Thickness -0.240 9.76 1.21
Ply6
Orientation -0.136 -0.0321 -1.45
Thickness -0.0713 -26.3 -3.31
Ply7
Orientation -0.238 -0.0722 3.27
Thickness -0.267 -12.0 -1.50
Ply8
Orientation -0.130 -0.0217 -1.97
Metamodels of the deformation are plotted against the stiffness of the first two skin
plies in Figures 5.3 and 5.4.
142 | P a g e
4.00
3.50
3.00
Deformation (mm)
2.50
2.00
1.50
1.00
0.50
0.00
86 88 90 92 94 96
Skin Ply 1 Orientation (deg)
Figure 5.3: Metamodel of net skin deformation against Ply 1 Orientation.
4.00
3.50
3.00
Deformation (mm)
2.50
2.00
1.50
1.00
0.50
0.00
0.11 0.115 0.12 0.125 0.13 0.135
Skin Ply 2 Thickness (mm)
Figure 5.4: Metamodel of net skin deformation against Ply 2 Thickness.
143 | P a g e
In both metamodels in Figures 5.3 and 5.4 the stiffest solution is the original panel with
the ply orientation of 90o and ply thickness of 0.125mm. The entries for these properties
appear as outliers in the data set as any variation of orientation or thickness increases
the deformation. The trend line also indicates higher values for these defined values.
It can be seen from the metamodels in Figures 5.3 and 5.4 that it is possible to reduce
the magnitude of the deformation, due to the curing process in the panels, by controlling
the quality of plies 1 and 2. Given that shell elements are used in the finite element
model, the stiffness components of the element are derived from CLPT, as described in
Chapter 2. Hence, each lamina orientation and thickness parameter contributes towards
the stiffness of each lamina. The positive gradient on the Sensitivity line suggests that
the amount of deformation from the curing increases proportionally with the stiffness of
the two laminae. This deformation is unavoidable and, therefore, stringent quality
control on the material quality during the layup process is required if less deformation
and scatter is sought.
5.1.3 Effect of geometrical imperfections on collapse of the D1 panel
This is an investigation into the effects of initial geometrical imperfections for the D1
panels. In order to accurately model the collapse of the D1 panel, the following
boundary conditions were used on the finite element model as shown in Figure 5.5.
144 | P a g e
Clamped edge: < 0, 0, 0 > Potting region: < , 0, 0 >

< 0, 0, 0 > < 0, 0, 0 >
Translations: < x, y, z > Loaded edge: < -x, 0, 0 >

Rotations: < θx, θy, θz > < 0, 0, 0 >
Figure 5.5: Boundary conditions for D1 finite element panel.
Finite element models were solved with MSC.Marc using a user-defined subroutine
created by Orifici et al. (2007) as part of the COCOMAT project. The subroutine
modified the model by offsetting the skin and stiffener and including user-defined
multiple-point constraints (MPCs) in the gap to model the disbond growth seen in the
experiment. Skin-stiffener disbonds were created as pre-damage by setting the MPCs to
the appropriate states as shown using the red regions in Figure 5.6. The pre-damaged
disbonded regions were taken from the thermographic scans of the damage for P29, and
were adapted to the regular grid mesh of the model, to match the area and shape of the
experimental damage sites. It was assumed that the cyclic loading only resulted in skin-
stiffener debonding, and any other damage, such as matrix cracking, that could have
been present in the panel prior to static loading, was not considered.
In the interlaminar damage growth model by Orifici (2007) pre-existing interlaminar

damage in the skin-stiffener interface was represented as a debonded region between the
skin and stiffener and the shell layers were connected with user-defined MPCs. The
user-defined MPCs were given one of three “states”, which were used to define the
intact (state 0), crack front (state 1) and debonded (state 2) regions. Gap elements were
145 | P a g e
used in any debonded region to prevent crossover of the two sublaminates. To model
the correct bending behaviour the shell layers were separated by a nominal distance and
the respective laminates offset using dummy plies.
At the end of every nonlinear analysis increment, the Virtual Crack Closure Technique
(VCCT) (Rybicki and Kanninen 1977) was used to determine the strain energy release
rates at all MPCs on the crack front. The VCCT equations accounted for arbitrary
element sizes, and an algorithm was written to determine the local crack front
coordinate system from the neighbouring crack front nodes, following
recommendations given by Krueger (2004). The onset of propagation was determined
using the B-K criterion (Benzeggagh and Kenane 1996), with modification for the
inclusion of the mode III component following the suggestion given by Camanho and
Dávila (2003), given by
GI GII GIII

1

GI C GII C GI C GII GIII

(5.1)
GI GII GIII
where G are the strain energy release rates in modes I, II and III,
GC are fracture toughness values and
is a curve fit parameter found from mixed-mode test data.
For crack propagation, an iterative method was applied that reduced the strain energy
release rate values based on the shape of the local crack front at each MPC. This was
developed as it was found that the local crack front affected the estimation of crack
opening displacement, which in VCCT is based on self-similar crack growth.
Modification factors were determined to account for the difference in crack opening
between the actual crack propagation and the assumed self-similar case. Further
information on this approach has been presented in publications such as Kelly et al.
(2009) and Orifici et al. (2009). In the analysis applied in this work, the interlaminar
damage is modelled at the skin-stiffener interface.
146 | P a g e
Figure 5.6 shows the disbond damage experienced by the panel and the finite element
model used to simulate the failure. The model consisted of 6004 nodes and 5772 shell
elements.
potting region
disbonded region
stiffener
potting region
(a) Thermographic reading (b) Finite element model with

from P29 disbond damage
Figure 5.6: Damage in skin-stiffener bond and the finite element model used to simulate
the failure.
The material and laminate properties used are the nominal values found in Table 5.4.
The initial geometrical imperfections were introduced by offsetting the nodes using the
curing deformation from the left finite element result shown in Figure 5.2. This ensured
that the maximum extent of realism was captured in the analysis, where the nominal
properties were applied to the panel before it was cured, and the resulting deformation
used as the initial geometry for the collapse analysis. Fracture properties for the model
are given in Table 5.3.
Table 5.3: Fracture properties for IM7/8552 carbon/epoxy unidirectional tape .

Fracture property Value
GI c [kJ/m2] 0.243
GII c [kJ/m2] 0.514
GIII c* [kJ/m2] 0.514
B-K coefficient, * 4.6
assumed values *
147 | P a g e
The load error convergence criteria for the analyses in MSC.Marc were set to 1×10-3
and quasi-static inertial damping was applied and the value was set to 1×10-6. The
resulting out of plane displacements for the analyses can be seen in Figure 5.7 and, the
corresponding load shortening curves can be found in Figure 5.8. The colour system
used to represent the out of plane displacements between the ARAMIS and FE results
have been reversed hence the blue regions in Figure 5.7(a) points out of the page while
the blue regions for the FE plots point into the page. It can be seen that the plots are in
good agreement up to local buckling, at about 0.5 mm shortening. The curves appear to
diverge after the local buckling stage and both finite element models have a higher
collapse load, compared to the benchmark experimental test. The geometrical
imperfections have resulted in the finite element models with the cure deformations
having a higher collapse load, compared to the finite element model with no geometrical
imperfections. The imperfections appear to increase the amount of energy required to
collapse the panel and this is a significant factor in accurately simulating the collapse of
the D1 panel.
(a) Panel P29 ARAMIS results
(g) 0.69 mm (h) 0.92 mm (i) 1.32 mm (j) 2.00 mm (k) 2.52 mm (l) 2.92 mm
(b) Damaged D1 panel using finite elements (no geometrical imperfections)
Figure 5.7: Out of plane displacement plots from ARAMIS and FEA.
148 | P a g e
120
FE result (no
100 geometrical
imperfections)
80 FE result (with
geometrical
imperfections)
60
Panel P28
40
20 Panel P29
0
0.00 1.00 2.00 3.00 4.00
Figure 5.8: Load shortening plots for stiffened panels compressed to collapse.
5.1.4 Conclusion
In this section, possible variations and defects have been used to investigate the initial
geometrical imperfections found in the undamaged D1 panels, using the ATOS system.
From the thermal analyses, it was found that the various deformation patterns observed
could be caused by variations in the individual plies. The outermost ply on the top of the
skin laminate was identified as the key parameter affecting the robustness of the D1
panel, under this thermal loading. It is acknowledged that while the finite element
analyses provide a qualitative match for the imperfections, other sources of
imperfections such as the layup process and the accuracy of the tooling cannot be
excluded.
Application of the progressive damage algorithm, developed by Orifici, allowed the
effects of curing deformations to be investigated. In this section, two panels, one with
geometrical offsets taken from a curing deformation analysis, were analysed and it was
clear that the geometrical imperfections positively affected the panel, in the deep
postbuckling region, by increasing the energy required to reach the collapse load.
149 | P a g e
5.2 Stochastic Analysis of D1 panel
In this section a stochastic analysis is conducted on the D1 panel. A range of thin

stiffened shell panels were manufactured and tested in the COCOMAT project as show
in Chapter 4. When the results of the tests were reviewed it was found that a number of
different buckling shapes were encountered in the experiments. In some cases it was not
possible to match the shapes in finite element analyses. The stochastic analysis
undertaken in this section is to identify all the buckling modes and understand the
source of the variation that has led to the uncertainty in the results.
5.2.1 Pre-processing of D1 panel for analysis
The panels were modelled using 5460 QUAD4 shell elements using the MSC.Patran
pre-processor. In order to achieve a minimum computational time for solving each
model, the skins and stiffener flanges were modelled as single laminates. Additionally,
offsets were applied in order to obtain panels that were geometrically accurate, having
stiffness coefficients that were representative of the physical panel. No failure criteria
were included in the models and, therefore, any sharp drops in the deep postbuckling
region on the load shortening plots were a result of a change in buckling mode. The
panel was then analysed using the MSC.Nastran nonlinear solver. Geometrical
properties for the panel can be found in Table 4.2.
The boundary conditions for the finite element model had to reflect the possible loading
and potting conditions experienced during the experiment. There was a strong
possibility that uneven loading in the through-thickness direction of the panel could
contribute to a variation in buckling responses and, therefore, this had to be included in
the stochastic analysis. A new node was created at the centre of curvature and a RBE2
multiple point constraint (MPC) was created to link all the nodes on the shortened edge
to the new node. This was done so that linear loading could be introduced on the panel,
via a displacement and rotation in the global axes at the new node. The rotation has
been defined so that the tip of the stiffener blade has an axial displacement range of
±0.05 mm, when compared to the skin. This through-thickness variation leads to non-
150 | P a g e
symmetric deformation due to the ply orientation in the laminate. Figure 5.9 shows an
isometric view of the finite element model, with the new node and MPC.
New node and MPC
Figure 5.9: Finite element model with created node and MPC.
A stochastic analysis was applied to a fuselage-representative multi-stiffener panel

design from the COCOMAT project. Boundary conditions similar to those shown in
Figure 5.5 and material properties within the stochastic boundary shown in Table 5.1,
were used in the analysis. The finite element models for the postbuckling analyses were
solved using MSC.Marc due to convergence problems experienced in the deep
postbuckling region for some of the postbuckling modes when using MSC.Nastran.
Each MSC.Marc finite element analysis for the D1 panel required an average
computational time of forty minutes to solve on an Intel® Core™2 Quad processor.
5.2.2 Results of stochastic analysis of the D1 panel
Table 5.4 shows the results obtained from the stochastic analysis, where the influence of
the input variables on stiffness parameters A11 for the skin and blade stiffeners is
monitored, and the effect on the maximum compressive load experienced by each panel
is determined.
151 | P a g e
From the Influence and Scaled Sensitivity parameters, it can be seen that the longitudinal
Young’s Modulus (taken from the fibre direction on each lamina) and the laminate
stiffness components A11, for both the skin and the stiffener blades, are most affected by
the imperfections introduced to the panels. The result of this analysis only takes the
compression load into account and not the postbuckling modes. From previous
experience in the COCOMAT project, this maximum compression load is dependent on
the postbuckling mode and this can affect the accuracy of the stochastic analysis
conducted. However, as this is a benchmarking study, the aim is to reasonably obtain
the maximum amount of data and variations in postbuckling responses.
Table 5.4: Results from stochastic analysis of D1 panel.

Property Influence Sensitivity Scaled Sensitivity
Material E11 0.337 6.25 × 10-4 89.3
E22 0.250 1.80 × 10-3 17.5
υ -0.0921 -1.23 × 101 -3.42
G12 0.498 3.22 × 10-3 16.4
-3
G23 0.0713 2.49 × 10 10.1
G31 0.498 3.22 × 10-3 16.4
Skin A11 0.314 5.63 × 10-4 34.1
B11 -0.172 -2.25 × 10-3 0.194
D11 0.0779 2.92 × 10-3 6.33
Blade A11 0.497 2.20 × 10-4 62.3
-4
B11 -0.110 -1.07 × 10 0.0245
D11 0.434 2.94 × 10-4 37.9
Angular Rotation Z 0.171 1.00 × 103 0.0251
Figures 5.10 and 5.11 show the metamodels of the key input variables affecting the
failure load; the Longitudinal Young’s Modulus and Blade stiffness component A 11,
respectively.
152 | P a g e
140
135
130
125
120
115
110
136000 138000 140000 142000 144000 146000 148000 150000 152000
Longitudinal Young's Modulus E11 (MPa)
Figure 5.10: Metamodel for maximum compression load against longitudinal Young’s
Modulus E11.
140
135
130
125
120
115
110
260000 265000 270000 275000 280000 285000 290000 295000 300000 305000
Blade Extentional Stiffness A11 (MPa)
Figure 5.11: Metamodel for maximum compression load against Blade extensional
stiffness A11.
153 | P a g e
From the stochastic analysis, five samples, representative of the results obtained, are
shown in Figures 5.12 and 5.13. As no failure criteria have been set for the finite
element analyses, the drops and changes in the plots are a function of changes in
geometry, as each panel progress through the various buckling modes. From the plots it
can be seen that the panels generally have a similar maximum loading capability but
this level of loading occurs at various compressive axial displacements. The area under
the load shortening curve represents the energy that each panel is able to withstand, as a
compressive displacement is applied.
140
120
100
Sample 2
80
Sample 8
60
Sample 32
40
20
0
0 0.5 1 1.5 2 2.5 3
Displacement (mm)
Figure 5.12: Sample load shortening curves for Samples 2, 8 and 32 obtained from
stochastic analysis using finite elements.
154 | P a g e
140
120
100
Sample 9
80
60 Sample 20
40
20
0
0 0.5 1 1.5 2 2.5 3
Displacement (mm)
Figure 5.13: Sample load shortening curves for Samples 9 and 20 obtained from stochastic
analysis using finite elements.
In Figure 5.14, the out-of-plane displacement plots, corresponding to the load

shortening curves in Figures 5.12 and 5.13, are shown. The plots show the panel from
the stiffener side, where the regions in white are displacements towards the centre of
curvature, while the red regions are in the opposite direction. It can be seen that, in
general, there are three different postbuckling mode shapes in the deep postbuckling
regions, as shown in the final plots for each sample. Also, it can be observed that,
although Samples 2, 8 and 32 have asymmetrically postbuckling mode shapes in the
deep postbuckling regions (at 2.72 mm, 2.24 mm and 2.62 mm compression,
respectively), the various buckling modes that the panels transition through are
different.
155 | P a g e
Sample 2
0.52 mm 0.96 mm 1.28 mm 1.36 mm 1.76 mm 2.72 mm

Sample 8
0.52 mm 0.96 mm 1.40 mm 1.48 mm 2.16 mm 2.24 mm

Sample 9
0.64 mm 0.96 mm 1.28 mm 1.64 mm 1.72 mm 2.80 mm

Sample 20
0.60 mm 0.96 mm 1.48 mm 1.76 mm 2.20 mm 2.88 mm

Sample 32
0.48 mm 0.96 mm 1.32 mm 1.40 mm 1.92 mm 2.72 mm

Figure 5.14: Out of plane displacement plots corresponding to the load shortening curves
in Figures 5.12 and 5.13.
156 | P a g e
Panel P23
0.51 mm 0.97 mm 1.31 mm 1.72 mm 2.72 mm

Panel P29
0.69 mm 0.92 mm 1.32 mm 1.99 mm 2.52 mm

Panel P30
0.40 mm 1.01 mm 1.50 mm 2.00 mm 2.85 mm

Figure 5.15: Out of plane displacement plots of the D1 panel from the DLR benchmark
and COCOMAT experiments.
Figure 5.15 presents the benchmark DLR D1 panel P23 which was tested on a voluntary
basis and two panels that the DLR tested as part of the COCOMAT project. These
panels were known internally as P23, P29 and P30, respectively. These results are in
good agreement with those obtained from the stochastic analysis. Good matches in
buckling and postbuckling modes can be observed between the benchmark panel P23
and FE Sample 32, P29 and FE Sample 20, and between P30 and FE Sample 9.
The introduction of imperfections and variations has caused the skin and blade
laminates in the panel to no longer be symmetrical. Coupled with the introduction of
nonlinear loading conditions, it can be seen that the panel no longer experiences pure
compression; instead there is a small bending component that also contributes to the
variations in buckling modes seen in the stochastic analysis. Figure 5.16 is an extract of
the results, where the postbuckling mode shapes at 3 mm axial compression are plotted
157 | P a g e
against the angular rotation introduced via the MPC. Note the bifurcation that exists in
the results of the plot. Within the range of variations included in the plots, the panel
undertook three postbuckling mode shapes indicating the level of instability that exists
in the design and, thus, its lack of robustness.
Postbuckling Mode Shape
-0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15
Angular Rotation (deg)
Figure 5.16: Plot of postbuckling mode shape at 3 mm compression against applied

rotation.
5.2.3 Coupling of bending and compression loads in D1 panels
As the curved panel is axially compressed, Poisson’s effect causes the panel to also
expand transversely. This expansion can lead to out of plane displacement, as the
constraint imposed by the end conditions resists an increase in the transverse dimension
of the plate. This behaviour is similar to the anticlastic bending of thin plates described
by Timoshenko and Woinowsky-Krieger (1959).
There are two ways in which a moment may be introduced onto the edge of the panel.
The first is a moment induced by a rotation of the RBE2 MPC, as shown in Figure 5.9,
and the second is via an asymmetrical skin lay-up, due to imperfections introduced into
158 | P a g e
the lamina orientations and thicknesses. Figure 5.17 illustrates this behaviour at the
mid-longitudinal plane of the panel. The outer surface of the panel has a lower
compressive force, compared to the inner surface. This difference in force induces a
moment about the mid-plane of the panel. As the moment increases, the panel deforms
into the shape shown in the figure in order to fulfil compatibility, as the inner surface
has a circumference greater than the outer surface.
My
Comp-
Comp+
Mx
Original Geometry
Deformed Geometry
Neutral Axis
Figure 5.17: Anticlastic bending of a curved panel.
In the case of the D1 panels that are anisotropic, the various skin laminae will have
different transverse expansions, due to the different thicknesses and orientations. The
result of this complex interaction between buckling and bending was observed in Figure
5.14. Sample 2 from Figure 5.14 is highlighted in Figure 5.18, where the panel is
observed from the through-thickness axis. As axial compression is applied, the panel
experiences the anticlastic bending up to 1.36 mm compression. The middle stiffener
(which is no longer symmetric, due to the introduction of imperfections in the lay-up)
begins to rotate after 1.36 mm axial compression and the moment created promotes the
asymmetry of the panel displacement.
159 | P a g e
0.96 mm
1.28 mm
1.36 mm
1.76 mm
2.72 mm
Figure 5.18: Rotational displacement of middle stiffener in finite element D1 panel.
5.2.4 Conclusion
The stochastic methodology developed in Chapter 3 has been used, in conjunction with
finite element analyses, to capture the variety of postbuckling responses observed
during the experiments in Chapter 4, for the D1 panel. The purpose of introducing
stochastic input variables was to show how the introduction of imperfections and
variations result in the various postbuckling modes that the D1 panel exhibits.
The prediction of which postbuckling modes the D1 panel will assume will always be
difficult, as it has been shown that the buckling and postbuckling failure is seldom a
function of a single variable; rather, it is always a combination of variables. Therefore,
through the introduction of stochastic variability, finite element analyses can take on
multiple cases for loading, geometrical and material properties. This is useful as it
enables the engineer to move away from curve-fitting, towards a scenario where the
possible results lie within a cloud or region, thereby replicating the stochastic nature of
the physical structure.
160 | P a g e
However, the success of this analysis depends on the pool of variables considered. In
practise, the selection of input variables for an analysis also depends on experience. A
review of literature was conducted on the buckling of panels in Chapter 2. This lead to
the conclusion that the dominant variables controlling the postbuckling modes and loads
were the boundary conditions, in addition to the material and geometrical properties.
This was the motive for including the applied rotation as well as variation in the
material and geometry.
As shown in the previous sections, the D1 panel exhibits various postbuckling modes,
due to imperfections in the material and laminate. In order to accurately interpret the
experimental results in Chapter 4, it would have been helpful if all the undamaged D1
panels had similar postbuckling mode shapes when they collapsed. This would ensure a
sufficient sample size for conclusive design standards to be derived. A redesign of the
composite panel is required if one repeatable failure sequence is to be achieved. This is
attempted Chapter 7 using the Robust Indices that have been developed in Chapter 3.
5.3 Conclusion
The final work package in the COCOMAT project involved the generation of design
standards for the degradation of postbuckling stiffened structures. It can be seen from
Section 5.1 that the initial geometrical imperfections have a considerable effect on the
profile of the load shortening curve. Therefore, in future collapse analyses, it is
important to include these initial geometrical imperfections so that better correlation
with the experimental results can be achieved.
For all analyses conducted in the COCOMAT program, a number of factors

considerably influenced comparison with experimental results. One aspect was the
difficulty in accurately capturing the correct buckling mode shapes and deformation
patterns, which is especially critical for crack growth in the region just ahead of any
crack front. The work by the authors has identified the significant effect of
manufacturing variability and uncertainty in the material properties. Wide variety was
encountered in the experimental postbuckling shapes. The procedure developed here
161 | P a g e
requires a survey of the variation encountered and implementation of FEA to a

stochastically determined family of panels. The work indicated the mode shapes
identified in the experiments could be recognised and categorised leaving no outliers in
the data set available to the authors.
It is also recommended that stochastic analyses be conducted before the experiments are
conducted, in order to assist in the design of the specimen. This is especially the case
for future projects which are similar in scale and context to the COCOMAT project.
The rationale behind this is so that any bifurcation behaviour in the postbuckling region,
that might otherwise occur, can be avoided at the design stage or, even better,
understood as it occurs during the experimental tests.
162 | P a g e
Chapter 6 Definition of Robust Indices for COCOMAT Panel Designs
Chapter 6
Definition of Robust Indices for

COCOMAT Panel Designs
This chapter presents the application of the Robust Indices developed in Chapter 3.
The Robust Indices provide two measures of robustness through the utilisation of the
Influence and Sensitivity parameters which quantitatively describe the metamodels that
have been generated.
Before the COCOMAT panels can be analysed, the initial assumptions for the material
imperfections and variations need to be validated using finite element analyses. It is a
common practice to apply an arbitrary coefficient of variation (CoV), say 5%, in
repeated sampling analyses in order to show the variations that emerge in the results. In
the COCOMAT project, materials characterisation was conducted in an early work
package and this data is used in the stochastic analyses to compare the differences that
may arise in robustness and response between assumed and measured scatter in the
input parameters for the stiffened composite panels. In Section 3.5 a hand calculation
was conducted on the effect of scatter in the input parameters of a simple structure. In
that particular case, it was found that quality had an insignificant effect on the
robustness of the structure. However this might not be the case for thin-walled
structures as such configurations are known to be significantly affected by material
imperfections under compressive loads. In Section 6.1 an investigation will be
conducted to determine whether stochastic analyses can be conducted before actual data
from materials characterisation is available.
163 | P a g e
The two COCOMAT panel designs, D1 and D2, which were described in Chapter 4 are
then analysed in Section 6.2. The scatter that was observed in the experimental results
is considered here and the responses between the two panels are quantified using the
Robust Indices. From the analyses conducted, it is found that the D1 panel has a more
robust skin layup while the D2 panel has a more robust blade layup.
6.1 Validation of Initial Assumptions
The review of literature in Chapter 2 revealed that thin-walled structures are highly
sensitive to imperfections in material properties, boundary conditions and geometry.
Therefore, the aim of this section is to compare the behaviour of the D1 panel when data
from the material characterisation in Chapter 4 is used. Results for an assumed CoV of
5% are also presented. The results will be used to decide which initial assumptions are
valid for use. The D1 P30 experimental panel is used here as its postbuckling mode
shape can be effectively captured in finite elements as seen in the benchmark plots in
Chapter 1.
No conclusive investigation was conducted to discover the actual orientations and

thickness of each lamina. Yurgatis (1987) reported a variation of ± 3 degrees about the
mean fibre direction, with a standard deviation of 0.693 to 1.936 degrees, while Piggott
(1995) tested PEEK specimens which resulted in a standard deviation calculated to be
ranging from about 1 to 6 degrees. The samples used in all the stochastic analyses in
this thesis will assume a standard deviation of 1.25 degrees, which is reasonable in
comparison to the measured values published by both Yurgatis and Piggott. This chosen
value assumes that manufacturing techniques have improved, which results in less
scatter. An arbitrary standard deviation of 2.5 %, or 0.003125 mm, was chosen for the
lamina thickness.
6.1.1 Initial assumptions and data from material characterisation
Table 6.1 shows the statistical scatter that will be used, including data measured from
characterisation testing and the assumed values with a 5% CoV. As the finite element
164 | P a g e
analyses are solved up to a level before any damage occurs, no strength and fracture
properties are required. The table summarises the essential parameters from Table 4.4.
The through-thickness shear modulus G13 is assumed to be similar to the in-plane shear
modulus G12 in each data set. The stochastic analyses will incorporate sufficient reality
into the simulations as each panel consists of multiple laminates containing variations in
the laminate stiffness matrix for the skin and stiffeners.
Table 6.1: Boundaries for stochastic analyses for Hexcel IM7/8552 material.
Characterisation Data Assumed Data
Standard Standard
Mean Mean
Deviation Deviation
E11 (GPa) 142.46 2.41 147.0 7.35
E22 (GPa) 9.75 0.47 11.8 0.59
Stiffness
G12 (GPa) 5.125 0.70 6.0 0.30
G23 (GPa) - - 4.0 0.20
Poisson’s
υ12 0.277 0.04 0.34 0.017
Ratio
Lamina Orientation (deg) - - - 1.25
Properties Thickness (mm) - - 0.125 0.003125
It can be seen that the assumed data values have a generally larger scatter compared to
those found in the material characterisation with the exception of Poisson’s Ratio.
6.1.2 Finite element model for D1 panel
The panels that were modelled for this set of analyses are similar to ones found in
Section 5.1 where the panel is modelled using 5460 QUAD4 shell elements. The skins
and stiffener flanges were modelled as single laminates with offsets to ensure that the
panels were geometrically accurate by having stiffness coefficients that were
representative of the physical panel. Boundary conditions for the panel are show in
Figure 6.1. Again, no failure criteria were included in the models and, therefore, any
sharp drops in the deep postbuckling region on the load shortening plots were a result of
165 | P a g e
a change in buckling mode. The panel was then analysed using the MSC.Nastran
nonlinear solver. Geometrical properties for the panel can be found in Table 4.2.
Potting region: < , 0, 0 >

Clamped edge: < 0, 0, 0 > < 0, 0, 0 >
< 0, 0, 0 >
Translations: < x, y, z > Loaded edge: < -x, 0, 0 >

Rotations: < θx, θy, θz > < 0, 0, 0 >
Figure 6.1: Finite element model of D1 panel.
Gaussian normal inputs were chosen for the input variables. Table 6.2 shows the
deterministic input values and the corresponding stochastic variation used. The values
in the defined range are from -3 to +3 standard deviations about the nominal mean.
Table 6.2: Stochastic boundary for analysis of D1 panel.

Both data sets
o
0 Lamina orientation (deg) 0 -3.38 - 3.38
-45o Lamina orientation (deg) -45 -41.6 - -48.4
166 | P a g e

Measured data
Poisson’s Ratio, 12 0.277 0.237 - 0.317
Shear Modulus, G12 (MPa) 5130 3014 - 7 240
Assumed data
Poisson’s Ratio, 12 0.3 0.255 - 0.345
There was a sample size of 40 panels per data set in this stochastic analysis, conforming
to the requirements set by Equation (3.7). The solver settings can be found in Table 6.3.
Table 6.3: Settings for MSC.Nastran nonlinear solver.

Static Nonlinear Iterations
Matrix update method Controlled Iterations
No. of iterations per update 1
No. of iterations per increment 175
Convergence Criteria
Load error 1×10-3
Work error 1×10-7
6.1.3 Results of analysis
Figure 6.2 shows load shortening curves for three representative D1 panels. These
include the experimental panel (P30) and finite element models with nominal measured
167 | P a g e
and assumed values. The profiles of the curves are in good agreement, while there are
noticeable differences in their load-carrying capability, due to variations in panel
stiffness. Figure 6.3 shows a comparison of the plots from the measured and assumed
variation panels, with their corresponding upper and lower bounds. It can be seen that
the plots with measured variations have a smaller boundary compared to the plots
obtained using assumed variations. This suggests that there has been an overestimation
of the amount of scatter in the values of the assumed data set.
120
100
Compressive Load (kN)
80 Experiment
(P30)
60 Measured
FEM
40
Assumed
FEM
20
0
0 0.5 1 1.5 2
Figure 6.2: Load shortening plots comparing P30 and FEM D1 panels using measured and
assumed mean values.
168 | P a g e
120
Measured
100 (nominal)
Measured
Compressive Load (kN)
80 (lower
bound)
60 Measured
(upper
bound)
40 Assumed
(nominal)
20 Assumed
(lower
0 bound)
0 0.5 1 1.5 2
Figure 6.3: Load shortening plots comparing upper and lower bounds of FEM D1 panels
using measured and assumed values.
Table 6.4 shows the results of the stochastic analyses. The properties are considered
with respect to the load carrying capability of the panels at a 2 mm axial compression.
This level of shortening was chosen as it was less computationally expensive compared
to running the model till collapse. The table highlights R.I.I for each panel in bold. A
total of 40 analyses were completed for each set of stochastic analyses. The results
revealed that the behaviour of the panels was mainly controlled by the longitudinal
Young’s Modulus, skin extensional stiffness A11 and blade extensional stiffness A11.
The R.I.I and R.I.II values of the measured data set were both greater when the
measured values were used. Figures 6.4, 6.5 and 6.6 show the plots taken from the
stochastic analyses, comparing the measured and assumed values. It can be seen that
there is a smaller variance in the measured values compared to the assumed values.
169 | P a g e
Table 6.4: Results of stochastic analyses.

Scaled
Property Influence Sensitivity [R.I.]Xi R.I.II
Sensitivity
Assumed Properties
E11 0.908 0.000609 90.5 0.0122
E22 0.181 0.00109 12.8 0.431
ν 0.108 25.2 8.53 1.09
Material
G12 0.272 0.005 29.8 0.123
G23 -0.091 -0.00255 -10.4 1.06
G31 -0.077 -0.000620 -3.72 3.48
0.0622
A11 0.853 0.00128 82.0 0.0143
Skin B11 -0.067 -0.000997 0.0883 -170
D11 0.557 0.0164 39.3 0.0457
A11 0.936 0.000307 91.1 0.0117
Blade B11 0.204 0.000336 -0.0767 -63.9
D11 0.855 0.000506 69.5 0.0168
Measured Properties
E11 0.631 0.000748 107 0.0148
E22 0.214 0.000893 8.66 0.540
ν -0.036 -1.54 -0.426 64.9
Material
G12 0.560 0.00213 10.8 0.165
G23 -0.053 -0.000538 -2.18 8.71
G31 0.560 0.00213 10.8 0.165
0.0778
A11 0.611 0.000985 59.7 0.0274
Skin B11 -0.090 -0.000726 -1.58 7.09
D11 0.328 0.008797 19.1 0.160
A11 0.877 0.000318 89.9 0.0127
Blade B11 0.093 0.000192 0.0441 244
D11 0.759 0.000377 48.6 0.0271
170 | P a g e
110
105
Assumed
100
Measured
95
Sensivity
90
(Assumed)
85 Sensitivity
(Measured)
80
135000 140000 145000 150000 155000 160000 165000
Figure 6.4: Metamodel of applied compression load against longitudinal Young’s

Modulus.
110
105 Assumed
100
Measured
95
Sensitivity
90 (Assumed)
Sensitivity
85 (Measured)
80
55000 60000 65000 70000 75000
Skin Stiffness A11 (MPa)
Figure 6.5: Metamodel of applied compression load against skin composite extensional
stiffness A11.
171 | P a g e
110
105
Assumed
100
Measured
95
Sensitivity
(Assumed)
90
Sensitivity
(Measured)
85
80
255000 265000 275000 285000 295000 305000 315000 325000
Blade Stiffness A11 (MPa)
Figure 6.6: Metamodel of applied compression load against blade composite extensional
stiffness A11.
6.1.4 Discussion and Conclusion
The Robust Index has been applied and shown to be useful in revealing the robustness
of each panel. Comparisons between the two sets of Robust Indices show that in the
case of the COCOMAT panels, the robustness is highly dependent on the input
parameters of the analyses. The measured E11 had a standard deviation that was one
third that of the assumed E11 and a nominal mean that was lower by 5 GPa. A
combination of these contributed to the results seen in Table 6.4. The differences in the
scatter of transverse Young’s Modulus, Poisson’s Ratio and the Shear Modulii do not
appear to affect the result, in both cases, and hence, can be considered less critical in
terms of contribution to the robustness of the panel. It is interesting to note how the
robustness of the skin laminate is affected while the robustness of the blade laminate
remains relatively unaffected, even with changes in the material properties. Therefore it
can be concluded that the lay-up chosen for the blade is less affected by properties of
the material, compared to that of the skin. From the study, it was also found that the
172 | P a g e
postbuckling behaviour of the D1 panels were mainly influenced by the stiffener blades,
given that R.I.I for both panels were taken from the extensional stiffness component A11
of the blade.
6.2 Study of Robustness for the D1 and D2 Panels
In order for the robustness of the D1 and D2 panels to be quantified, stochastic analyses
of the panels are conducted through means of numerical simulation. Both panels have
been modelled in MSC.Patran using QUAD4 elements and solved using the Nonlinear
Static SOL 106 solver in MSC.Nastran. The finite element model for the D1 panel can
be found in Section 6.1.2 and the finite element model for the D2 panel can be found in
Section 6.2.2. Particular attention is paid to how the load carrying capability of each
panel is affected by laminate stiffness. Similarly to previous stochastic analyses,
imperfections in each laminate are introduced via variations in material and lamina
properties. The panels may be considered different due to the additional edge supports
in the D2 design but this is acceptable for the purposes of understanding the behaviour
of the panels in the experiments and possibly in the implementation of these panel
designs for future use in designing aircraft fuselage structures. A total of 41 samples
were used for each panel.
The stochastic boundary for the lamina and material properties can be seen in Table 6.5.
Means and standard deviations for each material property have been taken from the
material characterisation. The lamina orientations and thicknesses have been assumed to
have a CoV that is 1.25% and 2.5% respectively. The panel was axially compressed to
2.0 mm and the corresponding loads were recorded. Each finite element analysis for the
D1 and D2 panels required an average computational time of twenty minutes to solve
on an Intel® Core™2 Quad processor.
173 | P a g e
Table 6.5: Stochastic boundary for the D1 and D2 panels.

Poisson’s Ratio, υ 0.277 0.237 - 0.317
Lamina thickness for D1 (mm) 0.125 0.116 - 0.134
Lamina thickness for D2 (mm) 0.152 0.141 - 0.163
6.2.1 Analysis of D1 panel
The finite element model for the D1 panel is similar to the one found in Figure 6.1. In
Chapter 5, a variety of postbuckling mode shapes were identified including non-
symmetric modes that have been attributed to the variation in boundary conditions on
the loaded edges. An observation of the experimental results from Chapter 4 led to the
conclusion that there was an improvement in the quality of the experiment which can be
attributed to experience gained in panel preparation. The experimental P30 and P31
panels developed symmetrical postbuckling modes similar to those obtained in the
original benchmark finite element models, where the end shortening was assumed to be
uniform. Therefore in this series of analyses, the number of variables was reduced by
removing the through thickness variation thereby matching the postbuckling mode
found in panels P30 and P31.
Figure 6.7 shows a comparison plot of the experimental out of plane displacement for
the P31 experimental panel, together with the plots taken from finite element analysis of
a D1 panel with nominal input parameters. In the experimental plots, the blue region
174 | P a g e
represents displacement that is out of the page while the red represents displacement
that is moving into the page. This is the opposite case for the results taken from finite
elements. It can be seen that there is a good match between the two results.
Experimental panel P31 result
Nominal finite element result

Figure 6.7: Comparison of results for D1 panel from experiment and finite element
analysis.
100
90
80
70
60
50
40
30
20
10
0
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2
Figure 6.8: Load shortening curves for D1 FEM with upper and lower bounds.
175 | P a g e
The stochastic analysis for D1 was completed and Figure 6.8 shows the predicted load-
shortening curve for the FE analysis of the panel, with input variables corresponding
with the nominal means. The upper and lower boundaries seen in the plot are from the
analysed panels with the highest and lowest load-carrying capability from the 41
samples. Table 6.6 shows the results with the Robust Index [R.I.]Xi of the input
variables together with the Robust Index (R.I.II) of the panel. It can be seen from the
Influence variable that longitudinal Young’s Modulus and the laminate in-plane
stiffness component A11 of the panel skin and blade contribute most to the load-carrying
capability of the panel. The R.I.I for this configuration is 0.0127, taken from the
extensional stiffness of the blade, while the R.I.II is 0.0778.
Table 6.6: Results of stochastic analysis for the D1 panel.

Scaled
Sensitivity
Material E11 0.631 0.000748 107 0.0148
E22 0.214 0.000893 8.66 0.540
ν12 -0.0362 -1.54 -0.426 64.9
G12 0.560 0.00213 10.8 0.165
G23 -0.0526 -0.000538 -2.18 8.71
G31 0.560 0.00213 10.8 0.165
0.0778
Skin A11 0.611 0.000985 59.7 0.0274
B11 -0.0895 -0.000726 -1.58 7.09
D11 0.328 0.008797 19.1 0.160
Blade A11 0.877 0.000318 89.9 0.0127
B11 0.0928 0.000192 0.0441 244
D11 0.759 0.000377 48.6 0.0271
6.2.2 Finite element model for D2 panel
The finite element model for the D2 panel can be seen in Figure 6.9 and the stochastic
boundary for the lamina and material properties are shown in Table 6.5. The D2 panels
were modelled using 3276 QUAD4 shell elements using the MSC.Patran pre-processor.
176 | P a g e
In order to achieve a minimum computational time for solving each model, the skins
and stiffener flanges were modelled as single laminates. Additionally, offsets were
applied in order to obtain panels that were geometrically accurate, having stiffness
coefficients that were representative of the physical panel. No material degradation was
included in the models and, therefore, any sharp drops in the deep postbuckling region
on the load shortening plots were a result of a change in buckling mode. The panel was
analysed using Solution Sequence 106 of the MSC.Nastran geometric nonlinear solver.
Geometrical properties for the panel can be found in Table 4.3. Due to the difficulty in
replicating the edge supports for the D2 design in the Cartesian coordinate system, a
decision was made to use a cylindrical coordinate system instead. Therefore, the
boundary conditions of the model shown in Figure 6.9 have been applied using the new
coordinate system.
Potting region: < 0, 0, >
Clamped edge: < 0, 0, 0 > < 0, 0, 0 >
< 0, 0, 0 >
t
z
r
Loaded edge: < 0, 0, -z >

< 0, 0, 0 >
Edge supports: < 0, , >

< , 0, > Translations: < r, t, z >
Rotations: < θr, θt, θz >
Figure 6.9: Finite element model of D2 panel.

177 | P a g e
6.2.3 Analysis of D2 panel
The stochastic analysis for D2 was completed and Figure 6.10 shows a comparison
between the out of plane displacement plots from the experiment and finite element
analysis. The finite element analysis used the nominated means for the variables in the
analysis. The corresponding load-shortening curves for the panel can be found in Figure
6.11. The experimental panel appears to be stiffer compared to the finite element model,
as evidenced from the load shortening plots. In the finite element model, the bifurcation
buckling occurs later compared to the experimental panel. From the out of plane
displacement plots, it can be seen that a difference emerges in the middle skin bay,
between the experimental result and finite elements. The bay in the experimental panel
remains unbuckled until 0.88 mm axial shortening while the skin bay in the finite
element model buckles at 0.74 mm. The differences in buckling load and mode shapes
can be attributed to variations and defects in the experimental panel which are non-
existent in the nominal finite element model that is shown.
Experimental panel P24 result
Nominal finite element result

Figure 6.10: Comparison of D2 buckling modes from experiment and finite element
analysis.
178 | P a g e
200
180
160
140
120
Panel P24
100
80 D2 FEM
60
40
20
0
0 0.5 1 1.5 2
Figure 6.11: Load shortening plots from experimental panel P24 and finite element
analysis.
The upper and lower boundaries seen in Figure 6.12 are from the analysed panels with
the highest and lowest load-carrying capability from the 41 samples. Table 6.7 shows
the results of the stochastic analysis conducted on the D2 panel. It can be seen that the
longitudinal Young’s Modulus, the laminate stiffness A11 of the panel skin and blade
contribute most to the load-carrying capability of the panel. The R.I.I of this
configuration is 0.0157, taken from the extensional stiffness of the skin, while the R.I. II
is 0.101.
179 | P a g e
180
160
140
120
100
80
60
40
20
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Figure 6.12: Load shortening curves for D2 FEM with upper and lower bounds.
Table 6.7: Results of stochastic analysis for the D2 panel.

Scaled
Sensitivity
Material E11 0.473 0.000776 110 0.0191
E22 0.091 0.000500 4.84 2.27
ν12 0.324 19.3 5.26 0.587
G12 0.406 0.00174 8.58 0.287
G23 0.096 0.000734 2.95 3.54
G31 0.406 0.00174 8.58 0.287
0.101
Skin A11 0.687 0.00126 92.9 0.0157
B11 0.057 0.000313 -0.0242 729
D11 0.510 0.00397 32.8 0.0599
Blade A11 0.592 0.000631 75.7 0.0223
B11 -0.130 -0.000252 0.00637 1210
D11 0.487 0.000888 31.0 0.0662
180 | P a g e
6.2.4 Discussion of results
A summary of the results from the analyses conducted in Sections 6.1.2 and 6.1.3 can
be found in Table 6.8. The D2 panel is more robust than the D1 panel, when R.I. I and
R.I.II are taken into consideration. On inspection of the [R.I.I]Xi values, it can be seen
that the robustness of the D1 panel is driven by the behaviour of the stiffener while the
behaviour of the D2 panel is driven by the skin. The skin laminate stiffness A11 is more
robust with the lay-up configuration of the D1 panel compared to the D2 panel. The
reason for this is the D1 panel having 90o plies on the outer surfaces compared to the D2
panel employing 45o plies which will be discussed in Chapter 7 where a redesign of the
panels is undertaken.
Table 6.8: Summary of results from analysis of the D1 and D2 panels.

Property Panel Robust Indices
Material D1 0.0148
E11 [R.I.]Xi D2 0.0191
Skin D1 0.0274
A11 [R.I.]Xi D2 0.0157 (R.I.I for D2)
Blade D1 0.0127 (R.I.I for D1)
A11 [R.I.]Xi D2 0.0223
D1 0.0778
Panel (R.I.II)
D2 0.101
The metamodels for the compression load against the skin and blade extensional
stiffness can be found in Figures 6.13 and 6.14, respectively. In the metamodel for
compression load against blade extensional stiffness, it can be seen the blade of the D1
panel is stiffer but less robust due to the higher influence that the stiffness has on the
compression load. This effect of stiffness and robustness needs to be investigated before
concrete conclusions can be drawn. It can also be seen that there are also visibly
observable differences in sensitivity between the blades of the two panels.
181 | P a g e
180
170
160
D1 Panel
150
140 D2 Panel
130
120 Sensitivity (D1)
110
Sensitivity (D2)
100
90
80
53400 58400 63400 68400 73400 78400
Skin Extensional Stiffness A11 (MPa)
Figure 6.13: Metamodel of compression load against skin extensional stiffness.
180
170
160
D1 Panel
150
140 D2 Panel
130
120 Sensitivity (D1)
110
Sensitivity (D2)
100
90
80
110000 160000 210000 260000 310000
Blade Extensional Stiffness A11 (MPa)
Figure 6.14: Metamodel of compression load against blade extensional stiffness.
182 | P a g e
6.2.5 Conclusions
From the analyses it can be seen that the factors affecting the robustness of each panel is
dependent on the buckling and postbuckling of each panel. The D1 panel was designed
to have a deep postbuckling region and the effect of this can be seen with the R.I. I
measure, which was obtained from the blades of the panel. In the deep postbuckling
region, the skin has already buckled and the majority of the load is transferred to the
stiffeners which can be seen in the large global buckling mode where the skin and
stiffener buckle as a single component. In the D2 panel, this did not occur; the
difference in stiffness between the panel skin and stiffeners were greater and this caused
the panel to buckle into squares.
From the analyses, it can be seen that there is a lack of knowledge about the effect of
laminate configurations. Laminates have traditionally been designed to satisfy stiffness
and strength criteria but it is also clear that the laminate configuration also has an effect
on the robustness of the panel. Therefore this gap in knowledge needs to be bridged and
a parametric study is conducted in the following chapter.
The Robust Indices have been shown to be useful in revealing the robustness of the
panels which were tested by the DLR. In this case, only two of the panel designs have
been analysed and compared. In a design scenario for structural components, there
would possibly be more than two designs to be studied. Hence, the Robust Indices
would provide more insight as the design selection of panels can be done based on a
quantified robustness.
6.3 Conclusion
In this chapter, the initial assumptions for the analysis of the COCOMAT panels were
validated. The results achieved in Section 6.1 were applied in the analysis of the
COCOMAT panels and these assumptions will also be applied in Chapter 7 when a
redesign of the D2 panel is undertaken.
183 | P a g e
The robustness properties of the experimental panels, D1 and D2, were also evaluated.
An analysis comparing the two panels tested by DLR found that the D1 panel was less
robust than the D2 panel design due to the design of the stiffener blade. This result can
be supported by the design criteria of each panel. The D1 panel was designed to be
sensitive so that more insight could be gained into postbuckling and the related collapse
mechanisms. The D2 panel was more stable due to the design criteria of the panel being
one that was representative of an actual panel that would be applied to an aircraft
fuselage.
When the results of the COCOMAT project are considered, it can be seen that the
Robust Indices are a key to determining the factors affecting the postbuckling response
of stiffened structures. Due to the sensitivity and variations in response of these highly
unstable structures, the robustness of such panels should be considered with the
assistance of the quantitative measures that have been developed.
184 | P a g e
Chapter 7 Redesign of COCOMAT Panels using Robust Indices
Chapter 7
Redesign of COCOMAT Panels

using Robust Indices
During the evaluation of structural components, it is common that more than two
designs are considered for the structural application. Whilst it is possible to make design
decisions based on qualitative measures, a problem emerges when multiple designs
need to be effectively compared. The application of the Robust Indices is demonstrated
here as evidence that the methodology can be used in quantitative design evaluations
where a family of components are considered.
Section 7.1 contains an analysis of various thin shelled composite laminates. These
composite laminates have been designed to reflect the type of boundary conditions
found in the COCOMAT testing. The panels that are tested in COCOMAT are complex
structures with multiple load paths and failure mechanisms. Therefore the aim of this
study is to simplify the problem so that there is no interaction between the buckling of
the skin and stiffener of the panel. The study revealed that the stacking sequence of the
laminate has an effect on the robustness of the panels under compression loading.
From the study conducted in Chapter 6, it was found that the D1 panel had a more
robust skin layup while the D2 panel had a more robust blade layup. The geometrical
design of the D2 panel was meant to be a realistic representation of fuselage panels for a
70-100 passenger aircraft. Therefore, the geometry is used here as a platform on which
parametric studies are conducted. In the parametric studies, the robustness of the panels
185 | P a g e
is considered with four separate laminate configurations. This parametric study can be
found in Section 7.2.
7.1 The Effect of Stacking Sequence on Laminate Robustness
This section presents three composite laminate configurations that are used to
investigate the effect that laminate stacking sequence has on the response of composite
plates in compression loading. This study has been undertaken so that the results in the
following sections of this chapter can be better understood. In the stochastic analyses,
only lamina orientation and thickness were varied while the material properties were
considered deterministic and kept constant. This was done so that the robustness of the
plates in buckling failure is purely a function of laminate configuration and not material
properties.
7.1.1 Design of laminates for the buckling study
The laminates that are used in the analyses are shown in Figure 7.1, together with the
equivalent bending stiffness of the laminate. These laminate configurations have been
applied on to three plates: Plates A, B and C. In a deterministic analysis where the
perfect laminates are applied on to the plates, it is obvious, from the figure, that in-plane
stiffness for all three should be identical while the resistance to buckling under
compression loading (or bending stiffness) will increase the from A to C.
186 | P a g e
Laminate Equivalent
Plate Example Laminate
Configuration Bending Stiffness
A [90,±45,0]s
B [±45, 0, 90]s
C [0, ±45, 90]s
0o ply 45o ply -45o ply 90o ply

Figure 7.1: Laminate configuration and equivalent bending stiffness of plates used in
study.
7.1.2 Geometry and boundary conditions of plates
The geometry of the flat plate is shown in Figure 7.2. It can be seen that the boundary
conditions have been selected in a way that they represent the behaviour of the panels
tested in the COCOMAT project. Potting has been included at both ends to ensure that
two edges are clamped while longitudinal edge stiffeners have been included so that the
edges are simply supported. The free edges of the plate were designed such that the ratio
of the longitudinal to traverse edge length was 1. The geometry of the plate can be
found in Table 7.1. Refer to Table 4.4 for the specifications of the material, Hexcel
IM7/8552, which is used for all analyses in this section.
187 | P a g e
clamped end
potting
L Lf K longitudinal
edge stiffener
potting
loaded end
Figure 7.2: Geometrical representation of flat plate.
Table 7.1: Nominal geometry for composite plates.

Parameter Value
Panel width, W (mm) 100
Edge Support Length, K (mm) 90
Lamina thickness (mm) 0.152
Material Hexcel IM7/8552
7.1.3 Finite element modelling of the laminate
A finite element model using the geometry shown in Figure 7.2 was created in
MSC.Patran with 720 QUAD4 shell elements and 777 nodes. The model can be seen in
Figure 7.3 with the boundary conditions that were applied. An axial load of 100 N was
applied on the loaded edge and the model was solved in MSC.Nastran using a linear
buckling solver, Solution 105. The laminate configurations shown in Figure 7.1 were
applied to the finite element model seen in Figure 7.3. In the application of the laminate
188 | P a g e
to the plates, care was taken to ensure that the mid-plane of the laminate was between
the fourth and fifth plies.
Potting region: < 0, 0, >

Clamped edge: < 0, 0, 0 >
< 0, 0, 0 >
< 0, 0, 0 >
Edge supports: < 0, , >

< , 0, > Loaded edge: < 0, 0, -z >
< 0, 0, 0 >
Translations: < x, y, z >
Rotations: < θx, θy, θz >
Figure 7.3: Finite element model of composite plate.
The first buckling mode experienced by the plate was a single buckle between the
potting boundary condition, as shown in Figure 7.4. This was a buckling pattern that
was observed in all the finite element models in the stochastic analyses.
189 | P a g e
Figure 7.4: Eigenvector plot of the panel at bifurcation buckling.
7.1.4 Analysis of laminate configurations
The three laminate configurations were analysed in with 25 samples for each
configuration. The stochastic boundary for the lamina properties can be seen in Table
7.2. The lamina orientations and thicknesses have been assumed to have a standard
deviation of 1.25o and 0.0038 mm, respectively. A 100 N load was applied to the edge
of the panel and the corresponding buckling factors were recorded. The buckling factors
were then used as an output response to measure the robustness, with respect to the
composite stiffness properties, which are shown in Table 7.2.
Table 7.2: Stochastic boundary for the plates.

Input Variable Mean Defined Range
o
-45 Lamina orientation (deg) -45 -41.6 - -48.4
190 | P a g e
Table 7.3: Results of stochastic analysis for the plates.

Scaled
Sensitivity
Panel A
A11 0.586 0.000764 56.1 0.0304
B11 -0.366 -0.000730 -0.0000927 29500 0.0550
D11 0.865 0.00841 32.5 0.0356
Panel B
A11 0.644 0.00126 92.9 0.0167
B11 -0.168 -0.000620 0.0882 -67.5 0.0378
D11 0.931 0.00557 45.8 0.0234
Panel C
A11 0.562 0.00123 90.9 0.0196
B11 0.0127 0.000130 0.00429 18400 0.0447
D11 0.956 0.00442 67.5 0.0155
The metamodels from the stochastic analyses are shown in Figures 7.5, 7.6, and 7.7
where the buckling factors are plotted against the laminate extensional, coupling and
bending stiffness, respectively.
191 | P a g e
90
80 Panel A
Panel B
70
Buckling Factor
Panel C
60
Sensitivity
50 (Panel A)
Sensitivity
(Panel B)
40
Sensitivity
(Panel C)
30
70000 71000 72000 73000 74000 75000 76000 77000 78000 79000
Laminate extensional stiffness A11 (MPa)
Figure 7.5: Metamodel of Buckling Factor against laminate extensional stiffness.
90
Panel A
80
Panel B
70
Buckling Factor
Panel C
60
Sensitivity
50 (Panel A)
Sensitivity
(Panel B)
40
Sensitivity
(Panel C)
30
-2000 -1500 -1000 -500 0 500 1000 1500 2000
Laminate coupling stiffness B11 (MPa)
Figure 7.6: Metamodel of Buckling Factor against laminate coupling stiffness.
192 | P a g e
90
Panel A
80
Panel B
70
Buckling Factor
Panel C
60
Sensitivity
50 (Panel A)
Sensitivity
40 (Panel B)
Sensitivity
30 (Panel C)
0 5000 10000 15000 20000
Laminate bending stiffness D11 (MPa)
Figure 7.7: Metamodel of Buckling Factor against laminate bending stiffness.
In Figure 7.5 it can be seen that Panel C has the largest range in terms of extensional
stiffness compared to Panels A and B. This is an interesting result as all the laminates
exist within a fixed range of variations, as shown in Table 7.2. This result is consistent
in the other metamodels, in Figures 7.6 and 7.7, where the laminate configuration has
contributed to a larger input variation in Panel C, followed by Panel B and Panel A. The
coupling stiffness of Panel B was found to have the lowest [R.I.] Xi compared to the
other two panels. This was due to the scaled sensitivity being highest among the three
designs. In Figure 7.7, where the buckling factors are plotted against the laminate
bending stiffness, it is noticeable that Panel A has the smallest range in terms of output
response compared to Panel B and C.
From the stochastic analyses that have been conducted, it is clear that Panel A is the
most robust of the three designs and this can be wholly attributed to the differences in
laminate configuration. Figure 7.8 shows a plot where the longitudinal Young’s
Modulus of the IM7/8552 unidirectional tape, and its first derivative, is plotted against
193 | P a g e
lamina orientation. It can be seen that a lamina having a scatter about the 45o orientation
will be more sensitive compared to a lamina having a scatter about a 90o orientation.
The change in longitudinal stiffness with respect to orientation will be least pronounced
in Panel A where the outermost plies have a nominal 90o orientation and most obvious
in Panel C where the outermost plies have a nominal 0o orientation.
160000 0
140000 -50000
1st Derivative of Young's Modulus

Young's Modulus, E11 (MPa)
-100000
120000
-150000
100000
Young's Modulus
-200000
80000
1st Derivative of -250000
Young's Modulus
60000
-300000
40000
-350000
20000 -400000
0 -450000
0 20 40 60 80
Global orientation Ex(degrees)
Figure 7.8: Plot of Young’s Modulus E11 and first derivative against global orientation.
7.1.6 Conclusions
This section has shown that laminate configurations do play an important role in the
stability and robustness of a thin plate in compression. Care has to be taken in laminate
selection in order for a robust response to be obtained.
7.2 Parametric Study of Four D2 Panel Configurations
The Design 2 panel geometry is used as in this section to evaluate possible laminate
configurations that may improve the robustness of the panel. Two skin laminate and
194 | P a g e
blade laminate configurations are considered here. In Section 7.1, it was found that the
most robust skin configuration was [90, ±45, 0]s. Therefore, this was the alternative skin
laminate that was applied to the panels together with the original [±45, 0, 90] s layup.
Another blade laminate configuration, [±45, 902, 02]s was introduced in this set of
analyses to investigate the behavior of the blade by bringing the zero degree plies
towards the mid-plane of the laminate.
The most robust panel in compressive loading using the D2 geometry had a skin
laminate of [90, ±45, 0]s and blade laminate of [±45, 902, 02]s.
7.2.1 Laminate design for hybrid D2 panels
As discussed in Sections 6.2 and 7.1, the laminate stacking sequence has an effect on
the strength and stiffness properties of the structure, and also on the overall robustness
of the design. In this section, a parametric study of the D2 panel configuration and three
hybrid designs are considered. Two laminate configurations of equivalent in-plane
stiffness have been created, one for the skin and the other for the stiffener, to be used in
the study. Table 7.5 shows the laminate configuration for the four panels that are
considered. In the COCOMAT experiments, only the D2 design was tested and this
panel was modelled as the D2(a) panel design in finite elements. The remaining panels
that are analysed are hybrids of the D2 design and are known in this study as the D2(b),
D2(c) and D2(d) panels. Therefore, this study extends the knowledge of the design by
introducing the various laminate configurations and investigating the resulting
differences in robustness. Example laminates and their equivalent bending stiffness are
Table 7.4: Laminate configurations for D2 panel parametric study.

Panel Parameter Value
D2(a) panel Stiffener web lay-up [±45, 02, 902]s
195 | P a g e
Panel Parameter Value

Skin lay-up [90, ±45, 0]s
D2(b) panel Stiffener web lay-up [±45, 02, 902]s
D2(c) panel Stiffener web lay-up [±45, 902, 02]s
Skin lay-up [90, ±45, 0]s
D2(d) panel Stiffener web lay-up [±45, 902, 02]s
Laminate Equivalent
Component Example Laminate
Configuration Stiffness
Skin (i) [±45,0,90]s
Skin (ii) [90,±45,0]s
Stiffener blade (i) [±45, 02, 902]s
Stiffener blade (ii) [±45, 902, 02]s
0o ply 45o ply -45o ply 90o ply
Figure 7.9: Laminate configurations and equivalent bending stiffness.
In the COCOMAT experiments, only the D2 panel design was tested and it has been
called the D2(a) panel is this study. Therefore, this study extends the knowledge of the
196 | P a g e
design by introducing the various laminate configurations and investigating the

resulting differences in robustness.
7.2.2 Stochastic analysis of hybrid D2 panels
The panels were re-analysed and the results for all the hybrid panels are presented in
Table 7.6. All the panels were axially compressed to 1.80 mm. There were 40 samples
for each of the hybrid D2 panel configurations, D2(b) to (d), conforming to the
requirement imposed by Equation (3.7). The results for the D2(a) panel can be found in
Table 6.7. Each finite element analysis for the D2 panel required an average
computational time of twenty minutes to solve on an Intel® Core™2 Quad processor.
Table 7.5: Results of stochastic analysis for the hybrid D2 panels.

Scaled
Property Influence Sensitivity [R.I.I]x R.I.II
Sensitivity
D2(b) panel
Material E11 0.354 0.000817 116 0.0243
E22 0.128 0.000773 7.48 1.0424
ν 0.248 23.4 6.35 0.634
G12 0.137 0.000462 2.28 3.21
G23 -0.158 -0.00249 -9.98 0.633
G31 0.137 0.000462 2.28 3.21
0.0966
Skin A11 0.697 0.00113 82.7 0.0173
B11 0.178 0.00169 0.0597 93.9
D11 0.625 0.00839 32.7 0.0490
Blade A11 0.479 0.000674 80.9 0.0258
B11 0.026 0.000107 0.00271 14400
D11 0.416 0.000942 32.9 0.0729
D2(c) panel
Material E11 0.330 0.000789 112 0.0270
E22 0.165 0.000532 5.15 1.18 0.0848
ν 0.219 15.6 4.23 1.08
197 | P a g e
Scaled
Property Influence Sensitivity [R.I.I]x R.I.II
Sensitivity
G12 0.329 0.00143 7.07 0.429
G23 -0.0760 -0.000491 -1.97 6.68
G31 0.329 0.00143 7.07 0.429
Skin A11 0.645 0.00132 97.0 0.0160
B11 0.0274 0.000528 -0.0407 -898
D11 0.543 0.00456 37.6 0.0490
Blade A11 0.712 0.000824 99.0 0.0142
B11 -0.0819 0.0000432 -0.000940 13000
D11 0.425 -0.000193 -3.76 -0.624
D2(d) panel
Material E11 0.192 0.000566 80.6 0.0645
E22 0.301 0.00236 22.9 0.145
ν 0.237 28.1 7.63 0.552
G12 0.206 0.000686 3.39 1.43
G23 -0.0478 -0.000690 -2.77 7.54
G31 0.206 0.000686 3.39 1.43
0.105
Skin A11 0.704 0.00122 89.7 0.0158
B11 -0.0946 -0.000110 0.00391 -2710
D11 0.664 0.00997 38.8 0.0388
Blade A11 0.544 0.000784 94.1 0.0195
B11 0.175 0.000681 -0.0148 -386
D11 0.366 -0.0000810 -1.58 -1.72
Figures 7.10, 7.11 and 7.12 show the load-shortening curves for the finite element
D2(b) to (d) panels with input variables corresponding with the nominal means. The
upper and lower boundaries seen in the plot are from the analysed panels with the
highest and lowest load-carrying capability. The corresponding plot for Panel D2(a) can
be found in Figure 6.12.
198 | P a g e
180
160
140
120
100
80
60
40
20
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Figure 7.10: Load shortening curves for D2(b) FEM with upper and lower bounds.
180
160
140
120
100
80
60
40
20
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Figure 7.11: Load shortening curves for D2(c) FEM with upper and lower bounds.
199 | P a g e
180
160
140
120
100
80
60
40
20
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Figure 7.12: Load shortening curves for D2(d) FEM with upper and lower bounds.
The R.I.I values originate from the extensional stiffness A11 of the skin laminates for all
the panels except the D2(c) design. Metamodels of the compression load at 1.8 mm
shortening, against the longitudinal Young’s Modulus, skin extensional stiffness and
blade extensional stiffness, are shown in Figures 7.13, 7.14 and 7.15, respectively. In
total there are 160 samples shown in these metamodels.
200 | P a g e
180
D2(a) Panel
175
D2(b) Panel
170
D2(c) Panel
165
D2(d) Panel
160
Sensitivity
155 (D2(a))
Sensitivity
150 (D2(b))
Sensitivity
145 (D2(c))
Sensitivity
140 (D2(d))
138000 140000 142000 144000 146000 148000
Figure 7.13: Metamodel of compression load against longitudinal Young’s Modulus.
180
D2(a) Panel
175
D2(b) Panel
170
D2(c) Panel
165
D2(d) Panel
160
Sensitivity
155 (D2(a))
Sensitivity
150 (D2(b))
Sensitivity
145 (D2(c))
Sensitivity
140 (D2(d))
68000 70000 72000 74000 76000 78000
Skin Extensional Stiffness A11 (MPa)
Figure 7.14: Metamodel of compression load against skin extensional stiffness.
201 | P a g e
180
D2(a) Panel
175
D2(b) Panel
170
D2(c) Panel
165
D2(d) Panel
160
Sensitivity
(D2(a))
155
Sensitivity
(D2(b)
150
Sensitivity
(D2(c))
145
Sensitivity
(D2(d))
140
114000 116000 118000 120000 122000 124000 126000
Blade Extensional Stiffness A11 (MPa)
Figure 7.15: Metamodel of compression load against blade extensional stiffness.
In Figure 7.16, the load shortening plot for the undamaged experimental D2 panel is
shown together with the plots obtained from the nominal finite element models of each
design. The finite element models appear to slightly under predict the stiffness of the
experimental panel. However, the results are reasonable as the D2(a) design, at an end
shortening of 1.8 mm, sustained a load that was only 6% less, which is a good match
once the effects of material and geometrical variation are considered. The D2(a) panel
was modelled to replicate the undamaged D2 experimental panel.
202 | P a g e
180
160
140
120 Panel P24
100 D2(a) FEM
80 D2(b) FEM
60 D2(c) FEM
40 D2(d) FEM
20
0
0 0.3 0.6 0.9 1.2 1.5 1.8
Figure 7.16: Load shortening curves for experimental D2 and hybrid D2 FEM.
Table 7.7 presents a summary of the results from the stochastic analysis of the D2
panels. From the results it can be seen that the D2(c) panel has the lowest robustness of
the four panels while the D2(d) panel has the lowest stiffness. The difference in skin
laminates between the D2(a) and D2(b) panels resulted in a panel design that was more
robust using the R.I.I measure. The R.I.II measures how representative the R.I.I value is
and, despite the D2(a) having a slightly larger value compared to the D2(b) panel, this
loss is offset by having a gain in R.I.I robustness. The D2(d) panel was the only design
that had an R.I.I and R.I.II that were higher than the D2(a) panel.
Table 7.6: Summary of results for stochastic analysis of the D2 panels.

Panel Mean Compression
R.I.I R.I.II
Configuration Load at 1.8 mm (kN)
D2(a) 0.0157 0.101 163.9
D2(b) 0.0173 0.0966 160.25
D2(c) 0.0142 0.0848 163.2
D2(d) 0.0158 0.105 154.8
203 | P a g e
The results of the stochastic analysis show that the [R.I.]Xi value of the skin laminate
stiffness A11 of the D2(b) panel is more robust compared to the D2(a) panel. A reason
behind this is the D2(b) panel having 90o plies on the outer surfaces compared to the D2
panel employing 45o plies, as explained in Section 7.1.4. The [R.I.I]x of E11 is higher in
all the hybrid panels. In the D2(d) panel it has a value more than twice that of the D2(a)
panel. This translates to the hybrid panels being less affected by scatter in material and
manufacturing quality, which could be converted into a cost saving.
Despite the D2(b) design having a lower R.I.II value, compared to the D2(a) and D2(d)
designs, the D2(b) design would be the design that was selected if robustness was the
key criterion in the design as it has the highest R.I.I value. The basis for this choice
would be the fact that the robustness of the structure would firstly be a function of its
least robust input variable, which in this case is the extensional stiffness of the panel
skin, followed by the R.I. II value which quantifies how representative the R.I. I value is
with respect to the other [R.I.I]x values.
7.2.4 Conclusion
The Robust Indices are useful in a design scenario where a manufacturer would be
interested in producing composite skins that are damage tolerant and composite
stringers that are less expensive. With the index, it is possible to use the quantified
values as design indicators so that robustness can be considered and included with other
criteria, such as cost and performance.
The considerations for the design of fuselage panels should also include other load
cases, such as torsion and tensile loading. If this is the case, then having the ±45° plies
on the outer layers of the laminates will ensure that adequate stiffness and strength are
retained in the event of high torsional shear loads and impact damage. The ±45° plies
will also ensure that stress concentrations from fasteners will be reduced when the
structure is undergoing tensile loading. In the study that has been conducted, the D2(a)
design remains the ideal choice for application in aerospace structures.
204 | P a g e
Among the pool of D2 designs that were analysed, it could be seen that there were
variations in robustness as well as stiffness. The design selection of aerospace structural
components is not a trivial activity. Other considerations such as damage tolerance,
stiffness, strength, cost etc, do exist and these play an important part in ensuring that the
final design meets the specifications and certification requirements.
7.3 Conclusion
In this chapter, the robustness properties of three composite plates and the D2 panel
together with three other hybrid D2 panels were evaluated. From analysis of the flat
composite plates under compression loading, it was found that having 90o plies on the
outer surface of the laminates result in the most robust plates while 0o plies result in the
least robust solution. However any gain in robustness for the plates is offset by a
decrease in the bifurcation buckling load.
Among the pool of D2 designs that were analysed, it could be seen that there were
variations in robustness as well as stiffness. The design selection of aerospace structural
components is not a trivial activity. Other considerations such as damage tolerance,
stiffness, strength, cost, etc do exist and these play an important part in ensuring that
final design meets the specifications and certification requirements.
It can be seen that the design of composite aerospace structures is a task that is made
difficult by the requirement to meet multiple objectives that can sometimes be
conflicting. However with composite structures, laminate properties can be tailored with
the use of fibre orientations to reach the desired outcome. The development and
application of an algorithm to achieve these multiple objectives is beyond the scope of
this thesis but some suggestions as to the direction that may be undertaken will be
covered in Chapter 8 under the heading of Future Work.
205 | P a g e
Chapter 8 Conclusion
Chapter 8
Conclusion
8.1 Summary of Key Findings and Achievements
The main goals of this research were to apply Stochastic Analysis to develop an
understanding of the sources and significance of the variability encountered in the
buckling of stiffened shell structures in the COCOMAT project. A natural extension of
the stochastic analysis procedure involved the development and validation of Robust
Indices to quantify the robustness of designs. The indices enable the design of robust
configurations that exhibit a reduction in the significant variations. These objectives
were accomplished through the completion of the work packages defined in Chapter 1
and the key achievements are summarised in the following sections.
8.1.1 Review of literature
This chapter summarised the state of the art for Stochastic Analysis and Robust Design
and their application in the design and analysis of structural components with focus on
the design of composite stiffened structures in postbuckling. A comprehensive review
of all relevant literature was provided and the key research developments in Stochastic
Analysis and Robust Design were summarised. The metamodel concept was introduced
in the Stochastic Analysis section while the current state of the art for Robust Design
was reviewed. A review of composite structures found that composite materials are
susceptible to many defects such as delamination, fibre misalignment, matrix voids,
skin-stiffener disbond, etc. These defects and variations ultimately affect the stiffness
and strength of the structure especially under compression loading and must therefore
206 | P a g e
be accounted for in the finite element analyses. Literature in the field of buckling and
postbuckling was reviewed and it was found that there were still challenges in
determining buckling modes and there were gaps of knowledge about quantifying the
stability of thin-walled structures.
8.1.2 Development of Theory and Formulation
The development and formulations for the Stochastic Analysis and the Robust Indices
were presented in Chapter 3. A Monte Carlo approach was used to generate
metamodels of the output responses against the input parameters for a sample set. Two
parameters, Influence and Sensitivity, were then developed and applied to quantify the
metamodels. Extending the Stochastic Analysis into the design environment, it was
found that the Influence and Sensitivity parameters are functions of robustness and these
were consequently used to derive two Robust Indices which are quantitative measures
of structural robustness. The indices are especially valuable in structural applications
where the input parameters of a structure are difficult to predict and in highly optimised
structures where the response is sensitive to changes in the input parameters.
8.1.3 Results of DLR and CRC-ACS COCOMAT experiments
Chapter 4 summarised and provided details for the experiments conducted in

collaboration with the German Aerospace Center (DLR) for the COCOMAT project.
Extensive investigations were conducted on the experimental panels, showing the
manufacturing defects and imperfections that emerged from the manufacturing process.
Experimental results that were relevant to the thesis were presented. The results of the
cyclically loaded panels were used to determine the level of axial compression that the
finite element analyses could accurately solve to without the use of any progressive
damage subroutines. This was an essential step in ensuring that computational costs
were contained.
207 | P a g e
8.1.4 Stochastic Analysis of COCOMAT D1 Panel
In Chapter 5, the Stochastic Analysis procedure developed in this thesis was applied on
to the COCOMAT D1 panel. Two sets of analyses were conducted in this chapter. The
first involved the residual thermal stresses in the panel causing the initial deformations
measured using the ATOS system. The stochastic methodology was used identify the
parameters affecting the final geometry during the curing process.
The final work package in the COCOMAT project involved the generation of design
guidelines for the degradation of postbuckling stiffened structures. It was found that the
initial geometrical imperfections have a considerable effect on the profile of the load
shortening curve. A finite element analysis of a nominal panel with initial geometrical
imperfections was completed and significant differences in stiffness and ultimate load
were observed when the results were compared to a finite element panel with no
imperfections. Therefore, in future collapse analyses, it is important to include these
initial geometrical imperfections so that better correlation with the experimental results
can be achieved.
The starting point of this thesis was an inability to achieve the postbuckling mode shape
in the experimental benchmark D1 panel using finite elements. With the introduction of
variations in the input parameters, the postbuckling mode shape seen was matched and
there was a good agreement with the experimental results. An area of concern for
numerical simulations is the ability to accurately simulate the possible responses which
enables any unexpected behaviour in the experimental regime to be anticipated ahead of
time. More importantly, the Stochastic Analysis provides insight into the behaviour of
the structures so that the observed phenomena can be understood.
8.1.5 Definition of Robust Indices for COCOMAT panel designs
The Robust Indices developed in the thesis were applied in Chapter 6 to first validate
the initial assumptions to be used for analysing the COCOMAT D1 and D2 panels.
Input parameters from the material characterisation completed in COCOMAT were
compared with an assumed set of input parameters that had a 5% Coefficient of
208 | P a g e
Variation. A significant difference in robustness was found between the two sets of data
and therefore a recommendation for future analyses is to apply input values which have
been measured beforehand.
Next, the two COCOMAT panel designs, D1 and D2, were analysed. The postbuckling
responses of the two panels were quantified using the Robust Indices. The input
variables having the lowest robustness for each design were identified while the
application of the Robust Indices provided evidence that the methodology developed
can be used in quantitative design evaluations where a family of components are
considered.
8.1.6 Redesign of COCOMAT panels using Robust Indices
In Chapter 7 the robustness properties of three composite plates, the D2 panel, as well
as three other hybrid D2 panels were evaluated. The composite plates were modelled in
finite elements with boundary conditions similar to the D2 panels. When variations
were introduced in the laminate, it was found that having 90o plies on the outer surface
of the laminates resulted in the plates being most robust in buckling, while 0 o plies led
to the least robust solution. The metamodels showed that the gains achieved in
robustness for the plates were offset by a decrease in the bifurcation buckling load.
From the analyses of the COCOMAT D1 and D2 panels in the previous chapter, it was
found that the D1 panel had a more robust skin layup while the D2 panel had a more
robust blade layup. Therefore the D2 panel design was identified as a viable platform
for parametric studies due to its design. The geometry and boundary conditions of the
D2 panel were used as a starting point for the parametric studies where various laminate
combinations were considered. Among the pool of D2 designs that were analysed, it
could be seen that there were variations in robustness as well as stiffness. The design
selection of aerospace structural components is not a trivial activity. Other
considerations such as damage tolerance, stiffness, strength, cost, etc, do exist and these
play an important part in ensuring that the final design meets the specifications and
certification requirements.
209 | P a g e
8.2 Future Work
The research presented in this Ph.D. has significantly contributed to the development of
a methodology for application of a stochastic analysis to the design of imperfection
sensitive curved postbuckling composite panels. To extend the work to design, a novel
methodology has also been developed that accounts for stability and robustness of
structures with the application of two quantitative measures of robustness. Throughout
this work, areas of future development can be identified, both as a consequence of the
scope of this thesis and in terms of extending the state of the art for structural analysis
and damage modelling. The conclusion of this work leads to the following suggestions
for future investigations.
8.2.1 Guidelines for the design of stiffened composite structures
The experimental program in the COCOMAT project was aimed at generating

improved design guidelines for the degradation and collapse of stiffened composite
structures. This leads to the requirement that in light of manufacturing defects and
operational variations, stiffened composite panels must be designed in such a way that
failure can be accurately predicted. This means in a collapse scenario, it is ideal that a
common knockdown factor can be applied to all stiffened panel designs. Research by
Huehne et al. (2008) have shown that a common knockdown factor can be applied for
all unstiffened cylinders using a perturbation method. In order for a reduction in
developmental and operational cost, by 20% and 50% to be achieved in the short and
long term, respectively, for stiffened composite structures a similar method should be
applied .
An indication of realistic knockdown factors that can be applied in stiffened composite

structures lies in the level of skin-stiffener disbond that can be allowed in bonded
stiffener applications. The D1 panels were designed to have a large postbuckling region,
so that the postbuckling degradation and growth of skin-stiffener disbonds in the CFRP
panels could be better observed and understood. From the experiments it was found that
the worst case scenario is a 16 % decrease in load, when loaded at 88 % of ultimate
collapse load. There was no degradation or growth observed when the panel was
210 | P a g e
cyclically loaded at the bifurcation buckling load. Therefore it can be concluded that the
D1 panel, when loaded at 85 % of ultimate collapse load, should experience a decrease
of no more than 15% in load carrying capability due to damage growth in the panel.
This translates to a knockdown factor of 0.73 from the ultimate collapse load.
An interesting find in the experimental testing of the undamaged D1 panel was that the
panels exhibiting asymmetric postbuckling mode shapes at collapse had a higher
ultimate load compared to the panels with symmetric postbuckling mode shapes. More
investigations should be conducted to ascertain if this was a function of higher material
stiffness or the energy state that the particular postbuckling mode possesses.
8.2.2 Redesigning the D1 panel for further testing
In all respects, the D1 panel was designed to have a deep postbuckling region before
collapse as well as a high sensitivity to disbond growth. In designing the panel, it was
not predicted that multiple postbuckling mode shapes would emerge under real-life
manufacturing conditions and experimental testing; this was an unwanted side effect of
the high sensitivity. Therefore if the panel were to be redesigned in the future, care has
to be taken in the design of the stiffener blade. In preliminary finite element modelling,
it was found that having a stiffer blade would reduce the level of sensitivity that the
panel has to imperfections. This would lead to a redesign of the laminate configuration
or even an increase in the height of the stiffener blade.
Another option for redesigning the panel means that more longitudinal stiffness has to
be provided to the inner skin. With the curved configuration, the panel already exhibits
significant geometrical imperfections and work needs to be done to see if re-orientating
some of the plies in the laminate will reduce the imperfections as well as the sensitivity
of the panel to outlier postbuckling modes. With the multiple requirements and the need
to maintain robustness in the panels it can be seen that there is a need for a stochastic
procedure that can assist in any future redesign process.
211 | P a g e
8.2.3 Pseudo Particle Swarm Optimisation (PSO) technique
The Particle Swarm Optimisation (PSO) (Kennedy and Eberhart 1995) is a stochastic,
population based technique used to find an optimal solution within a search space. The
optimisation algorithm is based on social-psychological principles which carry the
belief that a population can solve problems based on the actions of the individual and
interaction between the individuals. PSO was inspired by the actions of birds flocking
and fish schooling. The shifting of the population is based on the objectives of the
optimisation using two criteria. These are the individuals within the population which
satisfy the objectives and are known as the Population Best, Pbest, and the Global Best,
Gbest.
In the Stochastic Analysis procedure, multiple metamodels are generated in the process.
Each metamodel, defined using the Influence and Sensitivity parameters provides a
wealth of information as to the input variables most affecting the response of the
structure. It appears that the generation of these metamodels is similar to the initial step
in PSO where each marker in metamodel represents an individual and the cloud of
points in the metamodel represents the population. Using the Influence and Sensitivity
parameters that have been derived, it is theoretically possible to move the population so
that it becomes more robust. Criteria for quantifying robustness have been defined in
this thesis in the form of the Robust Indices. These indices can be used to measure if
any improvements have been made by shifting the input parameter mean values of each
metamodel.
8.2.4 Extending Stochastic Analysis and Robust Design to other fields
While the applications of this thesis have been focused primarily on the collapse of
stiffened thin-walled structures, there remain many applications that the formulations in
this thesis can address. Due to the open-form solution methodology that has been
developed, Stochastic Analysis and the Robust Indices can be used to measure the
variations of output responses of other composite failure modes as well as in fields such
as composites manufacture, the management of defects, etc.
212 | P a g e
8.2.5 Robust Design and Damage Tolerance
The damage tolerant design philosophy was introduced in the 1960s in order to reduce
the cost of operating aircraft. This was achieved via a rigorous maintenance strategy
where the growth of cracks from fatigue loading would be monitored and repairs
initiated when the rate of crack growth exceeds a certain tolerance. This was especially
the case for non fail-safe designs where there were no alternate loading paths. This
strategy has also been adopted by the United States Air Force where numerous
publications on damage tolerant design have originated, such as the Damage Tolerant
Design Handbook MIL-HDBK-1530.
Some common definitions of damage tolerance include:
(i) Damage tolerance in metallic structures has been demonstrated using fracture
mechanics to characterise crack growth under cyclic loading for the constituent
materials, predict the rate of crack growth in the structure under anticipated
service loads, and established inspection intervals and non-destructive test
procedures to ensure fail safety (O’Brien 1988).
(ii) Damage tolerance is concerned with the ability of the structure to contain
representative weakening defects under representative loading and
environment without suffering excessive reduction in strength for some
stipulated period of service (Baker, Jones and Callinan 1985).
(iii) The ability to resist failure due to the presence of cracks or other damage for a
specified period of time (Wilkins 1982).
(iv) Current damage tolerance philosophy states that a structure can sustain lifetime
design loads in a damaged condition up to the level where the barely visible
damage is detectable (Sela and Ishai 1989).
(v) Damage tolerance for aerospace composites is defined as the ability to sustain
an impact event with barely visible impact damage (BVID) and retain
appropriate residual strength (Department of Defence 2002).
With the exception of the last definition pertaining to composite structures, it appears
that damage tolerance relates to an ability to withstand a loss of stiffness and strength
213 | P a g e
subject to initial flaws and to contain those flaws so that failure does not occur. It
should be noted that the aim of all design philosophies that are currently in use lead
toward a similar goal of ensuring that premature failure does not occur.
While damage tolerant design may be suitable for metallic structural components in
aircraft, the level of its adequacy and application in composites is still questionable.
Even though composite materials have been applied to many secondary structures, their
application to primary structures such as wing boxes, main spars and ribs is still done
with some hesitation due to the variety of failure mechanisms and the inability to
effectively manage manufacturing defects and operational damage. Therefore it is
important to consider if further improvements can be made to the Damage Tolerant
Design philosophy and whether the Robust Design philosophy is applicable.
8.3 Closing Remarks
In the field of aerospace structures, there have been numerous theories of ranging
complexity from extensive research with a great deal of effort spent in experiments and
numerical simulations in order to improve performance with no decrease in safety. Even
though the areas of imperfection sensitivity in postbuckling analysis and designing for
robustness have been addressed in this thesis, it is recognised that this is but ‘a brick
within a large wall’ of research. However it is hoped that the result of this thesis will
provide insight and direction to inform and inspire future analysis of composite
structures for the next generation of aircraft and beyond, that one can be confident that
Robust Design presents a viable option for increasing safety and improving structural
design in the aerospace industry.
214 | P a g e
References
References
Abramovich, H., Weller, T., & Bisagni, C. (2008). Buckling Behavior of Composite
Laminated Stiffened Panels under Combined Shear-Axial Loading. AIAA Journal
of Aircraft , 45 (2), 402-413.
Adali, S., Lene, F., Duvaut, G., & Chiaruttini, V. (2003). Optimization of Laminated
Composites Subject to Uncertain Buckling Loads. Composite Structures , 62 (3-4),
261-269.
Allen, J. K., Seepersad, C., Choi, H. J., & Mistree, F. (2006). Robust Design for
Multiscale and Multidisciplinary Designs. Journal of Mechanical Design , 128 (4),
832-846.
Amazigo, J. C. (1969). Buckling under Axial Compression of Long Cylindrical Shells

with Random Axisymmetric Imperfections. Quarterly of Applied Mathematics , 26,
437-566.
Andsten, R. S., & Vaurio, J. K. (1992). Sensitivity, Uncertainty and Importance

Analysis of a Risk Assessment. Nuclear Technology , 98 (22), 160-170.
Annicchiarico, W. (2007). Metamodel-assisted Distributed Genetic Algorithms Applied

to Structural Shape Optimization Problems. Engineering Optimization , 39 (7),
757-772.
Antonio, C. C., & Hofbauer, L. N. (2008). From Local to Global Importance Measures
of Uncertainty. Composite Structures , 85 (3), 213-225.
Arbocz, J., & Hilburger, M. W. (2005). Toward a Probabilistic Preliminary Design

Criterion for Buckling Critical Composite Shells. AIAA Journal , 43 (8), 1823-
1827.
215 | P a g e
References
Arbocz, J., & Hol, J. M. (1995). Collapse of Axially Compressed Cylindrical Shells
with Random Imperfections. Thin-Walled Structures , 23 (1-4), 131-158.
Arbocz, J., & Hol, J. M. (1990). Koiter's Stability Theory in a Computer Aided
Engineering (CAE) Environment. International Journal of Solids and Structures ,
26 (9-10), 945-973.
Baker, A. A., Jones, R., & Callinan, R. J. (1985). Damage Tolerance of Graphite/Epoxy
Composites. Composite Structures , 4 (1), 15-44.
Baker, A., Dutton, S., & Kelly, D. (2004). Composite Materials for Aircraft Structures
(2nd ed.). American Institute for Aeronautics and Astronautics.
Baker, J. W., Schubert, M., & Faber, M. H. (2008). On the Assessment of Robustness.
Structural Safety , 30 (3), 253-267.
Bauer, W. F. (1958). The Monte Carlo Method. Journal for the Soceity of Industrial and
Applied Mathematics , 6 (4), 438-451.
Benzeggagh, M. L., & Kenane, M. (1996). Measurement of Mixed-mode Delamination

Fracture Toughness of Unidirectional Glass/Epoxy Composites with Mixed-mode
Bending Apparatus. Composites Science and Technology , 56 (4), 439-449.
Bisagni, C. (2000). Numerical Analysis and Experimental Correlation of Composite

Shell Buckling and Post-buckling. Composites: Part B , 31 (8), 655-667.
Bisagni, C., & Cordisco, P. (2003). An Experimental Investigation into the Buckling
and Post-buckling of CFRP Shells under Combined Axial and Torsion Loading.
Composite Structures , 60 (4), 391-402.
Bisagni, C., & Cordisco, P. (2006). Post-buckling and Collapse Experiments of

Stiffened Composite Cylindrical Shells subjected to Axial Loading and Torque.
Composite Structures , 73 (2), 138-149.
Bolotin, V. V. (1962). Statistical Methods in the Nonlinear Theory of Shells. From a

paper presented at the seminar in the Institute of Mechanics of Academy of
Sciences, USSR, 1957, NASA TTF-85.
216 | P a g e
References
Borg, U. (2004). Compressive Strength of Fibre Composite with Porosity. Proceedings

of the 21st International Conference of Theoretical and Applied Mechanics.
Warsaw, Poland.
Borgonovo, E. (2007). A New Uncertainty Importance Measure. Reliability

Engineering and System Safety , 92 (6), 771-784.
Borgonovo, E. (2006). Measuring Uncertainty Importance: Investigation and

Comparison of Alternative Approaches. Risk Analysis , 26 (5), 1349-1361.
Busse, G., Wu, D., & Karpen, W. (1992). Thermal Wave Imaging with Phase Sensitive
Modulated Thermongraphy. Journal of Applied Physics , 71 (8), 3962-3965.
Camanho, P. P., & Davila, C. G. (2002). Mixed-mode Decohesion Finite Elements for
the Simulation of Delamination in Composite Materials NASA/TM-2002-211737.
NASA Langley Research Center, Virginia, USA.
Camotim, D., Silvestre, N., Gonçalves, R., & Dinis, P. D. (2006). GBT-Based Structural
Analysis of Thin-Walled Members: Overview, Recent Progress and Future
Developments. In M. Pandey, W. -C. Xie, & L. Xu (Eds.), Advances in
Engineering Structures, Mechanics & Construction (Vol. 140, pp. 187-204).
Springer Netherlands.
Canisius, T., Sorensen, J. D., & Baker, J. W. (2007). Robustness of Structural Systems -
A New Focus for the Joint Committee on Structural Safety (JCSS). Applications of
Statistics and Probability in Civil Engineering: Proceedings of the 10th
International Conference. Tokyo: Taylor and Francis Group, London.
Carrere, N., Rollet, Y., Retel, V., Boubakar, L., & Maire, J. -F. (2009). Composites
Structural Modelling with Uncertain Data. Composites Science and Technology ,
69 (1), 60-66.
Cederbaum, G., & Arbocz, J. (1996). Reliability of Shells via Koiter Formulas. Thin-
Walled Structures , 24 (2), 173-187.
Chamis, C. C. (2004). Probabilistic Simulation of Multiscale Composite Behavior.

Theoretical and Applied Fracture Mechanics , 41 (1-3), 51-61.
217 | P a g e
References
Chan, R. (2008). A Statistical Measure of Robustness for Aerospace Structures. Sydney,

Australia: School of Mechanical and Manufacturing Engineering, The University
of New South Wales.
Chen, W., Allen, J. K., Tsui, K.-L., & Mistree, F. (1996). A Procedure for Robust
Design: Minimizing Variations caused by Noise Factors and Control Factors.
Journal of Mechanical Design , 118 (4), 478-485.
Chiang, Y. J. (1996). Robust Design of the Iosipescu Shear Test Specimen for
Composites. Journal of Testing and Evaluation , 24 (1), 1-11.
Choi, H.-J., Austin, R., Allen, J. K., McDowell, D. L., Mistree, F., & Benson, D. J.
(2005). An Approach for Robust Design of Reactive Metal Powder Mixtures.
Journal of Computer Aided Material Design , 12 (1), 57-85.
Chryassanthopoulos, M. K., & Poggi, C. (1995). Stochastic Imperfection Modelling in

Shell Buckling Studies. Journal of Thin-Walled Structures , 23 (1-4), 179-200.
Chryssanthopoulos, M. K., Giavotto, V., & Poggi, C. (1995). Characterization of

Manufacturing Effects for Buckling-Sensitive Composite Cylinders. Composites
Manufacturing , 6 (2), 93-101.
COCOMAT (Improved MATerial Exploitation at Safe Design of COmposite Airframe

Structures by Accurate Simulation of COllapse) . (2004). Retrieved October 1,
2008, from http://www.cocomat.de/COCOMAT_Flyer.pdf
Crisfield, M. A. (1981). A Fast Incremental/Iterative Solution Procedure that handles

“Snap-through”. Computers and Structures , 13 (1-3), 55-62.
Crisfield, M. A. (1996). Nonlinear Finite Element Analysis of Solids and Structures

(Vol. 1). Chichester, West Sussex, UK: John Wiley & Sons Ltd.
Daniel, I. M., & Ishai, O. (1994). Engineering Mechanics of Composite Materials. New
York: Oxford University Press, Inc.
Degenhardt, R. (2009). COCOMAT Final Activity Report. Braunschweig: German

Aerospace Center (DLR).
218 | P a g e
References
Degenhardt, R., Kling, A., Klein, H., Hillger, W., Goetting, H. C., Zimmermann, R., et
al. (2007). Experiments on Buckling and Postbuckling of Thin-walled CFRP
Structures using Advanced Measurement Systems. International Journal of
Structural Stability and Dynamics , 7 (2), 337-358.
Degenhardt, R., Kling, A., Rohwer, K., Orifici, A. C., & Thomson, R. S. (2008). Design
and Analysis of Stiffened Composite Panels including Postbuckling and Collapse.
Computers and Structures , 86 (9), 919-929.
Department of Defense . (2002). Composite Materials Handbook: MIL-HDBK17-3F

(Vol. 3 ).
Doltsinis, I., Kang, Z., & Cheng, G. (2005). Robust Design of Non-linear Structures
using Optimization Methods. Computer Methods in Applied Mechanics and
Engineering , 194 (12-16), 1779-1795.
Donnell, L. H. (1934). A New Theory for the Buckling of Thin Cylinders under Axial
Compression and Bending. Transections of the American Soceity of Mechanical
Engineers , 56 (AER-56-12), 795-806.
Doreille, M., & Merazzi, S. (2008). Post-buckling Analysis of Composite Structures

with B2000++. Proceedings of the 2nd International Conference Buckling and
Postbuckling Behaviour of Composite Laminated Shell Structures. Braunschweig:
German Aerospace Center (DLR).
Elishakoff, I. (1979). Buckling of Stochastically Imperfect Finite Column on a

Nonlinear Elastic Foundation - A Reliability Study. Journal of Applied Mechanics ,
46, 5411-5416.
Elishakoff, I. (1978). Impact Buckling of Thin Bar via Monte Carlo Method. Journal of
Applied Mechanics , 45, 586–590.
Elishakoff, I., Cai, G. Q., & Starnes, J. J. (1994). Non-linear Buckling of a Column with
Initial Imperfection via Stochastic and Non-stochastic Convex Models.
International Journal of Non-Linear Mechanics , 29 (1), 71-82.
219 | P a g e
References
Elishakoff, I., Li, Y. W., & Starnes, J. J. (1995). Buckling Mode Localization in Elastic
Plates due to Misplacement in the Stiffener Location. Chaos, Solitons & Fractals ,
5 (8), 1517-1531.
Elishakoff, I., van Manen, S., Vermeulen, P. G., & Arbocz, J. (1987). First-Order
Second-Moment Analysis f the Buckling of Shells with Random Imperfections.
AIAA Journal , 25 (8), 1113-1117.
Ellingwood, B. R. (2005). Strategies for Mitigating Risk of Progressive Collapse.

Proceedings of the 2005 Structures Congress and the 2005 Forensic Engineering
Symposium. New York : ASCE.
Falzon, B. G., & Cerini, M. (2007). An Automated Hybrid Procedure for Capturing
Mode-jumping in Postbuckling Composite Stiffened Structures. Composite
Structures , 73 (2), 186-195.
Falzon, B. G., & Hitchings, D. (2003). Capturing Mode-switching in Postbuckling

Composite Panels using a Modified Explicit Procedure. Composite Structures , 60
(4), 447-453.
Federal Emergency Management Agency. (2002). Report 403 - World Trade Centre
Building Performance Study: Data Collection, Preliminary Observations and
Recommendations. New York, New York: Federal Insurance and Mitigation
Administration.
Fraser, W. B., & Budiansky, B. (1969). The Buckling of a Column with Random Initial
Deflections. Jounal of Applied Mechanics , 36, 233-240.
Frey, C. H., & Patil, S. R. (2002). Identification and Review of Sensitivity Analysis
Methods. Risk Analysis , 22 (3), 553-571.
Fu, G., & Frangopol, D. M. (1990). Balancing Weight, System Reliability and
Reliability in a Multiobjective Optimization Framework. Structural Safety , 7 (2-4),
165-175.
Geier, B. (1998). The Imperfection Sensitivity of Axially Compressed Thin-Walled

Circular Cylinders - Attempts to Define a Measure. Proceedings of the 39th
220 | P a g e
References
AIAA/ASME/ASCE/AHS/ASC Structural Dynamics and Materials Conference.

AIAA 98-1771, pp. 604-612. Long Beach, California: AIAA.
Geier, B., Meyer-Piening, H. -R., & Zimmermann, R. (2002). On the Influence of

Laminate Stacking on Buckling of Composite Cylindrical Shells subjected to Axial
Compression. Composite Structures , 55 (4), 467-474.
General Services Administration. (2003). Progressive Collapse Analysis and Design

Guidelines for New Federal Office Buildings and Major Modernization Projects.
Washington, D.C., USA: General Services Administration.
GOM GmbH. (n.d.). Measurement Systems. Retrieved 2008, from Industrial 3D

Measurement Techniques: www.gom.com
Gonçalves, R., & Camotim, D. (2004). GBT Local and Global Buckling Analysis of
Aluminium and Stainless Steel Columns. Computers and Structures , 82 (17-19),
1473-1484.
Helton, J. C. (1993). Uncertainty and Sensitivity Techniques for Use in Performance

Assesment of Radioactive Waste Disposal. Reliability Engineering and System
Safety , 42 (2-3), 327-367.
Hexcel. (2005). HexPly 8552 Epoxy Matrix Product Data. Retrieved June 2009, from
http://www.hexcel.com/NR/rdonlyres/B99A007A-C050-4439-9E59-
828F539B03A4/0/HexPly_8552_eu.pdf
Hilburger, M. W., & Starnes, J. H. (2002). Effects of Imperfections on the Buckling

Response of Compression-Loaded Composite Shells. International Journal of Non-
Linear Mechanics , 37 (4-5), 623-645.
Hong, T., & Teng, J. G. (2002). Non-linear Analysis of Shells of Revolution.

Computers and Structures , 80 (18-19), 1547–1568.
Huehne, C., Rolfes, R., Breitbach, E., & Tessmer, J. (2008). Robust Design of
Composite Cylindrical Shells under Axial Compression - Simulation and
Validation. Thin-Walled Structures , 46 (7-9), 947-962.
221 | P a g e
References
Iman, R. L. (1987). A Matrix-based Approach to Uncertainty and Sensitivity Analysis

for Fault Trees. Risk Analysis , 7 (1), 21-33.
Kalsi, M., Hacker, K., & Lewis, K. (2001). A Comprehensive Robust Design Approach
for Decision Trade-Offs in Complex Systems Design. Journal of Mechanical
Design , 123 (1), 1-10.
Kelly, D. W., Lee, M. C., Orifici, A. C., Thomson, R. S., & Degenhardt, R. (2009).
Collapse Analysis, Defect Sensitivity and Load Paths in Stiffened Shell Composite
Structures. Computers Materials and Continua , In Press.
Kennedy, J., & Eberhart, R. (1995). Particle Swarm Optimization. Proceedings of the
IEEE International Conference on Neural Networks (pp. 1942–1948). Piscataway,
NJ, USA: IEEE.
Kleiber, M., & Hien, T. D. (1992). The Stochastic Finite Element Method, Basic
Perturbation Technique and Computer Implementation. Chichester: J. Wiley and
Sons.
Kleijnen, J. P., & Sargent, R. G. (2000). A Methodology for Fitting and Validating
Metamodels in Simulation. European Journal for Operational Research , 120 (1),
14-29.
Koiter, W. T. (1945). On the Stability of Elastic Equilibrium. Ph.D. Thesis .

Amsterdam, The Netherlands: T. H. Delft.
Krueger, R. (2004). Virtual Crack Closure Technique: History, Approach, and

Applications. Applied Mechanics Reviews , 57 (1-6), 109-143.
Krueger, R., & Minguet, P. J. (2007). Analysis of Composite Skin-stiffener Debond

Specimens Using a Shell/3D Modeling Technique. Composite Structures , 81 (1),
41-59.
Krueger, R., Cvitkovich, M. K., O'Brien, T. K., & Minguet, P. J. (2000). Testing and
Analysis of Composite Skin/Stringer Debonding under Multi-axial Loading.
Journal of Composite Materials , 34 (15), 1263-2000.
222 | P a g e
References
Krueger, R., Minguet, P. J., & O'Brien, T. K. (2000). A Method for Calculating Strain
Energy Release Rates in Preliminary Design of Composite Skin/Stringer
Debonding under Multiaxial Loading. ASTM Special Technical Publication (1383),
105-128.
Kulkarni, R., & Ochoa, O. (2006). Transverse and Longitudinal CTE Measurements of
Carbon Fibers and their Impact on Interfacial Residual Stresses in Composites.
Journal of Composite Materials , 40 (8), 733-754.
Lee, M. C. (2005). Stochastic Analysis for Robust Design. Sydney, Australia: School of
Mechanical and Manufacturing Engineering, The University of New South Wales.
Lee, M. C., Kelly, D. W., Thomson, R. S., & Degenhardt, R. (2008). Imperfection
Investigation of Stiffened Composite Fuselage Panels for Postbuckling Analyses.
Proceedings of the 5th International Conference on Thin Walled Structures. 1, pp.
155-162. Brisbane, Australia: Queensland University of Technology.
Legrende, A.-M. (1805). Nouvelles méthodes pour la détermination des orbites des
comètes - Sur la Méthode des moindres quarrés.
Levy, M., & Salvadori, M. (1992). Why Buildings Fall Down. New York, New York:
W. W. Norton and Company.
Li, G., Azarm, S., Farhang-Mehr, A., & Diaz, A. R. (2006). Approximation of
Multiresponse Deterministic Engineering Simulations: A Dependent Metamodeling
Approach. Structural and Multidisciplinary Optimization , 31 (4), 260-269.
Lundquist, E. E. (1933). Strength Tests of Thin-Walled Duralumin Cylinders in

Compression. NACA Report No.473.
Marczyk, J. (2005). Ontospace: A Complexity Management System. Ontonix s.r.l. White

Paper.
Marczyk, J. (2000b). Stochastic Multidiscipline Improvement: Beyond Optimisation.

Proceedings of the 8th AIAA/NASA/USAF/ISSMO Symposium on Multidisciplinary
Analysis and Optimisation. Long Beach: AIAA.
223 | P a g e
References
Marczyk, J. (2000a). Validation of FE Models Using Experimental Data and Monte

Carlo Simulation. Proceedings from MCS 2000 International Conference on Monte
Carlo Simulation. Monte Carlo.
Marczyk, J., Hoffman, R., & Krishnaswamy, P. (2000). Uncertainty Management in

Automotive Crash: From Analysis to Simulation. Proceedings from ASME Design
Engineering Technical Conference. Baltimore: ASME.
Megson, T. H. (Aircraft Structures for Engineering Students). 2007: Butterworth-

Heinemann.
Metropolis, N., & Ulam, S. (1949). The Monte Carlo Method. Journal of the American
Statistical Association , 44 (247), 335-341.
Meyer-Piening, H. -R., Farshad, M., Geier, B., & Zimmermann, R. (2001). Buckling
Loads of CFRP Composite Cylinders under Combined Axial and Torsion Loading
– Experiments and Computations. Composite Structures , 53 (4), 427-435.
Mikulik, Z. (2008a). Application of Fracture Mechanics to Predict the Growth of Single

and Multi-level Delaminations and Disbonds in Composite Structures. Ph.D.
Thesis . Sydney, New South Wales, Australia: University of New South Wales.
Mikulik, Z., Kelly, D. W., Prusty, B. G., & Thomson, R. S. (2008b). Prediction of
Flange Debonding in Composite Stiffened Panels Using an Analytical Crack Tip
Element-based Methodology. Composite Structures , 85 (3), 233-244.
Mohamed, O. A. (2006). Progressive Collapse of Structures: Annotated Bibliography.

Journal of Performance of Constructed Facilities , 20 (4), 418-425.
MSC.Software Corporation. (2004). MSC.Robust Design. Retrieved January 2009, from

http://www.mscsoftware.com/assets/2951_RD2004JUNZZZLTDAT.pdf
Nair, R. S. (2006). Preventing Disproportionate Collapse. Journal of Performance of

Constructed Facilities , 20 (4), 309-314.
National Institute of Standards and Technology (NIST). (2005). Final Report on the
Collapse of World Trade Center Towers. Gaithersburg, Md, USA.
224 | P a g e
References
National Safety Transport Board. (1985a). Safety Recommendations A-85-133 through -

137. Washington, D.C.: NTSB.
National Safety Transport Board. (1985b). Safety Recommendations A-85-138 through

A-85-140. Washington, D.C.: NTSB.
Noguera, J. H., & Watson, E. F. (2006). Response Surface Analysis of a Multi-product

Batch Processing Facility using a Simulation Metamodel. International Journal of
Production Economics , 102 (2), 333-343.
Noor, A. K., Starnes, J. H., & Peters, J. M. (1997). Curved Sandwich Panels Subject to
Temperature Gradient and Mechanical Loads. Journal of Aerosapce Engineering ,
10 (4), 143-161.
O'Brien, T. K. (1988). Towards a Damage Tolerance Philosophy for Composite

Materials and Structures. NASA TM 100584.
Ochoa, O. O., & Reddy, J. N. (1992). Finite Element Analysis of Composite Laminates.
Dordrecht, The Netherlands: Kluwer Academic Publishers.
Ontonix S.r.l. (2008). About Us: The Company. Retrieved January 2009, from
http://www.ontonix.com/index.php?page=91&sub=97
Ontonix S.r.l. (2009). OntoSpace Data Sheet. Retrieved January 2009, from Measuring
and Managing Complexity A Global First By Ontonix:
http://www.ontonix.com/ardocCM/fput.php/9456a5ee760b8ab9df58ac53ef124d41
383788037628be4989b8ebfd17ed91a5cbb9a3184be60d8456b6ce9ade210d145ddf
1a2982f1ebd54bb443a7119bc3029e432dec5ac9b660d82afa8e3dbe0d9c/OntoSpac
e_2009_DataSheet.pdf
Orifici, A. C. (2007). Degradation Models for the Collapse Analysis of Composite

Aerospace Structures. Ph.D. Thesis . Melbourne, Victoria, Australia: Royal
Melbourne Institute of Technoogy.
Orifici, A. C., Thomson, R. S., Degenhardt, R., Bisagni, C., & Bayandor, J. (2009). A
Finite Element Methodology for Analysing Degradation and Collapse in
225 | P a g e
References
Postbuckling Composite Aerosapce Structures. Journal of Composite Materials ,

(In Press).
Orifici, A. C., Thomson, R. S., Degenhardt, R., Kling, A., Rohwer, K., & Bayandor, J.
(2008). Degradation Investigation in a Postbuckling Composite Stiffened Fuselage
Panel. Composite Structures , 82 (2), 217-224.
Orifici, A., Lee, M., & Thomson, R. (2008). COCOMAT Design and Analysis
Guidelines, Appendix 2 - Derivation of Design and Analysis Guidelines.
Papoulis, A., & Pillai, S. U. (2001). Probability, Random Variables and Stochastic
Processes (4th Edition ed.). McGraw-Hill Science/Engineering/Math.
Park, G.-J., Lee, T.-H., Lee, K. H., & Hwang, K.-H. (2006). Robust Design: An
Overview. AIAA Journal , 44 (1), 181-192.
Patel, R. V., & Toda, M. (1980). Quantitative Measures of Robustness for Multivariable
Systems. Proceedings of the Joint Automatic Control Control Conference. San
Francisco, USA: ASME.
Payne, R. M. (2007). A Knowledge-Based Engineering Tool for Aiding in the

Conceptual Design of Composite Yachts. Ph.D. Thesis . Sydney, New South
Wales, Australia: University of New South Wales.
Payne, R. M., & Kelly, D. W. (2006). Knowledge-Based Engineering and Yacht

Design. Proceedings of the 2nd High Performance Yacht Design Conference.
Auckland, New Zealand.
Pearson, C., & Delatte, N. (2005). Ronan Point Apartment Tower Collapse and its
Effect. Journal of Performance of Constructed Facilities , 19 (2), 172-177.
Phadke, M. S. (1989). Quality Engineering Using Robust Design. Prentice Hall P T R.
Piggott, M. R. (1995). The Effect of Fibre Waviness on the Mechanical Properties of

Unidirectional Fibre Composites: A Review. Composites Science and Technology ,
53 (2), 201-205.
226 | P a g e
References
Potter, K., Khan, B., Wisnom, M., Bell, T., & Stevens, J. (2008). Variability, Fibre
Waviness and Misalignment in the Determination of the Properties of Composite
Materials and Structures. Composites: Part A , 39 (9), 1343-1354.
Potter, K., Langer, C., Hodgkiss, B., & Lamb, S. (2007). Sources of Variability in
Uncured Aerospace Grade Unidirectional Carbon Fibre Epoxy Preimpregnate.
Composites: Part A , 38 (3), 905-916.
Raj, B. N., Iyengar, N. G., & Yadav, D. (1998). Response of Composite Plates with
Random Material Properties using FEM and Monte Carlo Simulation. Advanced
Composite Materials , 7 (3), 219-237.
Reis dos Santos, M. I., & Nova, A. M. (2006). Statistical Fitting and Validation of Non-
linear Simulation Metamodels: A Case Study. European Journal of Operational
Research , 171 (1), 53-63.
Riks, E. (1979). An Incremental Approach to the Solution of Snapping and Buckling

Problems. International Journal of Solids and Structures , 15 (7), 529-551.
Riks, E., Rankin, C. C., & Brogan, F. A. (1996). On the Solution of Mode Jumping
Phenomena in Thin-walled Shell Structures. Computer Methods in Applied
Mechanics and Engineering , 136 (1-2), 59-92.
Rittweger, A., Schermann, T., Reimerdes, H. -G., & Öry, H. (1995). Influence of
Geometric Imperfections on the Load Capacity of Orthotropic Stiffened and
Composite Shells of Revolution with Arbitrary Meridians and Boundary Coditions.
Thin-Walled Structures , 23 (1-4), 237-254.
Rollet, Y., Bonnet, M., Carrere, N., Leroy, F. -H., & Maire, J. -F. (2009). Improving the
Reliability of Material Databases Using Multiscale Approaches. Composites
Science and Technology , 69 (1), 73-80.
Roorda, J., & Hansen, J. S. (1972). Random Buckling Behavior in Axially Loaded
Cylindrical Shells with Axisymmetric Imperfections. Journal of Spacecraft , 9 (1),
88-91.
Ross, S. M. (2003). Introduction to Probability Models. Academic Press.
227 | P a g e
References
Rybicki, E. F., & Kaninen, M. F. (1977). A Finite Element Calculation of Stress

Intensity Factors by a Modified Crack Closure Integral. Engineering Fracture
Mechanics , 9 (4), 931-938.
Saltelli, A. (2002). Sensitivity Analysis for Importance Assessment. Risk Analysis , 22

(3), 579-590.
Saltelli, A., Tarantola, S., & Chan, K. P.-S. (1999). A Quantitative Model-Independent
Method for Global Sensitivity Analysis of Model Output. Technometrics , 41 (1),
39-56.
Sanchez, S. M. (2000). Robust Design: Seeking the Best of All Possible Worlds.
Proceedings of the 32nd Conference on Winter Simulation, (pp. 69-76). Orlando,
Florida, USA.
Sandgren, E., & Cameron, T. M. (2002). Robust Design Optimization of Structures

through Consideration of Variations. Computers and Structures , 80 (20-21), 1605-
1613.
Sargent, R. G. (1991). Research Issues in Metamodeling. Proceedings of the 1991

Winter Simulation Conference (pp. 888-893). Phoenix, USA: IEEE.
Sela, N., & Ishai, O. (1989). Interlaminar fracture toughness and toughening of
laminated composite materials: a review. Composites , 20 (5), 423-435.
Shiao, M. C., & Cristos, C. C. (1999). Probabilistic Evaluation of Fuselage-type

Composite Structures. Probabilistic Engineering Mechanics , 14 (1-2), 179-187.
Silva, N. M., Camotim, D., Silvestre, N., & Degenhardt, R. (2008). On the use of
Generalised Beam Theory to Analyse Buckling and Post-Buckling Behaviour of
Laminated CFRP Cylindrical Stiffened Panels. Proceedings of the 2nd
International Conference on Buckling and Postbuckling Behaviour of Composite
Laminated Shell Structures. Braunschweig: German Aerospace Center (DLR).
Singer, J., Arbocz, J., & Weller, T. (1998). Buckling Experiments: Experimental
Methods in Buckling of Thin-Walled Structures (Vol. I). New York: John Wiley &
Sons, Inc.
228 | P a g e
References
Singer, J., Arbocz, J., & Weller, T. (2002). Buckling Experiments: Experimental
Methods in Buckling of Thin-Walled Structures (Vol. II). New York: John Wiley &
Sons, Inc.
Singh, B. N., Iyengar, N. G., & Yadav, D. (2002). Stability of Curved Composite Panels
with Random Material Properties. Journal of Aerospace Engineering , 15 (2), 46-
54.
Sobol, I. M. (2003). Global Sensitivity Indices for Nonlinear Mathematical Models and
their Monte Carlo Estimates. Mathematics and Computers in Simulation , 55 (1),
271-280.
Sobol, I. M. (1993). Sensitivity Estimates for Nonlinear Mathematical Models.

Mathematical Modelling and Computational Experiment , 2 (1), 407-414.
Spearman, C. (1904). The Proof and Measurement of Association between Two Things.
The American Journal of Psychology , 15 (1), 72-101.
Suemasu, H., Kurihara, K., Arai, K., Majima, O., & Ishikawa, T. (2006). Compressive
Property Degradation of Composite Stiffened Panel due to Debonding and
Delaminations. Advanced Composite Materials , 15 (2), 139-151.
Suh, N. P. (2001). Axiomatic Design: Advances and Applications. Oxford University

Press.
Sunderasen, S., Ishii, K., & Houser, D. R. (1995). A Robust Optimization Procedure
with Variations on Design Variables and Constraints. Design Optimization , 24 (2),
101-117.
Systems. (1968, Mary 23). Systems Built Apartments Collapse. Engineering News-
Record .
Taguchi, G. (1986). Introduction to Quality Engineering: Designing Quality into

Products and Processes. Quality Resources.
229 | P a g e
References
Teng, J. G., & Hong, T. (2006). Postbuckling Analysis of Elastic Shells of Revolution
Considering Mode Switching and Interaction. International Journal of Solids and
Structures , 43 (3-4), 551–568.
Timoshenko, S. P., & Woinowsky-Krieger, S. (1959). Theory of Plates and Shells (2nd
Edition ed.). New York: McGraw-Hill Book Company.
Tomblin, J. S., Barbero, E. J., & Godoy, L. A. (1997). Imperfection Sensitivity of Fiber
Micro-Buckling in Elastic-Nonlinear Polymer-Matrix Composites. Journal of
Solids and Structures , 34 (13), 1667-1679.
Tsouvalis, N. G., Zafeiratou, A. A., Papazoglou, Z. Z., Gabrielides, N. C., & Kaklis, P.
D. (2001). Numerical Modeling of Composite Laminated Cylinders in
Compression using a Novel Imperfections Modeling Method. Composites: Part B ,
32 (5), 387-399.
Varna, J., Joffe, R., Berglund, L. A., & Lundstrom, T. S. (1995). Effect of Voids on
Failure Mechanisms in RTM Laminates. Composites Science and Technology , 53
(2), 241-249.
Walker, M., & Hamilton, R. (2005). A Technique for Optimally Designing Fibre-
Reinforced Laminated Plates with Manufacturing Uncertainties for Maximum
Buckling Strength. Engineering Optimization , 37 (2), 135-144.
Weibull, W. (1951). A Statistical Distribution Function of Wide Applicability. Journal

of Applied Mechanics , 18 (3), 293-297.
Wilkins, D. J. (1982). Failure Analysis and Mechanisms of Failure of Fibrous

Composite Structures. NASA Conference Publication, (pp. 67-94).
Will, J., Moeller, J.-S., & Bauer, E. (2004). Robustheitsbewertungen des Fahrkomfort-
verhaltens an Gesamtfahrzeugmodellen mittels stochastischer Analyse. Weimarer
Optimierungs- und Stochastiktage.
Woo, C. H., & Pujara, L. R. (1988). Measures of Robustness with Reduced Order
Observers - An Example. IEEE Proceedings of the National Aerospace and
Electronics Conference (pp. 559-565). Dayton, USA: IEEE.
230 | P a g e
References
Wullschleger, L., & Meyer-Piening, H. -R. (2002). Buckling of Geometrically

Imperfect Cylindrical Shells — Definition of a Buckling Load. International
Journal of Non-Linear Mechanics , 37 (4-5), 645-657.
Yadev, B., & Verma, N. (1997). Buckling of Composite Circular Cylindrical Shells
with Random Material Properties. Composite Structures , 37 (3-4), 385-391.
Yap, J. W., Scott, M. L., Thomson, R. S., & Hachenberg, D. (2002). The Analysis of
Skin-to-Stiffener Debonding in Composite Aerospace Structures. Composite
Structures , 57 (1-4), 425-435.
Yap, J. W., Thomson, R. S., Scott, M. L., & Hachenberg, D. (2004). Influence of Post-
buckling Behaviour of Composite Stiffened Panels on the Damage Criticality.
Composite Structures , 66 (1-4), 197-206.
Yedavalli, R. K. (1985). Purtabation Bounds for Robust Stability in Linear Steady State
Space Models. International Journal of Control , 42 (6), 1507-1517.
Yurgartis, S. W. (1987). Measurement of Small Angle Fiber Misalignments in

Continuous Fiber Composites. Composites Science and Technology , 30 (4), 279-
293.
Zadeh, P. M., Toropov, V. V., & Wood, A. S. (2009). Metamodel-based Collaborative

Optimization Framework. Structural and Multidisciplinary Optimization , 38 (2),
103-115.
Zang, C., Friswell, M. I., & Mottershead, J. E. (2005). A Review of Robust Optimal
Design and its Application in Dynamics. Computers and Structures , 83 (4-5), 315-
326.
Zhang, Z., & Taheri, F. (2004). Dynamic Damage Initiation of Composite Beams
Subjected to Axial Impact. Composites Science and Technology , 64 (5), 719-728.
Zimmermann, H. -J. (2000). An Application-Oriented View of Modeling Uncertainty.

European Journal of Operational Research , 122 (2), 190-198.
231 | P a g e
References
Zimmermann, R., Klein, H., & Kling, A. (2006). Buckling and Postbuckling of Stringer
Stiffened Fibre Composite Curved Panels - Tests and Computations. Composite
Structures , 73 (3), 150-161.
232 | P a g e
Appendix A Markov Chain Methodology
Appendix A
Markov Chain Methodology
Another methodology for determining robustness was developed during the course of
this research project: The first method, that is introduced here, employs Markov Chains
to quantify robustness. This was developed in a Bachelor of Engineering thesis by Chan
(2008) and was co-supervised by the author. The Markov Chain method is shown in
Section A.1. It is akin to simulation methods for the operational life of structures where
repeated sampling is conducted, and probability theory is employed to (determine the
dependence between each time step), in order to establish the rate of failure. The
structural robustness is ascertained based on the resulting sample size of the failures that
occurred during sampling. Chan’s algorithm has been shown to be suitable for
designing truss structures and is demonstrated in Section A.2.
A.1 Chan’s Markov Chain Methodology
This section includes a brief introduction to the Markov Chain theory and its application
in structural robustness. The quantifying of robustness using Markov Chains captures
the probability of an event such as structural failure occurring, subject to variations and
uncertainties. The methodology here is useful, as both reliability and robustness can be
determined.
233 | P a g e
A.1.1 Theory of Markov Chains
The Markov Chain is a stochastic process that takes on a finite number of possible
values (Ross 2003) and has the ‘Markov property’. In the case of this research,
Equations (3.1) to (3.7) can be applied to obtain the sample space for the analysis. The
Markov property asserts that, given the present state, the future states are independent of
the past states. Future states are reached through a probabilistic process instead of a
deterministic process. Mathematically, this can be expressed as:
(A.1)
In the case of long slender structural members subjected to a compressive force, where
buckling failure either occurs or it does not, the system can be considered to comprise
of finite states. The finite states will have known probabilities pij, where pij is the
probability of moving from state j to state i. These probabilities are also known as
‘transitional probabilities’. When the states are finite, Equation (A.1) reduces to:
(A.2)
In the case where multiple finite states exist, it is possible to express the transitional
probabilities in a matrix form, as shown below:
(A.3)
The transitional probability matrix in Equation (A.3) has the following properties:
(i) the matrix will always be square as it represents all the possible probability
transitions and
(ii) the sum of the probability transitions has to be equal to 1.
234 | P a g e
A.1.2 Generation of Operational Flow Chart
In order to obtain the transitional matrix, it is required that an operational flow chart for
the structure is first generated. The Operational Flow Chart (OFC) has been illustrated
in Figure A.1. Each event in the flow chart, such as Variation 1 and Modification 1,
represents a state in the transitional matrix. According to Chan, ‘Reliability is the ability
for structures to remain in Operation 1 while robustness is the ability for structures to
remain in Operation 2’.
Variation 1
Operation 1
Failure from Modification 1

Operation 1
Operation 2
Variation 2
States in Operation 1
States in Operation 2
Failure from
Operation 2
Figure A.1: Operational Flow Chart for transitional matrix.
In Chapter 2 it was found that, for a structure to be considered robust, it had to have the
ability to withstand unexpected changes to input parameters, such as the boundary
condition. In the case of Figure A.1, this would translate to the paths that loop about
Operation 2. In order for the structure to be considered absolutely robust, it must not
235 | P a g e
transition to the ‘Failure from Operation 2’ state. Five main states can be assigned to
represent the five major events that the structure is likely to encounter during its
operational life. These include:
(i) remaining in Operation 1,

(ii) failure during Operation 1,
(iii) a requirement for modification and return to Operation 1,
(iv) experiencing an unexpected change and entry into Operation 2 and
(v) failure during Operation 2.
The transitional probabilities of these five main states can be incorporated into a
transition probability matrix as follows:
(A.4)
where a is the probability that the system remains in Operation 1,

b is the probability that the system fails during Operation 1,
c is the probability that the system requires modification and returns to
Operation 1,
1 – a – b – c is the probability that an unexpected change is experienced and the
system enters into Operation 2,
d is the probability that the system returns to Operation 1 and
e is the probability that the system remains in Operation 2.
A.1.3 Obtaining Reliability and Robustness Indices
According to Chan, ‘Reliability is defined as the average time that the process will
remain in Operation 1, while robustness is defined as the average time that the process
will remain in Operation 2’. The mean time that the processes spend in the transient
states can be determined using an algorithm outlined in Ross (2003). According to the
236 | P a g e
algorithm, the first step is to obtain the transition probability matrix for transient states
only, where transient states exclude all absorbing states. In this case, the transition
probability matrix from the previous section is reduced by eliminating States Two and
Five, as they represent failures in the operation flow chart or absorbing states. The
transient transition probability matrix reduces to:
(A.5)
where v is the probability that the system remains in Operation 1,

w is the probability that the system is modified and put back to Operation 1,
x is the probability that the system enters Operation 2 due to unexpected
changes,
y is the probability that the system returns to Operation 1 and
z is the probability that the system remains in Operation 2.
Note that in the transient transition probability matrix, the row sum does not necessarily
equal unity. The next step is to apply the formula outlined in Ross (2003) to determine
the expected number of time periods that the Markov chain spent in each transient state.
The formula is given by:
(A.6)
where S is a square matrix showing the expected number of time periods spent in each
transient state,
I is the Identity matrix and
PT is the transient transition probability matrix.
Using this formula, the matrix S is determined as:
(A.7)
237 | P a g e
As mentioned above, the average number of times spent in each operation can be used
to determine reliability and robustness. Therefore reliability and robustness can be
quantified, respectively, as:
(A.8)
and
(A.9)
A.2 Analysis of a Truss Structure using Markov Chains
A.2.1 Geometry and loading conditions
The methodology from the previous section is demonstrated here. A simple redundant
truss structure, Truss A, is shown in Figure A.2(a) with the geometry, applied load and
boundary conditions. This structure is redundant and hence, being statically
indeterminate, the force in each member has to be calculated using the Total
Complementary Energy Method from Megson (2007). It is assumed that the structure is
pin-jointed. This structure undergoes two stages of failure; the compressive cross-
member (member BD) first buckles, followed by a buckle in member CD. It is assumed
that member BD does not fail when it buckles but continues to be constantly loaded at
the critical buckling load to a state where equilibrium can no longer be maintained. A
second structure, Truss B, as illustrated in Figure A.2(b), features no compressive cross-
member. Both trusses have been designed so that they are of similar nominal weight. A
comparison between the results from the two structures is made after the analysis.
238 | P a g e
B L L
C C
PB P
L L
D A
D
(a) Truss A (b) Truss B
Figure A.2: Pin-jointed truss structures subjected to a horizontal force.
As Truss A is statically indeterminate, the internal forces in the members must be

resolved using the Total Complementary Energy Method. The internal forces in the
members for Truss B are resolved using the Method of Joints. The forces in the
members of both trusses are shown in Table A.1.
Table A.1: Loading of members in Truss A and Truss B.

Truss A Truss B
Member Loading Member Loading
AB 0.396P AB 0
BC 0.396P BC 0
CD -0.604 CD -P
DA 0.396P DA 0
AC 0.854P AC P
BD -0.560P
As the members above are long and slender, the critical buckling load Pcr can be found
using the Euler buckling equation:
239 | P a g e
(A.10)
where E is the Young’s Modulus,

I is the moment of inertia and
l is the length of the member.
In Truss A, member BD will buckle first, followed by member CD. This is similar to
Operation 1, as shown in the Section A.1.2. Once member CD buckles, the structure
will no longer be in a truss configuration and will fail, as it is now a mechanism. In this
calculation, length (l) and Young’s Modulus (E) are kept constant at 1000 mm and 73.1
GPa, respectively. Therefore, the occurrence of buckling in member BD is dependent on
the applied load exceeding the critical buckling load as follows:
(A.11)
where r is the nominal radius of the member.
Member CD will buckle at the following load, if member BD buckles and the applied
force exceeds:
(A.12)
N
If an unexpected change is experienced by Truss A and member BD is lost, then the

structure enters Operation 2, as shown in Section A.1.2, and the critical buckling load
for member CD becomes:
240 | P a g e
(A.13)
N
A.2.2 Operational flow charts for Trusses
The Operational Flow Chart for Truss A is shown in Figure A.3.
Variation 1
Operation 1
Modification 1
Failure from
Operation 1
Operation 2
Variation 2
Failure from Operation 2

N
Figure A.3: Operational Flow Chart for Truss A.
In Truss B, member CD, which was identified as the critical compressive member, has
been thickened from a nominal radius of 10mm to 15.54mm in order to ensure parity in
weight with Truss A. The structure will fail once member CD buckles as equilibrium
can no longer be maintained. The critical buckling load for member CD of Truss B is
similar to the value shown in Equation (A.13). The OFC for Truss B is shown in Figure
241 | P a g e
A.4. Since there is no redundancy in the design, it can be seen that the structure fails if it
enters into Operation 2.
Variation 1
Operation 1
Failure from
Operation 1
Operation 2
Variation 2
Failure from Operation 2
Figure A.4: Operational Flow Chart for Truss B.
A.2.3 Stochastic boundaries for trusses
In these analyses, there is an assumption that the structures will encounter an

unexpected change with a sudden loss of the critical member due to impact damage.
This also assumes that the unexpected change is irreversible and the lost members
cannot be replaced. Load is considered as a uniform distribution from 1000 N to 10000
N. Table A.2 shows the stochastic boundaries for the analyses.
242 | P a g e
Table A.2: Stochastic boundary for Truss A and Truss B.

Truss Member Input Variable Mean Possible Range
A, B - Length, L (mm) 1000 NA
A, B All Young’s Modulus, E (MPa) 73100 NA
A, B All Radius, r (mm) 10 7 - 13
B CD Radius, r (mm) 15.54 10.88 - 20.20
A, B - Load, P (N) NA 1000 - 10000
A.2.4 Results from Markov Chain analysis
The results for the analyses can be found in Table A.3.
Table A.3: Results from Markov Chain analysis.

Truss A Truss B
Applied Load (N) Reliability Robustness Reliability Robustness
1000 0.852 1.00 0.852 1
2000 0.852 1.00 0.852 1
3000 0.835 0.887 0.852 1
4000 0.802 0.786 0.852 1
5000 0.739 0.504 0.852 1
6000 0.681 0.379 0.852 1
7000 0.538 0.235 0.852 0.980
8000 0.439 0.156 0.852 0.980
9000 0.333 0.105 0.852 0.961
10000 0.274 0.075 0.852 0.942
243 | P a g e
1.2
Normalised Reliability and Robustness
1
0.8
0.6
Reliability (Truss A)
0.4 Reliability (Truss B)
Robustness (Truss A)
0.2
Robustness (Truss B)
0
0 2000 4000 6000 8000 10000
Load (N)
Figure A.5: Plots of reliability and robustness of Truss A and Truss B.
A.2.5 Discussion and conclusions
From Figure A.5, it is obvious that Truss B is more reliable and robust. It was initially
expected that Truss A, being redundant, would be the design that was more reliable and
robust, however, when member CD in Truss B was thickened in order to have a similar
nominal weight with Truss A, the critical buckling load was also increased and this
shifted the robustness and reliability in favour of Truss B.
One way to expand the analysis, in order for realistic conditions to be met, is to
introduce an unexpected event where the outcome is not dominated by the critical
buckling load. For example, the unexpected event may be a sudden change in the
direction of the force and also the consideration of fatigue due to cyclic loading. In such
cases, it is expected that Truss A will be more reliable and robust as the actual forces in
each member are lower, and cross members in Truss A are able to interchangeably
transfer both tension and compression forces while the cross member in Truss B will
change from being a tension to a compression member. With regards to future work
using the Markov Chains, the method needs to be expanded beyond its current
application to truss structures.
244 | P a g e
Appendix B Assessment of Robustness for Two Composite Yachts
Appendix B
Assessment of Robustness for Two

Composite Yachts
In this appendix the hull designs for two composite yachts are compared. Monocoque
hull designs are generally light, enabling higher sailing speeds and rely on the loads
being carried by the composite hull skins. This is generally not an issue up to the point
when damage occurs on the hull structure. A yacht with multiple frames in the forward
hull is thought to be more durable, as loads can travel through the frame members even
if damage occurs on the hull. However the frames introduce additional weight into the
structure via the stiffening members, and as a result the sailing speed is generally
reduced.
Payne and Kelly (2006) have previously conducted research into the design of
composite yachts. The key feature that was identified in the design process was that the
yachts could be designed for speed, safety or cost. With these design drivers in place,
two yacht hull designs were achieved. The first yacht hull was designed for speed with a
monocoque hull, while the second designed with durability as the design driver had
multiple frames in the forward portion of the hull, from the back bow to amidships. It
was identified however that safety was difficult to quantify and an attempt was made to
identify the robustness of the designs in the face of uncertainty in seaway loads.
Redundancy and robustness are studied with the two yacht designs subjected to
slamming loads which are highly unpredictable operating conditions.
245 | P a g e
B.1 Design specifications for composite yachts
This section provides information on the basic geometry and material properties for the
yacht designs. The composite yachts were designed as part of a Ph.D. thesis by Payne
(2007) and the configurations were optimised based the criteria of cost, speed and
durability. The hull designs for the monocoque and multiple frame hulls can be seen in
Sections B.2 and B.3 respectively. The principal dimensions for the yachts fall are
shown in Table B.1.
Table B.1: Principal dimensions for composite yachts.

Length at waterline (m) 12.86
Beam at waterline (m) 2.8
Maximum draft (m) 3.5
Displacement (t) 7.8
Both of the yacht hulls have been modelled in MSC.Patran. In both designs the common
critical failure mode has been identified as buckling in the foredeck as the yacht is
subjected to slamming loads. The models have been analysed using the SOL 105 linear
buckling solver in MSC.Nastran. Damage has been introduced on to the yacht hulls by a
90% reduction of stiffness on the bottom of the boats.
The following analyses detail the effect of the damage, enabling the influence and
sensitivity of the variables to be obtained. Figure B.1 shows the area that is damaged.
Damage here is taken as the hull area below the critical top deck panel. The stiffness of
the damaged area is reduced by 90% in order to investigate the behaviour of the top
deck panel when slamming loads are encountered.
246 | P a g e
Critical top deck panel
Direction of damage
Figure B.1: Region of structure that damage is simulated.
Both yachts have been modelled with the similar type of carbon fibre composite and the
details of the stochastic boundaries can be found in Table B.2. A total of 26 samples
including the original sample were used in both analyses. A CoV of 5% has been used
for the material properties and a CoV of 2.5% has been used for the lamina orientations
and thicknesses. The composite materials used in this application are Gurit UD
Carbon/Epoxy and P600 Corecell. Although the material properties are confidential,
ballpark figures can be found in Daniel and Ishai (1994) and these have been applied
here.
Table B.2: Stochastic boundary for the composite yachts.

Carbon fibre
Poisson’s Ratio, υ 0.27 0.2295 - 0.3015
247 | P a g e

o
45 Lamina orientation (deg) 45 41.625 - 48.375
Aluminium Core
Young’s Modulus, E (MPa) 117 99.45 - 134.55
Poisson’s Ratio, υ 0.4 0.34 - 0.46
Core thickness (mm) 10 9.25 - 10.75
B.2 Composite yacht with monocoque hull design
Monocoque hulls can generally be found in the racing yachts the structures are highly
optimised to minimise weight, with a total focus on maximising the yacht’s
performance. The monocoque hull designs only have localised structure in way of large
point loads such as the mast step, keel attachment, shrouds and mainsheet. The
structures are designed to provide only the minimum required level of redundancy.
Figure B.2 shows the structure of the monocoque hull that is analysed in this section. As
can be seen in this design, there are no transverse stiffening members forward of the
mast-step. All loads applied onto the keel are transferred and carried by the skin. In
Figure B.3 it can be seen that a buckling mode is induced in the deck of the yacht,
forward of amidships once a slamming load is applied. Specifications of the laminate
are as follows:
[±45, 0, ±453, 02, ±45, C]s
248 | P a g e
Figure B.2: Isometric view of monocoque hull.
Direction of slamming load
Figure B.3: Buckling occurring on the top deck of monocoque yacht.
249 | P a g e
The results for the stochastic analysis of the monocoque hull can found in Table B.3.
The robustness of the Shear Modulii in the damaged hulls has increased while the
robustness of the Stiffness Modulii has decreased. Overall the damage has resulted in a
decrease of R.I.II from 11.7 to 10.6. A summary of the actual change in response, where
the change in buckling load is used as a criterion instead of the actual buckling loads,
can be found in Section B.4.
Table B.3: Results for the analysis of the monocoque yacht hull.
Scaled
Sensitivity
Undamaged
Carbon E11 0.176 9.71×10-7 0.138 41.3
Fibre E22 -0.249 -1.79×10-5 -0.184 21.9
υ12 -0.110 -0.513 -0.137 66.4
G12 0.248 4.33×10-5 0.316 12.7
G23 0.216 2.06×10-5 0.144 32.0
G31 0.280 2.56×10-5 0.183 19.5 11.7
Core E11 0.817 0.00583 0.681 1.80
υ12 -0.270 -0.511 -0.205 18.1
-7
Stiffness A11 0.203 3.88×10 0.193 25.5
B11 -0.362 -4.74×10-7 0.00322 -857
D11 0.253 7.19×10-9 0.169 23.4
Damaged
Carbon E11 0.528 2.04×10-6 0.289 6.54
Fibre E22 -0.168 -1.04×10-5 -0.107 55.6
υ12 -0.224 -0.714 -0.190 23.4
G12 0.304 3.86×10-5 0.282 11.7
-6
G23 -0.067 -5.97×10 -0.0419 358 10.6
G31 0.047 6.32×10-6 0.0451 473
Core E11 0.743 0.00487 0.569 2.37
υ12 -0.099 -0.315 -0.127 79.7
-7
Stiffness A11 0.536 7.08×10 0.353 5.29
250 | P a g e
Scaled
Sensitivity
B11 -0.477 -7.35×10-7 0.00500 -420
D11 0.556 1.28×10-8 0.302 5.96
Damage size -0.257 -0.000842 -0.0105 369
The metamodels for the buckling factors against the Shear Modulus, G23 and
Extensional Stiffness, A11 are shown in Figures B.4 and B.5 respectively. In the plots, it
is apparent that the introduction of damage has caused the Sensitivity to change. The
metamodel for the Shear Modulus in Figure B.5 shows the Sensitivity changing signs
from positive to negative due to the introduction of damage.
1.24
Monocoque
1.22 Yacht
1.2 (Undamaged)
1.18 Monocoque
Buckling Factor
1.16 Yacht
(Damaged)
1.14
Sensitivity
1.12
(Undamaged)
1.1
1.08 Sensitivity
1.06 (Damaged)
1.04
6000 6500 7000 7500 8000
Shear Modulus G23 (MPa)
Figure B.4: Metamodel of buckling factor against Shear Modulus G23 before and after
damage has been applied.
251 | P a g e
1.24
Monocoque
1.22 Yacht
1.2 (Undamaged)
1.18 Monocoque
Buckling Factor
Yacht
1.16
(Damaged)
1.14
Sensitivity
1.12 (Undamaged)
1.1
1.08 Sensitivity
1.06 (Damaged)
1.04
440000 460000 480000 500000 520000 540000 560000
Extensional Stiffness A11 (MPa)
Figure B.5: Metamodel of buckling factor against skin extensional stiffness A11 before and
after damage has been applied.
B.3 Composite yacht with multiple frame hull design
The multiple frame yacht is similar in design to cruising yachts which are often made
completely of E-glass. As a result these designs require smaller panel sizes, which result
in an increased weight. In cruising classes, where safety is a priority, it is often
perceived that the additional frames and stringers provide additional load paths, thereby
increasing the robustness to variations in the loading. However, in order for these
frames to be fully effective, greater care needs to be given to their bonding to the hull
shell, which has been an issue in the past with yachts that exhibit slamming damage due
to poor workmanship. In the analysis in this section, it is assumed that the bonding is
perfect and the investigation only involves the decrease in buckling factor as the yacht
hull is damaged.
The yacht seen below in Figure B.6 has the same hullform as the monocoque, with the
exception of transverse members, or frames, being included in the forward region of the
hull. Here the frames act as redundant members, transferring loads across the hull by
252 | P a g e
providing alternative loading paths. The frames increase resistance to the deck buckling
once a slamming load is applied. The laminate specification for the multiple frame yacht
is as follows:
[±457, C]s
Figure B.6: Isometric view of multiple frame hull.
253 | P a g e
Direction of slamming load
Figure B.7: Buckling occurring on the top deck of multiple frame yacht.
In Figure B.7, the deck can be seen buckling between the main transverse member and
the first frame forward of the deckhouse as a slamming load is applied on the bottom
keel. As with the monocoque yacht, the multiple frame yacht has the laminate stiffness
A11 as having the highest influence on the deck buckling both before and after damage is
included.
The results for the stochastic analysis of the multiple frame hull can found in Table B.4.
Again, these results appear to have a similar trend to those in the previous section where
the monocoque hulls were analysed. The damaged structure appears to be more robust
due to the increase in R.I.I. Overall the damage has resulted in a decrease of R.I. II from
7.34 to 6.34. A summary of the actual change in response, where the change in buckling
load is used as a criterion instead of the actual buckling loads, can be found in Section
B.4.
254 | P a g e
Table B.4: Results for the analysis of the multiple frame yacht hull.
Scaled
Sensitivity
Undamaged
Carbon E11 0.180 1.29×10-6 0.186 29.94
Fibre E22 0.139 5.10×10-6 0.053 137
υ12 -0.214 -1.06 -0.287 16.3
G12 0.171 1.92×10-5 0.139 41.9
G23 -0.321 -5.78×10-5 -0.418 7.45
G31 0.0112 -4.45×10-6 -0.032 -2822 7.34
Core E11 0.910 0.00828 0.986 1.11
υ12 0.102 0.316 0.128 76.53
Stiffness A11 0.371 2.11×10-6 0.563 4.79
B11 -0.216 -3.82×10-7 0.000 9530
D11 0.438 3.77×10-8 0.418 5.46
Damaged
Carbon E11 0.162 1.26×10-6 0.182 33.9
Fibre E22 0.163 6.36×10-6 0.06547 93.7
υ12 -0.221 -1.08 -0.290 15.6
-5
G12 0.174 2.10×10 0.152 37.9
G23 -0.296 -5.71×10-5 -0.413 8.20
G31 0.0133 -4.90×10-6 -0.0348 -2150
6.34
Core E11 0.910 0.00821 0.978 1.12
υ12 0.0602 1.26×10-6 5.13×10-7 32400000
Stiffness A11 0.366 2.07×10-6 0.552 4.95
B11 -0.206 -3.90×10-7 -0.000495 9820
D11 0.417 3.67×10-8 0.406 5.91
Damage size -0.782 -0.00114 -0.0142 90.0
However, unlike the monocoque yacht, the scaled Sensitivity has not changed
significantly, and this is reflected in the Sensitivity lines seen in Figure B.8 and B.9.
They appear to be almost parallel. In this design, it can be seen that the buckling factors
of all samples are decreased once damage is introduced into the hull skins.
255 | P a g e
1.25
1.2
Multiple Frame
Hull (Undamaged)
1.15
Buckling Factor
Multiple Frame
Hull (Damaged)
1.1
Sensitivity
(Undamaged)
1.05
Sensitivity
1 (Damaged)
0.95
0.35 0.37 0.39 0.41 0.43 0.45 0.47
Core Poisson's Ratio υ12
Figure B.8: Metamodel of buckling factor against Poisson’s Ratio of the core material
before and after damage has been applied.
1.25
1.2
Multiple Frame
Buckling Factor
1.15 Hull
(Undamaged)
1.1 Multiple Frame
Hull (Damaged)
1.05 Sensitivity
(Undamaged)
1 Sensitivity
(Damaged)
0.95
240000 250000 260000 270000 280000 290000 300000
Extensional Stiffness A11 (MPa)
Figure B.9: Metamodel of buckling factor against skin extensional stiffness A11 before and
after damage has been applied.
256 | P a g e
B.4 Discussion of results
A summary of the results for the analyses of the composite yacht hulls can be found in
Table B.5. In this table, the Influence and Sensitivity for each variable has been found
with respect to the change in buckling load. This change has been found by finding the
difference in buckling loads between the undamaged and damaged hulls. This has been
done as a difference between the R.I.I and R.I.II values for the in Tables B.3 and B.4 are
not an accurate measure of the robustness within the context of this appendix as the
measures of robustness have to be derived from the change in output response due to
damage.
It can be seen that the monocoque hull is more affected by damage when compared with
the multiple frame hull. The [R.I.]Xi value increases from 46.1 to 514 in the monocoque
and multiple frame hulls respectively. R.I. I increases from 11.5 to 129 while and R.I.II
has increased from 23.8 to 482.
Table B.5: Summary of results for analysis of monocoque and multiple frame hulls.
Scaled
Sensitivity
Monocoque Hull
Damage -0.970 -0.00179 -0.0224 46.1
Material
Carbon E11 0.396 9.41×10-7 0.134 18.9
Fibre E22 0.172 4.45×10-6 0.0458 127
υ12 -0.181 -0.138 -0.0369 150
G12 0.000684 -5.21×10-6 -0.0381 -38400
23.8
G23 -0.484 -2.55×10-5 -0.180 11.5
G31 -0.326 -1.69×10-5 -0.120 25.4
Core E11 -0.0999 -0.000827 -0.0967 103
υ12 0.245 0.186 0.0748 54.7
Stiffness Components
A11 0.372 2.82×10-7 0.140 19.2
257 | P a g e
Scaled
Sensitivity
B11 0.156 -2.18×10-7 0.00148 4340
D11 0.360 4.48×10-9 0.105 26.4
Multiple Frame Hull
Damage -0.539 -0.000289 -0.00361 514
Material
Carbon E11 -0.116 -2.92×10-8 -0.00421 2060
Fibre E22 0.139 1.26×10-6 0.0130 553
υ12 -0.0167 -0.0136 -0.00367 16300
482
G12 0.127 1.70×10-6 0.0123 638
G23 0.152 7.73×10-7 0.00559 1180
G31 0.0346 -4.51×10-7 -0.00321 -9020
Core E11 -0.132 -6.97×10-5 -0.00830 912
υ12 -0.313 -0.0609 -0.0247 129
Stiffness Components
A11 -0.0848 -4.24×10-8 -0.0113 1040
B11 -0.0550 -7.39×10-9 -9.40×10-6 1940000
D11 -0.0578 -1.02×10-9 -0.0113 1540
From visual inspection of the change in buckling load against damaged area plot in
Figure B.10, it can be seen that the monocoque hull is more sensitive to damage
compared to the multiple frame hull once the change in buckling load for the two
designs are compared. In the figure, the Sensitivity plots for each design is shown and it
is apparent that the monocoque hull is less robust due to the higher gradient of the line
as well as the larger amount of scatter in the first six units of damage.
258 | P a g e
0.03
0.02
Change in Buckling Load
Monocoque Hull
0.01
Multiple Frame Hull
0
Sensitivity
-0.01 (Monocque Hull)
Sensitivity (Multiple
-0.02 Frame Hull)
-0.03
0 5 10 15 20 25
Damaged Area (unit area)
Figure B.10: Plots for change in buckling factor against damaged area between
monocoque and multiple frame hull designs.
The examples using the yacht hull designs have shown the multiple frame yacht hull to
be more robust compared to the monocoque yacht hull. In the multiple frame hull, the
frames acted as redundant members, allowing the slamming load to be transferred
across to the other structural members. This was not the case in the monocoque yacht
hull and this resulted in it being more sensitive to the application of damage. It can also
be noted that the influence of damage is more dominant in the monocoque hull as
compared to the multiple frame hull. This can be seen with the samples of the
monocoque hull closely following the sensitivity curve compared to that of the multiple
frame hull.
An interesting response from the monocoque hull was that the buckling factor actually
increased initially as damage was applied. A reduction of stiffness at the bottom of the
yacht caused the neutral axis to move upwards towards the deck. This caused the
buckling factor of the deck to increase as the distance between the neutral axis and the
extreme fibre on the deck decreased. This increase in buckling factor no longer occurred
once the damage exceeded seven units of area.
259 | P a g e
B.5 Conclusion
The analysis of the composite yacht hulls utilized two significantly different designs.
The monocoque yacht is a sleek and light design, meant to win races. The Achilles heel
of the structure is that it is highly sensitive to damage on the hull. Monocoque hulls
being damaged by grounding, collisions whilst racing and premature buckling from
over-tensioning of the rigging have lead to the premature retirement of many racing
teams during prestigious events such as the America’s Cup. On the other end of the
spectrum, multiple frame hull yachts seldom win speed trophies but they are able to
transverse great distances in most weather conditions due to the built-in robust design. It
appears that structural redundancies provide alternative paths for loads to travel once the
transfer of load is no longer possible through one member.
260 | P a g e

Whole

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Whole

Uploaded by

Copyright:

Available Formats

Stochastic analysis and robust design of stiffened composite

Downloaded from http://hdl.handle.net/1959.4/44217 in https://

B.E. Aerospace Engineering (UNSW)

A thesis submitted in fulfilment of the requirements for the degree of

School of Mechanical and Manufacturing Engineering

The European Commission 6th Framework Project COCOMAT (Improved MATerial

To all my fellow friends in research at UNSW, never give up.

1.1 Background ........................................................................................................ 1

Chapter 2 Review of Literature ................................................................................... 19

2.1 Introduction to Stochastic Analysis .................................................................. 20

2.1.1 Stochastic modelling ................................................................................. 20

2.2 Introduction to the Robust Design Concept ..................................................... 26

2.2.1 Definition of a robust design ..................................................................... 27

2.3.1 Progressive failure ..................................................................................... 37

2.4 Variability in Fibre Reinforced Composites..................................................... 47

2.4.1 Analysis of composite laminates............................................................... 48

2.5 Elastic Behaviour of Thin-walled Structures ................................................... 56

2.5.1 Bifurcation buckling and postbuckling of shells ....................................... 56

2.6 Placement of the Research in this Thesis ......................................................... 66

Chapter 3 Theory and Methodology ........................................................................... 70

3.1 Stochastic Analysis ........................................................................................... 71

3.1.1 Sensitivity Analysis (SA) procedures ....................................................... 71

3.2 Formulation for Indicators of Robustness ........................................................ 76

3.3 Formulation for Robust Indices ........................................................................ 80

3.3.1 Derivation of the Robust Index R.I.I ......................................................... 80

3.4 A Hand Calculation of the Stochastic Methodology ........................................ 82

3.4.1 Cantilevered beam with edge loading ....................................................... 82

3.5 Effect of Statistical Properties on Robust Indices ............................................ 87

3.5.1 Stochastic analysis of cantilevered beams with varying CoV .................. 88

3.6 Conclusion ........................................................................................................ 92

Chapter 4 Experimental Setup and Results ............................................................... 94

4.1 Experimental Setup .......................................................................................... 96

4.1.1 Buckling test facility ................................................................................. 96

4.2 Design of the COCOMAT Panels .................................................................. 101

4.2.1 COCOMAT Design 1 (D1) panel ........................................................... 101

4.3 Imperfections and Variations of the D1 Panel ............................................... 106

4.3.1 Variations in material properties ............................................................. 106

4.4 Results from Cyclic Loading of D1 Panel ...................................................... 111

4.5 Discussion of Experimental Results for the D1 Panel.................................... 126

4.5.1 Degradation of panel from cyclic loading ............................................... 126

4.6 Results from Cyclic Loading of the D2 panel ................................................ 131

4.6.1 Impact damage on the P27 D2 panel....................................................... 131

4.7 Conclusion ...................................................................................................... 135

Chapter 5 Stochastic Analysis of the COCOMAT D1 Panel .................................. 137

5.1 Investigation of Geometrical Imperfections in the D1 Panel ......................... 138

5.2 Stochastic Analysis of D1 panel ..................................................................... 150

5.2.1 Pre-processing of D1 panel for analysis.................................................. 150

5.3 Conclusion ...................................................................................................... 161

6.1 Validation of Initial Assumptions .................................................................. 164

6.2 Study of Robustness for the D1 and D2 Panels .............................................. 173

6.2.1 Analysis of D1 panel ............................................................................... 174

6.3 Conclusion ...................................................................................................... 183

Chapter 7 Redesign of COCOMAT Panels using Robust Indices .......................... 185

7.1 The Effect of Stacking Sequence on Laminate Robustness ........................... 186

7.1.1 Design of laminates for the buckling study............................................. 186

7.2 Parametric Study of Four D2 Panel Configurations ....................................... 194

7.2.1 Laminate design for hybrid D2 panels .................................................... 195

7.3 Conclusion ........................................................................................................... 205

Chapter 8 Conclusion ................................................................................................. 206

8.1 Summary of Key Findings and Achievements ............................................... 206

8.1.1 Review of literature ................................................................................. 206

8.2 Future Work ................................................................................................... 210

8.3 Closing Remarks ............................................................................................ 214