(Computational and Experimental Methods in Structures 6) Aliabadi, Mohammad H - Woven Composites (2015, Imperial College Press)

Woven Composites
P984_9781783266173_tp.indd 1 30/12/14 11:32 am

Computational and Experimental Methods in Structures
Series Editor: Ferri M. H. Aliabadi (Imperial College London, UK)
Vol. 1 Buckling and Postbuckling Structures: Experimental, Analytical and

Numerical Studies
edited by B. G. Falzon and M. H. Aliabadi (Imperial College London, UK)
Vol. 2 Advances in Multiphysics Simulation and Experimental Testing of MEMS

edited by A. Frangi, C. Cercignani (Politecnico di Milano, Italy),
S. Mukherjee (Cornell University, USA) and N. Aluru (University of
Illinois at Urbana Champaign, USA)
Vol. 3 Multiscale Modeling in Solid Mechanics: Computational Approaches

edited by U. Galvanetto and M. H. Aliabadi (Imperial College London, UK)
Vol. 4 Boundary Element Methods in Engineering and Sciences

by M. H. Aliabadi (Imperial College, UK) and
P. Wen (Queen Mary University of London, UK)
Vol. 5 Mathematical Methods and Models in Composites

edited by V. Mantic (University of Seville, Spain)
Vol. 6 Woven Composites

edited by M. H. Aliabadi (Imperial College London, UK)

Catherine - Woven Composites.indd 1 11/12/2014 2:51:12 PM

Computational and Experimental Methods in Structures – Vol. 6
Woven Composites
Editor
M H Aliabadi
Imperial College London, UK
Imperial College Press

ICP
P984_9781783266173_tp.indd 2 30/12/14 11:32 am

Published by
Imperial College Press
57 Shelton Street
Covent Garden
London WC2H 9HE
Distributed by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
Library of Congress Cataloging-in-Publication Data

Aliabadi, M. H.
Woven composites / M.H. Aliabadi.
pages cm. -- (Computational and experimental methods in structures ; volume 6)
Includes bibliographical references and index.
ISBN 978-1-78326-617-3 (hardcover : alk. paper)
1. Fibrous composites. 2. Reinforced plastics. 3. Textile fabric. I. Title.
TA418.9.C6A365 2015
677'.4--dc23
2014046840
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.
Copyright © 2015 by Imperial College Press

All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy
is not required from the publisher.
Typeset by Stallion Press

Email: enquiries@stallionpress.com
Printed in Singapore
Catherine - Woven Composites.indd 2 11/12/2014 2:51:12 PM

January 6, 2015 11:41 Woven Composites 9in x 6in b1954-fm page v
TABLE OF CONTENTS
Chapter 1 Micromechanical Modelling of Textile Composites 1
Chapter 2 Mechanical Response and Failure of 2D Woven

Composites Under Compression 75
Chapter 3 From a Virtual Textile to a Virtual Woven

Composite 109
Chapter 4 Modelling 3D Woven Composite Preform

Deformations 141
Chapter 5 Cryomechanics and Crack Behaviour of Woven

Polymer Matrix Composites 159
Chapter 6 Effects of Fabrication Processes and Nanoclay

Inclusion on the Performance of Woven
Carbon/Epoxy Composite Laminates 179
Chapter 7 Textile Composite Forming Simulations 215
Index 239
v
May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws
This page intentionally left blank

January 6, 2015 10:13 Woven Composites 9in x 6in b1954-ch01 page 1
Chapter 1
MICROMECHANICAL MODELLING
OF TEXTILE COMPOSITES
Omar Bacarreza∗,† , Pihua Wen‡ and M.H. Aliabadi∗,§

∗ Department of Aeronautics, Imperial College London,
London, SW7 2AZ, UK
† o.bacarreza-nogales@imperial.ac.uk
§ m.h.aliabadi@imperial.ac.uk
‡ School of Engineering and Material Sciences,
Queen Mary, University of London, London, UK, E1 4NS,

P.H.Wen@qmul.ac.uk
1.1 Introduction
Composite materials make use of the different properties of their con-
stituents, microstructure and interaction between constituents in order to
improve their mechanical behaviour at higher scales.
Fibre-reinforced composites, and in particular, unidirectional (UD)
composites, are widely utilised in different aerospace and automotive appli-
cations to reduce the overall weight of the components due to the high
strength-to-weight ratio that they exhibit.
In recent years, textile composites such as woven, braided, knitted and
stitched fabrics have increasingly been used as structural materials in indus-
trial applications because they are efficient at reinforcing more directions
within a single layer and their ability to conform to surfaces with com-
plex curvatures. Furthermore, textile composites provide improved impact
resistance, exceptional thermal, fatigue and corrosion resistance and better-
balanced properties. Moreover, textile composites are easier and cheaper
to handle and fabricate when compared with UD composites. The textile
weave processes used to form 3D woven fabric composites (WFCs) are able
1
2 O. Bacarreza, P. Wen and M.H. Aliabadi
to produce large volumes at an ever faster rate [1], allowing relatively easy
and cost-efficient manufacturing.
However, the complex architecture of these composites leads to diffi-
culties in predicting the mechanical response necessary for product design.
In particular, their complex structure, undulation and interfaces, and their
hard to understand behaviour under different loading conditions, ensures
that evaluating their constitutive properties for structural design remains
challenging.
Woven fabric composites are a network of fibre tows which are later
impregnated to create plies, before being stacked in predetermined ori-
entations and finally cured to form the composite material. This process
operates within the tow interlacing level of composite design [2]. The fibre
yarns are identified as warp and weft tows; warp tows can be considered as
lengthwise yarns (0◦ ), while the weft tows (or fill tows) (90◦ ) are inserted
over and under warp yarns to produce the weave pattern. It is worth noting
that these are not necessarily perpendicular to each other; the braid angle
can be varied to alter the properties.
A unique property of WFCs is their ability to be moulded to fit complex
curvatures while maintaining their desirable properties. The variation in
weave type controls the mechanical interlocking of the fibres; this is the
feature that defines the “drape” of a composite (the ability of the fabric to
conform to the shape of the mould [3]). Similarly, both surface smoothness
and stability are also characteristics affected predominantly by weave type.
In Table 1.1 some of the most commonly used weaves with a comparison
of a number of characteristics are listed.
Design remains difficult for WFCs due to the complex undulation in
the geometry involved. The ability to accurately evaluate these properties
numerically through computational methods such as finite element analysis
Table 1.1. Comparison of weave type and properties.
Weave
type Stability Drape Porosity Smoothness Symmetry Low crimp
Plain + + ++ ++ +++ ++ +++++ ++

Twill +++ + + ++ + + ++ +++ +++ +++
Satin ++ +++++ +++++ +++++ + +++++
Basket ++ +++ ++ ++ +++ ++
Leno +++++ + + + + ++
+++++ = Excellent, ++++ = Good, +++ = Acceptable, ++ = Poor, + = Very Poor

Micromechanical Modelling of Textile Composites 3
(FEA) will greatly aid the design process. It will help shape the future of
composite design, as well as provide a non-destructive means of testing.
The predictions of mechanical properties for textile composites have
been heavily researched with most studies focusing on plain woven com-
posites. Early models for the analysis of woven lamina can be traced back
to the 1970s, when Halpin et al. [5] investigated the stiffness of 2D and 3D
composites. Later, the theory was improved by Chou and Ishikawa [6, 7],
starting from modified classical laminate theories, developing the mosaic
model, fibre undulation model and bridging model. These models were fur-
ther improved by Naik and Ganesh [8], Shembekar and Naik [9], and Naik
and Shembekar [10] by considering different parameters such as yarn thick-
ness, undulation and the gap between adjacent layers. The unit cell was
divided into slices and the slices were arranged in different combinations —
series-parallel and parallel-series, to attain lower bound and upper bound
properties. Later, Jiang et al. [11] applied the method of cells to deter-
mine the effective properties of plain woven composites, and Tanov and
Tabiei [12] presented an efficient model with a simpler geometrical descrip-
tion. Tabiei and Ivanov [13] developed a model that allowed modelling of
progressive failure. Other methods based on classical energy principles were
applied by Kregers and Malbardis [14] for a random 3D reinforced compos-
ite, and Pastore and Gowayed [15] used a stiffness averaging technique.
Finite elements analysis has also been employed to evaluate the
mechanical properties on the micro-level or constituent level, Zhang and
Harding [16] presented the first simplified 2D numerical model for plain
woven composites, later Paumelle et al. [17] investigated the 3D behaviour.
Woo and Whitcomb [18] also proposed a 2D model, and Chapman and
Whitcomb [19] extended an improved model to a 3D model. Zeman and
Šejnoha [20] included imperfection in their analysis. A meshless imple-
mentation for the prediction of the mechanical behaviour of woven fabric
composites was proposed by Wen and Aliabadi [21].
Three-dimensional FEA was used by Whitcomb and Srirengan to pre-
dict the failure of plain weave composites [22]. Barbero et al. developed
a numerical model capable of capturing damage using continuum damage
mechanics (CDM) [23], Bacarreza et al. [24] used a semi-analytical method
to predict the failure behaviour of plain woven composites at different scales,
and Wen and Aliabadi [24] studied progressive failure of plain woven com-
posites using meshfree simulations.
Twill woven composites have also been studied widely. Chaphalkar and
Kelkar [25] proposed an analytical model based on the classical laminate
theory to predict the mechanical behaviour of twill composites. Numeri-

cal models were proposed by Kwon and Roach [26], and Goyal et al. [2].
Nicoletto and Riva [27] analysed failure in twill weave laminates as well as
Ng et al. [28].
For 3D woven composites in which tows are interlaced in multiple
directions, the methods for predicting mechanical properties are more com-
plicated than for 2D composites as 3D woven composites have a more com-
plex architecture. In addition, it is reported that the lack of mechanical
characterisation data and difficulties in testing methods have been signif-
icant issues for 3D woven composites. The level of complexity and lack
of mechanical characterisation have led to slow progress of research in the
field of predicting mechanical behaviour for 3D woven composites. However,
some prediction approaches for elastic properties using the finite element
method (FEM) have been introduced in recent years. Bogdanovich [29] pro-
posed an FE approach using a 3D mosaic model, Stig and Hallström [30]
performed FE analysis for fully interlaced 3D fabrics. Non-crimp fabric
laminates were studied by Heß et al. [31] and knitted fabric composites by
Huysmans et al. [32]. Meshless methods are proving to be very efficient in
modelling these complex architectures and behaviours [33, 34]. Researchers
are also working to combine extended finite element methods (XFEM) with
binary modelling to achieve results at a faster rate [35].
Much research has been carried out to determine the properties
of composites, and various methods and models have been developed.
Broadly, these models can be classified into: analytical, semi-analytical and
numerical.
Analytical methods include the rules of mixtures based on the Voigt [36]
and Reuss [37] upper and lower bounds respectively, and their improve-
ments such as those developed by Hashin and Shtrikman [38] which
applies variational principles (variational bounding method); the improve-
ments also contain asymptotic homogenisation [39, 40] and mean field
approaches [41–43], also called effective medium approximation.
Semi-analytical methods are those where the global constitutive equa-
tions are evaluated from the local scale using explicit relations that link
microscopic and macroscopic behaviour. Semi-analytical methods comprise
the method of cells [44–46], transformation of fields analysis (TFA) [47–49]
and its extension, non-uniform TFA [50].
Numerical approaches capable of dealing with the increasingly complex
architectures of composites include unit cell methods [51–54] and methods
based on computational homogenisation [55–66].
In this chapter, semi-analytical and numerical methods for the

micromechanical analysis of textile composites are presented in Section 1.2.
Computational methods presented include the FEM and newly developed
meshfree method. The analysis of a unit cell for 2D plain woven with two
different geometries is discussed in Section 1.3.1. The FEA of twill, satin
and 3D woven composites are detailed in Sections 1.3.2, 1.3.3 and 1.3.4
respectively. A newly more realistic mathematical representation of 2D and
3D woven composites is also reported. Finally, damage at the unit cell
scale of plain woven composites using meshfree methods is examined in
Section 1.4.1 and multi-scale progressive failure analysis of plain woven com-
posites using semi-analytical homogenisation with the FEM is explained in
Section 1.4.2.
In this chapter repetitive unit cell (RUC) is defined as the smallest
part of a woven fabric composite that includes all the features of the fabric
which can be constructed by tessellating the RUC, i.e. the RUC repeats
periodically. The representative volume element (RVE) is an approximation
of the RUC using symmetries and antisymmetries.
1.2 Calculation of Effective Elastic Properties

1.2.1 Homogenisation using semi-analytical methods
Paley and Aboudi [45] introduced the generalised method of cells for the
micromechanical analysis of fibrous composites. It is capable of modelling
multi-phase composites with different types of phase arrangements and
architectures. Tabiei and Jiang [67] presented a model for plain woven
composites which divides the unit cell into subcells and then each subcell is
homogenised through the thickness via iso-stress and iso-strain compatibil-
ity conditions, the composite is then represented as an array of 2D subcells
with homogenised properties. The homogenisation of the 2D subcell array
is equivalent to the generalised method of cells [68].
Homogenisation of the representative volume element’s (RVE) mechan-
ical properties can be performed using semi-analytical homogenisation
based on the four-cell model proposed by Tanov and Tabiei [12], which
showed good agreement with experimental results for obtaining effective
elastic moduli. This micromechanical model is based on the method of
cells, which consists of dividing the RVE into different blocks or cells, each
of which can be further divided into subcells. These subcells are an ideali-
sation of actual subvolumnes of the RVE, i.e., part of a UD yarn bundle or
a pure matrix region.
Fig. 1.1. Geometry of the RVE of the four-cell model for plain woven composites.
In Figure 1.1 the geometry and main features of the RVE or four-cell
model, which can be used for homogenising elastic and damaged properties
of plain woven composites, is presented. The four subcells or blocks can be
distinguished: (1) Subcell “FF ”; (2) Subcell “FM ”; (3) Subcell “MF ” and
(4) Subcell “MM ”; where “M ” stands for matrix and “F ” for fibre.
The following assumptions are used during homogenisation:
(i) The matrix is isotropic

(ii) The yarn bundles are transversely isotropic with the principal axis
along the yarn axis
(iii) The contact between the constituents is perfect.
In order to homogenise the subcells, through the thickness, the mate-

rial properties for yarn bundles, pure resin and the strains on the subcells
should be known. The homogenised stiffness matrices for the subcells in
their local co-ordinate systems are calculated based on the parallel-series
compatibility conditions. The in-plane relations for these materials are in a
parallel (Voigt) arrangement and the out-of-plane relations are in a series
(Reuss) arrangement.
1
∆εab a b
x = ∆εx = ∆εx ∆σxab = (∆σxa + ∆σxb )
2
1
∆εab a b
y = ∆εy = ∆εy ∆σyab = (∆σya + ∆σyb )
2
1
∆εab
z = (∆εaz + ∆εbz ) ∆σzab = ∆σza = ∆σzb
2
1 a b
∆εab a b
xy = ∆εxy = ∆εxy
ab
∆σxy = (∆σxy + ∆σxy )
2
(1.1)
1
∆εab
yz = (∆εayz + ∆εbyz ) ab
∆σyz a
= ∆σyz b
= ∆σyz
2
1
∆εab
zx = (∆εazx + ∆εbzx ) ab
∆σzx a
= ∆σzx b
= ∆σzx
2
a, b can be a fibre or matrix, i.e., “FF ”, “FM ” and “MF ”.
The above assumptions are independent of stress-strain relations and
have been widely used in the estimation of mechanical properties of com-
posite materials. They lead to the homogenised constitutive properties of
each subcell in its local co-ordinate system.
Once the equivalent strains in the subcells are computed, the com-
ponents of strains and stresses in the constituents, and the stresses on the
subcells can be determined. The stresses and strains in each subcell and the
average stress in the RVE can be calculated once the average incremental
strains in the RVE are known.
The homogenisation of the RVE is then performed by means of the
following micromechanical conditions based on the homogenisation of the
2D subcell array.
Continuity conditions of strains:
ab
a Va ∆εx = ∆ε̄x a = F, M ; b = F, M
ab
b Vb ∆εy = ∆ε̄y a = F, M ; b = F, M
∆εab
z = ∆ε̄z a = F, M ; b = F, M
∆εab
xy = ∆ε̄xy a = F, M ; b = F, M
ab
(1.2)
a b Va Vb ∆εyz = ∆ε̄yz a = F, M ; b = F, M
Fb
Mb
b Vb ∆εyz = b Vb ∆εyz b = F, M
ab
a b Va Vb ∆εxz = ∆ε̄xz a = F, M ; b = F, M
aF
aM
a Va ∆εyz = a Va ∆εyz a = F, M
Continuity conditions of stresses:

∆σxab = ∆σxa b a = F, M ; b = F, M ; a = a

b Vb ∆σxab = ∆σ̄x a = F ; b = F, M

∆σyab = ∆σya b a = F, M ; b = F, M ; b = b
ab
a Va ∆σy = ∆σ̄y a = F, M ; b = F
ab
a b Va Vb ∆σz = ∆σ̄z a = F, M ; b = F, M

a
ab
b Va Vb ∆σxy = ∆σ̄xy a = F, M ; b = F, M (1.3)

ab
∆σyz = ∆σyzab
a = F, M ; b = F, M ; b = b
ab
a Va ∆σyz = ∆σ̄yz a = F, M ; b = F

ab
∆σxz = ab
∆σxz a = F, M ; b = F, M ; a = a
ab
b Vb ∆σxz = ∆σ̄xz a = M ; b = F, M
where Va , Vb are the volumes of the constituents “F ” or “M ”.

Using the original continuity conditions from [12] leads to premature
failure of the RVE in out-of-plain shear failure, even when the specimen is
loaded purely in the normal direction. Equations (1.2) and (1.3) already
include the improvements in the continuity conditions of the out-of-plane
shear stresses and strains.
1.2.2 Homogenisation using FE Analysis

The procedure of determining the effective elastic properties using FE anal-
ysis can be performed using computational homogenisation and following
the four-step homogenisation scheme proposed by Suquet [55]:
(i) Identification of the RVE where the mechanical behaviour of each indi-
vidual constituent is known
(ii) Application of correct microscopic boundary conditions on the RVE
from the macroscopic variables (macro–micro transition)
(iii) Calculation of the macroscopic response from the deformed microstruc-
tural behaviour of the RVE (micro–macro transition)
(iv) Finding the implicit relationship between macroscopic input and out-
put variables.
1.2.2.1 Geometric modelling

The accuracy of the predicted properties and damage characterisation of
the numerical model is strongly dependent on the quality of the geometrical
description. Often the models rely on accurate information of the tow

geometric parameters such as cross-sectional shape, path and position
within the textile structure, undulation function, etc. It has to be noted
that in practical cases, the modelled tow or preform may not replicate the
actual ones, due to various manufacturing parameters encountered, start-
ing from the fibre manufacturing to moulding and curing. However, past
studies have demonstrated that the basic assumptions are quite sufficient
to characterise the properties and failure modes to acceptable levels [69].
The actual choice of the repetitive unit cell (RUC) is an important task.
The RUC should be large enough to represent the microstructure without
introducing non-existing properties, and at the same time, it should be small
enough to allow efficient computational modelling, it can be defined as the
smallest microstructural volume that represents the overall macroscopic
properties of the material accurately enough. Figure 1.2 shows the RUC
according to the previous definition for plain woven and 3D orthogonal
composites.
Periodicity is not the only aspect in woven composites, symmetries and
antisymmetries can also be exploited to reduce the size of the RUC [70, 71],
this reduced RUC or representative volume element (RVE) is widely used
Fig. 1.2. Repetitive unit cell.

Fig. 1.3. Representative volume element.
in the analysis of advanced composites like the ones shown in Figure 1.3
for different types of composites.
Complex geometries are difficult to model using standard pre-
processing software for FEM. In order to create these architectures CAD
systems are recommended. Several CAD codes provide associative interfaces
with commercial FEM codes. This feature is very useful while carrying out
design iterations.
Unfortunately, degradation of the geometry quality can be observed
during translation. This shortcoming is illustrated in Figure 1.4 where edges
of the tow and matrix overlap after the CAD assembly is imported to the
Fig. 1.4. Poor geometry after file transfer.
Fig. 1.5. Hexahedral fibre and tetrahedral matrix — incompatible tie constraint.
FEM code, giving poor results, even after the highest possible accuracy is
set. Therefore, the correct choice of exported file format and compatibility
between the programs should be checked.
1.2.2.2 Mesh generation

When meshing the fibres with hexahedral elements and the matrix with
tetrahedral elements, the difference in the mesh densities at the interface
lead to tie constraints that are not compatible. The poor or non-existent
constraint between the fibre and matrix surfaces can lead to bad results
as can be seen in Figure 1.5, where a gap is formed between the matrix
and the yarn during the analysis, instead of both surfaces having the same
displacements. A possible way to solve this is by meshing both matrix and
fibres with the same type of elements or using cohesive elements in their
interface.
Meshing complex geometries can be extremely time-consuming, partic-
ularly when trying to use hexahedral elements only. The geometry of the
Fig. 1.6. Matrix geometry: (a) Problematic places, (b) Possible solution.
matrix constituent is usually very complex and generating a good mesh is

challenging. The most problematic places are where two or more yarns cross,
and at the top and bottom of the RVE where sharps edges are generated
making it hard or almost impossible to produce a hexahedral mesh of good
quality (see Figure 1.6a). Figure 1.6b shows a possible solution, where a
gap between the yarns is introduced and filled with matrix material.
1.2.2.3 Boundary conditions

The Hill–Mandel condition (or the energy averaging theorem) states that
the microscopic volume average of the variation of work performed on the
RVE is equal to the local variation of the work on the macro-scale. Three
types of RVE boundary conditions are commonly used:
(i) Prescribed displacements

(ii) Prescribed tractions
(iii) Prescribed periodicity.
and satisfy the macro-homogeneity condition a priori [62, 70].

Displacement boundary conditions are most commonly used in the
homogenisation approach. When they are applied, six types of displacement
boundary conditions are selected in order to determine the stiffness matrix:
each case can determine a row or column of the stiffness matrix. The values
of the initial volume, all components of stress and strain for each element
are the output from FE analysis, and the macroscopic stresses and strains
are calculated by homogenisation of the micro-scale fields.
Table 1.2 illustrates the recommended displacement boundary condi-
tions to be used for obtaining each row of the stiffness matrix, where the
co-ordinates of the RVE are defined in Figure 1.7.
Table 1.2. Boundary condition with each load.
Face
Loading case Displacement direction x=0 x=a y=0 y=a z=0 z=t
Longitudinal U1 0 l/a — — — —
tensile (11)
U2 0 0 0 0 0 0
U3 0 0 0 0 0 0
Transverse U1 0 0 0 0 0 0
tensile (22)
U2 — — 0 l/a — —
U3 0 0 0 0 0 0
Z-direction U1 0 0 0 0 0 0
tensile (33)
U2 0 0 0 0 0 0
U3 — — — — 0 l/t
In-plane shear U1 0 0 0 0 0 0
(12)
U2 0 l/a — — — —
U3 0 0 0 0 0 0
Out-of-plane U1 — — — — 0 l/a
shear (13)
U2 0 0 0 0 0 0
U3 0 0 0 0 0 0
Out-of-plane U1 0 0 0 0 0 0
shear (23)
U2 0 0 0 0 0 0
U3 — — 0 l/a — —
1.2.2.4 Determination of effective mechanical properties

Firstly, it is important to transform the co-ordinate system of the con-
stituents from their local to the global co-ordinate system of the assembly
in order to get the stresses of each part (constituent) with respect to the
global directions rather than the local direction of the individual parts.
Stresses, strains and element volumes are extracted and post-processed to
find the homogenised stress and strains.
Fig. 1.7. RVE co-ordinate system.
The constitutive equations for the 3D homogeneous composite can be

expressed as below [72]:
    

 σ  Q Q12 Q13 Q14 Q15 Q16  ε11 
 11 
   11


 



 σ 
22   21 Q22 Q23
Q Q24 Q25 Q26 



 
ε 

  
22
   
σ33 Q Q32 Q33 Q34 Q35 Q36  ε33
=  31  (1.4)

 σ 
  Q Q42 Q43 Q44 Q45 Q46  
ε12  


12

  41
 


 σ 
  Q Q52 Q53 Q54 Q55 Q56  

ε23 


  
 
23 51
 
σ13 Q61 Q62 Q63 Q64 Q65 Q66 ε13
where, σij and εij are the macroscopic stress and strain components respec-
tively. To determine the stiffness coefficients Qij , only one constant non-
zero strain component is applied on a particular surface, and corresponding
compatible restraints are imposed on all other surfaces in the finite element
model. From the numerical results, the stresses in different directions under
the aforementioned boundary conditions are solved. The volume average of
the stresses and strains are given by Eq. (1.5) and Eq. (1.6) respectively.
V is the total volume of the model and dV is the volume of each element.

∗ 1
σM = σm dV (1.5)
V V

∗ 1
εM = εm dV (1.6)
V V
∗
The volume averaged stresses and strains (σM and ε∗M ) for each loading
direction yields six equations, which are solved to obtain the values of the
stiffness matrix (Qij ). Due to the orthotropic nature, the matrix can be
reduced to:
 
Q11 Q12 Q13 0 0 0
Q 0 
 12 Q22 Q23 0 0 
 
Q Q23 Q33 0 0 0 
[Qij ] =  13  (1.7)
 0 0 0 Q44 0 0 
 
 0 0 0 0 Q55 0 
0 0 0 0 0 Q66
However, to relate the coefficients with the engineering constants, find-
ing the compliance matrix (Sij ) is more straightforward, which is given in
Eq. (1.8) and Eq. (1.9). The compliance matrix is the inverse of the stiffness
matrix.
    

 ε11 
 S11 S12 S13 0 0 0   σ 
 
    11 


   
 22 

ε  
 
S 12 S 22 S 23 0 0 0 
σ22 

ε33 S S23 S33 0 0 0  σ33
=  13  (1.8)

 ε12 
  0 0 0 S44 0 0   σ12 


     


 ε23 

  0 0 0 0 S55 0  
 σ23 


    
 
ε13 0 0 0 0 0 S66 σ13
 
1
E1 − νE212 − νE313
 ν12 
− E 1
− νE323 
 1 E2 
 ν13 
 − E1 − νE232 1


[Sij ] = 
E3  (1.9)
1 
 
 G12

 1 
 G23 
1
G13
1.2.3 Homogenisation using Meshfree methods

Meshfree methods have gained certain popularity as an alternative to
the FEM and have found applications in various problems. Among dif-
ferent kinds of meshfree methods proposed so far, see [73–81], element
free Galerkin (EFG) [76] and meshless local Petrov–Galerkin (MLPG) [80]
have received the most attention, and both use the moving least square
(MLS) approximation as the shape function construction. More recently,
Liu and Gu [81] introduced a point interpolation method which uses radial
basis functions (RBFs) to construct the shape functions. Wen and Aliabadi
[82–85] applied radial basis functions to the analysis of elastodynamic, func-
tionally graded materials and fracture mechanics problems. This section is
Fig. 1.8. Sketch of domains and boundaries of 2D cases for the meshfree method.
based on the works of Wen and Aliabadi [21], and Li et al. [34], who pre-
sented the first application of the meshfree method to woven composites.
1.2.3.1 Shape function constructions for the Meshfree method

Radial Basis Function Interpolation
Consider a problem domain Ω which is represented by a set of nodes in the
domain Ω and on the boundary Γ, as shown in Figure 1.8. Let u(A) be the
function of the field variable defined in the domain Ω. The approximation
of displacement u at point A(xA , yA , zA ) can be expressed by:

n
u(A) = Ri (A)ai = RT (A)a(A) (1.10)
i=1
where n is the number of nodes in the subdomain ΩA of point A, and

{ai }ni=1 , are the unknown coefficients to be determined for the basis function
Ri (A) corresponding to point A. Ri (A) is a radial basis function related
to the distance between point A and Ai . A classical form of radial basis
functions is the multi-quadric (MQ) basis proposed by Hardy [86]. For 3D
cases, it can be written as:

2 2
Ri (A) = c2 + a2x (xA − xi ) + a2y (yA − yi ) + a2z (zA − zi )2 (1.11)
where c is a free parameter (set to unity), and ax , ay and az are scale

factors.
RT (A) = {R1 (A), R2 (A), . . . , Rn (A)} (1.12)

a(A) = {a1 , a2 , . . . , an } (1.13)
From the interpolation of Eq. (1.10) for the RBFs, a linear system for
the unknown coefficients a is obtained:
u = R0 a (1.14)
where:
uT = {u1 , u2 , . . . , un } (1.15)
 
R1 (A1 ) R2 (A1 ) . . . Rn (A1 )
 R1 (A2 ) R2 (A2 ) . . . Rn (A2 ) 
 
R0 =  .. .. .. ..  (1.16)
 . . . . 
R1 (An ) R2 (An ) . . . Rn (An )
The matrix R0 is invertible since the RBFs are positive definite. Therefore:
a = R−1
0 u (1.17)
Eq. (1.10) can be rewritten as:

n
u(A) = RT (A)R−1
0 u = Φ(A)u = φi (A)ui (1.18)
i=1
where the matrix of shape functions Φ(A) is defined by:
Φ(A) = RT (A)R−1
0 (1.19)
or in the matrix form:
Φ(A) = [φ1 (A), φ2 (A), . . . φi (A), . . . , φn (A)] (1.20)
Since the term R−1

0 contains no variable, the derivatives of shape functions
against xq (where xq denotes xA , yA or zA ) can be obtained as:
Φ,q = {R1,q (A), R2,q (A), . . . , Rn,q (A)}R−1

0 (1.21)
where (),q denotes ∂()/∂xq

The first-order derivative of the radial basis function is written:
xq − xiq
Ri,q (A) = (1.22)
Ri (A)
Moving Kriging Interpolation

In this section Kriging interpolation will be used. Kriging is a geostatistical
method of spatial data interpolation named after D.G. Krige, who first
introduced a version of this spatial prediction process [87]. Kriging has been
extensively described by Sacks et al. [88] who proposed the application of

Kriging in computer modelling. Similar to the MLS approximation [89], the
Kriging approach can be extended to any subdomain ΩA ∈ Ω; we call it
moving Kriging (MK) [90].
Consider the function u(A) which is defined in the domain Ω and let
the approximation be uh (A). To approximate the distribution function u
in subdomain ΩA , over a number of nodes {Ai }n(i=1) , the MK interpolation
uh (A) is defined by:

m
uh (A) = pj (A)aj + Z(A) (1.23)
j=1
where Z(A) is assumed to be the realisation of a stochastic process with

mean zero, variance σ 2 and non-zero covariance [90].
Or:
uh (A) = pT (A)β + eT (A)E−1 (u − Pβ ) (1.24)
where P, E, e(A) and β are given by:

 
p1 (A1 ) p2 (A1 ) ... pm (A1 )
p (A2 ) p2 (A2 ) ... pm (A2 )
 1 
P=  .. . .
.. .. .. 
 (1.25)
 . . 
p1 (An ) p2 (An ) . . . pm (An )
 
1 E(A1 , A2 ) . . . E(A1 , An )
E(A2 , A1 ) 1 . . . E(A2 , An )
 
E= .. .. .. ..  (1.26)
 . . . . 
E(An , A1 ) E(An , A2 ) . . . 1
 

E(A1 , A)
e(A) = .. (1.27)
 . 
 
E(An , A)
β = (PT E−1 P)−1 PT E−1 u (1.28)
p(A) is a polynomial basis and a is the vector of coefficients. m denotes the

number of terms in the basis, pT (A) = {p1 (A), . . . , pm (A)}, for example,
a linear basis in 3D is given by:
pT (A) = {1, x, y, z}, m=4 (1.29)

The correlation function E(Ai , Aj ) is chosen to be the Gaussian function:

2
E(Ai , Aj ) = e−θrij (1.30)
where θ is the correlation parameter used to fit the model and chosen to

be 1, and rij can be found using Eq. (1.11)
Introducing the notation:
M = (PT E−1 P)−1 PT E−1 (1.31)

−1
N=E (I − PM) (1.32)
where I is an n × n unit matrix. Eq. (1.24) can be rewritten as:

n
uh (A) = pT (A)M + eT (A)N u = φi (A)ui = Φ(A)u (1.33)
i=1
where the shape function φi (A) is defined by:

m
n
φi (A) = pj (A)Mji + E(Ak , A)Nki (1.34)
j=1 k=1
The partial derivatives of φi (A) against xq can be obtained as follows:

m
n
φi,q (A) = pj,q (A)Mji + E,q (Ak , A)Nki (1.35)
j=1 k=1
The partial derivatives of the correlation function E(Ak , A) against

xq are:
E,q (Ak , A) = −2θ(xq − xiq )E(Ak , A) (1.36)
1.2.3.2 Galerkin Method and Hooke’s Law

For 3D cases, Hooke’s law can be generally written in a matrix form as
Eq. (1.4). Consider a domain Ω bounded by Γ. The total potential energy
for the plane stress is obtained as follows:

1
Π=U −W = ε σdΩ −
T T T
u bdΩ + u tdΓ (1.37)
2 Ω Ω Γ
where U is the initial elastic strain energy and W is the external energy,
i.e., the sum of contributions from known interior and boundary forces.
b = {b1 , b2 }T is the body force vector, and t = {t1 , t2 }T , in which ti = σij nj
is the vector of traction on the boundary and nj denotes the component of
a unit outward normal vector. The minimum value of total potential energy
with respect to each nodal displacement is considered, i.e.,
δΠ = δU − δW = 0 (1.38)
So we have the minimum potential energy principle in Galerkin weak

form:

δεT σdΩ − δuT bdΩ − δuT tdΓ = 0 (1.39)
Ω Ω Γ
n n
By the use of shape function, we have u = i=1 φi ui , ε = i=1 Bi ui ,
where for 3D cases, uT = {uxi , uyi , uzi }, n is the total number of nodes in
the subdomain and:
 
∂φi
0 0
 ∂x 
 
 ∂φ 
 0 i
0 
 ∂y 
 
 ∂φ 
 0 i
 0
∂z 
 
B =  ∂φ  (1.40)
 i ∂φi 
 0 
 ∂y ∂x 
 
 ∂φi ∂φi 
 0 
 ∂y 
 ∂z 
 
∂φi ∂φi
0
∂z ∂x
The global discrete system of the equation is obtained by substituting
the expression for all the displacement related components into the weak
form Eq. (1.31) to give:
KU = F (1.41)
where U is the displacement vector for all the nodes in the entire problem
domain and K is the global stiffness matrix for the problem domain, which
is defined as:

K= BT C(x, y, z) BdΩ (1.42)
Ω
It is worth noting that the constitutive matrix C(x, y, z) depends on

different locations in the model, in the matrix or in yarns, for example,
C(x, y, z) also varies according to the orientation of the fibres in the yarns.
The nodal force vector F is given as:

T
F= Φ bdΩ + ΦT tdΓ (1.43)
Ω Γσ
where Γσ denotes the boundary on which the traction is given. For concen-
trated forces acting at the node i, the nodal force vector can be determined
directly by:
F = {F1i , F2i , F3i } (1.44)
For isotropic material (i.e., the matrix part of the woven composites),
the constitutive matrix C(x, y, z) can be simply obtained:
 
C11 C12 C12 0 0 0
C 0 
 12 C11 C12 0 0 
 
C12 C12 C11 0 0 0 
C=  (1.45)
 0 0 0 C44 0 0 
 
 0 0 0 0 C44 0 
0 0 0 0 0 C44
where C44 = (C11 + C12 )/2.

Assuming a transversely isotropic material with axis of transverse
isotropy along the yarn axis, which is x and y, the constitutive matrix
C(x, y, z) will be as follows:
 
C11 C12 C12 0 0 0
C 0 
 12 C22 C23 0 0 
 
C12 C23 C22 0 0 0 
Cx = 
 0
 (1.46)
 0 0 C44 0 0 

 
 0 0 0 0 C55 0 
0 0 0 0 0 C44
 
C22 C12 C23 0 0 0
C 0 
 12 C11 C12 0 0 
 
C23 C12 C22 0 0 0 
Cy = 
 0
 (1.47)
 0 0 C44 0 0 

 
 0 0 0 0 C44 0 
0 0 0 0 0 C55
Fig. 1.9. Example of constitutive matrices for undulated yarns.
In terms of the composites with fibres with an undulation angle, θ, such

as in the x-z plane or in the y-z plane, as shown in Figure 1.9, the consti-
tutive matrix can be transformed by:
C(θ) = TE · C · TTE (1.48)
where TE is the transformation matrix as given below:

 2 2 2

t11 t12 t13 2t11 t12 2t12 t13 2t11 t13
 t221 t222 t223 2t21 t22 2t22 t23 2t21 t23 
 
 t2 t232 t233 2t31 t32 2t32 t33 2t31 t33 
TE =
t t
31 

 21
11 t12 t22 t13 t23 t11 t22 + t12 t21 t12 t23 + t13 t22 t11 t23 + t13 t21 
t21 t31 t22 t32 t23 t33 t21 t32 + t22 t31 t22 t33 + t23 t32 t21 t33 + t23 t31 
t31 t11 t32 t12 t33 t13 t11 t32 + t12 t31 t12 t33 + t13 t32 t11 t33 + t13 t31
(1.49)
In the case of 3D orthogonal woven composites, the undulation angleθ,
which is a function of the co-ordinates x and y, determines the orientation
of the fibres in the yarns, and is related to the geometry of the unit cell
model. As is the case of wavy fibres, the constitutive matrices, Eq. (1.48),
are local in a sense that their components are functions of co-ordinates
x and y, which makes an obvious difference from the case of a respective
composite having straight fibres.
1.3 Verification by Experimental Data

Detailed and simplified FE models were made for plain, twill, satin and 3D
woven composites in order to evaluate their elastic properties. The elas-
tic mechanical properties obtained from the results of the FE model were
compared with experimental data, semi-analytical approaches and/or other
numerical methods.
1.3.1 Plain Woven Composite

Carbon/epoxy and graphite/epoxy plain woven composites were selected in
order to compare the experimental results with micromechanical analysis.
1.3.1.1 Four-cell mosaic model

Assuming that the yarns in both direction (the fill and warp) have the
same structure and properties, the woven representative volume element
(RVE) can be constructed using the four-cell model as shown in Figure 1.1.
Therefore, only a quarter of the composite’s RUC needs to be considered
to derive the homogenised stiffness properties of the whole woven compos-
ite. The homogenised stiffness properties of the woven composite’s RVE
are then used to predict the stress–strain behaviour of the plain woven
composite at the lamina level.
To obtain an analytical solution of the mechanical properties and keep
the formulation simple, the cross-section of the fill and warp yarns are
assumed to be rectangular and their undulating form is approximated by a
horizontal and an inclined section at angle θ, where θ denotes the average
undulation angle.
From the geometry of the four-cell model, the height of the RVE can
be written as:
H = 4(1 − Vy ) tan θ (1.50)
where Vy is the overall yarn volume fraction and θ is the undulation angle.
The RVE is assumed to have unit width and depth. Analytical solutions
for the homogenisation of the stiffness properties in [12] were derived based
on the parallel-series assumptions. In this approach some of the strain
components of the adjacent cells are assumed to work in parallel and the
corresponding strains are equal, while for the rest of the strains, the cells
are assumed to work in series, i.e., the corresponding stress components are
equal and the strains are averaged to obtain the whole resultant strains. A
similar parallel-series approach can be applied to the whole RVE and thus
the subcell strains can be determined.
1.3.1.2 Smooth fabric modelling

A 3D smooth woven fabric for plain woven composite unit cells was
proposed by Wen and Aliabadi [24] for the determination of its elastic
properties, as shown in Figure 1.10. In this case, the yarns are UD, smooth
and distributed along the fill and warp directions.
Fig. 1.10. RVEs with smooth fibres and geometry of yarn volume for the model.
We assume that the yarn fibres in the fill direction on section y = 0

follow the curve:
H H πx
z1 (x) = z0 − + cos (1.51)
4 4 2
where z0 denotes the location of a fibre at x = y = 0. For the bottom surface

of the yarn, z0 = H/2 and the fibre position in the xoz plane becomes:
H πx
z1 (x) = 1 + cos (1.52)
4 2
Suppose the configuration for the yarn in the warp direction to be:
z2 (x) = αx2 (1.53)
The coefficient α is determined by considering the location of the joint for

fibres of fill and warp:

H πλ
α= 1 + cos (1.54)
4λ2 2
where λ denotes the co-ordinate of the joint shown in Figure 1.10.

Considering the portion of yarn volumes in the RVE, we have the
following integral:

λ
H πx
Vy = 2 1 + cos − αx2 dx (1.55)
0 4 2
Then we have the following equation to evaluate the location of the

joint:

π πλ πλ
4λ = 6Vy + 3 λ − sin − λ 1 − cos (1.56)
2 2 2
The root of λ can be determined by the alternative RVE with smooth
fibres and geometry of yarn volume for the model. For instance, if Vy = 0.26,
λ = 0.3902. Therefore, the top and bottom surfaces of the yarn in the warp
direction can be written as:
H πy
zbottom (x, y) = 1 − cos + αx2
4 2
H πx πy
ztop (x, y) = 2 + cos − cos (1.57)
4 2 2
and the slope of the fibres is:
dz Hπ πy
= sin (1.58)
dy 8 2
where the slope is necessary to determine the elasticity energy stored in the
yarns. Similarly to the yarn in the fill direction, the locations of the top
and bottom surfaces of the yarn in the warp direction can be written as:
H πx
ztop (x, y) = H − αy 2 − 1 − cos
4 2
H πx πy
zbottom (x, y) = 2 + cos − cos (1.59)
4 2 2
and the slope of these fibres is:
dz Hπ πx
=− sin (1.60)
dx 8 2
1.3.1.3 FE analysis
The FE model for AS4 3k/3501-6 is shown in Figure 1.11 and the input
material properties used in the FE model for AS4 3k/3501-6 are shown in
Table 1.3. The configurations mentioned above were used for making FE
models for homogenised subcells in both the detailed and simplified models.
The input material properties used were those described in the Composite
Material Handbook .[91]. The input material properties of yarn were linearly
adjusted from the material properties for UD composite material according
to the volume fraction of the fibre.
Fig. 1.11. FE models for the carbon/epoxy plain woven composite: (a) Detailed model,
(b) Simplified model.
Table 1.3. Material properties of

constituents for the carbon/epoxy
woven composite.
Symbol Value
Elastic properties of the matrix

Em 4350 MPa
νm 0.36
Elastic properties of the yarns
E1 170000 MPa
E2 = E3 12835 MPa
ν12 = ν13 0.33
ν23 0.27
G12 = G31 9060 MPa
G23 4825 MPa
In Table 1.4, the comparison of elastic properties obtained from FE

analyses, for both detailed and simplified models, the semi-analytical
approach introduced by Tanov and Tabiei [12], and the experimental results
for AS4 3k/3501-6 reported in the Composite Material Handbook [91] are
shown.
Good agreement is achieved between the longitudinal and transverse
elastic modulus (E11 and E22 ) obtained from FE analyses and the experi-
mental value. The FE values are slightly higher than the experimental value
(about 3%). Moreover, all values obtained from FE analyses were almost
the same as those from the semi-analytical approach. However, the shear
Table 1.4. Effective mechanical properties of the carbon/epoxy plain woven

composite.
E11 =E22 E33 G12 G13 =G2

(GPa) (GPa) (GPa) (GPa) ν12 ν13 = ν23
Detailed model 69.646 11.289 7.007 4.372 0.042 0.417

Simplified model 68.318 11.726 6.495 5.511 0.040 0.411
Semi-analytical 68.886 11.625 7.194 5.120 0.038 0.414
approach [12]
Experimental [91] 67.5 N/A N/A N/A N/A N/A
Fig. 1.12. FE models for the graphite/epoxy plain woven composite: (a) Detailed model,
(b) Simplified model.
moduli obtained from the detailed numerical model were slightly smaller
than the analytical values but higher than those for the simplified model.
The out-of-plane shear modulus (G13 and G23 ) obtained from the detailed
numerical model was the smallest of all approaches. The discrepancy in
the results for the shear moduli can be explained by the sharp change of
the yarn orientation in the simplified model when compared with a smooth
variation used in the detailed model.
To allow further comparison with more experimental results, a gra-
phite/epoxy woven fabric composite introduced in [11, 12, 92] was selected.
FE models for the graphite/epoxy composite are shown in Figure 1.12 and
the input material properties used in the FE model for the graphite/epoxy
composite are shown in Table 1.5.
The differences between the carbon/epoxy and graphite/epoxy plain
woven composites in the FE model were the size of the subcell, volume
Table 1.5. Material properties

of constituents for the graphite/
epoxy woven composite.
Symbol Value

Em 4511 MPa
νm 0.38
Elastic properties of the yarn
E1 137300 MPa
E2 = E3 10790 MPa
ν12 = ν13 0.46
ν23 0.26
G12 = G31 5394 MPa
G23 3695 MPa
Table 1.6. Effective mechanical properties of the graphite/epoxy composite.
E11 =E22 E33 G12 G13 =G2

(GPa) (GPa) (GPa) (GPa) ν12 ν13 = ν23
Detailed model 45.105 9.975 3.641 2.780 0.055 0.471

Simplified model 45.019 10.103 3.528 2.969 0.055 0.469
Semi-analytical 45.055 10.099 3.813 2.845 0.055 0.414
approach [12]
Experimental [92] 49.8 N/A 3.83 N/A 0.068 N/A
fraction of yarn and input material properties. The yarn constituents were
also assumed as transversely isotropic.
In Table 1.6 the comparison of elastic properties obtained from FE
analysis, the semi-analytical method and experiments for a graphite/epoxy
composite are shown. The longitudinal and transverse elastic modulus (E11
and E22 ) obtained from experiments was smaller than the others (about 9%)
and all values obtained from FE analysis were almost the same as the values
obtained from the analytical approach, while the in-plane shear modulus
(G12 ) obtained from FE analysis was slightly smaller than analytical and
experimental values, and the out-of-plane shear modulus (G13 and G23 )
obtained from the detailed model was the smallest of all approaches; the
trend was similar to the case of AS4 3k/3501-6.
1.3.2 Study of 2 × 2 Twill Woven Composites

In [93] the size of the model geometry (RUC vs. RVE) and the element
type (tetrahedral vs hexahedral) used to model a carbon/carbon 2 × 2 twill
Table 1.7. Material properties of

constituents for the carbon/carbon
twill composite.
Symbol Value

Em 10590 MPa
νm 0.15
Elastic properties of the yarns
E1 199500 MPa
E2 = E3 10790 MPa
ν12 = ν13 0.34
ν23 0.24
G12 = G31 7350 MPa
G23 4330 MPa
Fig. 1.13. Model size: (a) RUC, 9b) RVE.
woven composite were studied. The material properties for the constituents
are shown in Table 1.7 [94].
The geometries of the RUC and RVE are illustrated in Figure 1.13 and
the different element types used are shown in Figure 1.14. The procedure
described above was applied in order to obtain the effective material prop-
erties of the composite for the different models and the results are reported
in Table 1.8.
The mesh size was the same for the tetrahedral meshes (0.01 mm) and
half as big for the hexahedral mesh (0.005 mm). The hexahedral mesh for
the RVE consisted of 149275 nodes and 124156 linear elements while the
one for the tetrahedral had 24336 nodes and 105685 linear elements, leading
to a shorter computational time using the tetrahedral mesh employing a
similar number of elements.
Fig. 1.14. Mesh type: (a) Hexahedral elements, (b) Tetrahedral elements.
Table 1.8. Effective mechanical properties of the carbon/carbon twill woven composite.
E11 E22 E33 G12 G13 G23

(GPa) (GPa) (GPa) (GPa) (GPa) (GPa) ν12 ν13 ν13
RUC tet. 38.245 38.175 10.964 5.407 5.217 5.216 0.055 0.263 0.262
elements
RVE tet. 37.043 37.242 10.969 5.413 5.067 5.102 0.057 0.261 0.260
elements
RVE hex. 37.160 37.310 10.967 5.416 5.063 5.094 0.057 0.262 0.260
elements
The RUC was meshed only with tetrahedral elements containing 228262
nodes and 1048281 elements which increased the computational time sub-
stantially.
As can be seen the results obtained with different models or different
elements give similar results, but the orthotropic behaviour of this type
of composite can be better captured by the RUC model. The use of more
accurate boundary conditions on the RVE models can improve these results.
The computational cost for the RVE is considerably lower than the one
for the RUC, leading to reduced use of computational resources and time.
More studies dealing with non-linear behaviour composites are rec-
ommended to investigate if the reduced model (RVE) using tetrahedral
elements is accurate enough to capture this type of response.
1.3.3 Satin Weaves

It is clear (see Table 1.1) that satin weave fabrics provide a number of
beneficial design features for WFCs. Satin weaves are very flat as they
permit the closest possible weave and possess outstanding drape properties.
Table 1.9. Model parameters proposed by Dow et al. for the CLT
analytical model [96].
Weave Parameter 5-Harness 8-Harness
Ply thickness, h [mm] 0.27 0.27

Total volume fraction, Vf 0.64 0.64
Length, L [mm] 1.411 1.411
Fibre area, Af [mm2 ] 0.113 0.113
Volume fraction of fibre bundle, Vf b 0.6857 0.6857
Volume fraction packing of fibre bundle, Vf p 0.9334 0.9334
Area, A [mm2 ] 0.1645 0.1645
Fig. 1.15. Fibre dimension illustration.
Two models were considered, an 8-harness satin weave and 5-harness

satin weave [95]. To obtain results comparable to published data it was vital
to ensure the model geometry remained relatively similar to that tested
experimentally and analytically. WFC models have a number of variable
parameters, most importantly; tow cross-section, undulation shape, weave
path and volume fibre fraction. However, due to the varied nature of manu-
facture it is impossible to model the complex geometry precisely, thus some
assumptions and simplifications are made.
Original data provided by Dow [96] did not summarise the exact exper-
imental dimensions, however it did outline a number of parameters used
in the design of the analytical model for both 5- and 8-harness examples
based on microscope images of the experimental test samples. These are
summarised in Table 1.9, however they proved insufficient in determining
the necessary geometry to model.
Raju and Wang [97] proposed a simplified methodology to determine
the approximate dimensions of the undulation length for satin weave mod-
els. This was utilised and refined iteratively to ensure the correct volume
fibre fraction, outlined in Table 1.9, was achieved. Some parameters required
to size the unit cell for a given composite (a, h and ht ) are required as inputs;
while the undulation length (au ) can be derived from the volume fraction.
Figure 1.15 illustrates how these dimensions relate to the weave properties.
Table 1.10. Constituent material properties [96].
Properties Epoxy Resin (934) T300 Fibre
E11 (GPa) 3.448 48.95 71.7 94.45 117.2 139.95

E22 (GPa) 3.448 4.875 5.65 6.6 7.76 9.2
E33 (GPa) 3.448 4.875 5.65 6.6 7.76 9.2
G12 (GPa) 1.276 1.9 2.28 2.785 3.44 4.343
G13 (GPa) 1.276 1.9 2.28 2.785 3.44 4.343
G23 (GPa) 1.276 1.9 2.28 2.785 3.44 4.343
ν12 0.35 0.3 0.3 0.3 0.3 0.3
ν13 0.394 0.381 0.367 0.355 0.347 0.394
ν23 0.394 0.381 0.367 0.355 0.347 0.394
pd — 0.2 0.3 0.4 0.5 0.6
Fig. 1.16. Tetrahedral mesh in ABAQUS (matrix part hidden).
1.3.3.1 Element Type Study

Tetrahedral elements were used as they are more suited to the complex
geometry under consideration. Both linear and quadratic elements were
used with consistent mesh densities of 0.05 mm, and the material elastic
properties evaluated. The results obtained are shown in Table 1.11 together
with experimental results reported in [96].
When also compared with experimental data, linear tetrahedral
elements showed slightly superior results for tensile moduli; with just 1.4%
error compared with 2.22% error for the quadratic tetrahedral elements.
This difference is even less significant for other material elastic properties,
showing little or no variation. Moreover, considering the CPU time of each
loading case job (six per simulation), as shown in Table 1.11, it is clear the
quadratic elements demonstrate poor computational efficiency relative to
Table 1.11. Experimental quadratic and linear tetrahedral element comparison.
E11 E22 E33 G12 G13 G23 CPU time

(GPa) (GPa) (GPa) (GPa) (GPa) (GPa) ν12 ν13 ν13 (secs)
Exp. [96] 69.29 69.57 — 5.24 — — 0.056 — — —

Linear 68.32 68.89 10.06 6.47 3.713 3.712 0.047 0.4439 0.4478 258.18
Quadratic 67.75 68.38 9.989 6.43 3.664 3.665 0.047 0.4451 0.4499 11506.0
the accuracy of the result and take almost 45 times longer per job. This
is in part due to the linear shape functions ability to allow analytical inte-
grations, while higher order elements generally require numerical solutions,
which are computationally expensive.
Consequently, tetrahedral linear elements were deemed a sufficient com-
promise in terms of computational efficiency and accuracy, and will be
applied throughout further investigations.
1.3.3.2 Mesh convergence study

As described in the previous section, tetrahedral linear elements were
selected for additional investigations due to their superior run-time. One of
the main benefits of this improved speed is the ability to use more refined
meshes, in theory the finer the mesh is the greater the accuracy of the result.
However, a compromise must again be made with regard to computational
expense and accuracy; this is done via a mesh convergence study. Consider-
ing a critical result, E11 in this case, a curve will be plotted comparing an
increasing number of elements, convergence should occur at a given mesh
density, i.e., the result will plateau.
Table 1.12 shows the results from the convergence study, for mesh den-
sities varying from 0.04 mm to 0.5 mm. In Figure 1.17 the convergence of
tensile modulus (E11 ) versus the number of elements is presented; it is clear
Table 1.12. Mesh convergence study results.
Mesh
Density E11 E22 E33 G12 G13 G23
(mm) (GPa) (GPa) (GPa) (GPa) (GPa) (GPa) ν12 ν13 ν13
0.5 70.85 70.69 10.87 4.235 4.142 4.14 0.0436 0.4399 0.4396
0.1 69.27 69.79 .3 6.575 3.813 3.812 0.0472 0.4349 0.4379
0.075 68.98 69.54 10.22 6.537 3.787 3.788 0.047 0.4367 0.4407
0.05 68.32 68.89 10.06 6.47 3.713 3.712 0.0468 0.4439 0.4478
0.04 68.54 69.12 10.11 6.482 3.723 3.723 0.0473 0.4366 0.4411
Fig. 1.17. Mesh convergence.
Fig. 1.18. CPU time.
from the plateau that the model converges at a mesh density of approxi-
mately 0.05 mm. Furthermore, Figure 1.18 presents the computational time
versus the number of elements, as expected, the computational time rises
significantly as the number of elements increases. An acceptable computa-
tional time of 258.18 seconds (4.3 minutes) is achieved with the converged
mesh density of 0.05 mm; consequently, this was selected as the optimal
mesh density for the subsequent investigations.
1.3.3.3 Validation of results and comparison to analytical models

A number of analytical models were developed and tested based on the
experimental model considered here, using both 5- and 8-harness. To
validate the results and the practicality of FEA for the purpose of evaluating
the composite material elastic moduli, it is vital the results are tested via
comparison to experimental data.
Furthermore, to emphasise the improved accuracy FEA can bring, the
results will be compared with the three analytical models; these include the
fabric analysis method [98], 5-harness CLT model [97] and crude mosaic
model [96]. Tables 1.13 and 1.14 show the full results of each model relative
to the limited experimental data for the 5- and 8-harness satin weaves
respectively.
Table 1.13. 5-harness satin weave validation.

E11 E22 E33 G12 G13 G23
Experi- 69.29 69.57 — 5.24 — — 0.06 — —

mental [96]
Fabric 65.99 65.99 11.37 5.03 5.03 4.96 0.03 0.32 0.32
analysis
[98]
CLT 64.77 64.77 — 4.894 — — 0.038 — —
model
[97]
Mosaic 79.7 79.7 — 6.16 — — 0.051 — —
model
[96]
FEA [95] 69.101 69.63 10.98 5.143 4.900 4.902 0.038 0.449 0.452
Table 1.14. 8-harness satin weave validation.

E11 E22 E33 G12 G13 G23
Experi- 73 71.353 — 5.561 — — 0.054 — —

mental
[96]
Fabric 66.734 66.734 11.444 5.03 5.03 4.96 0.03 0.32 0.32
analysis
[98]
CLT 65.82 65.82 — 4.894 — — 0.038 — —
model
[97]
Mosaic 80.04 80.04 — 6.2046 — — 0.052 — —
model
[96]
FEA [95] 72.631 72.687 10.799 5.1411 4.7105 4.7138 0.0374 0.4275 0.4279
Table 1.13 shows moderate correlations for all models, however FEA
produced significantly less error for longitudinal, transverse and shear
moduli. For the in-plane Poisson’s ratio FEA showed poor results, high-
lighting a potential flaw in the model possibly caused by the use of linear
tetrahedral elements. These are susceptible to locking, i.e., when the mesh
is excessively stiff and results in smaller, less realistic displacements.
The mosaic model tends to overestimate most properties as shown
by the negative percentage error, however all others including FEA,
underestimate the elastic properties. As all other methods appear to dis-
agree with the in-plane Poisson’s ratio result, it is possible the discrepancies
originate from experimental errors.
The data for 8-harness shows very similar trends, however the errors
are marginally larger than that of the 5-harness. Overall FEA has demon-
strated superior accuracy and proven to give very positive correlations to
experimental data, validating its use as a tool to evaluate elastic properties.
1.3.3.4 Undulation variation study

Distinctive characteristics of woven fabric composites are their unique undu-
lations, consequently it is vital to understand how these parameters affect
the elastic properties. Satin weaves are of particular interest due to their
tight weaves; these are considered the defining property of WFC.
A parametric sinusoidal function was used for the undulation trajectory.
This allowed for a simple modification of the weave trajectory; however as
a new CAD model is required for each iteration, the ABAQUS model must
also be repeatedly set up. Fortunately, ABAQUS software features a Python
scripting interface, allowing the code generated during the first model set
up to be reapplied to another geometry. Although this saved considerable
time, it was inconsistent in the selection of some faces due to geometry
changes and therefore, thorough checking of all models made using this
technique is advised.
A parametric investigation applying a progressive extension of the
undulation length (au ) was conducted during this study. Figure 1.19 plots
the variation of sinusoidal parametric equations used to define the undu-
lation, it can be seen that this increases au in 0.1 mm increments. Fur-
thermore, two additional parameters were recorded from the Pro/Engineer
models for comparison, firstly the crimp angle (θ), defined as the maximum
absolute value of inclination angle between an adjacent yarn path peak
and trough [99], see Figure 1.20; this represents the extreme of deviation
Fig. 1.19. Undulation variations.
Fig. 1.20. Crimp angle.
Fig. 1.21. Crimp ratio.
from the laminate plane and is therefore expected to influence strength and
stiffness properties. Secondly, the crimp ratio is defined as the ratio of the
excess length of yarn (woven yarn length) within a fabric, to the length of
the fabric (i.e., straight yarn length), see Figure 1.21.
This study was applied to both the 5- and 8-harness models, and the
results can be found in Tables 1.15 and 1.16 respectively. The crimp angle
January 6, 2015
38
10:13
O. Bacarreza, P. Wen and M.H. Aliabadi
Woven Composites
Table 1.15. 5-harness undulation variation results.
Sin Crimp Crimp E11 E22 E33 G12 G13 G23
(angle/2) angle (θ) Ratio (GPa) (GPa) (GPa) (GPa) (GPa) (GPa) ν12 ν13 ν13
0.1575 43.3 1.0588 69.1 69.63 10.99 5.1436 4.900 4.903 0.0379 0.4488 0.4525
0.1775 40.4 1.03814 70.48 70.4 10.86 5.1486 4.920 4.918 0.035 0.4512 0.4509
0.1975 37.5 1.01853 71.03 71.73 10.8 5.1498 4.931 4.952 0.0342 0.443 0.4515
9in x 6in
0.2175 35.3 1.00873 72.28 72.23 10.64 5.1488 4.956 4.951 0.0343 0.4501 0.4504
0.2375 32.6 1.00480 73.26 73.06 10.62 5.1613 4.993 5.000 0.0345 0.4465 0.4469
b1954-ch01
page 38
January 6, 2015
10:13
Micromechanical Modelling of Textile Composites
Woven Composites
Table 1.16. 8-harness undulation variation results.
Sin Crimp Crimp E11 E22 E33 G12 G13 G23
(angle/2) angle (θ) Ratio (GPa) (GPa) (GPa) (GPa) (GPa) (GPa) ν12 ν13 ν13
0.1575 43.3 1.0588 72.63 72.69 10.8 5.1411 4.710 4.714 0.0374 0.4275 0.4279
0.1775 40.4 1.03814 73.59 73.6 10.79 5.1482 4.752 4.753 0.0369 0.4268 0.4271
0.1975 37.5 1.01853 73.99 73.99 10.69 5.14295 4.752 4.754 0.0369 0.4256 0.4257
9in x 6in
0.2175 35.3 1.00873 74.62 74.60 10.61 5.1389 4.765 4.77 0.0367 0.4258 0.4254
0.2375 32.6 1.00480 75.11 75.11 10.57 5.1436 4.780 4.782 0.0367 0.4224 0.4225
b1954-ch01
39
page 39
Fig. 1.22. Crimp angle variation study.
and crimp ratio show little to no difference in trends, so for the purpose
of this study crimp angle will be the focus of the comparison. Figure 1.22a
to Figure 1.22f show the variation in properties with crimp angle; again
note the data for E22 , G23 and ν23 was omitted as it showed little variation
relative to the corresponding properties of E11 , G13 and ν13 respectively.
Initial results show similar trends for both 5- and 8-harness results
as anticipated, with greater magnitudes in the latter model. This pattern
highlights the increasing elastic properties resulting from warp yarns cross-
ing a greater number of weft yarns.
Figure 1.22a shows the longitudinal modulus, E11 , decreases linearly
with increasing crimp angle. This is as expected as the region of fibre
orientated in the loading direction is decreasing as a result of larger crimp

angles. Correspondingly, Figure 1.22b shows the through-thickness modu-
lus increasing with larger crimp angles; this is due to the same logic, as
the region of fibre orientating the out-of-plane loading direction increases.
Essentially this acts to transfer longitudinal load to transverse load carry-
ing; this is a feature that can be manipulated to give the desired properties
for specific applications.
Figure 1.22c illustrates an almost constant in-plane shear modulus; con-
sequently this property cannot be modified using undulation characteristics
alone, however it is more likely affected by the introduction of braid angles
in the weave structure. Out-of-plane shear, as displayed in Figure 1.22d,
decreases with increasing crimp angle as the vertically applied displace-
ments induce a vertical stress component. Consequently as the crimp angle
increases, a greater region of the load-carrying fibre is orientated vertically,
causing a reduction in the out-of-plane shear modulus and reduced resis-
tance to deformation.
Finally, Figures 1.22e and 1.22f plot the effects of the in-plane and
out-of-plane Poisson’s ratio respectively. The in-plane Poisson’s ratio
increases with crimp angle, as seen in the literature. This occurs due to
an increase in slack yarn used in the undulation. When under strain the
fibres are able to straighten, increasing the resulting strains; this process
is analogous to the stretching of a spring. Similarly, the out-of-plane Pois-
son’s ratio also increases, however the vertical-loading direction of the weft
fibres limit the achievable strain, applying resistance to the motion and
consequently the increase is small.
1.3.4 3D woven composite

1.3.4.1 Meshfree approach
Here, a more realistic representation of the 3D orthogonal woven fabric
model was formulated and implemented into the meshfree method as
described in [34]. As the undulations and the cross-sections of yarns are
formulated into curved shapes, it is referred to as the smooth fabric model.
For comparison, Figure 1.23 shows a representative smooth fabric model
with the same amount of yarns taken into consideration as the straight-
edge unit cell model. All the basic dimensions (listed in Table 1.17) of the
different yarns used for the smooth fabric model are kept the same as the
straight-edge unit cell model (SM1) shown in Figure 1.27b. The lower yarn
volume fraction in the smooth fabric model leads to some difference between
Fig. 1.23. Representative smooth fabric model of 3D orthogonal woven fabric.
Table 1.17. The determined characteristics of the geometric model (mm).
fx fz Fx Fy Fz Wx Wy Wz Zx Zy Zz Hz
3.082 0.292 4.326 5.08 0.614 4.618 4.171 0.637 0.292 0.909 0.292 2.472
the results of the straight-edge and smooth fabric model as the fibre volume
fraction in the yarns is kept the same in both models.
Due to the symmetry, while implementing the meshfree method, only
a quarter of the representative model in Figure 1.23 needs to be considered
as the unit cell model, as represented in Figure 1.24. In this approach,
the fibres in yarns are UD and smooth distributed along the filler/warp/z
directions. The geometry is formulated as below:
z-yarn
Suppose the variation of the z-yarn fibres are based on a part of the arc-
tangent function, a tan(x)|−15≤x≤15 , the functions of the bottom and the
Fig. 1.24. Smooth fabric unit cell model of 3D orthogonal woven fabric.
top fibres can be written:

Zz Zz
fzxb (x) = c0 a tan a0 x − + b0 + d0 − (1.61)
2 2

Zz Zz
fzxt (x) = c0 a tan a0 x + + b0 + d0 + (1.62)
2 2
where a0 , b0 , c0 , d0 are the parameters to determine the shape of the fibres

according to the dimensions of the unit cell model, as a0 = 15−(−15) Wx , b0 =
−15, c0 = a tan(15)−a hz−Zz
tan(−15) and d0 = Zz
2 − c 0 a tan(−15). Note that the
range of the “a tan” function is chosen to be from –15 to 15 in order to fit
the shape of inclined Z yarn.
We assume the configuration of the Z-yarn volume on section x=0 is a
half ellipse:
2
y − F y/2
fzy (y) = ±b1 1− + c1 (1.63)
a1
where the coefficients are determined according to the dimensions of

fzxt (x=0)−fzxb (x=0)
the unit cell model, as a1 = Zy
2 , b1 = 2 and c1 =
fzxt (x=0)+fzxb (x=0)
2
Therefore, the bottom and top surfaces of the Z-yarn volume can be
written as:
2
y − F y/2
fzbot (x, y) = −b1 1− + c1 + fzxb (x) (1.64)
a1
2
y − F y/2
fztop (x, y) = b1 1− + c1 + fzxt (x) (1.65)
a1
The slope of the Z-yarn fibre which starts at z = z0 is obtained as:
dfzx (x) c0 a 0
= 2 (1.66)
dx 1 + a0 x − Zz
2 + z0 + b0
Therefore, the rotation angle θ of the fibre, which is needed to determine

the constitutive matrix C(x, y, z), can be calculated by θ = a tan dfzx
dx
(x)
for different positions on the fibre.
f1-yarn
The undulation of the f1 yarn is constructed using two parts: a straight
part when 0 ≤ y ≤ F2y − 1.5Zy; and a curved part when F2y − 1.5Zy ≤ y ≤
Fy
2 . A sinusoidal function is used as the basic function of the curved part,
formulated as below:
ff 1y (y) = c2 sin [a2 (y − b2 )] + d2 + z0 (1.67)
where z0 denotes where the fibre starts. For example, when z0 = 0, ff 1y (y)
becomes the bottom fibre in the f1 yarn; when z0 = f z, ff 1y (y) becomes
the top fibre in the f1 yarn. a2 , b2 , c2 , d2 are coefficients determined as a2 =
fzxt (x=0)
1.5Zy , b2 = 2 −
π fy 1.5Zy
2 c2 = 2 and d2 = fzxt (x=0)
2
The top outline of the cross-section of the f1 yarn on the surface of
y = F2y is formulated to be:
ff 1x (x) = −a3 x2 + b3 (1.68)
b3 −fzxt (x= f2x )

where the coefficients are determined as a3 = ( f2x )2
and b3 = f z +
fzxt (x = 0).
Thus, the top and bottom surfaces of the f1 yarn can be expressed when
0 ≤ y ≤ F2y − 1.5Zy:
ff 1top (x, y) = −a3 x2 + f z (1.69)

ff 1bot (x, y) = fzxt (x) − fzxt (x = 0) (1.70)
when Fy
2 − 1.5Zy ≤ y ≤ Fy
2 :
ff 1top (x, y) = ff 1y (y)|z0 =f z − a3 x2 (1.71)

ff 1bot (x, y) = ff 1y (y)|z0 =0 + fzxt (x) − fzxt (x = 0) (1.72)
The slope of the fibre becomes:
dff 1y (y)
= c2 a2 sin [a2 (y − b2 )] (1.73)
dy
df f1y (y)
So the rotating angle θ = a tan dy
f2-yarn
The shape of the cross-section of the f2 yarn is constructed to be a half
ellipse:
2
x
ff 2 (x) = ±b4 1− + c4 (1.74)
a4
where a4 = F2x , b4 = F2z , c4 = hz2

hence, the top and bottom surfaces of filler yarn 2 (f2 yarn) can be
written as:
2
x
ff 2top (x, y) = b4 1 − + c4 (1.75)
a4
2
x
ff 2bot (x, y) = −b4 1 − + c4 (1.76)
a4
f3-yarn
For f3 yarn, the cross-section shape is symmetrical with the cross-section
of the f1 yarn. Since the fibres are straight along the y direction, the top
and bottom surfaces of the f3 yarn can be directly formulated as:
ff 3top (x, y) = fzxb (W x − x) + hz − fzxb (x = W x) (1.77)

ff 3bot (x, y) = hz − f z + a3 x 2
(1.78)
where the coefficient a3 has the same value as the one for f1 yarn.
w-yarns
The warp yarn fibres are straight along the x direction and the function of
the ellipse is used to formulate the cross-section. So the top and bottom
surfaces of w yarns can be written as:
2
y
fwtop (x, y) = b5 1 − + c5,6 (1.79)
a5
2
y
fwbot (x, y) = −b5 1 − + c5,6 (1.80)
a5
where a5 = W2y , b5 = W2z , c5 = f z + W2z for the lower warp yarn, and
c6 = f z + F z + 3W
2
z
for the upper warp yarn.
Homogenisation
Once the geometry of the yarns is defined, the global stiffness matrix K
for the whole unit cell can be obtained using Eq. (1.42). The undulated
yarns and matrix parts are expressed in Eq. (1.42) by different C(x, y, z)
of different integration points, as depicted in Figure 1.25. Therefore, when
obtaining the global discrete system equations, Eq. (1.41), each integration
point is considered as the field node and carries its own material proper-
ties while the nodes are only used to generate the shape functions of the
field nodes. In this case, it is not necessary to distribute nodes or integra-
tion points on the interfaces between matrix & yarn or yarn & yarn; also
there are no boundary conditions or discontinuous functions needed along
the interfaces since perfect bonding between constituents is assumed. The
aim of formulating the geometry in this meshfree approach is not only to
determine the position for each integration point but also to obtain the
undulation angle θ of C(x, y, z) when the integration point is located inside
the yarns. After the global discrete system equation, Eq. (1.41) is assembled
and boundary conditions are applied to the relevant nodes. Eq. (1.41) can
then be solved to obtain the displacement field. It can be seen that one
Fig. 1.25. Meshfree approach based on the Galerkin method for composites.
Fig. 1.26. Uniform distributions (11 × 11 × 9) of nodes in: (a) Straight-edge, (b) Smooth
fabric unit cell model.
significant advantage of the proposed meshfree approach is that the com-

plexity of the yarn shape hardly affects the simplicity and efficiency of the
numerical process; in other words, the approach can be a general method
for numerical homogenisation of composite materials.
The geometry of the unit cell models generated by the meshfree
method can be shown by plotting Gaussian integration points, as pre-
sented in Figure 1.8. The nodes are uniformly distributed (Nx× Ny× Nz
= 11 × 11 × 9 = 1089 nodes used to compare with FE), as shown in Fig-
ure 1.9, and the scale factors in the shape functions are selected to be
ax = N x/W x, ay = N y/F y and az = N z/Hz. The minimum amount of
support nodes inside subdomains is chosen to be 12.
The differences between experimental values and predicted values of
longitudinal and transverse moduli (E11 and E22 ) in the detailed model
were approximately 6% and 4%, respectively. The result from the detailed
model was slightly lower than the values in the simplified model 2, especially
out-of-plane shear moduli (G13 and G23 ). This trend for shear moduli was
similar to the results for plain woven composites. In addition, the modulus
of three-direction (E33 ) in the detailed model was smaller than the values of
the simplified models. Agreement between the FEM and meshfree results
are generally excellent, the biggest discrepancies between these methods
was in the out-of-plane shear directions.
1.3.4.2 FE approach
In [93] FE models for the 93 oz 3D woven S-2 Glass/Dow Derakane 8084
Epoxy-Vinyl Ester resin composite, shown in Figure 1.27, were created.
A detailed model and two simplified models were constructed in order
to compare the results of the elastic mechanical properties for 3D woven
composites.
The input material properties used in FE models for the S-2 glass
93 oz fabric composite are shown in Table 1.18. The parameters used for
the input material properties and geometry are from Bogdanovich [29]. In
the detailed model, the yarn geometries were slightly changed from the
simplified model by adapting the ellipsoidal shape of the yarn. In addition,
the angle of the Z-yarn had to be smooth in order to make the mesh of
brick elements. When the angle from the x3 axis was low, the model was
Fig. 1.27. FE models for 3D woven composite: (a) Detailed model, (b) Simplified models
1 and 2.
Table 1.18. Material properties of constituents

for the S-2 glass 93 oz fabric composite.
Symbol Value

Em 3170 MPa
νm 0.35
E1 53120 MPa
E2 = E3 14660 MPa
ν23 0.268
ν21 = ν31 0.266
G12 = G31 4240 MPa
G23 5780 MPa
not able to consist of only hexahedral brick elements or the model would
have negative elements (self-penetrating) which may have caused inaccurate
results in the FE analysis.
The comparison of elastic mechanical properties obtained from FE anal-
ysis [93], meshfree analysis [34] analytical method and experiments [29] for
the S-2 glass 93 oz fabric composite are shown in Table 1.19. The val-
ues obtained from the detailed model and simplified model 2 in which the
Z-yarns in the longitudinal direction were located above and below fill yarns
were close to the experimental results, while the values obtained from sim-
plified model 1 were much higher than the values of the experimental results.
1.4 Continuum Damage Mechanics Model for Woven

Composites
1.4.1 Progressive damage analysis of plain woven
composites by meshfree methods
Here, the CDM model developed by Wen and Aliabadi [24] for plain woven
composites will be described. When a woven composite is subjected to
uniaxial tension, the fundamental mode of damage usually involves matrix
cracking as the first sign of failure. In order to model damage propagation,
continuum constitutive models can feature internal variables representing,
directly or indirectly, density and/or distribution of the microscopic defects
that characterise damage.
Introducing the effective stress-strain relationship gives:
{σ} = [C f,w (ω)]{ε} (1.81)

Table 1.19. Effective mechanical properties of the S-2 glass 93 oz fabric composite.
E11 E22 E33 G12 G13 G23

Detailed 23.22 21.52 9.00 3.01 2.66 2.70 0.128 0.344 0.339
model [93]
Meshfree 24.61 22.38 10.78 3.11 3.37 3.69 0.132 0.316 0.319
results for
detailed
model [34]
Simplified 28.68 27.40 12.31 3.52 3.90 3.77 0.120 0.305 0.295
model 1
[93]
Mesh free 28.01 27.90 12.51 3.58 4.06 4.84 0.120 0.296 0.307
results for
simplified
model 1
[34]
simplified 24.17 22.86 9.28 3.11 3.13 3.08 0.125 0.347 0.335
model 2
[93]
FE analysis 27.54 24.89 9.72 3.16 3.11 2.98 0.093 0.402 0.402
(mosaic
model)
[29]
Experimental 24.68 20.75 N/A N/A N/A N/A 0.11 N/A N/A
[29]
where ω denotes the damaged state factor and [C(ω)] is the damaged stiff-
ness tensor, which can be written, by Tabiei and Aminjikarai [100], as:
−1
S 0
[C f,w (ω)] = [S f,w ]−1 = d (1.82)
0 Ss−1
where superscript f refers to the fill yarn and w to the warp yarn respec-
tively, and the direction stress compliance matrix, in terms of the mechan-
ical properties, is assumed to be:
[Sdf,w (ω)]
 
1 νLT νLT
 (1 − ω1 )EL − −
 EL (1 − ω1 )(1 − ω2 ) EL (1 − ω1 )(1 − ω3 ) 

 1 νT T 
=
 − 
 (1 − ω2 )ET ET (1 − ω1 )(1 − ω2 ) 

 1 
Symm.
(1 − ω3 )ET
(1.83)
and the shear stress compliance matrix:

 
1
0 0
 (1 − ω4 )GLT 
 
 1 
[Ssf,w (ω)] =  0  (1.84)
 (1 − ω5 )GT T 
 1 
Symm.
(1 − ω6 )GLT
y y
where GT T = (C22 − C23 )/2. The Weibull [101] distribution function of
defects in the fibres is selected and the damage factors ωi are evaluated by:
mi
1 |εi |
ωi = 1 − exp − i = 1, 2, 3 (1.85)
mi e ε0i
where mi are material constants,
ε1 = ε11 , ε2 = ε22 , ε3 = ε33 ,

(1.86)
ε01 = σ1t/c /EL , ε02 = ε03 = σ2t/c /ET
and amended strengths are defined:
σ1t/c = Xt/c /bt/c (1.87)
in which Xt/c present the tensile/compressive strength of the pure fibres

and bt/c is a reduction factor. For in-plane shear, the damage factors are:
ms
αs |ε12 |
ω4 = 1 − exp − ,
ε4m
ms
αs |ε23 |
ω5 = 1 − exp − , (1.88)
ε5m
ms
αs |ε31 |
ω6 = 1 − exp −
ε4m
where ε4m , ε5m are the ultimate shear strains of the matrix material and
αs is a factor that is used to control the damage initiation strain and the
rate of damage evaluation [100].
There are many CDM models and assumptions for the stress com-
pliance matrices. For instance, the direct stress–strain behaviour of the
constituents is assumed elastic up to failure, while non-linear shear stress–
strain behaviour is assumed for in-plane shear for both UD and pure matrix
constituents. Tabiei and Ivanov [13] assumed that the generalised tangen-
tial shear modulus Gy of the yarn is obtained using the three-parameter
equation of Ramberg–Osgood by:

τ py 1+1/py
Gy = Gy0 1 − (1.89)
S
where Gy0 is the initial shear modulus, τ is the shear stress, S is the shear
strength and py is a fitting parameter to be tuned against experimental
results.
1.4.1.1 Progressive failure anlayisis of a Carbon-Epoxy (T300/913)

woven composite
The meshfree method is compared with the semi-analytical formulation
described in Section 1.2.1 for the case of the mosaic model described in
Section 1.3.1.1 and the smooth RVE of Section 1.3.1.2. For the meshfree
mode, the 3D uniform distribution of nodes is adopted. The total number
of nodes is (Nx + 1) × (Ny + 1) × (Nz + 1) (11 × 11 × 7 in all examples) and
the scale factors are selected by α1 = α2 = 1; α3 = Nz /(Nx × H).
A Carbon-Epoxy (T300/913) woven composite, which has the prop-
erties for the yarn EL = 132 GPa, ET = 8.8 GPa, GLT = 4.6 GPa,
νLT = 0.021, νTT = 0.4, Xt = 2005 MPa, Xc = 1355 MPa was stud-
ied. It was assumed that σ2t = 68 MPa, σ2c = 255 MPa, bt = bc = 1,
ε4m = ε5m = 0.15, m1 = m2 = 5, αs = 2.25 and ms = 2. In addition, the
isotropic matrix is of epoxy with Em = 3 GPa, νm = 0.4, σ 0 = 100 MPa
and m = 5. The yarn volume fraction is assumed to be Vy = 0.58 and the
average undulation angle θ = 1.4◦ . Therefore, the thickness of RVE from
Eq. (1.50) is H = 0.0411, and the location of the joint of fill yarn and warp
yarn λ =0.8824 from Eq. (1.56).
From the semi-analytical solution, the variations of stress σxab in each
subcell are shown in Figure 1.28 against the RVE strain ε̄x in the case of
the Weibull model for fibres and von Mises criteria for the matrix in each
subcell. Apparently, the order of the subcell’s failure can be seen in this
figure. If the von Mises criterion is used for matrix materials, the jumps
show the break points.
Figure 1.29 shows the stress distribution of stress σxab in each subcell
with RVE strain ε̄x using the Weibull model in the matrices and yarns in
each subcell. In addition, the variations of stress σxab of fill yarn in subcells
“mf” and “ff” are shown in Figure 1.30. With different CDM models, the
orders of failure are different (see Figures 1.28 and 1.29).
Fig. 1.28. Variations of stress σxab in each subcell with RVE strain εy by the Weibull
model for fibres and the von Mises criteria for the matrix in each subcell.
Fig. 1.29. Variations of stress σxab in each subcell with RVE strain ε by the Weibull
model for the matrix and fibres in each subcell.
Fig. 1.30. Variations of stress σxab in subcells “mf” and “ff” with RVE strain ε.
When the RVE strain ε̄x = 0.021, total failure will happen due to
failure of the fill yarn. Therefore, for von Mises criteria, the failure order
should be warp in “fm” and “ff” (almost at the same time), then fill in
subcells “fm” and “ff” at the same time, and finally, the failure of matrices
in “mm”, “mf” and “fm” respectively.
Figure 1.31 shows the relationship between RVE stress σ̄x and RVE
strain ε̄x using the analytical formulation and the meshfree Galerkin
method. Using the meshfree Galerkin method, we only consider the Weibull
model for fibres and matrices with the four-cell model and smooth fibres
model. The maximum value of RVE stress using the analytical solution is
5% higher than that using the meshfree method, as shown in Figure 1.29.
The failure RVE strain is about 0.021 for all methods and models.
From these results, we can see that the agreement between the analytical
solution and meshfree Galerkin method is satisfied. In addition, the four-cell
and smooth fibre models share the same accuracy.
1.4.2 Multi-scale progressive damage of plain woven

composite structures
In this section, progressive failure analysis reported by Bacarreza et al. in
[102] is described. This numerical model is capable of dealing with the
Fig. 1.31. Variations of RVE stress σ̄x with RVE strain ε using different formulations.
progressive degradation of plain woven composite structures in a com-

putationally efficient manner. It uses the semi-analytical homogenisation
method presented in Section 1.2.1 to derive effective properties of plain
woven composites from the material properties of its constituents. The pro-
gressive failure is described using non-local CDM where the driving internal
variable for the damage is the non-local strain.
1.4.2.1 Progressive failure

Bogdanovich used nine failure modes for each brick in the mosaic
model, based on the ultimate strain failure criterion which uses an
elastic-properties-discount scheme in order to introduce a sequence of
damaged states into the model [29]. An anisotropic damage model applied
to the yarn and perfect elastic matrix was employed by Barbero et al. [23]
to analyse the progressive failure of plain woven composites. Discount coef-
ficients were used by Tabiei and Ivanov [13] to model material degradation
in their proposed model.
The progressive failure of the RVE is modelled using non-local CDM
[103]. Once a failure criterion is met, damage propagation in the failing con-
stituent begins. Damage parameters are defined to degrade the compliances
of the failing constituents.
Apart from the assumptions for homogenisation, the following assump-

tions are used by the model:
• The damage in the matrix is isotropic and depends on the load conditions
• The damage in the yarn bundles is anisotropic.
Yarn bundles material damage

The failure criteria used in [102] for predicting yarn bundle failure differ-
entiates between the type and direction of the load, i.e., it is different for
tension, compression and shear, and it is also different for different direc-
tions of the load whether it is aligned to the fibres or not.
Damage initiation refers to the onset of degradation at a material point.
In this model, the damage initiation criteria (failure criteria) for the yarn
material are based on Hashin’s theory [104]. The damage propagation is
then modelled with an energy-based damage mechanics approach. The com-
pliance matrix is:
 
1 νT L νT L
− − 0 0 0
 h21 EL h1 h2 ET h1 h3 ET 
 
 
 νLT 1 νT T 
− − 0 0 0 
 h2 h1 EL h22 ET h2 h3 ET 
 
 
 νLT νT T 1 
− − 0 0 0 
 h3 h1 EL h3 h2 ET h23 ET 
 
[D]dam
yarn =  
 1 
 0 0 0 0 0 
 h24 GLT 
 
 1 
 0 0 0 0 0 
 h25 GT 
 
 
 1 
 0 0 0 0 0 
h26 GLT
(1.90)
where hi is defined as:

hi = (1 − di ) (1.91)
and finally, the constitutive stiffness matrix is defined as:
[C] = [D]−1 (1.92)

The damage evolution functions have the exponential form:



 0 κi ≤ ε0i
 !
di (κ) = ε 0
κ − ε 0 (1.93)

 i
1 − κ exp − f
i i
κi > ε0i
i εi − ε0i
where ε0i is the strain at onset of damage and εfi is a material parameter
controlling the slope of the softening curve:
gF ε0
εfi = 0 + i (1.94)
σi 2
gF is the fracture energy density (also called specific fracture energy and
is the area under the uniaxial stress-strain curve), σi0 is the peak stress.
Figure 1.32 shows the meaning of the parameters [105].
The mode of failure for loading in the normal direction depends on the
sign of the loading, i.e., it is different for tensile and compressive loads,
while the modes of failure due to shear loads are not affected by their sign.
Matrix Material Damage

Experiments show that polymer epoxy matrices exhibit asymmetry between
tensile and compressive behaviours [106, 107]. The compressive strengths
of polymer resins are always higher than the tensile strengths. In order to
Fig. 1.32. Stress-strain with exponential softening.

take that into account Bacarreza et al. [102] adopted the isotropic damage
model proposed by Mazars [108].
The constitutive matrix is:
Em
matrix = (1 − d)
[C]dam
(1 + νm )(1 − 2νm )
 
1 − νm νm νm 0 0 0
 ν 1 − νm 0 
 m νm 0 0 
 
 νm νm 1 − νm 0 0 0 
× 
 0 0 0 1 − 2νm 0 0 
 
 0 0 0 0 1 − 2νm 0 
0 0 0 0 0 1 − 2νm
(1.95)
d is calculated as a linear combination of the damage parameters, gt and gc

take into account the type of loading, i.e., whether the material is loaded
under tension or under compression:
d = αt gt + αc gc (1.96)
where:
!β !β

3
εtI εI
3
εtI εI
αt = , αc = 1− (1.97)
ε̃2m ε̃2m
I=1 I=1
and εtI , I = 1, 2, 3, are the principal strains due to positive stresses, i.e.,
the principal values of εt = Ce : De : εm , in which Ce = D−1
e is the elastic
compliance matrix and β = 1.06, which slows the evolution of damage
under shear loading.


 0 κm ≤ ε0m
 !
gt (κ) = ε0m κm − ε0m (1.98)


1 − κ exp − f κm > ε0m
m εm − ε0m

 0 κm ≤ ε0m

gc (κ) =
 0
1 − (1 − Ac ) εm − Ac exp[−Bc (κm − ε0 )] κm > ε0
m m
κm
(1.99)
The definition for gt (κ) has been modified, since the stress obtained using
the original formula asymptotically approaches a limit value of (1 − At )Eε0
while Eq. (1.98) approaches 0. The stress obtained using gc (κ) approaches
√
(1 − Ac )Eε0 /( 2ν) but it is limited to 0 in this model and κm (t) =
maxτ ≤t ε̃m (τ ).
These parameters are computed from the so-called Mazars definition
of equivalent strain:

ε̃m = εm = εm : εm (1.100)
where . are the McAuley brackets and denote the positive part operator.
Non-local Continuum Damage

Progressive damage leads to damage localisation due to strain softening
which can be modelled by various constitutive laws. Application of these
models to numerical methods produces mesh sensitivity and leads to an
infinitely small softening region, snapback of the load displacement response
and a total amount of dissipated energy equal to zero when the mesh is
refined.
In their pioneering work, Pijaudier-Cabot and Bazant [109, 110] pro-
posed a non-local damage theory where the variable controlling strain soft-
ening is treated as non-local, while the elastic part of the strain is treated
as local, this theory can be used as a localisation limiter.
Several non-local formulations can be found in the literature [103] which
average the damage energy release rate, the damage variable, the strain,
inelastic strain or the specific fracturing strain. The non-local variable used
in this model is the strain, and is calculated as:

ε̄(x) = α(x, ξ)ε(ξ)dξ (1.101)
V
where α(x, ξ) is the non-local weight function. In an infinite body, the

weight only depends on the distance between the source point ξ and the
target point x. Near a boundary the function is rescaled such that the non-
local operator does not alter the uniform field. This is achieved by [110]:
α0 (x − ξ)
α(x, ξ) = " (1.102)
α (x − ς)dς
V 0
here α0 is the quartic spline function [111]:

 2

 r2
1− 2 0≤r≤R
α0 (x − ξ) = α0 (r) = R (1.103)

 0 r>R
and R is called the interaction radius or radius of influence.
1.4.2.2 Verification using experimental data

The damage in each constituent is described using CDM and non-local
continuum in order to dissipate the correct amount of energy while avoiding
mesh sensitivities, i.e., that the damage parameters have to be calculated
using a non-local variable, which leads to homogenisation of the local and
non-local fields. Using a semi-analytical homogenisation method with a
small number of unknowns ensures an efficient way to perform a multi-scale
analysis of plain woven composites. The main feature of the proposed model
is that it can efficiently predict not only the failure of woven composite
materials, but also the failure of structures made of plain woven composites.
The experimental results for the plain woven composite as well as its
constituents’ properties, UD yarn bundles and matrix materials, reported
in Table 1.20, are taken from the Composite Materials Handbook [91].
The yarn bundles material properties were linearly adjusted according
to the fibre volume fraction that was used in the experiment and the numer-
ical model. It is necessary to get the correct total fibre volume fraction in
the RVE and it is achievedby using a yarn bundle material with fibre
RV E UD .
volume fraction F Vyarn = F Vyarn
The material tested was the AS4 3k/3501-6 plain weave fabric, which
is a plain weave fabric with an areal weight of 193 g/m2 , typical cured resin
content of 37−41% and a typical cured ply thickness of 0.188−0.218 mm.
All the experiments are at room temperature and moisture content.
The AS4 fibres (Hercules AS4 3k W) are continuous carbon filaments
made from a PAN precursor, and surface treated to improve handling
characteristics and structural properties. The filament count is 3000 fila-
ments/tow, the typical tensile modulus is 234.4 GPa and the typical tensile
strength is 3790 MPa.
The Matrix 3501-6 (Hercules 3501-6) is an amine-cured epoxy resin. It
will retain light tack for a minimum of 10 days at room temperature.
Tension along axis-1

The experiment is a [0f ]8 specimen made from AS4 3k/3501-6 plain weave
fabric with a normalised batch fibre volume to 57% and a cured ply thick-
ness of 0.188 mm was loaded in tension along direction 1 (along the main
fibre orientation) according to ASTM D 3039-76.
The numerical model is a single element loaded by displacement in the
direction parallel to axis-1 on one side and supported on the other as shown
in Figure 1.33.
Table 1.20. Material properties.
Symbol Value Description

Em 4350 MPa Young’s modulus of the pure matrix
νm 0.36 Poisson’s ratio of the pure matrix
E1 170000 MPa Longitudinal Young’s modulus of the yarn
E2 = E3 12835 MPa Transversal Young’s modulus of the yarn
ν23 = ν32 0.33 Poisson’s ratio of the yarn (Symmetry plane)
ν21 = ν31 0.27 Poisson’s ratio of the yarn
G12 = G31 9060 MPa Out-of-plane Shear Modulus
G23 4825 MPa In-plane Shear Modulus (Symmetry plane)
Geometry of the yarn
θ 10.86◦ Undulation angle of the yarn
H 0.188 mm Height of the RVE
Vf 0.755 Yarn volume fraction
Damage evolution parameters of the yarn
σ10t 2523 MPa Longitudinal strength of the yarn in tension
σ10c 1831 MPa Longitudinal strength of the yarn in compression
Gc1 60 J/mm Fracture toughness in longitudinal direction
σ20t = σ30t 72 MPa Transversal strength of the yarn in tension
σ20c = σ30c 286.9 MPa Transversal strength of the yarn in compression
Gc2 = Gc3 0.4 J/mm Fracture toughness in transversal direction
0 = σ0
σ12 151 MPa Out-of-plane shear strength of the yarn
31
Gc12 = Gc31 1.5 J/mm Fracture toughness in out-of-plane shear direction
0
σ23 88. MPa In-plane shear strength of the yarn
Gc23 1 J/mm Fracture toughness in in-plane shear direction
Damage evolution parameters of the matrix
0
σm 58.5 MPa Tensile strength of the matrix
Gcm 1.1 J/mm Fracture toughness of the matrix
Ac 1.9971 Parameter affecting the softening in compression
Non-local continuum
R 3 mm Radius of influence of non-local continuum
The stress-strain response is illustrated in Figure 1.34. The maxi-

mum applied stress was found to be around 950 MPa and the initial
modulus of elasticity was around 68.9 GPa. The experimental results are
855 ± 35.74 MPa for the strength of the lamina and a modulus of elasticity
of 67.6 ± 2.03 GPa.
It can be seen that there is a very good agreement between the numer-
ical model and the experimental results.
Fig. 1.33. Geometry and load.
Fig. 1.34. Stress-strain response.
The failure of each constituent in each subcell is shown in Figure 1.35,

leading to the failure of the RVE. As shown, the failure starts in the direc-
tion perpendicular to the load in the warp yarn in subcell FF, followed
almost immediately by the failure of the yarn in subcell FM which is the
continuation of the warp yarn in subcell FF. Then, the matrix in subcell
MF and later FM begin to fail too. The yarn in subcell FM begins to fail,
and it is followed very quickly by the failure in the fill yarn in subcell FF,
leading to the final collapse of the RVE, since there is full failure in subcells
FF and FM, “breaking” of the RVE occurs. Notice that subcell MM, starts
to damage, but never reaches failure.
Fig. 1.35. Evolution of damage parameters.
Open Hole Tension [/±45f /0f /90f ]2S

A [±45f /0f /90f ]2S specimen made from AS4 3k/3501-6 plain weave fabric
with a normalised batch fibre volume to 60% and a cured ply thickness of
0.211 mm was investigated here. The specimen was loaded in tension along
direction 1 (along the main fibre orientation) according to SACMA SRM 5-
88 (1).
Exploiting symmetries along all the planes, only 1/8th of the geometry
is modelled and the numerical model is loaded by increasing displacement.
The geometry and boundary conditions are represented in Figure 1.36.
The stress-strain response of the numerical model is shown in Fig-
ure 1.37. As it can be seen the response is very brittle, almost linear until
peak load, followed by a fast drop in carrying capacity.
The ultimate stress attained in the numerical model is around 375.5
MPa while the experimental results give a value of 393 ± 13.36 MPa.
Figures 1.38 to 1.40 show the equivalent strains (defined as εeq =
√
ε : ε) at different load levels, illustrating the very complex way in which
the relatively uniform strain field changes into a highly localised one, show-
ing the zones where the maximum damage is expected.
The progressive failure for this type of specimen is very complex, involv-
ing the combination of failure of the constituents in different modes (ten-
sion, shear, etc.) and the delamination between layers. Total collapse is
achieved by delamination between layers with different orientations, total
Fig. 1.36. Geometry and boundary conditions.
Fig. 1.37. Stress-strain response.
failure of the 0 and 90 layers in tension and the shear failure of the ±45
layers. See Figures 1.41 to 1.43.
Also note that the adhesive layers between similar orientations (0 and
90; −45 and 45) do not fail, since these layers in a plain woven composite
are equivalent and do not generate big out-of-plane shear stresses.
Fig. 1.38. Equivalent strains in elastic range.
Fig. 1.39. Equivalent strains at peak load.
1.5 Summary
Textile composites are increasingly being used as a structural material
because of their balanced properties, higher impact resistance, and eas-
ier handling and fabrication compared with UD composites. However, the
complex architecture of these composites leads to difficulties in predict-
ing the mechanical response necessary for product design. In this chapter,
Fig. 1.40. Equivalent strains at failure.
Fig. 1.41. Pure matrix damage parameter at failure.
different methods for micromechanical analysis were presented. Microme-

chanical modelling analysis using FE and meshfree analyses were used
for predicting the homogenised (effective) elastic mechanical properties
in plain, twill, satin and 3D woven composites. Evaluation and compar-
ison of mechanical properties obtained from the numerical models with
experimental data showed good agreement in the accuracy of the predic-
tions; damage mechanics analysis of 2D plain woven composites were also
presented. Finally, multi-scale progressive failure analysis, using non-local
CDM, where the driving internal variable for the damage is the non-local
strain, was presented using a model implemented into ABAQUS/Explicit,
Fig. 1.42. Fibre damage parameter in the principal ply load direction at failure.
Fig. 1.43. Adhesive damage parameter at failure.
and the failure of longitudinal tension and open hole tension specimens
were simulated and verified against experimental results.
Acknowledgements
The authors of this chapter would like to thank Daisei Abe, Sanjay Patel
and Navanee Kopula Ragavan for their valuable contributions and find-
ings in modelling work performed during their MSc projects at Imperial
College.
References
[1] Ansar, M., Xinwei, W. and Chouwei, Z. (2011). Modeling strategies of 3D
woven composites: A review, Composite Structures, 93, 1947–1963.
[2] Goyal, D., Tang, X., Whitcomb, J.D. and Kelkar A.D. (2005). Effect of
Various Parameters on Effective Engineering Properties of 2 × 2 Braided
Composites, Mechanics of Advanced Materials and Structures, 12, 113–128.
[3] Strong, A.B. (2008). Fundamentals of composites manufacturing: materials,
methods and applications: SMe.
[4] Cripps, D. Woven Fabrics; Available from: http://www.netcomposites.com/
guide/woven-fabrics/40.
[5] Halpin, J., Jerine, K. and Whitney, J. (1971). The laminate analogy for 2
and 3 dimensional composite material, Journal of Composite Materials, 5,
36–49.
[6] Ishikawa, T. and Chou, T.W. (1982). Stiffness and strength behaviour of
woven fabric composites, Journal of Materials Science, 17, 3211–3220.
[7] Ishikawa, T. and T.W. Chou, (1982). Elastic behavior of woven hybrid
composites, Journal of Composite Materials, 16, 2–19.
[8] Naik, N. and Ganesh, V. (1992). Prediction of on-axes elastic properties
of plain weave fabric composites, Composites Science and Technology, 45,
135–152.
[9] Shembekar, P. and Naik, N. (1992). Elastic behavior of woven fabric com-
posites: II—Laminate analysis, Journal of Composite Materials, 26, 2226–
2246.
[10] Naik, N.K. and Shembekar, P.S. (1992). Elastic behavior of woven fab-
ric composites: III—Laminate design, Journal of Composite Materials, 26,
2522–2541.
[11] Jiang, Y., Tabiei, A. and Simitses, G.J. (2000). A novel micromechanics-
based approach to the derivation of constitutive equations for local/global
analysis of a plain-weave fabric composite, Composites Science and Tech-
nology, 60, 1825–1833.
[12] Tanov, R. and Tabiei, A. (2001). Computationally efficient micromechan-
ical models for woven fabric composite elastic moduli, Journal of Applied
Mechanics, 68, 553–560.
[13] Tabiei, A. and Ivanov, I. (2004). Materially and geometrically non-linear
woven composite micro-mechanical model with failure for finite element
simulations, International Journal of Nonlinear Mechanics, 39, 175–188.
[14] Kregers, A. and Melbardis, Y.G. (1978). Determination of the deformabil-
ity of three-dimensionally reinforced composites by the stiffness averaging
method, Mechanics of Composite Materials, 14, 1–5.
[15] Pastore, C. and Gowayed, Y. (1994). A self-consistent fabric geometry
model: modification and application of a fabric geometry model to predict
the elastic properties of textile composites, ASTM J. Compos. Technol.
Res.(USA), 16, 32–36.
[16] Zhang, Y. and Harding, J. (1990). A numerical micromechanics analysis of
the mechanical properties of a plain weave composite, Computers & Struc-
tures, 36, 839–844.
[17] Paumelle, P., Hassim, A. and Léné, F. (1990). Composites with woven rein-
forcements: Calculation and parameter analysis of the properties of the
homogeneous equivalent, Rech. Aerosp., 1–12.
[18] Woo, K. and Whitcomb, J. (1994). Global/local finite element analysis for
textile composites, Journal of Composite Materials, 28, 1305–1321.
[19] Chapman, C. and Whitcomb, J. (1995). Effect of assumed tow architecture
on predicted moduli and stresses in plain weave composites, Journal of
Composite Materials, 29, 2134–2159.
[20] Zeman, J. and Šejnoha, M. (2004). Homogenization of balanced plain weave
composites with imperfect microstructure: Part I — Theoretical formula-
tion, International Journal of Solids and Structures, 41, 6549–6571.
[21] Wen, P.H. and Aliabadi, M.H. (2009). Mesh-free micromechanical model
for woven fabric composite elastic moduli, Journal of Multiscale Modelling,
1, 303–319.
[22] Whitcomb, J. and Srirengan, K. (1996). Effect of various approximations on
predicted progressive failure in plain weave composites, Composite Struc-
tures, 34, 13–20.
[23] Barbero, E., Lonetti, P. and Sikkil, K. (2006). Finite element continuum
damage modeling of plain weave reinforced composites, Composites Part B:
Engineering, 37, 137–147.
[24] Wen, P.H. and Aliabadi, M.H. (2012). Damage mechanics analysis of plain
woven fabric composite micromechanical model for mesh-free simulations,
Journal of Composite Materials, 46, 2239–2253.
[25] Chaphalkar, P. and Kelkar, A.D. (2001). Classical laminate theory model
for twill weave fabric composites, Composites Part A: Applied Science and
Manufacturing, 32, 1281–1289.
[26] Kwon, Y. and Roach, K. (2004). Unit-Cell Model of 2/2-Twill Woven Fabric
Composites for Multi-Scale Analysis, CMES: Computer Modeling in Engi-
neering & Sciences, 5, 63–72.
[27] Nicoletto, G. and Riva, E. (2004). Failure mechanisms in twill-weave lam-
inates: FEM predictions vs. experiments, Composites Part A: Applied Sci-
ence and Manufacturing, 35, 787–795.
[28] Ng, S.P., Tse, P.C. and Lau, K.J. (2001). Progressive failure analysis of 2/2
twill weave fabric composites with moulded-in circular hole, Composites
Part B: Engineering, 32, 139–152.
[29] Bogdanovich, A. (2006). Multi-scale modeling, stress and failure analyses
of 3-D woven composites, Journal of Materials Science, 41, 6547–6590.
[30] Stig, F. and Hallström, S. (2009). Assessment of the mechanical proper-
ties of a new 3D woven fibre composite material. Composites Science and
Technology, 69, 1686–1692.
[31] Heß, H., Roth, Y. and Himmel, N. (2007). Elastic constants estimation of
stitched NCF CFRP laminates based on a finite element unit-cell model,
Composites Science and Technology, 67, 1081–1095.
[32] Huysmans, G., Verpoest, I., and Van Houtte, P. (2001). A damage model for
knitted fabric composites, Composites Part A: Applied Science and Manu-
facturing, 32, 1465–1475.
[33] Wen, P.H. and Aliabadi, M.H. (2010). Elastic Moduli of Woven Fabric
Composite by Meshless Local Petrov-Galerkin(MLPG) Method, Computer
Modeling in Engineering & Sciences(CMES), 61, 133–154.
[34] Li, L., Wen, P.H. and Aliabadi, M.H. (2011). Meshfree modeling and homog-
enization of 3D orthogonal woven composites, Composites Science and Tech-
nology, 71, 1777–1788.
[35] Haasemann, G., Kastner, M. and Ulbricht V. (2009). A new modelling
approach based on Binary Model and X-FEM to investigate the mechanical
behaviour of textile reinforced composites, Computer Modeling in Engineer-
ing and Sciences (CMES), 42, 35.
[36] Voigt, W. (1889) Ueber die Beziehung zwischen den beiden Elas-
ticitätsconstanten isotroper Körper, Annalen der Physik, 274, 573–587.
[37] Reuss, A. (1929). Berechnung der Fließgrenze von Mischkristallen auf
Grund der Plastizitätsbedingung für Einkristalle, ZAMM — Journal of
Applied Mathematics and Mechanics/Zeitschrift für Angewandte Mathe-
matik und Mechanik, 9, 49–58.
[38] Hashin, Z. and Shtrikman, S. (1962). On some variational principles in
anisotropic and nonhomogeneous elasticity, Journal of the Mechanics and
Physics of Solids, 10, 335–342.
[39] Papanicolau, G., Bensoussan, A. and Lions, J.-L. (1978). Asymptotic Anal-
ysis for Periodic Structures, North Holland: Elsevier.
[40] Sánchez-Palencia, E. (1980). “Non-Homogeneous Media and Vibration The-
ory” in Lecture Notes in Physics, New York: Springer-Verlag.
[41] Hill, R. (1965.) Continuum micro-mechanics of elastoplastic polycrystals,
Journal of the Mechanics and Physics of Solids, 13, 89–101.
[42] Eshelby, J.D. and Eshelby, J. (1957). The determination of the elastic field of
an ellipsoidal inclusion, and related problems, Proceedings of the Royal Soci-
ety of London. Series A. Mathematical and Physical Sciences, 241, 376–396.
[43] Mori, T. and Tanaka, K. (1973). Average stress in matrix and average elas-
tic energy of materials with misfitting inclusions, Acta metallurgica, 21,
571–574.
[44] Aboudi, J. (1989). Micromechanical analysis of composites by the method
of cells, Applied Mechanics Reviews, 42, 193.
[45] Paley, M. and Aboudi, J. (1992). Micromechanical analysis of composites
by the generalized cells model, Mechanics of Materials, 14, 127–139.
[46] Bogdanovich, A.E. and Pastore, C.M. (1996). Material-smart analysis
of textile-reinforced structures, Composites Science and Technology, 56,
291–309.
[47] Dvorak, G.J. and Rao, M. (1976). Axisymmetric plasticity theory of fibrous
composites, International Journal of Engineering Science, 14, 361–373.
[48] Dvorak, G.J. and Benveniste, Y. (1992). On transformation strains and
uniform fields in multiphase elastic media, Proceedings of the Royal Society
of London. Series A: Mathematical and Physical Sciences, 437, 91–310.
[49] Fish, J., Yu, Q. and Shek, K.L. (1999). Computational damage mechanics
for composite materials based on mathematical homogenization, Interna-
tional Journal for Numerical Methods in Engineering, 45, 1657–1679.
[50] Michel, J.C. and Suquet, P. (2003). Nonuniform transformation field anal-
ysis, International Journal of Solids and Structures, 40, 6937–6955.
[51] Christman, T., Needleman, A. and Suresh, S. (1989). An experimental and
numerical study of deformation in metal-ceramic composites, Acta metal-
lurgica, 37, 3029–3050.
[52] Bao, G., Hutchinson, J. and McMeeking, R. (1991). Particle reinforcement
of ductile matrices against plastic flow and creep, Acta metallurgica et mate-
rialia, 39, 1871–1882.
[53] Nakamura, T. and Suresh, S. (1993). Effects of thermal residual stresses and
fiber packing on deformation of metal-matrix composites, Acta metallurgica
et materialia, 41, 1665–1681.
[54] Van der Sluis, O., Schreurs, P. and Meijer, H. (2001). Homogenisation of
structured elastoviscoplastic solids at finite strains, Mechanics of Materials,
33, 499–522.
[55] Suquet, P. (1985). Local and global aspects in the mathematical theory of
plasticity, Plasticity Today: Modelling, Methods and Applications, 279–310.
[56] Ghosh, S., Lee, K. and Moorthy, S. (1995). Multiple scale analysis of het-
erogeneous elastic structures using homogenization theory and Voronoi cell
finite element method, International Journal of Solids and Structures, 32,
27–62.
[57] Smit, R., Brekelmans, W. and Meijer, H. (1998). Prediction of the mechani-
cal behavior of nonlinear heterogeneous systems by multi-level finite element
modeling, Computer Methods in Applied Mechanics and Engineering, 155,
181–192.
[58] Miehe, C., Schotte, J. and Schröder, J. (1999). Computational micro-macro
transitions and overall moduli in the analysis of polycrystals at large strains,
Computational Materials Science, 16, 372–382.
[59] Miehe, C., Schröder, J., and Schotte, J. (1999). Computational homogeniza-
tion analysis in finite plasticity simulation of texture development in poly-
crystalline materials, Computer Methods in Applied Mechanics and Engi-
neering, 171, 387–418.
[60] Feyel, F. and Chaboche, J.L. (2000). FE2 multiscale approach for mod-
elling the elastoviscoplastic behaviour of long fibre SiC/Ti composite mate-
rials, Computer Methods in Applied Mechanics and Engineering, 183,
309–330.
[61] Böhm, H.J., Eckschlager, A. and Han, W. (2002). Multi-inclusion unit cell
models for metal matrix composites with randomly oriented discontinuous
reinforcements, Computational Materials Science, 25, 42–53.
[62] Kouznetsova, V., Geers, M. and Brekelmans, W. (2002). Multi-scale con-
stitutive modelling of heterogeneous materials with a gradient-enhanced
computational homogenization scheme, International Journal for Numeri-
cal Methods in Engineering, 54, 235–1260.
[63] González, C., Segurado, J. and Llorca, J. (2004). Numerical simulation
of elasto-plastic deformation of composites: evolution of stress microfields
and implications for homogenization models, Journal of the Mechanics and
Physics of Solids, 52, 1573–1593.
[64] Lomov, S.V., Ivanov, D.S., Verpoest, I., Zako, M., Kurashiki, T., Nakai, H.
and Hirosawa, S. (2007). Meso-FE modelling of textile composites: Road
map, data flow and algorithms, Composites Science and Technology, 67,
1870–1891.
[65] Zhang, F., Lisle, T., Curtin, W.A. and Xia, Z. (2009). Multiscale modeling
of ductile-fiber-reinforced composites, Composites Science and Technology,
69, 1887–1895.
[66] Coenen, E., Kouznetsova, V. and Geers, M. (2010). Computational homoge-
nization for heterogeneous thin sheets, International Journal for Numerical
Methods in Engineering, 83, 1180–1205.
[67] Tabiei, A. and Jiang, Y. (1999). Woven fabric composite material model
with material nonlinearity for nonlinear finite element simulation, Interna-
tional Journal of Solids and Structures, 36, 2757–2772.
[68] Bednarcyk, B.A. (2001). “Discussion of Woven fabric composite material
model with material non-linearity for non-linear finite element simulation”
by Tabiei and Jiang (Int. J. Solids Struct. 36 (1999) 2757–2771), Interna-
tional Journal of Solids and Structures, 38, 8585–8588.
[69] Adumitroaie, A. and Barbero, E.J. (2011). Beyond plain weave fabrics–I,
geometrical model, Composite Structures, 93, 1424–1432.
[70] Tang, X. and Whitcomb, J.D. (2003). General techniques for exploiting
periodicity and symmetries in micromechanics analysis of textile compos-
ites, Journal of Composite Materials, 37, 1167–1189.
[71] De Carvalho, N., Pinho, S. and Robinson, P. (2011). Reducing the domain
in the mechanical analysis of periodic structures, with application to woven
composites, Composites Science and Technology, 71, 969–979.
[72] Jones, R.M. (1975). Mechanics of composite materials. Vol. 2, London:
Taylor & Francis.
[73] Gingold, R.A. and Monaghan, J.J. (1977). Smoothed particle hydro-
dynamics-theory and application to non-spherical stars, Monthly Notices
of the Royal Astronomical Society, 181, 375–389.
[74] Oñate, E., Idelsohn, S., Zienkiewicz, O.C. and Taylor, R.L. (1996). A finite
point method in computational mechanics. Applications to convective trans-
port and fluid flow, International Journal for Numerical Methods in Engi-
neering, 39, 3839–3866.
[75] Nayroles, B., Touzot, G. and Villon, P. (1992). Generalizing the finite ele-
ment method: Diffuse approximation and diffuse elements, Computational
Mechanics, 10, 307–318.
[76] Belytschko, T., Lu, Y.Y. and Gu, L. (1994). Element-free Galerkin methods,
International Journal for Numerical Methods in Engineering, 37, 229–256.
[77] Liu, W.K., Jun, S. and Zhang, Y.F. (1995). Reproducing kernel particle
methods, International Journal for Numerical Methods in Fluids, 20, 1081–
1106.
[78] Duarte, C.A. and Oden, J.T. (1996). An h-p adaptive method using
clouds, Computer Methods in Applied Mechanics and Engineering, 139,
237–262.
[79] Yagawa, G. and Yamada, T. (1996). Free mesh method: A new meshless
finite element method, Computational Mechanics, 18, 383–386.
[80] Atluri, S.N. and Zhu, T. (1998). A new Meshless Local Petrov-Galerkin
(MLPG) approach in computational mechanics, Computational Mechanics,
22, 117–127.
[81] Liu, G. and Y. Gu. (1999). A point interpolation method, in Proc. 4th Asia-
Pacific Conference on Computational Mechanics, December, Singapore.
[82] Wen, P.H. and Aliabadi, M.H. (2007). Meshless method with enriched radial
basis functions for fracture mechanics, SDHM Structural Durability and
Health Monitoring, 3, 107–119.
[83] Wen, P.H. and Aliabadi, M.H. (2007). Applications of meshless method
to fracture mechanics with enriched radial basis functions, Durability of
Structures and Health Monitoring, 3, 107–119.
[84] Wen, P.H. and Aliabadi, M.H. (2008). An improved meshless collocation
method for elastostatic and elastodynamic problems, Communications in
Numerical Methods in Engineering, 24, 635–651.
[85] Wen, P.H., Aliabadi, M.H. and Liu, Y.W. (2008). Meshless method for
crack analysis in functionally graded materials with enriched radial base
functions, CMES — Computer Modeling in Engineering and Sciences, 30,
133–147.
[86] Hardy, R.L. (1971). Multiquadric equations of topography and other irreg-
ular surfaces, Journal of Geophysical Research, 76, 1905–1915.
[87] Matheron, G. (1963). Principles of geostatistics, Economic Geology, 58,
1246–1266.
[88] Sacks, J., Welch, W.J., Mitchell, T.J. and Wynn, H.P. (1989). Design and
analysis of computer experiments, Statistical Science, 4, 409–423.
[89] Lancaster, P. and Salkauskas, K. (1981). Surfaces generated by moving least
squares methods, Mathematics of Computation, 37, 141–158.
[90] Gu, L. (2003). Moving kriging interpolation and element-free Galerkin
method, International Journal for Numerical Methods in Engineering,
56, 1–11.
[91] MIL-HDBK-17-2F Polymer Matrix Composites — Material Properties, in
Composite Materials Handbook (2002) Department of Defense.
[92] Ishikawa, T., Matsushima, M. Hayashi, Y. and Chou, T.W. (1985). Exper-
imental confirmation of the theory of elastic moduli of fabric composites,
[93] Bacarreza, O., Abe, D. Aliabadi, M.H. and Kopula Ragavan, N. (2012).
Micromechanical modeling of advanced composites, Journal of Multiscale
Modelling, 4, 1250005.
[94] Walrath, D.E. and Adams, D.F. (1985). Finite Element Micromechanics and
Minimechanics Modeling of a Three-Dimensional Carbon–Carbon Compos-
ite Material, DTIC Document.
[95] Patel, S., Bacarreza, O. and Aliabadi, M. (2014). Evaluation of Elas-
tic Properties of Satin Composites, Key Engineering Materials, 577,
229–232.
[96] Dow, N.F., Ramnath, V. and Rosen, B.W. (1987) Analysis of Woven Fabrics
for Reinforced Composite Materials, Virginia: NASA.
[97] Raju, I.S. and Wang, J.T. (1994). Classical laminate theory models for
woven fabric composites, Journal of Composites Technology and Research,
16, 289–303.
[98] Foye, R. (1992). Finite element analysis of the stiffness of fabric rein-
forced composites, National Aeronautics and Space Administration, Langley
Research Center.
[99] Summerscales, J. (1998). Microstructural Characterisation of Fibre-
Reinforced Composites, Oxford: Woodhead Publishing.
[100] Tabiei, A. and Aminjikarai, S.B. (2009). A strain-rate dependent micro-
mechanical model with progressive post-failure behavior for predicting
impact response of unidirectional composite laminates, Composite Struc-
tures, 88, 65–82.
[101] Weibull, W. (1951). A statistical distribution function of wide applicability,
Journal of Applied Mechanics, 18, 293–297.
[102] Bacarreza, O., Aliabadi, M. and Apicella, A. (2012). Multi-scale failure
analysis of plain-woven composites, The Journal of Strain Analysis for Engi-
neering Design, 47, 379–388.
[103] Bažant, Z. and Jirásek, M. (2002). Nonlocal integral formulations of plas-
ticity and damage: survey of progress, Journal of Engineering Mechanics,
128, 119.
[104] Hashin, Z. (1980). Failure criteria for unidirectional fiber composites,
ASME, Transactions, Journal of Applied Mechanics, 47, 329–334.
[105] Jirasek, M., Rolshoven, S. and Grassl, P. (2004). Size effect on fracture
energy induced by non-locality, International Journal for Numerical and
Analytical Methods in Geomechanics, 28, 653–670.
[106] Walrath, D.E. (1980). Fatigue Behavior of Hercules 3501-6 Epoxy Resin,
DTIC Document.
[107] Littell, J.D., Ruggeri, C.R., Goldberg, R.K., Roberts, G.D., Arnold, W.A.
and Binienda, W.K. (2008). Measurement of epoxy resin tension, compres-
sion, and shear stress–strain curves over a wide range of strain rates using
small test specimens, Journal of Aerospace Engineering, 21, 162.
[108] Mazars, J. (1986). A description of micro-and macroscale damage of con-
crete structures, Engineering Fracture Mechanics, 25, 729–737.
[109] Pijaudier-Cabot, G. and Bazant, Z. (1987). Nonlocal damage theory, Jour-
nal of Engineering Mechanics, 113, 1512–1533.
[110] Bazant, Z.P. and Pijaudier-Cabot, G. (1988). Nonlocal continuum damage,
localization instability and convergence, Journal of Applied Mechanics, 55,
287–293.
[111] Bazant, Z.P. and Ozbolt, J. (1990). Nonlocal microplane model for fracture,
damage, and size effect in structures, J. Eng. Mech., 116, 2485–2505.
Chapter 2
MECHANICAL RESPONSE AND FAILURE OF 2D WOVEN

COMPOSITES UNDER COMPRESSION
Nelson V. De Carvalho∗ and Silvestre T. Pinho†

∗ National Institute of Aerospace, 100 Exploration Way Hampton,
Hampton VA 23666, USA
† Department of Aeronautics, Imperial College London,
Prince Consort Road, South Kensington, London SW7 2AZ, UK
2.1 Introduction
The ability to accurately predict and model the mechanical and structural
response of composite materials is essential to optimise their use, to reduce
the number of experiments needed to validate new designs/materials, and
to support the development of improved materials. In the last decades,
the response of unidirectional (UD) composites under different loadings
and their failure mechanisms have been studied by numerous authors.
Although good agreement can be found concerning the description of the
failure mechanisms, their full understanding and accurate modelling are
not yet accomplished [1]. The variability of woven composites, inherent to
their reinforcement architecture, promotes the interaction between differ-
ent micro-mechanical failure mechanisms, increasing the difficulty of their
failure study. Experimental studies on compressive failure of woven compos-
ites have so far focused on final failure morphology and its characterisation
[2–4]. Few works have attempted to study in detail the damage process
leading to failure, and identify key mechanisms [5]. Regarding modelling,
both analytical and numerical, most of the literature focus on predicting
failure and damage under tensile loading e.g. [6–12], and only very few have
attempted to model failure under compressive loading [13].
75
76 N.V. De Carvalho and S.T. Pinho
The present chapter will start with a discussion of experimental obser-

vations aimed at identifying key features of the compressive failure process
of 2D woven composites. This aims at providing experimental evidence that
can aid guiding and validating existing and future approaches. This discus-
sion will be followed by a detail account of how to exploit periodicity in
structures (materials) to gain further insight into their mechanical response
and aid in their detailed modelling. Using this strategy, detailed numerical
and analytical models are developed. These are capable of capturing key
features of the compressive response, and show good agreement with exper-
iments. The numerical and analytical methodology used can be applied to
other weaves and geometries.
2.2 Experimental Observations

The detailed study of the compressive failure mechanisms in woven compos-
ites is hindered by: (i) the typically sudden nature of compressive failure,
and (ii) the need to understand the effect of the reinforcement geometry on
the failure process. Therefore, it is not surprising that most of the literature
has focused on characterising the final failure morphology rather than the
failure process itself. In this sections, a summary of the key experimental
findings regarding the compressive failure process of woven composites is
presented and discussed.
2.2.1 Damage mechanisms

Kink-band formation, intra-ply and inter-ply delamination, matrix cracking
and transverse tow cracking are the main compressive failure mechanisms
of orthogonal 2D woven composites [2–4]. Inspecting the edges of standard
compression specimens just before failure, Reifsnider et al. [5] observed
kinked load-aligned tows, concluding that the latter was the predominant
damage mechanism controlling the failure process. In [14] different rein-
forcement architectures were tested under compression: 2 × 2 twill and
five-harness (5H) satin. The load-aligned tows were seen to fail by kink-
band formation, Figure 2.1. Furthermore, regions of cracked matrix and/or
transverse tows connected adjacent tows failed by kink-band formation,
forming multiple damage bands, Figure 2.2.
2.2.2 Structural role of the tows

In [14] it was observed that tows failed individually with evident out-of-
plane movement (Figure 2.3). Moreover, even in areas where adjacent tows
Mechanical Response and Failure of 2D Woven Composites Under Compression 77
1 mm
100 µm
Fig. 2.1. Detail of a tow failed by kinking in a 2 × 2 twill specimen [14].
Fig. 2.2. Damage band formed by kinked load-aligned tows, matrix and transverse tow
cracking.
fail in neighbouring regions, tows were seen to fail individually (Figure 2.3).
The latter suggests that load transfer between the first tow failing and the
adjacent tow was the mechanism responsible for the local damage propa-
gation, prior to final failure. This observation suggests the tows behave as
structural elements.
3 mm
Fig. 2.3. Several tows failing individually [16].
2.2.3 Effect of the weave architecture and geometry

The damage morphology is also seen to vary with weave architecture and
geometry. Figure 2.4 illustrates a difference in failure morphology between
2 × 2 twill and 5H satin. While in 2 × 2 twill tows tended to fail very
close to the centre of the crimp region d = 0, in the 5H satin tows failed
systematically at a distance d = 0 from it.
2.2.4 Effect of stacking: out-of-plane support

Upon compression, adjacent layers will interact. Breilling et al. [15], study-
ing the compressive failure of a 5H satin carbon-epoxy, concluded that vary-
ing the stacking configuration (i.e. the relative position of peaks and valleys
of adjacent mats with the same orientation), could lead to a reduction of
up to 32% on the ultimate strength. De Carvalho et al. [14], investigated
this interaction further, observing differences in the damage morphologies
obtained by varying stacking configuration. Several specimens were pro-
duced with carefully aligned adjacent layers in an in-phase (IP) configura-
tion, i.e. all load-aligned tows of adjacent layers are IP Figure 2.5. Figure 2.6
shows that the failure morphology can change significantly with the support
provided by the adjacent layers. As highlighted previously, tows behave as
structural elements within the reinforcement architecture; under compres-
sion, the out-of-plane support provided by the adjacent layers affects the
bending of the tows and consequently the failure morphology. This effect
was also seen to be a function of the reinforcement architecture [14]. At
the microscale, compressive failure of woven composites occurs in a similar
d 0
d
d 0
(a) (b)
Fig. 2.4. Differences in failure location between different reinforcement architectures:
2 × 2 twill and 5H satin [14]. (a) 5H satin: tows failed at a distance d = 0 from the centre
of the crimp region. (b) 2 × 2 twill: tows failed at a distance d = 0 from the centre of the
crimp region.
fashion to UD composites: microcracking/plasticity of the matrix between

fibres (within the tows or at the interface between tows and pure matrix
regions) leading to kinking/splitting [16].
2.2.5 Summary
From the experimental observations it is possible to conclude that: (i) tows
behave as structural elements at the reinforcement level, (ii) damage mor-
phology is affected by the weave architecture and geometry, (iii) kinking
of the load-aligned tows followed by inter/intra-ply delamination, matrix
cracking and transverse tow cracking are the predominant damage propa-
gation mechanisms in compression (iv) the support provided by adjacent
layers (stacking configuration), affects the damage mechanisms.
Fig. 2.5. Cross sections of (a) random-stacked, (b) in-phase (IP) and (c) out-of-phase
(OP) laminates.
Failure criteria developed for UD are often applied to predict failure of

2D woven composites. However, the present study indicates that, to capture
the physics of the compressive failure of 2D woven composites, the weave
architecture needs to be considered, both at lamina and laminate level. Unit
cell (UC) meso-scale finite element models have been increasingly used for
predicting the failure strength of 2D woven composites. In this approach,
the details of the weave architecture are included. However, these models
are often single ply models, and therefore do not account explicitly for the
effect of the support given by the adjacent layers. Nonetheless, experimental
observations suggest the effect of the support is not negligible.
Transverse
tow cracking Kink Band
Matrix
cracking
(a)
Equivalent regions
on adjacent layers
Splitting
Delamination Fibre breaks

(b)
Fig. 2.6. Longitudinal compression in a 2 × 2 twill composite [14]. (a) Random stacking
and (b) IP stacking.
2.3 Periodic Boundary Conditions and Reduced Unit Cells

The application of periodic boundary conditions (PBC) to representative
regions has been discussed in various works, e.g. [17–19]. For periodic struc-
tures, the unit cell (UC) is generally used as the representative region, and
the analysis (numerical/analytical) is performed by applying periodic dis-
placement boundary conditions. These types of analysis are very useful to
determine key deformation mechanisms under different loading conditions,
as a function of the structures internal architecture. According to the peri-
odicity definition given in [20], a UC is the smallest sub-domain that allows
a periodic structure to be reconstructed by tessellation of sub-domains that
are physically equivalent to the UC and have parallel local coordinate sys-
tems (LCS). Nevertheless, in most applications the UC is defined such that
the LCS are not only parallel but orthogonally translated. However, equiv-
alent mechanical analysis can in general be performed in domains smaller
then the UC if non-orthogonal translations are considered and/or if sym-
metries operations are applied. Reducing the analysis domain aids in the
identification of key features of the mechanical response of a periodic struc-
ture. Additionally, it can lead to significant reductions in both modelling
and analysis time. This concept has been applied to the numerical analysis
of woven composites [21, 22]. Recently, a generic framework for deriving
PBC for reduced unit cells (rUC) has been proposed [20]. Its application
to a 2 × 2 twill weave is illustrated next.
2.3.1 Reduced unit cells

To assess whether a part of the UC domain can be used to determine the
response of a periodic structure to a given loading two steps are required:
(i) determine whether the selected domain is admissible, i.e select a rUC,
(ii) determine the displacement constraint equations that need to be set at
the boundary to apply the intended loading.
Sub-domain admissibility
A given sub-domain E is admissible for the analysis of a periodic structure
under a given loading εE , if the transformation matrix Ti between the
LCSs of Êi and E, and the load reversal factor γi correspondent to any
sub-domain Êi are such that, for all Êi :
εE = γi Ti εE Tti (2.1)

Equation 2.1 can be used to, for a given applied loading, determine the
load reversal factors γi = ±1 associated with each of the sub-domains. The
admissibility of a subdomain for structural analysis leads to the definition
of a rUC.
Determining periodic boundary conditions

Having identified a potential rUC, and confirmed its load admissibility, the
periodic boundary conditions can be readily derived. The first step is to
determine the geometrical relations between equivalent points at boundary
of the rUC. The latter is obtained applying:

OÊ
E = T xE − xE
xA Â
(2.2)
O
to the boundary of the sub-domain, where xEÊ is the position vector of the
origin of the LCS of the sub-domain Ê given in the LCS of the sub-domain
E; xÂ
E is the position vector of the points at the common boundary of E
and Ê given the LCS of E, and xA E is its equivalent position vector. Having
established the geometrical relation between all equivalent points at the
boundary, the PBCs can be obtained from:
u(A) − γTu(Â) = −εTxOÊ . (2.3)
where A and Â refer to the equivalent points at the boundary, determined by

Eq. 2.2. Once a displacement constraint equation is associated to all points
at the boundary of the sub-domain E, loading can be applied by prescribing
a volume average strain ε. It is relevant to notice that the displacement
constraint equation traditionally used to impose periodic boundary condi-
tions on a UC, see [23] for example, is a particular case of Eq. 2.3 where
the matrix T is equal to the identity matrix I, since the LCS of the UC are
parallel by definition and consequently, from the sub-domain admissibility
evaluation, the load reversal factor γ is equal to one. The application of
Eqs 2.2 and 2.3 will be exemplified next.
2.3.2 Applications
In the present section two applications of the formulation presented pre-
viously are illustrated. The first concerns a particular type of UCs, here
named offset-reduced unit cells (OrUCs), and the second with determining
the minimum rUC. The two cases will be illustrated using a 2 × 2 twill
geometry.
2.3.2.1 Offset-reduced unit cells

As referred in most applications the UC is defined such that the LCS are
not only parallel but orthogonally translated. However, smaller UCs can in
general be defined if non-orthogonal translations are considered, Figure 2.7.
A UC obtained considering non-orthogonal translation is referred to as an
offset-reduced unit cell (OrUC), since it leads to a reduction in the domain
of the traditionally defined UC [20], Figure 2.7. An important feature of
OrUC is that all loading combinations are admissible. Using the present
formulation this feature comes as a natural result: since the LCS of all sub-
domains are parallel, they relate to each other by the identity matrix, i.e.
T = I, as a consequence Eq. 2.1 is always verified and therefore all loading
cases are admissible. The only variables needed to fully define Eq. 2.3,
and prescribe the periodic boundary conditions to the OrUC, are the load-
ing and the geometric relations between equivalent points at its boundary.
The latter are obtained applying Eq. 2.2 to the equivalent domains at the
boundary of the OrUC, and are provided in Table 2.1 and illustrated in
Figure 2.1.
Fig. 2.7. (a) shows a 2 × 2 twill OrUC E and its equivalent domains (b) shows the
geometrical relations between equivalent points at the boundary of E.
January 6, 2015
Mechanical Response and Failure of 2D Woven Composites Under Compression
10:13
Table 2.1. Geometrical relations between equivalent points at the boundary for the 2 × 2 twill OrUC. l, w and t are
respectively, the length width and thickness of the OrUC.
Ê1 Ê2 Ê3 Ê4 Ê5 Ê6
Woven Composites
2 3 2 3 2 3 2 3
w w 2 3 −w −w 2 3
6 l7 6 3l 7 0 6 l 7 6 3l 7 0
x OÊ 6− 7 6 7 4l5 6 7 6− 7 4−l5
4 45 445 4 4 5 4 45
0 0
0 0 0 0
2 w 3 2 w 3 2 w w3 2 w 3 2 w 3 2 w w3
x= x= − ≤x≤ x=− x=− − ≤x≤
6 7 2 6 7 2 6 2 27 6 7 2 6 7 2 6 2 27
6 7 6 7 6 7 6 7 6 7 6 7
6 l 7 6 l l 7 6 7 6 l 7 6 l 7 6 7
6− ≤ y ≤ l 7 6 7 6 l 7 6− ≤ y ≤ l 7 6− ≤ y ≤ − l 7 6 y=−l 7
xÂ 6 2 7 6 4 ≤y≤ 2 7 6 y= 7 6 4 7 6 2 7 6 7
6 4 7 6 7 6 2 7 6 2 7 6 4 7 6 2 7
9in x 6in
4 t t5 4 t t5 4 t t 5 4 t t5 4 t t 5 4 t t 5
− ≤z≤ − ≤z≤ − ≤z≤ − ≤z≤ − ≤z≤ − ≤z≤
2 2 2 2 2 2 2 2 2 2 2 2
2 3 2 3 2 3 2 3 2 3 2 3
1 −w
xÂ 1 −w
xÂ xÂ 1 +w
xÂ 1 +w
xÂ xÂ
6 l7 6 3l 7 6 1 7 6 l7 6 3l 7 6 1 7
6 Â 7 6 Â 7 6xÂ − l7 6 Â 7 6 Â 7 6xÂ + l7
xA 6x2 + 7 6x2 − 7 4 2 5 6x2 − 7 6x2 + 7 4 2 5
4 45 4 45 4 45 4 45
xÂ xÂ xÂ
3 xÂ xÂ xÂ
3
3 3 3 3
b1954-ch02
85
page 85
FACE 2
Ê2
FACE 1
FACE 3
Ê3 E Ê1
Ê4
FACE 4
(a) (b)
Fig. 2.8. (a) shows a 2 × 2 twill rUC E and its equivalent domains (b) shows the
geometrical relations between equivalent points at the boundary of E.
2.3.2.2 Reduced unit cells

Using symmetry operations it is in general possible to define sub-domains
considerably smaller then the UC. Figure 2.8 shows a rUC with the smallest
possible volume that can be used to model a 2 × 2 twill weave. With this
rUC it is also possible to model the out-of-plane support. Two extreme
cases IP, and OP are illustrated in Figure 2.9. All variables used in the
definition of Eq. 2.3 are provided in Tables 2.2 and 2.3.
For the admissible loading corresponding to Case 1 in Table 2.2, the
application of Eq. 2.3 to the boundary of E leads to:
(1) for the boundary between E and Ê1 :
     
u 1   u1  ε11 w
u + u2 = ε21 w , (2.4)
 2    
u3 x=xA −u3 x=xÂ 0
with {xÂ | x = w2 , − 2l ≤ y ≤ 2l , − 2t ≤ z ≤ 2t } and {xA | x = w

2, − 2l ≤
−y ≤ 2l , − 2t ≤ z ≤ 2t }
     
u 1   u1  ε12 l
u + u2 = ε22 l , (2.5)
 2    
u3 x=xA −u3 x=xÂ 0
with {xÂ | − w2 ≤ x ≤ w2 , y = 2l , − 2t ≤ z ≤ 2t } and {xA | − w2 ≤ −x ≤

2 , y = 2 , − 2 ≤ z ≤ 2 }.
w l t t
Ê6-IP
Ê5-IP
(a) rUC used (sub-domain E) and equivalent adjacent sub-domains

in an IP configuration, through-thickness.
(b) Geometric relations between equivalent

points at top/bottom boundaries of the rUC
used in an IP configuration.
Ê6-OP
E
Ê5-OP
(c) rUC used (sub-domain E)and equivalent adjacent sub-domains

in an OP configuration, through-thickness.
(d) Geometric relations between equivalent

points at top/bottom boundaries of the rUC
used in an OP configuration.
Fig. 2.9. rUC used (sub-domain E) adjacent equivalent domains, and geometric
relations between equivalent points at the top and bottom boundaries for IP and OP
laminates.
Table 2.2. rUC1/16 : admissible loadings cases

and respective value of the load reversal factor.
Admissible loading γi
2 3
ε11 ε12 0 ˆ ˜
Case 1 4ε21 ε22 0 5 1 1 1 1
0 0 ε33
2 3
0 0 ε13 ˆ ˜
Case 2 4 0 0 ε23 5 1 1 −1 −1
ε31 ε32 0
Table 2.3. rUC1/16 : geometrical relations between equivalent points at the bound-
aries; l, w and t are respectively, the length width and and thickness of the rUC.
Ê1 Ê2 Ê3 Ê4

2 3 2 3 2 3 2 3
−1 0 0 −1 0 0 −1 0 0 −1 0 0
T 4 0 −1 0 5 4 0 −1 0 5 4 0 −1 05 4 0 −1 05
0 0 −1 0 0 −1 0 0 1 0 0 1
2 3 2 3 2 3 2 3
w 0 −w 0
xOÊ 405 4l5 4 0 5 4−l5
0 0 0 0
2 w 3 2 w w3 2 w 3 2 w w3
− ≤x≤ − − ≤x≤
6 2 7 6 2 27 6 2 7 6 2 27
6 7 6 7 6 7 6 7
6 l 7 6 7 6 l 7 6 7
xÂ 6− ≤ y ≤ l 7 6 l 7 6− ≤ y ≤ l 7 6 −
l 7
6 2 7 6 7 6 2 7 6 7
6 2 7 6 2 7 6 2 7 6 2 7
4 t t5 4 t t 5 4 t t5 4 t t 5
− ≤z≤ − ≤z≤ − ≤z≤ − ≤z≤
2 2 2 2 2 2 2 2
2 3 2 3 2 3 2 3
−xÂ + w −xÂ −xÂ − w −xÂ
6 1 7 6 1
7 6 1 7 6 1
7
xA 6 −xÂ 7 6−xÂ + l7 6 −xÂ 7 6−xÂ − l7
4 2 5 4 2 5 4 2 5 4 2 5
−xÂ
3 −xÂ3 xÂ
3 xÂ
3

     
u 1  u1  −ε11 w
u2 + u2 = −ε21 w , (2.6)
     
u3 x=xA u3 x=xÂ 0
with {xÂ | x = − w2 , − 2l ≤ y ≤ 2l , − 2t ≤ z ≤ 2}
t
and {xA | x =
− w2 , − 2l ≤ −y ≤ 2l , − 2t ≤ −z ≤ 2t }.

     
u 1  u 1  −ε12 l
u2 + u2 = −ε22 l , (2.7)
     
u3 x=xA u3 x=xÂ 0
with {xÂ | − w2 ≤ x ≤ w2 , y = − 2l , − 2t ≤ z ≤ 2}
t
and {xA | − w2 ≤
−x ≤ w2 , y = − 2l , − 2t ≤ −z ≤ 2t }.
(5) for the boundary between E and Ê5 ,
(a) IP case:
     
u 1  u 1   0 
u − u2 = 0 , (2.8)
 2    
u3 x=xA u3 x=xÂ ε33 t
with {xÂ | − w2 ≤ x ≤ w2 , − 2l ≤ y ≤ 2l , z = − 2t } and {xA | − w2 ≤

x ≤ w2 , − 2l ≤ y ≤ 2l , z = 2t }.
(b) OP case:
     
u 1   u1   0 
u − u2 = 0 , (2.9)
 2    
u3 x=xA −u3 x=xÂ −ε33 t
with {xÂ | − w2 ≤ x ≤ w2 , − 2l ≤ y ≤ 2l , z = − 2t } and {xA | − w2 ≤

x ≤ w2 , − 2l ≤ y ≤ 2l , z = − 2t }. Since, xÂ ≡ xA , Eq. 2.9 can be
simplified to:
−ε33 t
u3 = . (2.10)
2
(a) IP case:
     
u 1  u 1   0 
u − u2 = 0 , (2.11)
 2    
u3 x=xA u3 x=xÂ −ε33 t
with {xÂ | − w2 ≤ x ≤ w2 , − 2l ≤ y ≤ 2l , z = 2t } and {xA | − w2 ≤

x ≤ w2 , − 2l ≤ y ≤ 2l , z = − 2t }. Notice that Eq. 2.8 is equal to
Eq. 2.11, since they couple the same boundaries (between E and
Ê5 , and E and Ê6 ).
(b) OP case:
     
u 1   u1   0 
u − u2 = 0 , (2.12)
 2    
u3 x=xA −u3 x=xÂ ε33 t
with {xÂ | − w2 ≤ x ≤ w2 , − 2l ≤ y ≤ 2l , z = 2t } and {xA | − w2 ≤

x ≤ w2 , − 2l ≤ y ≤ 2l , z = 2t }. Since, xÂ ≡ xA , Eq. 2.12 can be
simplified to:
ε33 t
u3 = . (2.13)
2
2.3.3 Summary
The theoretical framework has been illustrated which enables the derivation
of periodic boundary conditions (PBC) for the analysis of domains smaller
then the unit cells (UC), named reduced unit cells (rUC), by exploiting
non-orthogonal translations and symmetries. The use of rUC can greatly
facilitate the detailed modelling of complex periodic structures. Further-
more, it can be useful to provide additional insight into the mechanical
response of a particular periodic arrangement, by identifying its building
blocks and key features.
2.4 Numerical Modelling

Numerical modelling has been extensively used to study the mechanical
properties and failure of woven composites [24]. Different strategies can be
found in literature, e.g. [25–27]. Typically, a finite element model (FEM)
of a representative region is developed at the meso-scale, distinguishing
between tows and matrix, e.g. Figure 2.10. These models can be used as part
of multiscale algorithms [28, 29], and/or as virtual labs to gain insight into
the material deformation mechanisms at the reinforcement level, e.g. [30].
This information can in turn be used in the development and optimisation
of materials, and in the search of alternative/complementary modelling
strategies. In these models, failure is generally simulated through a non-
linear analysis that couples a damage progression scheme, to capture the
nonlinear response of tows and/or matrix prior to failure, and failure cri-
teria applied at the constituent (tow/matrix) level [6–8, 10, 11, 30, 31].
Regarding loading, the majority of the works have focused on tensile fail-
ure prediction, [6–8, 10–12, 30]. Very few studies present compressive failure
Fig. 2.10. Finite element model of the 2 × 2 twill reduced unit cell (rUC). The matrix
was only partially included to enable the visualisation of the tows [37].
predictions [13]. Recently, increased attention has been given to the effect
of the support provided by the adjacent layers on damage and failure, e.g.
[14, 32]. Previous works, both experimental [15, 33] and numerical [34–36],
suggested that this effect, albeit relatively small in terms of stiffness, should
not be neglected when performing strength predictions of woven composites.
Despite these findings, this issue has been often overlooked. The FEM of a
2 × 2 twill reduced unit cell illustrated in Figure 2.10 was developed in [37].
In the out-of-plane direction, two sets representing different cases of out-of-
plane support were applied: (i) in-phase (IP), where all load-aligned tows
of adjacent layers are IP, and (ii) out-of-phase (OP), where all load-aligned
tows of adjacent layers are OP. These correspond to the cases illustrated in
Figure 2.9. These two cases define practical limits of support that any layer
can have within a laminate, where adjacent layers are randomly shifted.
The model was used to investigate the mechanical response and failure in
compression, tension, and under bi-axial loading [37]. Particular attention
was given to the effect of the support provided by the adjacent layers. The
mechanical response and failure in compression will be discussed next.
2.4.1 Geometry, boundary conditions and constitutive

modelling
The internal geometry was defined based on detailed micrographs. The
tows are modelled as an orthotropic material, with material orientations
following the central path of the tow. The matrix is considered to be an
elasto-plastic material and its response is modelled using a linear Drucker–
Prager plasticity model. Additionally, debonding between tows and matrix
is accounted for and modelled through the definition of cohesive contact.
The periodic displacement boundary conditions applied have been derived

in the previous section and are given in Eqs 2.4 to 2.13.
2.4.2 Failure prediction

The failure of the tow is determined using two different sets of failure crite-
ria: maximum stress and physically-based. The material is assumed to fail
when failure of the load-aligned tows is detected using the referred criterion.
The model was also used to predict tensile failure and the bi-axial loading
envelope, the interested reader is referred to [37]. The material is assumed
to fail when the failure of the load-aligned tow is detected.
2.4.2.1 Maximum stress

The maximum stress criteria compare directly the measured strengths of a
given material with the applied stresses. Therefore, compressive failure is
predicted when:
σ11
F IC = − = 1, for σ11 < 0, (2.14)
XC
where XC is the longitudinal compressive strength of the tow.
2.4.2.2 Physically-based
In physically-based criteria, different failure modes are modelled separately
and derived from the physics of the failure process. As suggested by exper-
imental evidence, Section 2.2, longitudinal compressive failure is predicted
using a kinking criterion [38]:

2
2 m
2
m
τ23 m
τ12 σ22 +
F IKIN K = + + = 1, (2.15)
ST − ηT σ22
m SL − ηL σ22
m YT
where ST and SL are the transverse and longitudinal shear strengths, and
YT is the transverse tensile strength. The variables ηT and ηL are the slope
or friction coefficients. The latter increase the respective shear strengths in
the presence of a compressive normal traction and reduce the respective
shear strengths in the presence of a tensile normal traction. The stress
components in the misaligned frame in Eq. 2.15 are given by:

 2 2
σ22 = σ11 sin ϕ + σ22 cos ϕ − 2τ12 sin ϕ cos ϕ
 m

m
τ12 = −σ11 sin ϕ cos ϕ + σ22 sin ϕ cos ϕ + τ12 cos2 ϕ − sin2 ϕ . (2.16)


τ m = τ cos ϕ − τ sin ϕ
23 23 31
The angle ϕ is the sum of an initial misalignment angle ϕ0 , a material

property, and the shear strain γm0 expressed in a coordinate system aligned
with ϕ0 [38]:
ϕ = sign (τ12 ) ϕ0 + γm0 . (2.17)
In the present work the shear strain γm0 is considered to be a linear function
of the shear stress applied, and is obtained by [38]:
ϕ0 G12 + |τ12 |
γm0 = . (2.18)
G12 + σ11 − σ22
2.4.3 Results
LCS
Figure 2.11 compares the values for the direct stress σ11 , (along the LCS
of the tow, see Figure 2.11), obtained with IP and OP cases. Both magni-
tude and gradient (across the thickness) are greater for the IP case. This is
due to the larger tow bending verified in the IP case. In the latter, all load-
aligned tows from adjacent layers displace in a similar fashion. In contrast,
for the OP case, the through-thickness deformation of the load-aligned tows
of adjacent layers is equal and opposite, mutually reducing their effects.
The comparison between IP and OP for other stress components can be
found in [37].
The numerical model captures well the stiffer constitutive responses,
Figure 2.12a. As referred to previously, two different criteria were used
to predict compressive failure: maximum stress and physically-based [38].
Figure 2.12a shows that, in the IP case, the physically-based criterion is
Fig. 2.11. Comparison between the stress field σ11LCS obtained, close to the crimp region
of a load-aligned tow, for IP and OP cases. The stress components are given in the LCS,
represented in the figure, whose direction 1 follows the central-line of the tow. The
homogenised strain applied was εGCS
11 = −0.005 given in the GCS represented in the
figure.
-700 -700 OP
St. Dev.
-600 -600
OP
-500 -500
IP
-400 XC -400
GCS
11
IP
Experiments
[MPa] -300 [MPa] -300
Numerical, IP -543 -564
-485
-200 Numerical, OP -200
Max stress
-100 Physically-based
-100
0 0
-0.001 -0.006 -0.011 -0.016 Max Stress Physically-based Experimental
GCS
11
(a) Constitutive responses (b) Compressive strengths

Fig. 2.12. Comparison between constitutive responses and compressive strengths pre-
dicted numerically and obtained experimentally. Failure is determined using two different
criteria: maximum stress and physically-based.
slightly more conservative. The opposite occurs in the OP case where the
maximum stress criterion is more conservative. Therefore, the physically-
based failure criterion predicts a wider range of failure stresses and strains,
Figure 2.12a. Averaging the IP and OP results, the physically-based failure
criterion and the maximum stress criterion under-predict the compressive
strength by ∼ 4% and 14%, respectively, Figure 2.12b.
The maximum stress criterion predicts similar failure locations for both
LCS
cases, IP and OP, corresponding to the regions where σ11 is minimum,
Figure 2.13a. However, the physically-based criterion predict different fail-
ure locations for the IP and OP models. Figure 2.13b shows that in the
IP case F IKINK , Eq. 2.15, is one at the centre of the tow, while in the
OP case F IKINK is one at the surface of the tow near the crimp region,
LCS
where σ11 is minimum. Modelling damage prior to failure (plasticity of
the matrix and intra-ply delamination) had, for this material, a small effect
on the predicted compressive strength. However, the effect on the tensile
strength prediction was more significant, [37].
2.4.4 Summary
A numerical model consisting of a rUC of a 2 × 2 twill geometry has been
developed. The reduced size of the analysis domain enabled the detailed
modelling of tows and matrix. Two cases of support were considered, IP
and OP, that bound the support a given layer can have within a laminate. It
was observed that the support provided by the adjacent layers (IP vs OP)
affects significantly the stress fields obtained and can not be neglected.
Compressive strength predictions show good agreement with experiments,
particularly when physically-based failure criteria are used.
Fig. 2.13. Failure indexes at failure in compression of the two set of criteria used:
maximum stress (F IC ) and physically-based (F IKINK ).
2.5 Analytical Modelling

Analytical models to predict failure of woven composites can be broadly
divided into meso- and macro-scale models. In macro-scale models, no dis-
tinction is made between reinforcement and regions of pure matrix. The
composite lamina or laminate is regarded as an orthotropic material, defined
by its homogenised properties. In meso-scale models, reinforcement and
matrix are distinguished, and their geometry and properties considered
independently. Few macro-scale models can be found specifically devel-
oped for woven composites, e.g. [39, 40]. Macro-scale models have as main
advantages their simplicity, as well as capability to be adapted to different
reinforcements geometries and types, provided the mechanical tests that
define them are performed. Their main disadvantage is that, since the rein-
forcement is not modelled explicitly, the actual damage mechanisms are
not captured, leading to an arguable lack of physical representativeness.
Various analytical models have been proposed to capture the mechanical
response of 2D woven composites and predict their failure [3, 41–46]. In
general, they are able to provide insight on the stress and strain fields
within the reinforcement. Knowing the strains and stresses within tows
and matrix (or their equivalent), failure prediction is normally made using
a failure criteria applied to tows and matrix, sometimes coupled with a pro-
gressive damage approach. Tows are typically regarded as a UD composite,
and the matrix as an isotropic material. In these type of approaches, the
effect of the internal reinforcement on the damage mechanisms and failure
is explicitly determined. However, the detailed modelling often leads to a
complex formulation and narrower range of application (e.g only one weave
type or loading). Recently, an analytical modelling strategy for 2D woven
composites has been proposed [47]. Following the works from [44] and [45],
the analytical model is based on a beam supported by an elastic founda-
tion. In [47] the elastic foundation is considered to provide not only normal
support, as in [44, 45], but also shear support to the load-aligned tow.
Its characteristics are derived from kinematic models for the deformation
of the weave, and are a function of: (i) weave architecture, (ii) the case of
through-thickness support provided by the adjacent layers being considered
(IP or OP), and (iii) properties of both matrix and transverse tow. The
kinematic models used lead to a formulation where the distributed force
exerted on the tow is proportional not only to the deflection of the tow,
as in [44, 45], but also to the first and second derivative of the deflection.
The model was used to study both the tensile and compressive response
of a 2 × 2 twill weave [47]. The results obtained for compression will be
discussed next.
2.5.1 Model geometry and boundary conditions

Exploiting existing symmetries, it is possible to define a 2D representative
model of a n × n twill weave consisting of half a sinusoidal beam connected
with a straight beam, Figure 2.14. The size of each beam, LA and LB , and
the vertical distance from the axis of the straight beam to the origin of the
global coordinate system (GCS), v in Figure 2.14, are determined so as to
represent the geometry of the load-aligned tow. The centre line of the tow,
given in the GCS of Figure 2.14, can be written as:

w (x) = v sin xπ 0 ≤ x ≤ LA
0 2LA
(2.19)
w0 (x) = v LA < x ≤ L
where L = LA +LB . The boundary and continuity conditions corresponding

to two connected beams simply supported in A and roller-clamped in B are
given in Table 2.4.
Region
g A Region B
2
y,2 1
LCS
CS
v
w0
GCS
x,1
LA LB
L
Fig. 2.14. Geometry of the analytical model of a n × n twill weave, and definition of the
coordinate systems used: global coordinate system (GCS) and local coordinate system
(LCS). The LCS is defined such that direction 1 follows the tow centre-line. Compressive
loading is shown in the figure, but the model is equally valid for tension [47].
Table 2.4. Boundary conditions of the model represented

in Figure 2.14.
x=0 x = LA x=L
dwA dwB dwB
wA = 0 wA = wB , dx
= dx dx
=0
2 2 2 3 3 3
d wA d wA d wB d wA d wB d wB
dx2
=0 dx2
= dx2
, dx3
= dx3 dx3
=0
2.5.2 Weave effect

The weave effect is accounted for by considering that the IP adjacent tows
(in-plane) affect the deflection of each other through the shearing of the
matrix connecting them, Figure 2.15. The same effect is neglected in the
OP case, since the overall tow deflection is much smaller. This effect is
Tl g Tc Tr
wc
wl cl cr wr
Fig. 2.15. Shearing of the matrix connecting two in-plane adjacent tows, (of the same
woven layer) [47].
obtained by realising that the deflection of adjacent tows can be related,

taking into account that equivalent positions in adjacent tows displace in
the same fashion [20]. Having determined the shearing of the connecting
matrix, the distributed load exerted in a given tow by the adjacent tows
can be approximated by:
Gm w (x)
pweave ≈ d+ (2.20)
g
where Gm is the shear modulus of the matrix, g is the gap between adjacent
tows, w(x) the deflection of the tow, and d+ is a geometrical constant
given by:
d+ = 2(v + t), (2.21)
where t is the average thickness of the tow.
2.5.3 Out-of-plane support

2.5.3.1 In-phase
The kinematic model that accounts for the IP support is illustrated in
Figure 2.16a. In this case, the support is essentially provided by the shearing
of the material (matrix and transverse tows) between load-aligned tows.
Assuming that all load-aligned tows from adjacent layers displace in an
identical fashion, the shear strain of the material between tows can be
Undeformed Deformed Undeformed Deformed
dw
dw
hTop hTop − 2 .dw

t dw
dw
t
δu
h
hBot h
Bot
+ 2 .dw
dw
dw
dx dx
(a) IP (b) OP
Fig. 2.16. (a) IP and (b) OP kinematic models.
obtained from geometrical considerations.

t dw
γIP = 1+ (2.22)
h dx
where t is the average tow thickness over the cross section, and h is the
average vertical distance between two load-aligned tows of adjacent plies.
The shear stress applied to the tow can be obtained by multiplying the
shear strain by the homogenised shear modulus Gh of the material between
load-aligned tows:

t dw
τIP = Gh 1 + . (2.23)
h dx
Additionally, the gradient of Eq. 2.23 leads to a direct pressure applied to

the tow given by:
dτ
p dτ = − h · b, (2.24)
dx dx
where b is the tow width and h is the average thickness of the region
between tows.
2.5.3.2 Out-of-phase
In the OP configuration (Figure 2.16b), the support is mainly provided
by the straining of the material between load-aligned tows because of
their deflection in opposite directions. Similar to the IP case, the strain
of the region between tows can be obtained from geometrical considera-
tions, Figure 2.16b. Knowing the strain, the pressure padj , applied by the
load-aligned tows of the adjacent layers, is obtained by:

EhTop (1 − νhTop ) EhBot (1 − νhBot )
padj = 2w(x)b +
hTop (1 + νhTop )(1 − 2νhTop ) hBot (1 + νhBot )(1 − 2νhBot )
(2.25)
T op
where EhBot and νhBot are the homogenised Young’s moduli and Poisson’s
Top
ratios, and h Bot is the distance between the adjacent tows [47]. The super-
scripts ‘Top’ and ‘Bot’ refer to the regions above and below a given tow,
respectively.
2.5.4 Analytical model

The governing differential equation is derived in a general fashion from the
analysis of the equilibrium of a beam element, Figure 2.17, and can be
written as:
d4 w (x) 2 d2 w (x) 2
2
2 d w0 (x)
+ λ IP + λ IP w (x) = −λ3 , (2.26)
dx4 1− OP dx2 2− OP dx2
where λi− IP are a function of the case of support (IP or OP) considered,
OP
and w0 (x) is the initial shape of the tow.
2.5.5 Results
The numerical and analytical results obtained for the direct stresses at
the upper and bottom surface of the tow, and the shear stress at its cen-
tre, obtained under compressive loading are compared in Figure 2.18. The
numerical model is the same as in Section 2.4. Numerical and analytical
results show, in general, good agreement. The analytical model developed
captures well the difference in response between IP and OP cases. The
range of stresses for all cases is well predicted as well as their local trends.
P dM
V M+ dx
dx
t wT (x)
P dV
M V+ dx
dx
x dx
Fig. 2.17. Equilibrium of a beam element.
x/L x/L
0 0
-200 0 0.2 0.4 0.6 0.8 1 -200 0 0.2 0.4 0.6 0.8 1
-400 Analytical Top
Top -400 Top
Analytical (T)
Numerical Bottom Bottom
Series5
-600 Bot -600
LCS -800 LCS -800
σ11 σ11
[MPa] -1000 [MPa] -1000
-1200 -1200
-1400 -1400
-1600 -1600 Analytical
Numerical
-1800 Numerical Analytical -1800
(a) IP - σ1LCS (b) OP - σ1LCS
100 Numerical
25 Analytical
Numerical
50 Analytical
LCS LCS
τ12 τ12 5
[MPa] [MPa]
0 0.2 0.4 0.6 0.8 1
0 -15 x/L
0 0.2 0.4 0.6 0.8 1
x/L
-50 -35
(c) IP - τ12
LCS (d) OP - τ12
LCS
Fig. 2.18. Comparison between analytical and numerical results for σ1LCS and τ12 LCS
given in the LCS following the tow centre line (Figure 2.14). The results were obtained
for a compressive strain applied of εGCS
1 = −0.008. σ1LCS was determined along the
tow and at the centre of the top and bottom surfaces (a) and (b). τ12
LCS was determined
along the tow centre line (c) and (d).
Nevertheless, numerical and experimental results show, in general, better

agreement for IP than for OP.
The analytical model captures well the stiffer experimental curves,
apart from the nonlinear region near failure. The latter leads to a slight
-700 Std. Dev.

OP
-700 -600
-600 OP
-500
-500
XC -400 IP IP
GCS -400 [MPa]
σ11
Experiments -300 -564
[MPa] -300 -527
Numerical, IP -459
-200 Numerical, OP -200
Max stress
-100 -100
Physically-based
0 0
-0.001 -0.006 -0.011 -0.016 Analytical
Max Stress Analytical
Physically-based Experimental
ε11
GCS Max Stress Physically-based
(a) Constitutive responses (b) Strengths

Fig. 2.19. Comparison between (a) the compressive constitutive responses, and (b)
strengths obtained numerically and predicted analytically. Failure was determined using
maximum stress and physically-based criteria. In (b), the limits of the error bars of
the experimental results correspond to the minimum/maximum values of strength regis-
tered. In the analytical predictions, the limits of the error bars correspond to the results
obtained with the IP and OP cases.
under-prediction of the failure strains. As in the numerical case, two dif-

ferent criteria were used to predict compressive failure: maximum stress
and physically-based, see Section 2.4.2. Both criteria have a similar failure
prediction for the IP case. In the OP case, the physically-based failure
criteria predict failure for higher stresses than the maximum stress crite-
ria. Therefore, the physically-based failure criteria predict a wider range of
failure stress and strains. Averaging the predictions for the two cases, IP
and OP, physically-based and maximum stress criteria under-predict the
strength by ∼6.6% and ∼18.6%, respectively, Figure 2.19b.
2.5.6 Summary
An analytical model, based on a beam on elastic foundation, has been
developed. The elastic foundation is considered to provide both normal
and shear support. Its properties are derived from kinematic models for the
deformation of the weave and account for: (i) weave architecture, (ii) sup-
port provided by the adjacent layers and (iii) properties of matrix and
transverse tows.
The local stress predictions obtained analytically compare well with
the predictions made by an equivalent numerical model, both in terms
of maximum/minimum stresses predicted, and local trends. This agree-
ment confirms that the essential physics of the deformation process are
well captured.
The proposed model accurately predicts the tensile and compressive

constitutive response and failure strengths, particularly when coupled with
physically-based failure criteria. Additionally, it also enables the analytical
determination of a range of values for the failure strengths, as a function
of the support provided by the adjacent layers.
The present work focuses on obtaining a closed-form solution for the
generic model proposed. In its current form, the analytical model has a
negligible run time, which can be extremely valuable to perform para-
metric/sensitivities studies. Nevertheless, a numerical solution can also be
implemented, requiring fewer approximations and enabling the inclusion of
non-linear material response.
2.6 Conclusions
Experimental observations show that woven composites share with UD com-
posites the same fundamental damage mechanisms in compression, such as
fibre kinking or matrix cracking. However, evidence suggests that to capture
the physics of the compressive failure of 2D woven composites, the weave
architecture and internal geometry needs to be considered explicitly, both
at lamina and laminate level.
The application of a systematic framework for determining reduced
unit cells (rUC) and respective periodic boundary conditions (PBC) has
been illustrated. Using smaller domains (rUC), enables the detailed numer-
ical/analytical modelling of 2D woven composites at the reinforcement level.
Furthermore, the application of the framework provides additional insight
into the mechanical response of any periodic structure, as it helps identify-
ing its building blocks and key features.
Unit cell FE models of woven composites can be used as part of multi-
scale approaches, or as virtual test labs. By modelling the weave archi-
tecture and internal geometry explicitly, they are capable of providing
invaluable insight into the deformation mechanisms. In the present chapter
emphasis was given to a 2 × 2 twill rUC. In the methodology used, the
effect of the support given by the adjacent layers was taken into account by
considering the practical limits of support that any layer can have within a
laminate, where adjacent layers are randomly shifted. This model was then
combined with physically-based compressive failure criterion, showing good
agreement with the experiments.
Analytical models are key to obtain quick parametric studies, gather
insight into key mechanisms, and validate more complex approaches.
An analytical model that accounts for: (i) weave architecture, (ii) the case
of through-thickness support provided by the adjacent layers being consid-
ered (IP or OP), and (iii) properties of both matrix and transverse tow has
been discussed in detail. The local stress predictions obtained analytically
compare well with the predictions made by an equivalent numerical model,
both in terms of maximum/minimum stresses predicted, and local trends.
Combined with physically-based compressive failure criterion the model is
capable of predicting the compressive strength. The methodology proposed
can be extended to other weaves.
References
[1] Hinton, M. J., Soden, P., Kaddour, A. S. (ed.) (2004). Failure Criteria in
Fibre Reinforced Polymer Composites: The World Wide Failure Exercise
(Elsevier Science).
[2] Wilkinson, E., Parry, T. V. and Wronski, A. S. (1986). Compressive fail-
ure in two types of carbon fibre-epoxide laminates, Composites Science and
Technology, 26, 17–29.
[3] Karayaka, M. and Kurath, P. (1994). Deformation and failure behavior of
woven composite laminates, Journal of Engineering Materials and Technol-
ogy (Transactions of the ASME) (USA), 116, 222–232.
[4] Gyekenyesi, A. L. (1998). Isothermal fatigue, damage accumulation, and life
prediction of a woven PMC, Technical report, NASA/CR-1998-206593.
[5] Reifsnider, K. L. and Mirzadeh, F. (1988). Compressive strength and mode
of failure of 8H celion 3000/PMR15 woven composite material, ASTM J.
Compos. Technol. Res., 10, 156–164.
[6] Blackketter, D. M., Walrath, D. E., and Hansen, A. C. (1993) Modeling
damage in a plain weave fabric reinforced composite material, Journal of
Composite Technology & Research, 15, 136–142.
[7] Dasgupta, A. and Bhandarkar, S. M. (1994). Effective thermomechanical
behavior of plain-weave fabric-reinforced composites using homogenization
theory, Journal of Engineering Materials and Technology (Transactions of
the ASME) (USA), 116, 99–105.
[8] Guagliano, M. and Riva, E. (2001). Mechanical behaviour prediction in plain
weave composites, The Journal of strain analysis for engineering design, 36,
153–162.
[9] Carvelli, V. and Poggi, C. (2003). A numerical approach for the failure anal-
ysis of textile composites, in ICCM-14, (San Diego, CA, USA).
[10] Zako, M., Uetsuji, Y. and Kurashiki, T. (2003). Finite element analysis of
damaged woven fabric composite materials, Composites Science and Tech-
nology, 63, 507–516.
[11] Nicoletto, G. and Riva, E. (2004). Failure mechanisms in twill-weave lami-
nates: FEM predictions vs. experiments, Composites Part A: Applied Science
and Manufacturing, 35, 787–795.
[12] Daggumati, S., Paepegem, W. V., Degrieck, J., Xu, J., Lomov, S. and Ver-
poest, I. (2010). Local damage in a 5-harness satin weave composite under
static tension: Part II — Meso-FE modelling, Composites Science and Tech-
nology, 70, 1934–1941.
[13] Kollegal, M. G. and Sridharan, S. (2000). Strength prediction of plain woven
fabrics, Journal of Composite Materials, 34, 240–257.
[14] Carvalho, N. V. D., Pinho, S. T. and Robinson, P. (2011). An experimental
study of failure initiation and propagation in 2D woven composites under
compression, Composites Science and Technology, 71, 1316 – 1325.
[15] Breiling, K. B. and Adams, D. O. (1996.) Effects of layer nesting on
compression-loaded 2-D woven textile composites, Journal of Composite
Materials, 30, 1710–1728.
[16] Pinho, S. T., Gutkin, R., Pimenta, S., De Carvalho, N. V. and Robinson, P.
(2012). On longitudinal compressive failure of carbon-fibre-reinforced poly-
mer: from unidirectional to woven, and from virgin to recycled, Philosophical
Transactions of the Royal Society A: Mathematical, Physical and Engineering
Sciences, 370, (1965), 1871–1895.
[17] Xia, Z., Zhang, Y. and Ellyin, F. (2003). A unified periodical boundary
conditions for representative volume elements of composites and applications,
International Journal of Solids and Structures, 40, 1907–1921.
[18] Xia, Z., Zhou, C., Yong, Q. and Wang, X. (2006). On selection of repeated
unit cell model and application of unified periodic boundary conditions in
micro-mechanical analysis of composites, International Journal of Solids and
Structures, 43, 266–278.
[19] Sun, C. T. and Vaidya, R. S. (1996). Prediction of composite properties
from a representative volume element, Composites Science and Technology,
56, 171–179.
[20] Carvalho, N. V. D., Pinho, S. T. and Robinson, P. (2011). Reducing the
domain in the mechanical analysis of periodic structures, with application to
woven composites, Composites Science and Technology, 71, 969–979.
[21] Whitcomb, J. D., Chapman, C. D. and Tang, X. (2000). Derivation of bound-
ary conditions for micromechanics analyses of plain and satin weave compos-
ites, Journal of Composite Materials, 34, 724–747.
[22] Tang, X. and Whitcomb, J. D. (2003). General techniques for exploiting
periodicity and symmetries in micromechanics analysis of textile composites,
[23] Suquet, P. (1987). Elements of homogenization theory for inelastic solid
mechanics, in Sanchez-Palencia, E. and Zaoui, A. (eds.), Homogenization
Techniques for Composite Media, Lecture Notes in Physics (Springer-Verlag,
Berlin), 194–275.
[24] Lomov, S. V., Ivanov, D. S., Verpoest, I., Zako, M., Kurashiki, T., Nakai, H.
1870–1891.
[25] Li, L., Wen, P. and Aliabadi, M. (2011). Meshfree modeling and homogeniza-
tion of 3D orthogonal woven composites, Composites Science and Technology,
71, 1777–1788.
[26] Potter, E., Pinho, S. T., Robinson, P., Iannucci, L. and McMillan, A. J.
(2012). Mesh generation and geometrical modelling of 3D woven compos-
ites with variable tow cross-sections, Computational Materials Science, 51,
103–111.
[27] Gagera, J. and Pettermannb, H. E. (2011). FEM homogenization of tex-
tile composites based on shell element discretization, in 16th International
Conference on Composite Structures, ICCS 16.
[28] Whitcomb, J., Srirengan, K. and Chapman, C. (1995). Evaluation of homog-
enization for global/local stress analysis of textile composites, Composites
[29] Fish, J., Yu, Q. and Shek, K. (1999). Computational damage mechanics for
composite materials based on mathematical homogenization, Int. J. Numer.
Meth. Engng, 45, 1657–1679.
[30] Carvelli, V. and Poggi, C. (2001). A homogenization procedure for the numer-
ical analysis of woven fabric composites, Composites Part A: Applied Science
and Manufacturing, 32, 1425–1432.
[31] Yang, Q. D. and Cox, B. (2010). Predicting failure in textile composites using
the binary model with gauge-averaging, Engineering Fracture Mechanics, 77,
3174–3189.
[32] Ivanov, D. S., Lomov, S. V., Ivanov, S. G. and Verpoest, I. (2010). Stress
distribution in outer and inner plies of textile laminates and novel boundary
conditions for unit cell analysis, Composites Part A: Applied Science and
Manufacturing, 41, 571–580.
[33] Mirzadeh, F. and Reifsnider, K. L. (1992). Micro-deformations in
C3000/PMR15 woven composite, Journal of Composite Materials, 26,
185–205.
[34] Woo, K., Suh, Y. W. and Whitcomb, J. D. (2002). Phase shift effect on the
stress distribution for satin weave composites, Journal of Composite Mate-
rials, 36, 271–286.
[35] Rupnowski, P. and Kumosa, M. (2003). Meso- and micro-stress analyses in an
8HS graphite/polyimide woven composite subjected to biaxial in-plane loads
at room temperature, Composites Science and Technology, 63, 785–799.
[36] Le Page, B. H., Guild, F. J., Ogin, S. L. and Smith, P. A. (2004). Finite
element simulation of woven fabric composites, Composites Part A: Applied
Science and Manufacturing, 35, 861–872.
[37] De Carvalho, N. V., Pinho, S. T. and Robinson, P. (2012). Numerical mod-
elling of woven composites: Biaxial loading, Composites Part A: Applied
Science and Manufacturing, 43, 1326–1337.
[38] Pinho, S. T., Darvizeh, R., Robinson, P., Schuecker, C. and Camanho, P. P.
(2012). Material and structural response of polymer-matrix fibre-reinforced
composites, Journal of Composite Materials 46, 19–20, pp. 2313–2341.
[39] Daniel, I. M., Luo, J. J. and Schubel, P. M. (2008). Three-dimensional

characterization of textile composites, Composites Part B: Engineering, 39,
pp. 13–19.
[40] Hochard, C., Payan, J. and Bordreuil, C. (2006). A progressive first ply
failure model for woven ply CFRP laminates under static and fatigue loads,
International Journal of Fatigue, 28, 1270–1276.
[41] Ishikawa, T. and Chou, T. W. (1982). Stiffness and strength behaviour of
woven fabric composites, Journal of Materials Science, 17, 3211–3220.
[42] Ishikawa, T. and Chou, T. W. (1983). Nonlinear behavior of woven fabric
composites, Journal of Composite Materials, 17, 399–413.
[43] Naik, N. K. and Ganesh, V. K. (1996). Failure behavior of plain of weave fab-
ric laminates under on-axis uniaxial tensile loading: II-analytical predictions,
[44] Tan, P., Tong, L. and Steven, G.P. (1999). Micromechanics models for the
elastic constants and failure strengths of plain weave composites, Composite
[45] Naik, N. K., Tiwari, S. I. and Kumar, R. S. (2003). An analytical model for
compressive strength of plain weave fabric composites, Composites Science
and Technology, 63, 609–625.
[46] Edgren, F., Mattsson, D., Asp, L. E. and Varna, J. (2004). Formation of
damage and its effects on non-crimp fabric reinforced composites loaded in
tension, Composites Science and Technology, 64, 675–692.
[47] De Carvalho, N. V., Pinho, S. T. and Robinson, P. (2012). Analytical mod-
elling of the compressive and tensile response of woven composites, Composite
Structures, 94, 2724–2735.

Chapter 3
FROM A VIRTUAL TEXTILE TO A VIRTUAL

WOVEN COMPOSITE
Stepan V. Lomov
Department MTM, KU Leuven, Kasteelpark Arenberg 44
3.1 Introduction
The future of fibre-reinforced composites, in general, and textile composites
in particular, seems bright. Four industrial branches — aeronautic, automo-
tive, sporting goods and wind energy — are now major users of composites,
and their demands will shape composite material science for decades ahead.
With the Airbus-380, Airbus-350 and Boeing-787 flying, and being
produced by the hundreds, the aeronautics industry has a high demand for
further materials, manufacturing and quality control improvement. All lead-
ing car manufacturers have developed concept solutions of composite cars,
with several of them already on the roads or expected there shortly. The
specific price requirements and recyclability regulations of the car industry
shape the research directions in somewhat different ways than in aeronau-
tics. The demands of wind energy industry for composites are extreme,
both in production volumes and in material performance. Turbine blades
of 90 metres, composed fully of composite materials need to run for 25 years
in extreme, off-shore conditions at the same time, meeting the competitive
cost limits of the energy market. And finally, the sporting goods industry
has become almost a “traditional” user of composites, both because they
allow for weight reduction and hence reduce the energy consumption by the
athletes (in cycling . . . ) and they improve the control and efficiency (in ski-
ing, tennis . . . ). Sporting goods have always been, and will remain, an ideal
testing ground for new composites concepts, because the consumer demands
them and the developers are not hindered by too many regulations.
109
110 S.V. Lomov
All these applications are, to a large extent, textile composites, prefer-

ably out-of-autoclave.
This defines a strong industrial demand for adequate predictive models
and simulation tools, which are experimentally validated and based on phys-
ically sound theoretical principles. The demand is driven by the desire to
resolve the conflict between the attractive weight reduction and material
tailoring on the one hand, and the need of costly part/material charac-
terisation and testing on the other hand. The main industries providing
this “pull” are the same aeronautic, automotive, sports and wind energy
companies.
Not only end product manufacturers are pulling for increased modelling
initiatives. It is also supported by the strong need for scientific under-
standing and predictive modelling of the behaviour of the individual con-
stituents and their properties within a composite. Several industrial sectors
are involved in the search for fundamental understanding and modelling:
the chemical and materials industry (matrices and nanoreinforcements), the
fibre manufacturers (carbon, glass, natural fibres) and the technical textile
industry (playing an important part in “resurrecting” the traditional tex-
tile industries in industrial countries). This “industrial pull” is also com-
bined with an “academic push” of fundamental research. In parallel, there
is an important “push” from the community working on the development
of numerical methods and the software industry.
Modelling of textile composites may be seen as an integrated simulation
process which involves two integration paths. On one hand, the behaviour
of a composite (be it in processing or in performance) is determined by its
reinforcement fibrous architecture. This dictates the necessity to develop
a scale integration of the models: from micro-scale, representing the local
behaviour of dry or impregnated fibrous yarns and plies, through meso-
scale, corresponding to a unit cell (representative volume element, (RVE))
of textile reinforcement up to macro-scale analysis of a composite part. On
the other hand, processing of a composite (involving reinforcement defor-
mation during draping, its impregnation and the material consolidation)
defines the final reinforcement architecture, as well as its defects (voids,
fibre misplacement . . . ). Hence the performance models on all the scale
levels should be integrated with process models (Figure 3.1).
This chapter is dedicated to the meso-level modelling of internal archi-
tecture, mechanical behaviour, and damage of 2D and 3D woven compos-
ites, focusing on meso-level models of a textile reinforcement unit cell. The
chapter is intended to represent the personal views and approach of the
From a Virtual Textile to a Virtual Woven Composite 111
Fig. 3.1. Integration of textile composite models.
author in the context of the state-of-the-art rather than to be a review of

the current literature and the history of the subject; therefore, the reference
list is by no means exhaustive.
3.2 General Organisation of a Meso-Level Textile

Pre-Processor
Meso-level (unit cell) modelling of textile composites is an established field,
with a plethora of works published, and in-house, commercial and open
source software tools available, for example, WiseTex [1, 2] and TexGen [3].
A meso-level textile processor (MLTP) can be defined as a numerical tool,
which:
— Accepts information of textile reinforcement parameters (such as weave

structure, yarn spacing, yarn dimensions etc.), parameters of the com-
posite (overall fibre volume fraction, ply thickness, nesting of the plies
etc.) and local (in relation to a scale of a composite part) overall defor-
mation of the reinforcement (shear, compression etc.)
— Creates a geometrical model of the reinforcement in the given textile
composite.
The geometrical model can be further processed by different models

to produce such parameters as permeability of the textile reinforcement,
112 S.V. Lomov
stiffness matrix, thermal conductivity or dumping parameters of the con-

solidated composites etc. Moreover, the unit cell geometrical model can
be transformed into a “general purpose” meso-level finite element (FE)
model of the unit cell, allowing further in-depth simulation of the com-
posite/reinforcement properties and behaviour. The reader is referred to
[1, 4–6] for more detailed discussion of these integrations.
Given an MLTP, the user gains access to its functionality via its
“native” data formats and user interface. Plugging the MLTP into an inte-
grated simulation chain, however, most often requires interaction and active
collaboration with the author(s) of the MLTP, and is not a trivial task.
The aim of the present chapter is to demonstrate the possibilities, given
by the open data exchange format (extensible markup language (XML))
and command line scripting possibilities, on an example of WiseTex, an
MLTP developed in KU Leuven [1]. These features allow integration of the
MLTP into custom modelling systems, addressing such tasks as: (1) para-
metric studies of composite properties, depending on the parameters of
the textile reinforcement, (2) integration with meso-level finite FE mod-
elling, (3) upstream integration with textile process models, (4) upstream
integration with simulations of composite processing and (5) downstream
integration with structural and impregnation analysis of composite parts.
3.2.1 Data exchange and data formats

The general data flow in an MLTP is shown in Figure 3.2. An MLTP can be
seen as a processor of textile data (TEX) and deformation data (DEF) into
Specifications
and
measurements
Permeability
model
Textile
processing
model Textile
Geometrical Micro-mechanical
data (TEX)
MLTP model model
(GEO)
Specifications Textile
and deformations Meso-FE
measurements (DEF)
Thermal,
electromagnetic
etc. models
Composite
processing
model
Fig. 3.2. Data flow of a meso-level textile processor.

a geometrical model (GEO) — these abbreviations will be used in the rest

of the chapter for brevity. If the DEF is absent (no deformation, relaxed
state), then the deformation of the dry textile can be modelled during
post-processing of the GEO, for example, via meso-FE models. TEX and
DEF input are either user specified, or are a result of (pre-) processing in
textile and composite process models. Examples of the former are models
of braiding (e.g., [7–9]), which deliver such parameters as braiding angle
and braiding density depending on the machine settings. For weaving, soft-
ware packages as ScotWeave and its likes are widely used in the apparel
and technical fabrics industry. The latter, composite processing models, are
represented by draping simulations [10, 11], which compute local in-plane
deformations and the thickness of a reinforcement during forming.
TEX data is organised in a hierarchy of data levels for fibres, yarns and
the fabric [12]. The easy and open way of implementing this hierarchy is
use of XML [13], which defines a set of rules for encoding documents in a
format that is both human- and machine-readable. Figure 3.3 shows TEX
data, defined in WiseTex for yarns and woven fabrics in XML format (tree
view), as displayed by the XML Notepad editing tool. Note that the woven
fabric description (Figure 3.3b) includes the yarn data (Figure 3.3a), which,
in turn, includes the fibre data. The data fields are briefly described below,
to the extent needed for understanding the rest of the chapter. The full
Fig. 3.3. WiseTex TEX data for: (a) Yarn, (b) Woven fabric in XML format (tree view),
as displayed by the XML Notepad editing tool.
114 S.V. Lomov
description of the TEX data fields can be found in the WiseTex software
documentation (WiseTex User’s Manual, KU Leuven, 2012 ).
The fibre data (not shown in Figure 3.3) include fibre geometry and
transversely orthotropic elastic constants for the fibre material.
The yarn data (Figure 3.3a) include general yarn parameters such as
linear density, shape and dimensions of the cross-sections, data on the
fibrous structure of the yarn and also a description of the yarn mechan-
ical behaviour during compression, bending and tension. The latter group
of the data is not compulsory. For example, if the compression data is
omitted, the sections of the yarn will retain the specified dimensions in the
fabric; if the bending data is omitted, then crimp of the yarns in the fabric
must be specified by the user, as the crimp balance calculation will not be
possible etc. Note that the yarn data describe the yarn per se — some of
the specified parameters can be modified after the fabric geometry model
has been built, for example, the yarn cross-section dimensions in the fabric
can be different because of the yarn compression. The reader is referred to
[4, 14–18] for details of use of the yarn data in the fabric geometry and
deformation models.
The woven fabric data (Figure 3.3b) contain data on the yarn placement
density (the distance between the yarn centrelines p), the weave interlacing
pattern and the placement of different yarns in warp weft. The Weave data
section contains a weave code matrix W, which allows coding of 2D and
3D weaves (the detailed explanation of the weave codes can be found in
[12, 14–16]. Finally, the Modelling Parameters data section holds informa-
tion on the computational parameters used for building the geometrical
model.
The XML TEX data can be modified either manually, using an XML
editor or directly using the text representation of the XML file, or via
a custom program, hence it is open for integration in custom simulation
software. The freeware and open source libraries for manipulation of XML
files are available for all programming languages, for example, [19]. The
TEX XML data is also open in a sense that it can be augmented for different
MLTPs, but the presence of already defined fields will ensure compatibility
of the formats with WiseTex.
3.3 Sources of the TEX Input

Collection of reliable TEX input may present a problem, because of two rea-
sons. First, the manufacturers’ specifications do not contain data necessary
for the mechanical characterisation of yarns (for example, compression

diagram or bending rigidity). Second, during composite production the
preform is deformed in-plane and compacted through-the-thickness, which
changes the dimensions and shapes of the yarn cross-sections. Acquiring
these data can involve time-consuming and costly experiments, and depen-
dency of the data on the specific case of the composite fibre volume frac-
tion or thickness lowers the predictive capabilities of the model. However,
there are numerous “shortcuts” which allow estimation of the missing data
using certain assumptions on the internal structure symmetry or mechanical
behaviour of the yarns, or “master curves” supplying the typical parameters
for yarns/tows of given linear density or fibre count.
The most important “shortcut” is for calculations of the internal geom-
etry of a composite reinforcement using direct measurements or estimations
of the yarn cross-sectional shape and size for a composite of given fibre vol-
ume fraction, and assuming a certain crimp balance (equal crimp of warp
and weft, straight weft etc.), or adjusting it to fit a measurable parameter,
for example, a ply thickness. Table 3.1 and Table 3.2 show the data sources
for TEX data and applicable shortcuts.
3.4 Result of the Geometrical Modelling: GEO Data

The result of the geometrical modelling is shown in Figure 3.8. The GEO
data organisation is shown in Figure 3.9. The midline of a yarn is given
by an array of the spatial positions of the centre of the yarn cross-sections
O: r (s), where s is the co-ordinate along the midline and r is the radius-
vector of the point O. Let t(s) be the tangent to the midline at point O. The
cross-section of the yarn, normal to t, is defined by its dimensions d1(s)
and d2(s) along axes a1(s) and a2(s). These axes are “glued” with the
cross-section and rotate around t(s) if the yarn is twisted along its path.
Because of this rotation the system [a1, a2, t] may differ from the natural
co-ordinate system along the spatial path [n,b,t] (t = dr /ds, n = dt/ds,
b = t × n). The shape of the cross-section can be assumed to be elliptical,
lenticular etc., as defined in the Yarn section of the TEX data. The shape
type does not change along the yarn, but dimensions d1 and d2 can change
because of the varying compression of yarn in the contact zones and between
them. The reader is referred to [24] for more details of the yarn geometry
description.
Once GEO data is generated by MLTP, the TEX data in principle is not
needed any more for use in subsequent processing: the node WovenFabric is
116 S.V. Lomov
Table 3.1. TEX data sources: geometrical data.
Parameter Source “Shortcut” Comment
Fibre
Fibre diameter MS*
Yarn
Linear density MS
Cross-section measurement assume ellipse see Figure 3.4
shape
Cross-section measurement use typical data for the see Figure 3.5
dimensions given yarn type
thickness, balanced
fabric: 1/2 of the ply
thickness
Woven fabric
Areal density MS
Ends/picks count MS measurement on the
fabric/composite
surface image
Crimp balance calculation need for yarn
by MLTP mechanical data
(Table 3.2)
estimation special cases: balanced no yarn mechanical
fabric, straight warp or data specified
weft
fitting the given ply
thickness
∗ MS = manufacturer’s specifications/data sheets
Table 3.2. TEX data source: yarn mechanical data.
Parameter Source “Shortcut” Comment
Bending rigidity measurement calculation as sum of fibre-bending Figure 3.6c

rigidities, Figure 3.7
Compressibility measurement use typical behaviour Figure 3.6b
Tension diagram measurement none: simple linear approximation Figure 3.6a
leads to errors — initial
non-linearity must be included
Friction measurement use typical data [22, 23] Figure 3.6d
not present in the GEO XML file (Figure 3.9a). GEO data still holds Yarn
nodes with information about yarns, which is kept from the TEX data set.
In this sense the GEO data format is generic and is the same for woven,
braided or knitted textiles: all of them are represented as a set of tube-like
Fig. 3.4. Measurement of yarn cross-section shape: (a) On a micrograph [20], (b) On
micro-computed tomography (CT) [21].
8.00 1.2
0.5117
y = 0.0883x y = 1.790E-04x + 1.691E-01
7.00 2
R = 0.6743 1 2
R = 9.896E-01
6.00
0.8
Thickness, mm
Tow width, mm
5.00
4.00 0.6
3.00
0.4
2.00
0.2
1.00
0.00 0
0 1000 2000 3000 4000 5000 6000 0 1000 2000 3000 4000 5000 6000
Linear density, tex Linear density
(a)
5 2
4.5 1.8
4 1.6
Thickness, mm
3.5 1.4
Width, mm
3 1.2
2.5 1
2 0.8
1.5 0.6
1 0.4
0.5 0.2
0 0
0 20 40 60 80 100
К (thousands of fibres)
(b)
Fig. 3.5. Thickness and width of typical glass (a) and carbon (b) tows.
volumes. However, it could be beneficial in some cases to keep specific textile

data (TEX data sections) together with the GEO data created for them.
In this case the corresponding XML elements (e.g., WovenFabric) will be
kept in the XML file.
118 S.V. Lomov
(a) (b)
(c) (d)
Fig. 3.6. Textile measurements: (a) Tension, (b) Compression with a glass plate for the
simultaneous measurement of yarn thickness, (c) Bending: cantilever method and KES-F
device, (d) Friction, twisted strand method.
Fig. 3.7. Comparison of bending rigidity B of typical glass and carbon tows, measured
on a KES-F device and estimated as a sum of their fibre-bending rigidities.
Fig. 3.8. Textile geometrical models: (a) Laminate of carbon/aramid woven composite
reinforcement, (b) Multi-layered fabric, (c) Three-axial braided fabric, (d) Weft-knitted
fabric, (e) Unidirectional laminate, (f) 3D woven fabric.
3.5 Deformations of the Woven Unit Cell

The mechanical nature of the model of internal geometry of a woven and
braided fabric allows upgrading them into analytical models for calculation
of resistance of these fabrics to deformation during forming (compression,
biaxial tension and shear), and the prediction of geometry of a deformed
unit cell of the fabric. Note that the deformed configuration of the unit
cell (GEO data) is described in the geometrical model using the same data
structure as per the undeformed fabric, as explained in the previous section.
120 S.V. Lomov
Fig. 3.9. WiseTex GEO data: (a) General parameters of the geometrical model,
(b) Geometry of the yarn volumes, (c) Schematics of the yarn cross-sections.
Table 3.3. Textile deformation data (DEF).
Deformation type Data
Shear Shear angle

Biaxial tension Deformations in two directions
Compression (in the thickness direction) Applied pressure or
Final thickness/final fibre volume fraction
DEF data (see Figure 3.2), which is an input for the calculation of deformed
geometry, is summarised in Table 3.3.
As shown in Figure 3.2, DEF data can just be given as an input to cal-
culate deformation of the unit cell, without reference to the fabric draping,
or represent local deformations of the fabric draped over a mould, either

measured or calculated using a draping modeller [10].
In this section a short description is given of the analytical models
of fabric deformability. Geometry of the deformed unit cell can also be
calculated using meso-level finite element modelling. The reader is referred
to the book [10], which gives a good overview of the state-of-the-art.
A model of compression of woven fabrics implemented in WiseTex [17]
accounts for two physical phenomena associated with fabric compression:
change of yarn crimp and compression of the individual yarns. The inter-
nal structure of laminated preforms after compression is also affected by a
relative shift and nesting of the layers in lay-up [25].
The biaxial tension of a woven fabric [18] is characterised by change
of the fabric dimensions in warp (x-axis) and weft (y-axis) directions
X = X0 (1 + ex ), Y = Y0 (1 + ey ), where X and Y are sizes of the fab-
ric repeat, subscript “0” designates the undeformed state and ex , ey are
technical deformations of the fabric. The internal structure of the fabric
is described based on weft crimp heights, and weft and warp cross-section
dimensions at the intersections. These values change after the deformation.
Tension of the yarns induces transversal forces, which compress the yarns,
changing their dimensions. The same transversal forces change the equi-
librium conditions between warp and weft, which leads to a redistribution
of crimp and change of crimp heights. When the values of the crimp and
yarn dimensions in the deformed configuration are computed, the internal
geometry of the deformed fabric is built using the same algorithm as for
the undeformed fabric.
The model of shear of woven fabrics [18] accounts for the following
mechanisms of yarn deformation, determining the shear resistance: friction,
(un)bending, lateral compression, torsion and vertical displacement of the
yarns. The geometrical model of the sheared fabric is similar to the model
of a biaxial braid (non-orthogonal unit cell), with the additional compli-
cation of the change of yarn cross-section dimensions induced by lateral
compression of the yarns during the fabric shear. When a fabric is sheared,
the deformation is resisted by the friction between yarns, and bending and
compression of the yarns. Friction forces are estimated in the model using
normal forces of the yarn interaction, tension being a pretension normally
employed in the shear test. The transversal forces are increased by the inter-
nal pressure, developed inside yarns due to their lateral compression in the
sheared structure. This is taken into account using the experimental com-
pression diagrams of the yarns. Resistance due to bending is estimated using
122 S.V. Lomov
the difference in bending energies in deformed and undeformed configura-

tions, the latter computed with algorithms for non-orthogonal structures.
3.6 Beyond a Geometrical Model

3.6.1 Permeability
Calculation of permeability is based on a voxel representation of the unit
cell volume (Figure 3.10a). A voxel is either an empty (pore) or filled with
fibres. The flow of the fluid in the pores is governed by Navier–Stockes
equations (NS-voxels) and inside the permeable tows is governed by the
Brinkmann equation (B-voxels). In the latter case, the local permeability
(micro-level) is calculated with the formulae of Gebart and Berdichevsky
for a unidirectional array of fibres. These equations are solved by numer-
ical schemes based on lattice Boltzmann [26], or finite difference [27, 28]
solver of Navier–Stokes or the Stokes equation. The homogenised perme-
ability of a unit cell is then determined using an average flux of the fluid
through the unit cell under periodic boundary conditions for the given pres-
sure difference on the unit cell facets (Figure 3.10b). These algorithms are
implemented in FlowTex software. Figure 3.10c shows the comparison of
the experimental and predicted permeability of reinforcements of different
structure: woven, NCF and random fibres.
If a laborious study of the actual textile geometry within the preform is
made, errors on the permeability of 20–50% are obtained. On the contrary,
Fig. 3.10. Calculation of permeability of a reinforcement: (a) Voxel model of the unit
cell, (b) Boundary conditions and flow lines, (c) Comparison of the experimental and
calculated permeability of different reinforcements.
without detailed information, errors of 100% and even more were found.
In [29] it is shown that for a specific reference specimen, with a well-defined
and fixed geometry, the permeability computed by the Stokes solver lies
within the range of the 10% scatter of the experimental results. This allows
us to conclude that accurate textile permeability computation is possible:
the necessary modelling and simulation tools are available, but precise and
carefully obtained geometrical data of the textiles is indispensable.
Because the geometrical description (GEO data) has the same format
in deformed and undeformed fabric, nothing prevents applying the solu-
tion of Stokes equations to the case of the deformed unit cell, calculating
homogenised permeability locally in a draped preform [28]. Figure 3.11
shows the results of such a calculation for different shear angles and differ-
ent compaction of the preform. The results are compared with the Kozeny–
Carman formula:
r2 (1 − Vf )3
K=
4c Vf2
where K is the fabric permeability, r is the fibre/yarn radius and Vf is

the fibre volume fraction. In Figure 3.11, the Kozeny–Carman formula is
Fig. 3.11. The permeability in the X-direction of a plain woven fabric as a function
of the volume fraction as computed by the Stokes solver. Different volume fractions are
obtained by a combination of shearing (0◦ , 10◦ , 20◦ , 40◦ ) and compaction (0, 0.003,
0.028, 0.053, 0.078 MPa applied pressure). The full line is the Kozeny–Carman perme-
ability for r = 0.2 mm and c = 0.4.
124 S.V. Lomov
applied to inter-yarn permeability (as if they are impermeable — this is

how the calculations were performed), hence the value of r corresponds to
the yarn characteristic dimension and the value of c is chosen to fit the
calculation of an undeformed fabric. The comparison justifies the use of
simple approximations for fast accounting of the preform deformation in
permeability predictions.
The unit cell scale permeability of textile reinforcements is an impor-
tant input parameter for the simulation of impregnation of the preform.
As discussed in [28], the meso-scale simulations should of course be suffi-
ciently fast. The designer of a mould will only use the meso-scale simulation
software if the time it takes to compute the permeabilities is a reasonable
part of the whole design process and if the effort pays off against the use
of experiments. The speed requirements depend on the choice of whether
on-the-fly computations are used, or a look-up table is made and used.
For 101 samples of plain woven fabric, the Stokes solver needs 25 min on
a 2 GHz computer, or about 15 s for one simulation. If the computations
are performed on-the-fly for each element of a mesh with 10000 elements,
the computation time is prohibitive. However, if we use a look-up table
with 600 elements, the Stokes solver needs about 2.5 h, which is acceptable
considering the fact that the table can be reused for every mould design,
and in view of the total time the macro-modelling process takes.
3.6.2 Fast micromechanical models: orientation averaging

and method of inclusions
Calculation of the homogenised stiffness must take into account the fact
that the fibres are organised into impregnated yarns, hence the homogeni-
sation calculation should consist of two steps: first, homogenise the stiffness
of the impregnated yarns, second, homogenise the stiffness of the unit cell,
taking into account their orientation.
3.6.2.1 Orientation averaging

Once the homogenisation of the impregnated yarns is performed, the
homogenised stiffness of the textile composite can be calculated using dif-
ferent methods. The simplest one, orientation averaging (OA) (also called
the fabric geometry model) [30] uses the iso-strain assumption for the
unit cell of the textile composite. Based on the geometry of the reinforce-
ments, N groups of yarn segments are defined inside the unit cell, so that
inside a group i the orientation of the fibres (or centreline of the yarn) is
approximated using the co-ordinate system CSi , aligned with the fibres. Vyi
is the volume fraction of the impregnated yarn segments belonging to the
group i. In the geometrical model described in the previous section, the
segments represent the yarn volumes between the subsequent yarn cross-
sections. The homogenised stiffness of the composite is then calculated as:
N
N

Ceff [GCS ] = Cm [GCS ] · (1 − Vy ) + Cyi [CSi → GCS] · Vyi ; Vy = Vyi
i=1 i=1
where Ceff is the averaged stiffness matrix of the composite, Cm is the

stiffness of the matrix, Cy is the homogenised stiffness of the impregnated
yarns and Vy is the total volume fraction of the impregnated yarns. All the
stiffness matrices are written in or transformed to the global co-ordinate
system (GCS).
With all the simplicity of the OA method it produces reasonable predic-
tions for fibre-dominated properties and low crimp, when deviations from
the iso-strain assumption play a minor role. For off-axis properties, or for
cases when transverse parts of the reinforcement (for example, weft yarns for
warp direction loading) play an important role, or in the presence of high-
crimp yarns (for example, 3D reinforcements with a significant fraction of
binder) the iso-strain assumption is not valid any more, and quality of the
OA predictions decreases. There are approaches which combine iso-strain
and iso-stress formulations [31], but these approaches leave the choice of
the combination rules open, which makes their predictive abilities limited.
3.6.2.2 Method of inclusions

A more advanced method of the stiffness homogenisation for textile compos-
ites is based on the Eshelby solution of the elastic problem for an ellipsoidal
inclusion and Mori–Tanaka homogenisation scheme. The approach was pro-
posed by Huysmans et al. [4, 5, 32, 33] and is successfully applied to very
different textile composites, woven [34], braided [35] and knitted [32, 33].
In short, the approach will be called the “method of inclusions” (MoI).
MoI homogenisation of a heterogeneous medium, which should have a
homogeneous matrix with a second (discontinuous) phase, or more phases
of reinforcement embedded in it, is achieved via the following steps:
1. Build a geometrical model of the RVE of the reinforcement

2. Subdivide the reinforcement into elements, which somehow could be rep-
resented as ellipsoids
126 S.V. Lomov
3. Consider the assembly of the ellipsoidal inclusions in the matrix

4. Using properties of the reinforcement, assign stiffness tensors to the
inclusions (micro-homogenisation may be performed in this step)
5. Apply the Mori–Tanaka/Eshelby theory to calculate the equivalent stiff-
ness of the RVE.
The application of MoI is straightforward if the reinforcing particles/fibres

are short, slender and straight, and can be effectively approximated as ellip-
soids. For textile composites, with a well-organised structure (vs random-
ness in short fibre composites) and consist of crimped yarns (vs straight
short fibres), a mechanically equivalent assembly of inclusions should be
constructed first. This is based on the segmentation of the yarn geometry
in GEO data, created by MLTP.
First, the elastic properties of each impregnated yarn segment are cal-
culated using homogenisation formulae for the unidirectional array of fibres,
using the local fibre volume fraction at the segment, properties of the fibres
and elastic properties of the matrix. The result is the stiffness matrix Cyi ,
expressed in the local 123 co-ordinate system. Then, each yarn segment is
represented by an ellipsoidal inclusion with axis:
√
a1 = αd1 ; a2 = αd2 ; a3 = λ a1 a2
where α is the volume correction coefficient and the elongation λ depends

on the curvature of the yarn segment:
R
λ = 3.14 √
d1 d2
where R is the radius of curvature of the yarn path in the segment
(Figure 3.12). The volume correction factor α ensures that the volume of
the inclusion is equal to the volume of the yarn segment. The stiffness
of the inclusion in the local co-ordinate system CSi is equal to Cyi . Note
that the spatial location of the inclusions is irrelevant for the Mori–Tanaka
homogenisation; the only important geometrical factors are orientation and
elongation of the inclusions.
Figure 3.12b illustrates the inclusion assembly for a very rough seg-
mentation of yarns in a 3D woven reinforcement.
Once the inclusion assembly has been constructed, the homogenised
stiffness matrix of the composite is calculated as follows. First, calculate
the Eshelby tensors S i for the inclusions [36, 37] in local co-ordinates CSi .
Transform the result in the global co-ordinate system GCS. Then, calculate
Fig. 3.12. Equivalent ellipsoidal inclusions for yarn segments: (a) A scheme of the inclu-
sion parameters, (b) An inclusion set for a 3D woven unit cell.
the strain concentration tensors for all the inclusions:

Ai = Ami (cm I + cj Am
j )
−1
where I is the unit tensor, ci are relative volumes of the inclusions and cm
is the relative volume of the matrix:

cm + cj = 1
i
128 S.V. Lomov
and Am
i are calculated as:
Am −1
i = [I + Si Cm (Ci − Cm )]
−1
.
The homogenised stiffness matrix Ceff of the composite is then calcu-

lated as:

Ceff = Cm + ci (Ci − Cm )Ai .
i
For the iso-strain assumption Si = 0 (strain in the inclusions equals

the far field strain), Ai = I, and these formulae are reduced to formulae for
orientation averaging.
3.6.2.3 Example
Examples of fast homogenisation methods for woven composites can be
found in [34], where orientation averaging and the MoI are applied for
2D plain weave laminate and 3D woven composite. Experimental data for
these composites is described in [38, 39]. Figure 3.13 and Table 3.4 show
the architecture of the reinforcement and parameters of the composites.
Composite material samples were fabricated in a vacuum bag with Dow
Derakane 8084 Epoxy-Vinyl Ester resin. WiseTex models (Figure 3.13) for
the composites were built and then the OA and MoI homogenisation algo-
rithms were applied, as described in the previous section.
Table 3.5 shows the calculated and measured mechanical properties of
these two composites: Young’s moduli (E), shear moduli (G) and Poisson’s
ratios (ν). The comparison shows that for the 2D woven composite, the
OA and MoI give results very close to the experimental data for all the
considered engineering constants. All theoretical values of Poisson’s ratio
Fig. 3.13. Unit cells of woven composites, modelled with OA and MoI.
Table 3.4. Parameters of woven composites, modelled with OA and MoI.
Parameter 2D woven laminate 3D woven composite
Fibres Hybon 2022 E-glass Hybon 2022 E-glass

Number of plies 4 1
Total areal density, g/m2 3260 3255
Ends per 1 cm per layer 5.08 2.76
Picks per 1 cm per layer 6.19 2.64
Z-yarns per 1 cm n/a 2.76
Fibre volume fraction 0.52 0.489
Yarns linear density, tex

Warp 2275 2275 (layers 1,3)
1100 (layer 2)
Weft 2275 1470
Z-yarns n/a 276
Table 3.5. Calculated and measured mechanical properties of woven composites.
2D plain weave 0/90/90/0

Property composite 3D weave composite
exp OA MoI exp OA MoI

E1 , GPa 26.0 ± 1.5 25.1 25.4 24.3 ± 1.2 22.7 24.17
E2 , GPa 26.0 ± 1.5 25.1 25.4 25.1 ± 2.34 22.8 24.21
E3 , GPa n/a 12.2 8.63 n/a 10.1 9.11
G12 , GPa n/a 4.4 4.2 n/a 3.38 3.22
ν12 0.264 ± 0.148 0.118 0.117 0.141 ± 0.071 0.109 0.161
ν13 n/a 0.372 0.396 n/a 0.377 0.370
ν23 n/a 0.372 0.396 n/a 0.380 0.368
E45◦ , GPa 12.2 ± 0.4 13.6 13.0 12.9 ± 0.5 10.7 10.4
G45◦ , GPa n/a 11.2 11.4 n/a 10.3 10.8
ν45◦ 0.610 ± 0.148 0.524 0.549 0.502 ± 0.21 0.581 0.618
Notes: “±” — standard deviation in nine tests; subscripts designate loading directions:
“1” warp direction, “2” weft direction, “3” through-the-thickness direction, “45◦ ” bias
in-plane direction.
ν12 are consistently lower (about two-fold) than the average experimental
value, but they are near the edge of the experimental scatter, which is
very large for this characteristic. For the 3D woven composite OA provides
considerably lower values for E1 , E2 and ν12 than experimental data. The
MoI results are considerably closer to experimental data (especially for E1 ),
however it overestimates ν12 . MoI gives sufficiently accurate predictions of
the in-plane effective elastic properties for both 2D and 3D woven compos-
ites, including properties in the off-axis loading.
130 S.V. Lomov
It has to be pointed out again that OA and MoI methods were originally
developed for effective elastic property predictions. They do not provide
stress-/strain fields and cannot be used for failure initiation, progression and
strength predictions. Contrary to that, the finite element method (discussed
in the next section) is not primarily aimed at effective elastic property pre-
dictions. Its computational expense is incomparably higher compared with
the other two aforementioned methods. Hence, its application is advisable
only in those cases when those much simpler methods produce obviously
inadequate or inaccurate results.
3.7 From a Geometrical Model to Finite Element Analysis∗

Meso-scale (unit cell) finite element modelling (meso-FEM) is an important
research direction in the numerical simulation of the mechanical behaviour
of textile composites and for the prediction of their stiffness, strength and
damage properties [3, 6, 40]. It is easy to transform the yarn volumes from
GEO description into an FE package; GEO also provides information of
local fibre directions, allowing the assignment of anisotropic properties to
the impregnated yarns (Figure 3.14).
The difficulties start from this moment. The first of the difficulties is
inter-penetration of the yarn volumes defined by approximate GEO data.
The reader is referred to the papers cited above, especially to [40] for an in-
depth discussion of inter-penetrations. This section concentrates on another
problem, namely meshing of complex volumes of the matrix, full of difficult
wedge-like regions, which create very bad finite elements. It is based on the
recent paper [41].
The common approach in meso-FEM is the conventional continuous
mesh method, called “full modelling” for brevity. The matrix and rein-
forcement volumes, for example, yarns in textile composites, unidirectional
fibres in a ply and randomly placed short fibres, are merged to form a unit
cell. There are two different approaches that are commonly used to create
the matrix and the reinforcement parts inside the unit cell volume (“box”):
(a) the reinforcement volume is subtracted from the unit cell box, then
the “holed” matrix and the reinforcement parts are combined together to
create the partitioned unit cell volume, (b) partitioning the spatial locations
(surfaces/volumes) of the reinforcement parts from the matrix without cut-
ting the matrix. Then, reinforcement material is assigned to the partitioned
∗ Section 3.7 was written together with S.A. Tabatabaei.

Fig. 3.14. Transformation of a geometrical model of a woven fabric into the ABAQUS
FE model. Below: one yarn out of the fabric with the assigned local co-ordinate systems.
sections. The former method (subtraction) is widely used in meso-FE mod-

elling of fibre-reinforced composites but it has major technical drawbacks
as: (1) after assemblage of the parts, the previously defined local co-ordinate
systems/material orientations are deleted and should be redefined, (2) dif-
ficult quality meshing of the matrix volume near the matrix and reinforce-
ment part interfaces in the final model. The latter method (partitioning) is
more appropriate for 2D models since it is difficult to separate the volume
of the complex reinforcement parts from the matrix in 3D models.
Non-continuous meshing techniques present a radical solution of the
meshing problems in FE models of heterogeneous materials. Fish [42] intro-
duced the superposition method (a sequence of overlapping FE meshes) or
“s-version” of the finite element mesh and applied [43] the superposition
technique to hierarchical modelling of laminated composites and discon-
tinuous stress-strain fields. The superposition method was applied to the
meso-FE modelling of textile composites by two research groups who called
it the “superimposed mesh”, “M3 method” [44, 45] or domain superposition
technique [46]. We apply and study the “embedded element” (EE) feature
of the widely used ABAQUS package as a mesh superposition technique.
132 S.V. Lomov
In the EE method, the reinforcement part (“embedded”) mesh is placed

inside the matrix part (“host”) mesh and the “embedding equation” is
created for defining a relationship between the degrees of freedom (DOF)
of the two meshes. The host part is the main part, which is considered as an
independent model from the point of view of translational DOF. ABAQUS
creates geometric relationships between the nodes of the embedded and
host elements. If an embedded element node lies within a host element,
the translational DOFs at the node are eliminated and the node becomes
an “embedded node”. The translational DOF of the embedded node are
constrained to the inter-polated values of the corresponding DOF of the host
element. Embedded elements are allowed to have rotational DOF, but these
rotations are not constrained by the embedding. In the EE method, the
reinforcement parts have a relationship with the host part; at the same time
the EE method provides an opportunity to define a contact relationship (as
a solution for the inter-penetration problem) between the reinforcement
parts.
The EE method has the potential to provide an effective solution of
the problems inherent to the full method in meso-FE modelling of fibre-
reinforced composites:
— Inter-penetration of the reinforcement parts, which in principle can even

be meshed separately; however, in the present work inter-penetrations
were eliminated using a contact algorithm applied to the yarns before
the matrix volume was constructed
— Difficulties in matrix meshing, as the volume geometry of the separately
meshed parts is much simpler than in a full model
— Necessity of redefinition of the local co-ordinate systems after Boolean
operations of partitioning, as no partitioning is performed.
As an example of the use of EEs, the model of a 5H satin reinforced

carbon/epoxy composite is chosen. The full model calculations are reported
in [47, 48]. We have used the ABAQUS models kindly provided by the
authors of [47, 48] to benchmark the new EE calculations. The reader is
referred to the original papers for details of building the full models.
In the EE method, the unit cell modelling was performed as follows.
The WiseTex geometrical model (the same as used in the original papers)
was directly imported to ABAQUS using the Python script, resulting in
creation of the yarn volumes, as in Figure 3.14. The yarn volumes were
corrected to eliminate the inter-penetrations using the contact algorithms of
ABAQUS. Then the “matrix box” was created in ABAQUS and assembled
with the yarns. The yarns and matrix were meshed separately using 3D
8-node linear brick elements (C3D8R) and were put together to create the
proper unit cell. The ratio of the total number of matrix elements to yarn
elements is nearly 10. In all models the parts are meshed in a way that the
aspect ratio of the elements is in the range of 1 . . . 3. This is a significant
improvement of the mesh quality compared with the full models, where the
element aspect ratio reaches 200 (at the surface of the unit cell). The mesh
is illustrated in Figure 3.15.
Table 3.6 compares the homogenised elastic properties of the 5H satin
composite, calculated using a full mesh and the EEs. There is a reasonable
agreement between the results of the EE and full methods. The maximum
difference between the EE and full methods is about 12% for Poisson’s
ratios.
(a) (b)
(c) (d)
Fig. 3.15. Embedded element model of 5H satin composite: (a) Full model, (b) Mesh
in the yarns, (c) One yarn, (d) Mesh of one yarn embedded in the matrix mesh.
134 S.V. Lomov
Table 3.6. Elastic properties of carbon/epoxy 5H satin composite calculated using

full mesh and embedded elements.
Embedded
Mechanical Full method element Experiment Difference (%)
properties (FM)[19] method(EEM) (EXP)[20] (EEM vs FM)
E XX (GPa) 56.49 56.63 57 ± 1 0.25

E YY (GPa) 56.41 56.63 -NA- 0.39
E ZZ (GPa) 10.53 10.55 -NA- 0.15
νXY 0.08 0.09 0.05 ± 0.02 12.5
νXZ 0.41 0.46 -NA- 12.4
νYZ 0.41 0.46 -NA- 12.4
G XY (GPa) 4.28 4.77 4.17 11.4
G XZ (GPa) 3.05 3.26 -NA- 6.96
G YZ (GPa) 3.05 3.26 -NA- 6.96
Fig. 3.16. 5H satin model, stress profiles in warp yarns, six profiles for different yarns
in the unit cell are shown together: full model (a,d) and EE model (b,e), comparison full
vs embedded methods, one yarn (c,f).
The stress fields in warp and weft yarns are compared using the local
longitudinal and transverse component of stress (S11 and S22), shown in
Figure 3.16 and Figure 3.17. In all stress profiles, the stresses are calculated
at the centroid of the elements in the middleline of the yarn and the origin
Fig. 3.17. 5H satin model, stress profiles in weft yarns, six profiles for different yarns
in the unit cell are shown together: full model (a,d) and EE model (b,e), comparison full
vs embedded methods, one yarn (c,f).
of the X axis is the apex element of the yarn. Warp yarns are numbered 1,
2, 3, 4, 5 and 1∗ and the weft yarns are 6, 7, 8, 9, 10 and 6∗ . The yarn 1∗ /1
and yarn 6∗ /6 are two parts of the yarns at the border of the unit cell. The
profiles for the different yarns, with co-ordinates shifted according to the
position of the yarns in the weave, coincide with a good accuracy — this
confirms the correct application of the periodic boundary conditions.
From the stress patterns and profiles, it can be seen that there is an
acceptable agreement between the full and EE method. The overall trend
of the profiles is similar, and the maximum and minimum locations in both
methods coincide. These locations are important in damage analysis of the
reinforced composites, since they correspond to the most probable loca-
tions for damage initiation. The maxima of longitudinal stress in warp
(loading direction) are different by about 8%, transverse stress in warp and
weft differ by about 5% — this gives an estimation of probable differences
between the two methods in the damage initiation calculation for fibre and
matrix damage. The importance of these differences, for calculations of pro-
gressive damage modelling and strength predictions, should be investigated
in future work.
136 S.V. Lomov
3.8 Outlook
Modelling of textiles and textile composites is a mature research direction,
which has achieved important goals in industrial applications to practical
design problems. It can be envisaged that the following goals will be reached
in the near future:
— Modelling of textile structures: “push button” software tools for the

transformation of micro-CT images of textile composites into geometri-
cal models and of GEO data of geometrical models into consistent, good
quality finite element meshes, ready for a solution
— Modelling of damage: validated software tools for damage prediction in
textile composites in quasi-static and fatigue loading, with an effective
link between micro-, meso- and macro-simulations
— Modelling of manufacturing: validated software tools for the deforma-
bility and permeability of textile reinforcements, also with an effective
link between meso- and macro-simulations.
References
[1] Verpoest, I. and Lomov, S.V. (2005). Virtual textile composites software
Wisetex: Integration with micro-mechanical, permeability and structural
analysis, Composites Science and Technology, 65, 2563–2574.
[2] Lomov, S.V., Verpoest, I., Cichosz, J., Hahn, C., Ivanov, D.S. and Verleye B.
(2014). Meso-level textile composites simulations: open data exchange and
scripting, Journal of Composite Materials, 48, 621–637.
[3] Sherburn, M. (2014). TexGen open source project: Accessed from:
http://texgen.sourceforge.net/.
[4] Lomov, S.V., Gusakov, A.V., Huysmans, G., Prodromou, A. and Verpoest, I.
(2000). Textile geometry preprocessor for meso-mechanical models of woven
composites, Composites Science and Technology, 60, 2083–2095.
[5] Lomov, S.V., Huysmans, G., Luo, Y., Parnas, R., Prodromou, A., Verpoest, I.
and Phelan, F.R. (2001). Textile composites: Modelling strategies, Compos-
ites Part A, 32, 1379–1394.
[6] Lomov, S.V., Ivanov, D.S., Verpoest, I., Zako, M., Kurashiki, T., Nakai, H.
1870–1891.
[7] Pastore, C.M. and Ko, F.K. (1990). Modelling of textile structural compos-
ites. Part I: Processing-science model for three-dimensional braiding, Journal
of the Textile Institute, 81, 480–490.
[8] Kessels, J.F.A. and Akkerman, R. (2002). Prediction of the yarn trajectories
on complex braided preforms, Composites Part A 33, 1073–1081.
[9] Pickett, A.K., Sirtautas, J. and Erber, A. (2009). Braiding simulation

and prediction of mechanical properties, Applied Composite Materials, 16,
345–364.
[10] Boisse, P., editor (2011). Composite Reinforcements for Optimum Perfor-
mance, Oxford: Woodhead Publishing.
[11] Long. A., editor (2007). Composite Forming Technologies, Cambridge: Wood-
head Publishing,
[12] Lomov, S.V., Huysmans, G. and Verpoes, I. (2001). Hierarchy of textile struc-
tures and architecture of fabric geometric models, Textile Research Journal,
71, 534–543.
[13] W3C. Extensible Markup Language (XML) 1.0 (Fifth Edition). 2008 [cited
2012]. Available from: http://www.w3.org/TR/REC-xml/.
[14] Lomov, S.V., Verpoest, I. and Robitaille, F. (2005). “Manufacturing and
Internal Geometry of Textiles”, in Design and Manufacture of Textile Com-
posites, Long, A.C., editor. Oxford: Woodhead Publishing., 1–60.
[15] Lomov, S.V., Perie, G., Ivanov, D.S., Verpoest, I. and Marsal, D. (2011).
Modelling 3D fabrics and 3D reinforced composites: Challenges and solutions,
Textile Research Journal, 81, 26–41.
[16] Lomov, S.V. (2005). “Modelling the Geometry of Textile Reinforcements for
Composites: WiseTex”, in Composite Reinforcements for Optimum Perfor-
mance, Boisse, P., editor, Oxford: Woodhead Publishing, 200–238.
[17] Lomov, S.V. and Verpoest, I. (2000). Compression of woven reinforcements:
A mathematical model, Journal of Reinforced Plastics and Composites, 19,
1329–1350.
[18] Lomov, S.V. and Verpoest, I. (2006). Model of shear of woven fabric and
parametric description of shear resistance of glass woven reinforcements,
[19] Thomason, L. (2012). TinyXML [cited 2012]. Available from: http://www.
grinninglizard.com/tinyxml/.
[20] Karahan, M., Lomov, S.V., Bogdanovich, A.E., Mungalov, D. and Verpoest,
I. (2010). Internal geometry evaluation of non-crimp 3D orthogonal woven
carbon fabric composite, Composites Part A, 41, 1301–1311.
[21] Pazmino, J., Carvelli, V., Lomov, S.V. and Verpoest, I. (2012). “Meso-
Scale Analysis of the Deformability OF A Non-Crimp 3D Orthogonal Weave
e-Glass Composite Reinforcement”, in The 4th World Conference on 3D
Fabrics and Their Applications: Aachen, electronic edition.
[22] Cornelissen, B., Rietman, B. and Akkerman, R. (2013). Frictional behaviour
of high performance fibrous tows: Friction experiments, Composites Part
A — Applied Science and Manufacturing, 44, 95–104.
[23] Liu, L., Chen, J., Zhu, B., Yu, T.X., Tao, X.M. and Cao, J. (2006). “The
Yarn-To-Yarn Friction of Woven Fabrics”, in Proceedings of ESAFORM-9,
Glasgow, electronic edition.
[24] Lomov, S.V., Mikolanda, T., Kosek, M. and Verpoest, I. (2007). Model
of internal geometry of textile composite reinforcements: Data struc-
ture and virtual reality implementation, Journal of the Textile Institute,
98, 1–13.
138 S.V. Lomov
[25] Lomov, S.V., Verpoest, I., Peeters, T., Roose, D. and Zako, M. (2002). Nest-
ing in textile laminates: Geometrical modelling of the laminate, Composites
Science and Technology, 63, 993–1007.
[26] Belov, E.B., Lomov, S.V., Verpoest, I., Peeters, T., Roose, D., Parnas, R.S.,
Hoes, K. and Sol, V. (2004). Modelling of permeability of textile reinforce-
ments: Lattice Boltzmann method, Composites Science and Technology, 64,
1069–1080.
[27] Verleye, B., Croce, R., Griebel, M., Klitz, M., Lomov, S.V., Morren, G.,
Sol, H., Verpoest, I. and Roose, D. (2008). Permeability of textile rein-
forcements: Simulation; influence of shear, nesting and boundary conditions;
validation, Composites Science and Technology, 68, 2804–2810.
[28] Verleye, B., Lomov, S.V., Long, A.C., Verpoest, I. and Roose, D. (2010).
Permeability prediction for the meso-macro coupling in the simulation of
the impregnation stage of Resin Transfer Moulding, Composites Part A, 41,
29–35.
[29] Morren, G., Bottiglieri, M., Bossuyt, S., Sol, H., Lecompte, D., Verleye, B.
and Lomov, S.V. (2009). A reference specimen for permeability measure-
ments of fibrous reinforcements for RTM, Composites Part A, 40, 244–250.
[30] Kregers, A.F. and Melbardis, Y.G. (1978). Determination of the deforma-
bility of three-dimensionally reinforced composites by the stiffness averaging
method, Polymer Mechanics, 14, 3–8.
[31] Bogdanovich, A.E. and Pastore, C.M. (1996). Mechanics of Textile and Lam-
inated Composites, London: Chapman and Hall.
[32] Huysmans, G., Verpoest, I. and Van Houtte, P. (1998). A poly-inclusion
approach for the elastic modelling of knitted fabric composites, Acta Mate-
rials, 46, 3003–3013.
[33] Huysmans, G., Verpoest. I. and Van Houtte, P. (2001). A damage model for
knitted fabric composites, Composites Part A, 32, 1465–1475.
[34] Lomov, S.V., Ivanov, D.S., Verpoest, I., Bogdanovich, A.E., Mungalov, D.,
Zako, M., Kurashiki, T. and Nakai, H. (2008). Predictive analyses and experi-
mental validations of effective elastic properties of 2D and 3D woven compos-
ites, in The 13th European Conference on Composite Materials (ECCM-13),
Stockholm.
[35] Birkefeld, K., Roder, M., Von Reden, T., Bulat, M. and Drechsler, K.
(2012). Characterization of biaxial and triaxial braids: Fiber architecture
and mechanical properties, Applied Composite Materials, 19, 259–273.
[36] Eshelby, J.D. (1957). The determination of the elastic field of an ellip-
soidal inclusion and related problems, Proceedings of Royal Society, A-241,
376–396.
[37] Mura, T. (1987). Micromechanics of Defects in Solids, Dordrecht: Kluwer
Academic Publishers.
[38] Lomov, S.V., Bogdanovich, A.E., Ivanov, D.S., Mungalov, D., Karahan, M.
and Verpoest, I. (2009). A comparative study of tensile properties of non-
crimp 3D orthogonal weave and multi-layer plain weave e-glass composites.
Part 1: Materials, methods and principal results, Composites Part A, 40,
1134–1143.
[39] Ivanov, D.S., Lomov, S.V., Bogdanovich, A.E., Karahan, M. and Verpoest, I.
(2009). A comparative study of tensile properties of non-crimp 3D orthogonal
weave and multi-layer plain weave E-glass composites. Part 2: Comprehensive
experimental results. Composites Part A, 40, 1144–1157.
[40] Ivanov, D.S. and Lomov, S.V. (2013). Modeling of 2D and 3D Woven Com-
posites in Polymer Composites in the Aerospace Industry, Oxford: Woodhead
Publishers.
[41] Tabatabaei, S.A., Lomov, S.V. and Verpoest, I. (2014). Assessment of embed-
ded element technique in meso-FE modelling of fibre reinforced composites,
Composite Structures, 107, 436–446.
[42] Fish, J. (1992). The s-version of the finite element method, Computers &
[43] Fish, J., Markolefas, S., Guttal, R. and Nayak, P. (1994). On adaptive mul-
tilevel superposition of finite element meshes for linear elastostatics, Applied
Numerical Mathematics, 14, 135–164.
[44] Kurashiki, T., Hamada, K., Honda, S., Zako, M., Lomov, S.V. and Ver-
poest, I. (2009). Mechanical behaviors of non-crimp fabric composites based
on multi-scale analysis, In The 17th International Conference on Composite
Materials (ICCM-17), Edinburgh: IOM Communications Ltd.
[45] Kurashiki, T., Watanabe, N., Matsushima, M., Zako, M., Lomov, S.V. and
Verpoest, I. (2012). Effects of stitching parameters on damage development
for non-crimp fabric composites under tensile loading, in 15th European Con-
ference on Composite Materials (ECCM-15), electronic edition, Venice.
[46] Jiang, W.-G., Hallett, S.R. and Wisnom, M. (2008). “Development of
Domain Superposition Technique for the Modeling of Woven Fabric Com-
posites”, in Mechanical Response of Composites, Camanho, P.P., editor. New
York: Springer.
[47] Daggumati, S., Voet, E., Van Paepegem, W., Degrieck, J., Xu, J., Lomov,
S.V. and Verpoest, I. (2011). Local strain in a 5-harness satin weave compos-
ite under static tension: Part I — Experimental analysis, Composites Science
and Technology, 71, 1171–1179.
[48] Daggumati, S., Voet, E., Van Paepegem, W., Degrieck, J., Xu, J.,
Lomov, S.V. and Verpoest, I. (2011). Local strain in a 5-harness satin weave
composite under static tension: Part II — Meso-FE analysis, Composites

Chapter 4
MODELLING 3D WOVEN COMPOSITE

PREFORM DEFORMATIONS
Stephen R. Hallett, Steve D. Green and Bassam S.F. El Said

Advanced Composite Centre for Innovation and Science (ACCIS),
University of Bristol, UK
4.1 Introduction
The term “textile composites” covers a wide range of composite materials,
but the common feature is a textile-based preform which is infused with a
matrix material. The textile preform can be manufactured using a number
of different techniques, the most common of which for structural composite
applications are braiding and weaving. The use of textile composites has
advantages over other composite material formats in that they can be more
easily formed to complex shapes, and the textile nature of the preform
means it is held together during handling and manufacture. In the latter
case this is particularly so for materials known as 3D woven composites,
where the weaving process is used to include yarns or tows in the 3rd or
through-thickness direction to create a multi-layer near net-shape preform.
A second advantage of having the yarns in the through-thickness
direction, sometimes referred to as “binder yarns”, is that they provide
through-thickness reinforcement. Conventional 2D composites suffer from
inherently poor out-of-plane properties due to the lack of through-thickness
reinforcements, making them susceptible to delamination, with poor impact
performance [1, 2]. In contrast, the reinforced nature of 3D woven preforms
gives them load-carrying capacity in the through-thickness direction.
Three-dimensional woven composites therefore exhibit enhanced inter-
laminar fracture toughness [1, 3], and better impact and energy absorption
performance [2, 4].
141
142 S.R. Hallett, S.D. Green and B.S.F. El Said
An important factor limiting the use of 3D woven composites is a lack

of material data for use in design. Each set of material properties is only
applicable to a specific architecture and fibre/matrix type. Currently, deter-
mining the mechanical properties of these materials is performed experi-
mentally. Setting up a loom for a specific 3D fabric can take several days
and preforms must then be infused and tested. There is therefore a signifi-
cant need for accurate modelling techniques, particularly in the early stages
of design and preform selection.
Meso-scale finite element (FE) analysis, in which the individual yarns
and matrix material are explicitly modelled, is commonly used for pre-
diction of mechanical properties and performance [5]. Efficient modelling is
based on consideration of the smallest possible region that is still representa-
tive of a larger structure. The nature of textile composites is such that they
exhibit a regular, periodic structure both at the micro- and meso-scales.
Therefore, the entire material can be represented as assembled copies of a
“unit cell” without rotation or reflection. Finite element models of the unit
cell along with the application of appropriate periodic boundary conditions
can therefore be used to model an infinite portion of material [6, 7].
The internal architecture of the meso-scale unit cell is a function of the
weave style and the weaving process. Textile pre-processors such as Tex-
Gen [8] and WiseTex [9] from the Universities of Nottingham and Leuven
respectively, can create geometry of the unit cells, based on knowledge of the
weave style. The weaving process itself is, however, only one factor affecting
the final fabric architecture in a 3D woven composite component. Further
complex deformations can occur during draping when localised deforma-
tions arise as the fabric shears and yarns slide relative to one another [10].
Compaction also deforms the textile further from the idealised assump-
tions. Such deformations can have a detrimental effect on the mechanical
properties of the material. Mahadik et al. demonstrated this effect for an
angle interlock 3D woven fabric, where the amount of waviness present in
the in-plane yarns varied significantly with the level of compaction [11].
When using FE models to predict the mechanical performance of 3D
woven composites, it is thus important that these models consider accu-
rate textile geometry. There have been a number of approaches developed
to generate such geometry. These have included experimental character-
isation of the internal tow geometry, either from cut-ups [12] or X-ray
micro-computed tomography (CT) scanning [13]. Since these techniques
are labour intensive and also require a physical sample of the desired pre-
form material to exist, for a more predictive approach researchers have
Modelling 3D Woven Composite Preform Deformations 143
Fig. 4.1. Workflow for the creation of models with realistic internal architecture for 3D
woven preforms.
developed modelling techniques to determine accurate deformed unit cell

geometry. Modelling techniques employed include using continuum elements
[14], multi-filament methods [15, 16] and hollow shell element tubes [17] for
the tows to determine their interactions with each other and the tooling.
The work presented in this chapter shows how a numerical approach
can be used to obtain an accurate representation of the internal architecture
of 3D woven composites, firstly at the unit-cell scale, and then for larger
structural features. This latter scale is important since once one moves
beyond the level of a single unit cell, more complex interactions with the
tool and overall geometry occur, which can result in variations from a repet-
itive unit cell geometry at different locations in the structure. Figure 4.1
shows the workflow, which has been implemented to achieve a prediction
of deformed geometry at a feature level scale through kinematic modelling
of the individual yarns and internal fabric architecture. This consists of:
• An initial idealised TexGen model is generated to define the preform unit

cell geometry
• From this, a beam element model of the loosely interwoven yarns is
created
• The beam model is solved in the explicit finite element code, LS-Dyna,
to generate a more accurate unit cell model
• The beam model geometry is returned to TexGen

• From this new, more accurate geometry a simplified shell element unit
cell model is created
• The unit cell model is tessellated to form a fabric
• The fabric model is solved in LS-Dyna for large, feature level deforma-
tions, e.g., drape.
Each of the processes in this workflow is described in more detail in the

subsequent sections.
4.2 Unit Cell Beam Model

A multi-filament method for preform deformation modelling has been
implemented in the commercial FE solver, LS-Dyna [18]. Each yarn is rep-
resented by chains of beam elements, where the number of chains in a yarn
is less than the number of actual fibres. Contact between the chains of a
yarn defines the yarn cross-sectional shape and contact between each yarn
assembly defines the yarn paths.
To begin the process an idealised unit cell model of the as-woven fabric
is generated in TexGen. Figure 4.2a, b and c show an orthogonal 3D woven
fabric schematic and the initial geometric model that will be used here as
an example. A beam FE model is then generated from this geometry using
a conversion script (Python). The script interrogates the idealised TexGen
unit cell model in order to extract key parameters of the textile such as
yarn spacing in warp and weft directions, and position of yarns through
the fabric thickness. This yarn is initially assumed to have acircular cross-
sectional shape. Beam elements are arranged in a pattern of concentric
circles to fill the cross-section. The number of beam elements is selected
depending on the level of fidelity required. The script then automatically
generates the beam element chains for the yarns themselves. The geome-
try is then expanded to a loosely woven state to ensure there is no inter-
penetration of the different yarn cross-sections (Figure 4.2d) in this initial
configuration.
A contact formulation is applied to all of the beam elements within the
model so that the interaction of the beam elements with a yarn and also
between adjacent yarns is correctly captured. Before running the model,
periodic boundary conditions are applied through the use of multi-point
constraints. Equivalent nodes at each end of a yarn are constrained to move
together. This is, however, not sufficient for full periodicity as during loading
there will be interaction of the edge yarns with their neighbours, which exist
(a) (b)
(c) (d)
Fig. 4.2. Three-dimensional woven orthogonal fabric geometry (a) Fabric top view and
unit cell with periodic translations, (b) Cross-section A-A schematic, (c) Idealised geom-
etry from TexGen model, (d) Loosely woven beam element model.
outside of the unit cell. To model this, an extra column of warp and weft
yarns is added at each edge of the unit cell, and the nodes of these yarn
elements are constrained to follow the displacement of the equivalent yarn
that exists within the unit cell. Thus, there is the correct contact condition
at the unit cell edges as well as within the unit cell.
In order to compact the loosely woven fabric an artificial coefficient of
thermal expansion and a thermal load in the form of a temperature drop
were applied to the binder yarns. This caused them to contract and hence
apply compaction forces to the fabric, bringing the fabric to an as-woven
thickness of around 7 mm. Rigid plates were then gradually brought into
contact with the fabric to model the mould surfaces. Compared with exper-
imental observations, it was noted that the binder yarns in the model were
relatively tight at the as-woven thickness. Therefore, after initial contrac-
tion of the fabric, a small increase in temperature was applied to these
yarns to better represent the real fabric. During this phase, the plates were
held in position at the as-woven thickness of 7 mm. The final loading stage
in the model was a simulation of mould closure by applying a prescribed
Table 4.1. Material properties applied to yarns in the model com-

pared with carbon fibre.
Binder yarns Warp/weft yarns Carbon fibre
ρ (tonne/mm3 ) 0.05 0.05 1.75e-9

E (GPa) 20 20 230
σy (MPa) 40 10 3400
displacement to the plates in order to compact the fabric to the desired

fibre volume fraction.
The bending behaviour of yarns is important in textile mechanics and
can be assessed experimentally with standard tests such as Kawabata
KES-F [19]. It is not computationally feasible to model a 24k yarn with
24,000 chains of beam elements, thus the discretisation was reduced to 61
chains. Due to this simplification an elastic model of a yarn would lead to
a ∼400 fold increase in bending stiffness compared with the real yarn. In
order to address this, an elastic-perfectly plastic material model was used to
limit the maximum stress which could be sustained and hence aid flexural
deformation of the yarns.
The input parameters used for the model are listed below, with material
properties shown along with those of the actual carbon fibres in Table 4.1.
These values were chosen from a parametric study, which is presented in
full in [20].
• Number of chains per yarn: 61
• Initial cross-sectional shape: Circular
• Friction: µ = 0.35.
It was not attempted to directly determine the model inputs based

on a mechanical approach by, for example, calibrating the model at the
individual yarn level to produce an accurate yarn bending response. Instead
a pragmatic approach was applied at the global fabric level, with the best
input values selected from the parametric study. This selection was based on
the accuracy of correlation to experimental results and speed of the solution
obtained from the model. The model accurately captures the kinematics
of the yarn interactions with each other, the tooling and the boundary
conditions, but it is not representative of the mechanical forces required to
deform and compact the fabric.
The model was run on a high performance computing system using
eight 2.8 GHz cores taking 8 hours and 55 minutes to complete. Most of
the computation time was due to the significant number of contacts in
Fig. 4.3. Comparison of model predictions with X-ray CT scans at ∼56% volume frac-
tion (a) Model binder yarn cross-section, (b) CT scan binder yarn cross-section, (c) Warp
yarn cross-section, CT scan with model yarn paths overlaid, (d) Fabric top view with
weft yarn path highlighted, (e) CT scan fabric top view with weft yarn path highlighted.
the model which were handled by a single contact definition. Figure 4.3
shows some comparisons between the model results and X-ray CT scans of
the orthogonal fabric compressed to the same overall fibre volume fraction.
It can be seen that the model captures a number of key features in the
fabric deformation that are not achieved with this initial idealised geomet-
ric model. Most importantly for mechanical performance, the out-of-plane
waviness or crimp is accurately modelled. Figure 4.3c shows the model
predictions overlaid with the CT scan, whilst Figures 4.3a and 4.3b show
the model and CT data separately. The horizontal shift of the weft stack
of yarns causes local deformation in the warp yarn path. This is caused by
the path of the binder yarn through the thickness of the fabric, which also
causes some in-plane deformation, as shown in Figures 4.3d and 4.3e.
4.3 Conversion from Beam Element Model to Geometry

The output of the deformation models, with yarns represented as a bundle
of beam elements (Figure 4.4a), is not immediately useful for mechanical
performance models, where yarns require definition as solid geometry. Con-
version of this output to a graphical format, such as that used by the textile
pre-processor, TexGen, is therefore necessary before an FE model can be
built. There are three distinct stages required to generate a geometric model
Fig. 4.4. Reconstructed deformed geometry (a) Beam element model, (b) Yarn centre-
lines, (c) TexGen geometry, (d) Final trimmed unit cell.
of a textile unit cell in TexGen; define yarn paths, define yarn cross-sections,
and assign yarn repeats and domain.
The path of a yarn in TexGen is represented by a 1D line defined in 3D
space. This line is described by master nodes (positioned along the length
of the yarn) in conjunction with a spline to offer a smooth interpolation of
the path between these master nodes. Periodicity of the spline can also be
enforced to ensure continuity over a periodically repeating unit cell. The
deformation models were meshed in a regular manner, with the nodes of
each chain of beam elements within a yarn being grouped to form sections
along the yarn length. In order to define the yarn path from the deformed
model, a script was used to march along each yarn, from section to section,
calculating a yarn path master node as the centroid of the section by aver-
aging the x, y and z co-ordinates of each node in the section. Extracted
yarn paths are shown in Figure 4.4b.
Yarn cross-sections are defined as 2D shapes in a plane perpendicular
to the yarn tangent. A plane was defined at each section using the yarn
path master node. A set of nodes on this cross-section plane was defined
from the nearest finite element nodes using a mapping algorithm. The next
step was to assign the yarn cross-sections through the use of polygons; a
Fig. 4.5. (a) Convex hull “gift wrapping” analogy, (b) and (c) Convex hull modified
algorithm solutions for different yarn shapes.
closed chain consisting of straight lines defined by points. A unique cross-

section was defined at each master node to account for the variation in
cross-sectional shape along the yarn path.
The polygon sections require definition in a way which produced a
smooth outline around an arbitrary arrangement of points. Convex hull
algorithms are widely used in computational geometry to define the smallest
convex polygon around a set of points e.g., [21]. A polygon is considered
convex when all internal angles are less than or equal to 180◦. This can be
visualised using an analogy in which the points are represented by pins in
a board. By tying a piece of string to a starting pin which is at an extreme
position in the set (e.g., furthest pin to the right), the convex hull is the
shape formed by wrapping the string around the set of pins until it returns
to the starting pin (Figure 4.5a). However, the shape of a yarn cross-section
is not always convex so it is necessary to account for some large local concave
regions in a set of points, while smoothing over the small concave regions.
Goffman [22] proposed a modification of the convex hull to take into
account local properties of a set of points. Here, a similar algorithm has
been implemented. In this approach, a string of finite length is employed.
In practice, this means searching for nodes which lie within a radius of
the current point. Each subsequent point is selected as the node which lies
within this radius and creates a line with the smallest external angle from
the previous polygon section. A closed polygon is formed once the initial
node is reselected. Since the nodes within the yarn cross-section represent
the centre of a cylindrical beam element, each point in the polygon was
subsequently translated by the beam radius in a direction perpendicular to
the tangent of the polygon, thus expanding the section. Figure 4.5b shows
Fig. 4.6. Voxel mesh of deformed unit cell geometry, matrix elements removed for visu-
alisation only.
two extreme shapes found in the fabric with the thick cross-section being
a surface weft yarn at low compaction and the thin cross-section being a
centre weft yarn at high compaction. The convex hull produces good results
for the thick yarn but poor results for the thin yarn which is in an S-shape
due to the out-of-plane waviness of warp yarns. The modified algorithm
(Figure 4.5c) produces good results in both situations, still managing to
smooth out the irregular surface of the poorly compacted thick yarn.
Since the output of the deformation models is a fully periodic unit cell,
specification of yarn repeat vectors corresponding to the tessellation of the
unit cell can be utilised to define an infinite fabric. Two vectors were used
to prescribe the repeats for the fabric; one in the warp direction, as well as
one in the weft direction with an offset due to the stagger in tessellation.
This infinite fabric can then be trimmed to a finite domain to create the
unit cell shown in Figure 4.4d. This new geometry of the deformed unit
cell can then be used directly in the generation of finite element meshes
for modelling of mechanical performance, as has been done using the voxel
meshing technique in [23]. Figure 4.6 shows an example of the mesh that
was generated in this process.
4.4 Reduced Fidelity Modelling

The multi-filament method, which was used to find the as-woven and com-
pacted geometry in the previous sections, represents each yarn in the model
with bundles of beam elements. The presence of numerous contacts between
large numbers of beams is a main reason behind the high computational

expense associated with this method. In order to be able to model deforma-
tion of woven fabrics at a structural feature scale, that includes explicit def-
inition of the individual yarn paths, a simplified representation is required.
For this, an approach is adopted using a single contact surface for each
yarn, built from shell elements for model size reduction and more efficient
contact algorithms [24]. In order to capture the fabric behaviour with a
reduced geometry, special care needs to be given to the yarn mechanical
modelling. In the reduced fidelity approach shell elements trace the yarn
surface, giving an accurate representation of the yarn geometry. The initial
yarn geometry used is from the high-fidelity multi-filament beam model at
the as-woven stage. From the detailed geometry, a reduction of the res-
olution in terms of the number of nodes along the yarn centre line and
in each cross-section definition is required. As a final step, the yarn cross-
section points are offset in the section plane to account for the shell element
thickness using a generic polyline curves offset algorithm.
Using the points defining the yarn cross-section, shell elements are
meshed to represent the yarn surface. These act as contact surfaces dur-
ing the simulation, representing the yarn interactions during compaction.
A yarn model constructed from hollow shell elements will only, however,
exhibit unrealistic cross-section deformation. Increasing the shell element
stiffness by using stiffer material properties or thicker elements will increase
the flexural stiffness as well as the in-plane stiffness. This will lead to
an overall stiff fabric behaviour which is unrealistic. As a result, it is
desirable to separate the set of material properties controlling the yarn
cross-section deformation and the yarn flexural deformation. This can be
achieved by introducing a cross-section support at each centre-line node
(see Figure 4.7c). The cross-section support fills the yarn cross-section at a
given point by connecting all the yarn surface nodes using 2D shell elements.
Each node on the yarn surface is thus connected to a yarn cross-section,
ensuring that the entire yarn deformation is constrained by the internal
cross-section supports.
The cross-section deformation is dominated by fibres sliding against
each other. It is assumed that the yarn cross-section deformation can be
modelled as shearing deformation. This can be achieved by assigning vis-
coelastic material properties to the yarn cross-section support. By choosing
a relatively high bulk modulus, the cross-section support and hence the
entire yarn cross-section will exhibit shear dominated deformations. The
viscoelastic behaviour of dry fibres has already been observed and studied
Fig. 4.7. (a) Yarn geometric representation, (b) Cross-section meshed with shell ele-
ments, (c) Close up of cross-section supports.
in the literature [25]. The high-fidelity beam element models can capture
both inter- and intra-yarn volume fraction variations. In the reduced fidelity
approach, models only capture the inter-yarn volume fraction variation dur-
ing compaction. The intra-yarn volume fraction is carried over from the
digital element model when the single shell surface is created and remains
almost constant at each yarn cross-section because of the shear dominated
deformation assumption.
In order to validate this reduced fidelity modelling approach, a unit cell
compaction model was built based on the as-woven geometry of the orthogo-
nal 3D woven preform. As with the digital element models, the model input
properties were selected based on accuracy of results, not physical material
properties, since it is the kinematic interaction of the tows in the model
that determine the final deformed shape, more strongly than the material
constitutive law. Periodic boundary conditions were applied to the model,
which was compacted between two moving rigid plates. The fabric volume
fraction (VF) changed from 45% to 56% and from a thickness of 7 mm to
a thickness of 5.5 mm. Figure 4.8 shows a comparison between CT scans,
digital element results and the reduced fidelity model. The results from the
unit cell compaction were evaluated based on accuracy and run-time. In
terms of yarn path and crimp, the reduced fidelity representation shows
good agreement with both experiments and high-fidelity models. From a
run-time point of view, the reduced model compared with the digital ele-
ment model gives a significant reduction in CPU time, up to 90%. Since the
number of unit cells is squared as fabric dimensions increases, the run-time
reduction will be further amplified for larger fabric models.
Fig. 4.8. Comparison of high-fidelity and reduced fidelity modelling results for com-
paction of 3D woven fabric.
4.5 Application to Feature Scale Models

Creating a feature scale fabric model is achieved by tessellating the as-
woven unit cell geometry to form a full-scale fabric. This process involves
copying the unit cell geometry multiple times until the required size is
reached. Next, to maintain mechanical and contact consistency, yarns are
merged to similar yarns from the adjacent unit cells to form continuous
yarns throughout the fabric. Finally, an equivalence algorithm merges any
overlapping geometry points as a result of yarn merging.
The “humpback bridge” specimen is one that has been designed
for determining composites’ through-thickness tensile strength [26]. Test
samples have been previously prepared from the same orthogonal fabric
described earlier in this chapter. Dry preforms were laid on a rigid tool
and enclosed in a vacuum bag. The samples were then infused with liq-
uid epoxy resin and cured. The cured composite was cut into 12 mm wide
strips. Selected samples were CT scanned to show the internal fibre archi-
tecture. Figure 4.9 shows the sample dimensions and a final infused sample
(Figure 4.9b). The compaction process, without any consideration of the
resin infusion, was modelled using a vacuum bag approach (Figure 4.9c
and 4.9d). This required first deforming the fabric over the rigid tool with
an “intermediate tool” (Figure 4.9c) to mimic the hand layup process in
which the fabric is physically placed and manipulated before bagging. This
intermediate tool is stopped when the distance between it and the other
rigid tool is equal to the as-woven fabric thickness. A deformable shell
element vacuum bag is then used with an applied pressure to simulate the
compaction phase (Figure 4.9d).
(a) (b)
(c) (d)
Fig. 4.9. Humpback bridge specimen (a) Dimensions, (b) Infused specimen, (c) Model
set-up, (d) Final deformed model result.
Fig. 4.10. Humpback bridge specimen cross-sections (a) X-ray CT scan, (b) Model
prediction.
The model run-time was 18 hours on a Linux cluster, running on 8

cores. A comparison between the CT scan results and the simulation is
shown in Figure 4.10. The model was found to be in good agreement with
the experimental results showing a good match for yarn paths. Additionally,
gaps between yarns in the model correspond well to the resin pockets in
the CT scan. Marker “a” on Figure 4.10 points to the gap formed as a
result of the binder yarn interaction with the tool apex while marker “b”
points to the gap forming around the tool corners. Marker “c” points to
weft yarn overlap as a result of the interaction with each other and the
vacuum bag, which is also in good agreement with the CT scan. One of the
common challenges with vacuum bag manufacturing is thickness variability

especially around corners. Results show that the relative thickness varia-
tion along the sample length, as predicted by the model, compares well to
the experimental results with the model and the experiment being thicker
around the corner and thinner at the apex.
As a final demonstration of the feature scale model capabilities a hemi-
spherical forming test has been modelled. This problem has been widely
used in literature for 2D fabrics, with several analytical and experimental
results available e.g., [27–29]. A model of a preform with an in-plane dimen-
sion of 360 mm in the warp direction and 300 mm in the weft direction was
assembled. The fabric was laid in contact with a female tool with a fabric
guide on top. A spherical indenter was used to form the fabric into the
female tool cavity. The indenter has a radius of 72 mm and the cavity has a
radius of 77 mm. The fabric guide and tools were modelled as rigid shells.
The model run-time was 37 hours on 64 2.6 GHz cores of an HPC Linux
cluster. The analysis offered a detailed view of the compaction where the
through-thickness deformation at each location is described in detail, while
the global fabric behaviour is still captured. Figure 4.11 shows the deformed
fabric model. The fabric edge deformation shown is consistent with dome
compaction results for a 2D fabric e.g., [27]. Figure 4.11b and c show cross-
sections at selected locations with the yarn paths in the through-thickness
direction clearly visible. The results show that the tool-fabric interaction
(b)
(a)
(c)
Fig. 4.11. Dome-forming simulation results; (a) Overall deformation, (b) Warp direction
cross-section, (c) Weft direction cross-section.
has affected the yarn paths, waviness and crimp at each location differ-
ently. While the initial unit cell geometry is constant throughout the fab-
ric, the deformed fabric shows that significant variation from the unit cell
geometry occurs between locations. When loaded, damage will initiate at
weak points that develop within the structure due to these localised defor-
mations. In future, failure modelling for complex 3D woven components
should take the compacted fabric architecture, as can now be predicted,
into consideration.
4.6 Conclusion
For the prediction of mechanical performance of woven and textile com-
posites it is necessary to account for the internal fabric architecture, since
failure initiates from localised features such as yarn crimp and resin-rich
regions. It is thus important to include in mechanical performance models
the deformation of the internal yarn architecture, which occurs during the
manufacture of textile composites from processes such as weaving loom ten-
sion, mould closure and tool-fabric interaction. These deformations will also
have an influence on the fabric permeability during infusion of the matrix
resin. A method and workflow have been shown in this chapter as to how the
geometry arising from such deformation mechanisms can be predicted using
FE analysis. As an example, a complex 3D woven orthogonal preform has
been used to show how both in-plane and through-thickness deformations
can be captured. This then paves the way for simulation of resin flow and
mechanical performance using more realistic geometry in further analyses.
Acknowledgements
The authors acknowledge the support of Rolls-Royce plc., through the Com-
posites University Technology Centre at the University of Bristol and the
Engineering and Physical Sciences Research Council (EPSRC) through the
ACCIS DTC.
References
[1] Guénon, V.A., Chou, T.W. and Gillespie, J.W. (1989). Toughness proper-
ties of a three-dimensional carbon-epoxy composite, Journal of Materials
Science, 24, 4168–4175.
[2] Brandt, J., Drechsler, K. and Arendts, F.J. (1996). Mechanical perfor-
mance of composites based on various three-dimensional woven-fibre pre-
forms, Composites Science and Technology, 56, 381–386.
[3] Chou, S., Chen, H.C. and Chen, H.E. (1992). Effect of weave structure
on mechanical fracture behavior of three-dimensional carbon fiber fabric
reinforced epoxy resin composites, Composites Science and Technology, 45,
23–35.
[4] Baucom, J.N. and Zikry, M.A. (2003). Evolution of failure mechanisms in
2D and 3D woven composite systems under quasi-static perforation, Journal
of Composite Materials, 37, 1651–1674.
[5] Wang, X.F., Wang, X.W., Zhou, G.M. and Zhou, C.W. (2007). Multi-scale
analyses of 3D woven composite based on periodicity boundary conditions,
[6] Tang, X. and Whitcomb, J.D. (2003). General techniques for exploiting peri-
odicity and symmetries in micromechanics analysis of textile composites,
[7] Li, S., Warrior, N., Zou, Z. and Almaskari, F. (2011). A unit cell for FE anal-
ysis of materials with the microstructure of a staggered pattern, Composites
Part A: Applied Science and Manufacturing, 42, 801–811.
[8] Lin, H., Zeng, X., Sherburn, M., Long, A.C. and Clifford, M.J. (2012). Auto-
mated geometric modelling of textile structures, Textile Research Journal,
82, 1689–1702.
[9] Vandeurzen, P., Ivens, J. and Verpoest, I. (1996). A three-dimensional
micromechanical analysis of woven-fabric composites: I. Geometric analysis,
[10] Charmetant, A., Orliac, J.G., Vidal-Sallé, E. and Boisse, P. (2012). Hyper-
elastic model for large deformation analyses of 3D interlock composite pre-
forms, Composites Science and Technology, 72, 1352–1360.
[11] Mahadik, Y., Brown, K.A.R. and Hallett, S.R. (2010). Characterisation of 3D
woven composite internal architecture and effect of compaction. Composites
[12] Zhou, E., Mollenhauer, D. and Iarve, E. (2007). Image Reconstruction Based
Modeling of 3D Textile Composites, 48th AIAA/ASME/ASCE/AHS/ASC
Structures, Structural Dynamics, and Materials Conference, Honolulu.
[13] Djukic, L.P., Herszberg, I., Walsh, W.R., Schoeppner, G.A. and Prusty, B.G.
(2009). Contrast enhancement in visualisation of woven composite architec-
ture using a Micro CT Scanner, Part 2: Tow and preform coatings, Compos-
ites Part A: Applied Science and Manufacturing, 40, 1870–1879.
[14] Potluri, P. and Sagar, T.V. (2008). Compaction modelling of textile preforms
for composite structures, Composite Structures, 86, 177–185.
[15] Zhou, G.M., Sun, X.J. and Wang, Y.Q. (2004). Multi-chain digital ele-
ment analysis in textile mechanics, Composites Science and Technology, 64,
239–244.
[16] Durville, D. (2010). Simulation of the mechanical behaviour of woven fab-
rics at the scale of fibers, International Journal of Material Forming, 3,
1241–1251.
[17] Stig, F. and Hallström, S. (2012). Spatial modelling of 3D-woven textiles,
Composite Structures, 94, 1495–1502.
[18] Mahadik, Y. and Hallett, S.R. (2010). Finite element modelling of tow geom-
etry in 3D woven fabrics, Composites Part A: Applied Science and Manufac-
turing, 41, 1192–1200.
[19] Kawabata, S. (1980). The Standardization and Analysis of Hand Evaluation
2nd ed. Osaka: The Textile Machinery Society.
[20] Green, S.D., Long, A.C., El Said, B.S.F. and Hallett, S.R. (2014). Numerical
modelling of 3D woven preform deformations, Composite Structures, 108,
747–756.
[21] Sklansky, J. (1982). Finding the convex hull of a simple polygon, Pattern
Recognition Letters, 1, 79–83.
[22] Gofman, Y. (1993). Outline of a set of points, Pattern Recognition Letters,
14, 31–38.
[23] Green, S.D., Matveev, M., Long, A.C. and Hallett, S.R. (2013). Modelling
of 3D woven composites with realistic unit cell geometry, 19th International
Conference on Composite Materials, Montreal.
[24] El Said, B.S.F., Green, S.D. and Hallett, S.R. (2014). Kinematic modelling of
3D woven fabric deformation for structural scale features, Composites Part
A: Applied Science and Manufacturing, 57, 95–107.
[25] Bickerton, S., Buntain, M.J. and Somashekar, A.A. (2003). The viscoelastic
compression behavior of liquid composite molding preforms,Composites Part
A: Applied Science and Manufacturing, 34, 431–444.
[26] Wisnom, M.R. and Jones, M.I. (1996). Size effects in interlaminar tensile
and shear strength of unidirectional glass fibre-epoxy, Journal of Reinforced
Plastics and Composites, 15, 2–15.
[27] Badel, P., Vidal-Sall, E. and Boisse, P. (2008). Large deformation analysis of
fibrous materials using rate constitutive equations, Computers & Structures,
86, 1164–1175.
[28] Cherouat, A. and Billoët, J.L. (2001). Mechanical and numerical modelling
of composite manufacturing processes deep-drawing and laying-up of thin
pre-impregnated woven fabrics, Journal of Materials Processing Technology,
118, 460–471.
[29] Dong, L., Lekakou, C. and Bader, M.G. (2001). Processing of composites:
simulations of the draping of fabrics with updated material behaviour law,
Chapter 5
CRYOMECHANICS AND CRACK BEHAVIOUR OF

WOVEN POLYMER MATRIX COMPOSITES
Yasuhide Shindo, Tomo Takeda and Fumio Narita

Department of Materials Processing, Graduate School of Engineering,
Tohoku University, Aoba-yama 6-6-02, Sendai 980-8579, Japan
5.1 Introduction
Woven composite materials have been used in many kinds of engineering
applications because of their unique characteristics [1]. In particular, super-
conducting magnets of cryogenic power systems, such as the International
Thermonuclear Experimental Reactor (ITER), may use large quantities
of woven glass fibre reinforced polymer (GFRP) composite laminates as
electrical and thermal insulation, and structural support [2]. Also, carbon
fibre reinforced polymer (CFRP) composites have been considered for liquid
hydrogen and oxygen storage tanks in future space vehicles, such as reusable
launch vehicles (RLVs), and it was demonstrated that woven CFRP mate-
rials are possible candidates for the cryogenic storage systems [3]. For the
effective use of woven composites in cryogenic applications it is important
to understand their performance in a variety of conditions.
Extensive studies have been performed on the physical and mechan-
ical properties of fibre reinforced composites at cryogenic temperatures,
and several review articles were published which summarise the research
work on this subject [4, 5]. Also, our research group has investigated the
cryogenic behaviour of woven composite laminates. For woven GFRP com-
posite laminates, the interlaminar shear [6, 7], tensile [8, 9] and compressive
[10] responses at cryogenic temperatures were investigated. The cryogenic
translaminar crack behaviour in woven GFRP laminates was also char-
acterised [11]. In addition, the cryogenic interlaminar fracture behaviour
159
160 Y. Shindo, T. Takeda and F. Narita
of woven GFRP laminates was examined under Mode I [12, 13], Mode II
[14–16], Mode III [17], mixed-mode I/II [18, 19], mixed-mode I/III [20]
and mixed-mode II/III [21] loadings. Concerning the composite response
to cyclic loading, the tensile fatigue behaviour of woven GFRP laminates
at cryogenic temperatures was studied [22]. The cryogenic translaminar
fatigue crack growth in woven GFRP laminates was also examined [23].
Furthermore, the cryogenic Mode I [24], Mode II [25], Mode III [26] and
mixed-mode I/II [27] fatigue delamination growth was discussed.
Polymer matrix composite materials are often processed (cured) at ele-
vated temperatures and subsequently cooled to room temperature. When
composites are subjected to cooling from the stress-free temperature, resid-
ual thermal stresses are induced in the materials due to the mismatch in
coefficients of thermal expansion (CTEs) of the reinforcement and matrix.
Also, the residual thermal stresses induced in these composites increase
with decreasing temperature to cryogenic temperatures. The non-zero state
of residual thermal stresses at cryogenic temperatures are the underlying
cause of microcracking in composites, and the microcracks could lead to
degradation of their performance [28]. To secure the integrity of cryogenic
composite structures, understanding of the crack behaviour in woven com-
posites at cryogenic temperatures is of great importance.
In this chapter, we present our theoretical studies on the cryogenic
crack behaviour in woven composite laminates. The basic woven fabric
structures are plain, twill, satin and basket [29]. Here we focus on woven
GFRP/CFRP laminates under tensile loading. First, two-dimensional finite
element analyses of woven GFRP laminates, reinforced by plain weave fab-
rics, with transverse fibre bundle cracks are conducted and the numerical
results for the elastic properties, the stress distributions near the crack tip
and the stress intensity factors (SIFs) are discussed. The cross sections of
the woven composite laminates are modelled as a generalised plane strain
problem, and the model considers undulation of fibre bundles only along
the loading direction. Also, temperature-dependent constituent properties
are used in the analysis. Next, the three-dimensional finite element model
of cracked woven GFRP laminates is developed. This model incorporates
the fibre bundle undulation in both longitudinal (loading) and transverse
directions in order to determine more accurately the elastic properties and
stress distributions near the crack front of the woven composite laminates.
Finally, the modelling of cracked woven CFRP laminates reinforced with
satin weave fabrics is described.
Cryomechanics and Crack Behaviour of Woven Polymer Matrix Composites 161
5.2 Two-dimensional Crack Problem in GFRP Laminates

In this section, we present our investigations on the two-dimensional crack
problem in woven GFRP laminates. Here, National Electrical Manufactur-
ers Association (NEMA) G-11 woven GFRP laminates reinforced with plain
weave fabrics are considered. We first examine the elastic properties and
the stress distributions near the transverse fibre bundle cracks for the two-
layer [30] and multi-layer [31] woven laminates, and then discuss the SIFs
for the two-layer [32] and multi-layer [33] woven laminates.
5.2.1 Geometry of cracked plain weave composite laminates

Figure 5.1 shows the 2N -layer plain weave fabric laminates (N =
1, 2, 3, . . .). Let the coordinate axes x and y be chosen such that they coin-
cide with the middle plane of the woven laminate and the z-axis is perpen-
dicular to this plane. The woven laminates occupy the region (|x| < ∞,
|z| ≤ 2N h), and comprise three constituents: the impregnated warp and fill
Fig. 5.1. 2N -layer plain weave fabric laminates with cracks.

fibre bundles, and the pure matrix regions. To keep the fabric architecture
simple, the undulating form of the warp fibre bundles is assumed to be
linear. In addition, we assume that the woven laminates are in the gener-
alised plane strain state in the y-direction. The basic parameters describing
the weave geometry are the thickness of the fibre bundles (h), the warp
fibre bundle wavelength (L) and the length of flattened warp region (d),
which are then used to define the constant weave curvature for the warp
fibre bundles (θ). Each lamina has the thickness 2h. The length of the
internal and edge cracks are 2a and a (a ≤ h), respectively, and the follow-
ing cases are considered: 2N -layer woven laminates with collinear cracks
in x = ±2nL (n = 0, 1, 2, . . .) planes of the fill fibre bundles (Case A-1),
woven laminates with collinear cracks in x = ±(2n + 1)L planes of the fill
fibre bundles (Case A-2), and woven laminates with cracks in all fill fibre
bundles (Case B). The surface layers of the woven laminates with an odd
number of N (N = 1, 3, 5, . . .) for Case A-1 and an even number of N
(N = 2, 4, 6, . . .) for Case A-2 contain the internal cracks, and those of
the woven laminates with an even number of N for Case A-1 and an odd
number of N for Case A-2 contain the edge cracks.
The 2N -layer woven laminates are subjected to a thermal load Φ − Φs
∗
and a mechanical mean stress σxx in the warp direction, i.e., x-direction,
such that the woven laminates undergo the uniform displacement in the
x-direction along the planes x = ±(2n + 1)L (n = 0, 1, 2, . . .) except along
the cracks for Cases A-2 and B. The thermal load Φ − Φs represents the
difference between the stress-free temperature Φs , which is usually taken
to be the cure temperature, and the current temperature Φ.
5.2.2 Temperature-dependent material properties

The material properties are temperature dependent. The matrix (bisphenol-
A epoxy) and fibre (E-glass fibre) of the NEMA G-11 woven laminates are
isotropic materials with two independent elastic constants, i.e., Young’s
moduli EM (Φ), EF (Φ) and Poisson’s ratios νM (Φ), νF (Φ), and the shear
moduli GM (Φ) = EM (Φ)/[2{1 + νM (Φ)}], GF (Φ) = EF (Φ)/[2{1 + νF (Φ)}].
The subscripts M and F refer, respectively, to the matrix and the fibre.
The elastic properties EM (Φ), EF (Φ), νM (Φ), νF (Φ), GM (Φ), GF (Φ) are
approximated by the exponential functions [31] using the experimental
data [34]. The warp and fill fibre bundles are taken as homogeneous trans-
versely isotropic unidirectional composites. Once the constituent material
(matrix and fibre) properties and fibre volume fraction in the fibre bundles
VF are obtained, composite cylinder assemblage (CCA) theory [35] can be

applied to calculate the equivalent elastic properties of the unidirectional
composites ELB (Φ), ETB (Φ), GB B B B B
T L (Φ), GLT (Φ), GT T (Φ), νLT (Φ), νT L (Φ),
B
νT T (Φ). The superscript B denotes the fibre bundle, and the subscripts L
and T represent the fibre and transverse directions, respectively. The CTEs
of the matrix and fibre bundles αM (Φ), αB B
L (Φ), αT (Φ) are also determined
using the CCA model, and approximated by the exponential functions.
5.2.3 Finite element analysis

The constitutive equations for the warp fibre bundles, fill fibre bundles and
matrix in the generalised plane strain state can be written as:
    
σxxδ (x, z) C11δ (Φ) C12δ (Φ) C16δ (Φ) εxxδ (x, z) − εxx0δ (Φ)
 σzzδ (x, z)  = C12δ (Φ) C22δ (Φ) C26δ (Φ)  εzzδ (x, z) − εzz0δ (Φ) ,
σxzδ (x, z) C16δ (Φ) C26δ (Φ) C66δ (Φ) 2εxzδ (x, z) − 2εxz0δ (Φ)
(δ = w, f, M ), (5.1)
where σxxδ (x, z), σzzδ (x, z), σxzδ (x, z) are the stress components, εxxδ (x, z),
εzzδ (x, z), εxzδ (x, z) are the strain components, C11δ (Φ), C12δ (Φ), . . . ,
C66δ (Φ) are the components of the stiffness matrix and εxx0δ (Φ), εzz0δ (Φ),
εxz0δ (Φ) (δ = w, f, M ) are the initial strain components. The subscripts w
and f denote the warp and fill fibre bundles, respectively. The components
of the stiffness matrix C11δ (Φ), C12δ (Φ), . . . , C66δ (Φ) (δ = w, f, M ) are
given by

C11w (Φ) = C11w (Φ) cos4 θ + 2{C12w (Φ) + 2C66w (Φ)} cos2 θ sin2 θ 

4 

+ C22w (Φ) sin θ, 




C12w (Φ) = {C11w
(Φ) + C22w
(Φ) − 4C66w (Φ)} cos2 θ sin2 θ 



4 4
+ C12w (Φ)(cos θ + sin θ), 





C16w (Φ) = {C11w (Φ) − C12w (Φ) − 2C66w (Φ)} cos3 θ sin θ 



+ {C12w (Φ) − C22w (Φ) + 2C66w (Φ)} cos θ sin3 θ,

C22w (Φ) = C11w
(Φ) sin4 θ + 2{C12w
(Φ) + 2C66w 
(Φ)} cos2 θ sin2 θ




+ C22w (Φ) cos θ,4 





C26w (Φ) = {C11w (Φ) − C12w (Φ) − 2C66w (Φ)} cos θ sin θ
3




+ {C12w (Φ) − C22w (Φ) + 2C66w (Φ)} cos θ sin θ,
3 

2 

C66w (Φ) = {C11w (Φ) + C22w (Φ) − 2C12w (Φ) − 2C66w (Φ)} cos θ sin θ
2




+ C66w (Φ)(cos4 θ + sin4 θ),
(5.2)

C11f (Φ) = {1/ELf (Φ)}{1/ET f (Φ) − νLT f (Φ)2 /ELf (Φ)}/Hf (Φ), 


C12f (Φ) = {1/ELf (Φ)}{νLT f (Φ)νT Lf (Φ)/ET f (Φ) 





+νT T f (Φ)/ET f (Φ)}/Hf (Φ), 

C16f (Φ) = 0, (5.3)


C22f (Φ) = {1/ET f (Φ)}{1/ELf (Φ) − νT Lf (Φ) /ET f (Φ)}/Hf (Φ),
2 




C26f (Φ) = 0, 


C66f (Φ) = GT T f (Φ),

C11M (Φ) = EM (Φ){1 − νM (Φ)}/[{1 + νM (Φ)}{1 − 2νM (Φ)}],



C12M (Φ) = EM (Φ)νM (Φ)/[{1 + νM (Φ)}{1 − 2νM (Φ)}], 



C (Φ) = 0, 
16M
(5.4)
C22M (Φ) = C11M (Φ), 





C26M (Φ) = 0, 


C66M (Φ) = GM (Φ),
where


C11w (Φ) = {1/ET w (Φ)}{1/ET w (Φ) − νT T w (Φ)2 /ET w (Φ)}/Hw (Φ),




C12w (Φ) = {1/ET w (Φ)}{νLT w (Φ)νT T w (Φ)/ELw (Φ) 


+νT Lw (Φ)/ET w (Φ)}/Hw (Φ),



C22w (Φ) = {1/ELw (Φ)}{1/ET w (Φ) − νLT w (Φ)2 /ELw (Φ)}/Hw (Φ),  




C66w (Φ) = GT Lw (Φ),
(5.5)

Hw (Φ) = {1 − 2νLT w (Φ)νT Lw (Φ)νT T w (Φ) − νT T w (Φ)2 



−ET w (Φ)νLT w (Φ)2 /ELw (Φ) 



−ELw (Φ)νT Lw (Φ) /ET w (Φ)}/{ELw (Φ)ET w (Φ) },
2 2
Hf (Φ) = {1 − 2νLT f (Φ)νT Lf (Φ)νT T f (Φ) − ELf (Φ)νT Lf (Φ)2 /ET f (Φ)


−ET f (Φ)νLT f (Φ)2 /ELf (Φ) 



−νT T f (Φ) }/{ELf (Φ)ET f (Φ) },
2 2
(5.6)

ELw (Φ) = ELf (Φ) = ELB (Φ), ET w (Φ) = ET f (Φ) = ETB (Φ), 


GT T f (Φ) = GB B 

T T (Φ), GT Lw (Φ) = GT L (Φ), 
B
νLT w (Φ) = νLT f (Φ) = νLT (Φ), (5.7)


B
νT Lw (Φ) = νT Lf (Φ) = νT L (Φ), 



B 
νT T w (Φ) = νT T f (Φ) = νT T (Φ).
The initial strain components εxx0δ (Φ), εzz0δ (Φ), εxz0δ (Φ) (δ = w,
f, M ) are
 
εxx0δ (Φ)
 εzz0δ (Φ) 
2εxz0δ (Φ)
  

 ET w (Φ)

 ε T
−1  LLw (Φ) − ν LT w (Φ){ε yy0 (Φ) − ε T
(Φ)} 

  ELw (Φ) TTw
,

 T  

 εT T w (Φ) − νT T w (Φ){εyy0 (Φ) − εT T w (Φ)}
T T



 0



 (δ = w),





  

 ELf (Φ)

 ε T
(Φ) − ν (Φ){ε (Φ) − ε T
(Φ)}
  TTf ET f (Φ)
T Lf yy0 LLf 
=  ,
  εT T f (Φ) − νLT f (Φ){εyy0 (Φ) − εLLf (Φ)}
T T 



 0



 (δ = f ),





  T 



 εM (Φ) − νM (Φ){εyy0 (Φ) − εTM (Φ)}

 εTM (Φ) − νM (Φ){εyy0 (Φ) − εTM (Φ)} ,





 0


(δ = M ),
(5.8)
where the superscript −1 denotes the matrix inverse and [T ] is a transfor-

mation matrix given by
 
cos2 θ sin2 θ cos θ sin θ
[T ] =  sin2 θ cos2 θ − cos θ sin θ . (5.9)
−2 cos θ sin θ 2 cos θ sin θ cos2 θ − sin2 θ
The thermal strains εTLLw (Φ), εTT T w (Φ), εTLLf (Φ), εTT T f (Φ), εTM (Φ) are given
as follows:
Φ 

T T
εLLw (Φ) = εLLf (Φ) = αL (φ)dφ, 
B 


Φs 

Φ 


B
εTT T w (Φ) = εTT T f (Φ) = αT (φ)dφ, . (5.10)


Φs 

Φ 



εTM (Φ) = αM (φ)dφ. 

Φs
The uniform strain in the y-direction εyy0 (Φ) can be found from the con-
dition:

σyyw (x, z)dxdz + σyyf (x, z)dxdz + σyyM (x, z)dxdz
Ωw Ωf ΩM
= 0,
Ω w + Ωf + ΩM
(5.11)
where Ωδ (δ = w, f , M ) denote the regions of the warp fibre bundles, fill
fibre bundles and matrix, respectively. The stresses σyyδ (x, z) (δ = w, f, M )
are given by
σyyδ (x, z)

 ET w (Φ)

 ET w (Φ){εyy0 (Φ) − εTT T w (Φ)} + νLT w (Φ)σLLw (x, z)

 ELw (Φ)





 +νT T w (Φ)σT T w (x, z), (δ = w),



 ELf (Φ)
= ELf (Φ){εyy0 (Φ) − εTLLf (Φ)} + νT Lf (Φ)σxxf (x, z)
 ET f (Φ)



 +νLT f (Φ)σzzf (x, z), (δ = f ),





 EM (Φ){εyy0 (Φ) − εTM (Φ)} + νM (Φ){σxxM (x, z) + σzzM (x, z)},




(δ = M ),
(5.12)
where the stresses σLLw (x, z) and σT T w (x, z) can be obtained from trans-
formation equations written as
 
2 σxxw (x, z)
σLLw (x, z) cos θ sin2 θ 2 cos θ sin θ 
= σzzw (x, z)  . (5.13)
σT T w (x, z) sin2 θ cos2 θ −2 cos θ sin θ
σxzw (x, z)
The SIF is defined from the singular stress field in front of a crack, and
the SIFs KI at the tips of the fill fibre bundle cracks in the kth lamina
(1 ≤ k ≤ N ) can be written as
Case A-1:

 lim [2π{z − 2(k − 1)h − a}]1/2 σxxf (0, z), (a < h),
z→{2(k−1)h+a}+
KI =
 lim (2π)1/2 {z − (2k − 1)h}ω σxxw (0, z), (a = h),
z→{(2k−1)h}+
(k = 1, 3, 5, . . .), (5.14)

 lim {2π(2kh − a − z)}1/2 σxxf (0, z), (a < h),
z→(2kh−a)−
=
 lim (2π) 1/2
{(2k − 1)h − z} σxxw (0, z), (a = h),
ω
z→{(2k−1)h}−
(k = 2, 4, 6, . . .), (5.15)
Case A-2:

 lim {2π(2kh − a − z)}1/2 σxxf (L, z), (a < h),
z→(2kh−a)−
KI =
 lim (2π) {(2k − 1)h − z} σxxw (L, z), (a = h),
1/2 ω
z→{(2k−1)h}−
(k = 1, 3, 5, . . .), (5.16)

 lim [2π{z − 2(k − 1)h − a}] 1/2
σxxf (L, z), (a < h),
z→{2(k−1)h+a}+
=
 lim (2π) 1/2
{z − (2k − 1)h} σxxw (L, z),
ω
(a = h),
z→{(2k−1)h}+
(k = 2, 4, 6, . . .), (5.17)
where ω is the order of stress singularity at the interface tip of the crack. The
order of stress singularity at the tip of a crack in a homogeneous anisotropic
medium is 0.5. The stress singularity at the tip of a crack which is normal
to and ends at the warp/fill interface is calculated using the analytical
method [32].
5.2.4 Numerical results and discussion

The geometry of the plain weave fabric laminates is defined by L = 4h/
3 tan θ and d = L/8. Based on the dimensions of NEMA grade G-11 woven
GFRP laminates measured with a scanning electron microscope (SEM),
the parameters 2h and L are taken to have the values 2h = 0.19 mm and
L = 0.81 mm. The warp angle is then estimated to be θ = 8.9 deg. The
value of a fibre volume fraction in the fibre bundles (VF ) is taken as 0.75,
corresponding to that of actual G-11 woven laminates. The stress-free tem-
perature Φs is assumed to be 395 K.
The predicted Young’s modulus increases as temperature decreases.
A similar trend was observed for Young’s moduli from experiments [36].
Also, the fill fibre bundle cracks reduce the Young’s modulus.
∗
Figure 5.2 shows the normalised stresses σxxw /σxx in the warp fibre
bundle at x = 0 (h ≤ z ≤ 2h) as a function of the distance from the crack
∗
tip (z − h)/h for N = 1, Case A-1, and a/h = 1.0 under σxx = 350 MPa
at Φ = 293 K (room temperature), 77 K (liquid nitrogen temperature), 4 K
(liquid helium temperature). The stress σxxw near the crack tip increases
Fig. 5.2. ∗ (N = 1, Case A-1, a/h = 1.0).

Distributions of normalised stress σxxw /σxx
∗ (N = 1, 2, 3, 4, ∞, Case A-1,
Fig. 5.3. Distributions of normalised stress σxxw /σxx
a/h = 1.0).
with the decrease in the temperature while the opposite trend is seen for
that away from the crack tip. Figure 5.3 displays the normalised stresses
∗
σxxw /σxx at x = 0 (h ≤ z ≤ 2h) versus the distance from the crack
tip (z − h)/h for N = 1, 2, 3, 4, ∞, Case A-1, and a/h = 1.0 under
∗
σxx = 350 MPa at Φ = 4 K. Except for N = 1, the stress σxxw ahead of the
crack tip increases and tends toward the result for N = ∞ as N increases.
Fig. 5.4. Normalised stress intensity factor KI /K0 versus crack length to fibre bundle
thickness ratio a/h (N = 1, Case A-1).
The normalised SIF KI /K0 , where K0 = ETB (Φs )αB T (Φs )Φs (πa)
1/2
,
is shown in Figure 5.4 as a function of the crack length to fibre bundle
∗
thickness ratio a/h for N = 1 and Case A-1 under σxx = 350 MPa at
Φ = 293, 77, 4 K. The SIF increases with the decrease in the temperature.
Also, the SIFs at 77 K and 4 K are dependent on crack length, whereas the
SIF at 293 K is almost independent of crack length.
Table 5.1 presents the normalised SIFs KI /K0 for N = 1, Cases A-1,
∗
A-2, and a/h = 1.0 under σxx = 350 MPa at Φ = 293, 77, 4 K. The SIF for
a/h = 1.0 is normalised by K0 = ETB (Φs )αB T (Φs )Φs π
1/2 ω
a . The values of
the order of stress singularity at the crack tip ω at 293 K, 77 K and 4 K are
0.41, 0.45 and 0.46, respectively. The SIF for Case A-2 (edge crack) is large
compared to that for Case A-1 (internal crack). Furthermore, although not
shown here, the SIFs at the tips of the internal crack and the edge crack for
Case B are very similar to those for Case A-1 and Case A-2, respectively.
This indicates that the interaction between the internal and edge cracks for
Table 5.1. Normalised stress intensity fac-

tors KI /K0 for two-layer woven laminates
∗ = 350 MPa (N = 1, a/h = 1.0).
under σxx
293 K 77 K 4K
KI /K0 (Case A-1) 2.72 3.40 3.59

KI /K0 (Case A-2) 3.06 4.25 4.67
Table 5.2. Normalised stress intensity factors

KI /K0 for 2N -layer woven laminates under
∗ = 350 MPa at Φ = 4 K (a/h = 1.0).
σxx
N 1 2 3 4
KI /K0 (Case A-1) 3.59 3.44 3.71 3.80

KI /K0 (Case A-2) 4.67 3.52 3.62 3.77
Case B seems to be negligible. Table 5.2 lists the normalised SIFs KI /K0
at the tip of the crack in the first layer (k = 1) for N = 1, 2, 3, 4, Cases A-1,
∗
A-2, and a/h = 1.0 under σxx = 350 MPa at Φ = 4 K. Except for the SIF
for N = 1, the SIF increases as N increases.
5.3 Three-dimensional Crack Problem in GFRP Laminates

This section describes a study on the three-dimensional crack problem in
woven GFRP laminates [37]. Consider a two-layer plain weave GFRP lam-
inate, as shown in Figure 5.5, where O-xyz is the Cartesian coordinate
system. The two-layer woven laminate of thickness 2H occupies the region
(|x| < ∞, |y| < ∞, |z| ≤ H). In the figure, Lx and Ly stand for the length of
the repeating unit in the x- and y-directions, lw and lf denote the width
of the warp and fill fibre bundles, and hw and hf represent the thickness
of the warp and fill fibre bundles. It is assumed that the cracks of length
c lie in x = ±2nLx (n = 0, 1, 2, . . .) planes of the fill fibre bundles. The
undulations of the fibre bundles are assumed to be sinusoidal. Lower and
Fig. 5.5. Two-layer plain weave fabric laminate with cracks.

upper surface coordinates of the fibre bundles (0 ≤ x ≤ Lx , 0 ≤ y ≤ Ly ,

0 ≤ z ≤ H) are as follows:

 hf πx hw πy

 1 + cos + 1 − cos ,

 2 Lx 2 Ly




 (0 ≤ x ≤ Lx , 0 ≤ y ≤ lw /2),
L
ψw (x, y) = (5.18)

 hf πx hw πy

 1 − cos + 1 + cos ,

 2 Lx 2 Ly




(0 ≤ x ≤ Lx , Ly − lw /2 ≤ y ≤ Ly ),

 hf πx hw πy

 1 + cos + 1 + cos ,

 2 Lx 2 Ly




 (0 ≤ x ≤ Lx , 0 ≤ y ≤ lw /2),
U
ψw (x, y) = (5.19)

 hf πx hw πy

 1 − cos + 1 − cos ,

 2 L 2 L

 x y


(0 ≤ x ≤ Lx , Ly − lw /2 ≤ y ≤ Ly ),



hf
−
πx hw
−
πy

 1 cos + 1 cos ,

 2 Lx 2 Ly



 (0 ≤ x ≤ lf /2, 0 ≤ y ≤ Ly ),
L
ψf (x, y) = (5.20)


 hf 1 + cos πx
 +
hw
1 + cos
πy
,

 2 Lx 2 Ly




(Lx − lf /2 ≤ x ≤ Lx , 0 ≤ y ≤ Ly ),

 hf πx hw πy

 1 + cos + 1 − cos ,

 2 L 2 L


x y


 (0 ≤ x ≤ lf /2, 0 ≤ y ≤ Ly ),
U
ψf (x, y) = (5.21)

 hf πx hw πy

 1 − cos + 1 + cos ,

 2 Lx 2 Ly




(Lx − lf /2 ≤ x ≤ Lx , 0 ≤ y ≤ Ly ),
where superscripts L and U indicate the lower and upper boundaries of the
fibre bundles, respectively. The fibre orientation within the bundles varies
continuously as a function of x and y for the warp and fill fibre bundles,
respectively. The local angle θw (x) between the warp fibre bundle and the
x-axis is given by
L U
∂ψw (x, y) ∂ψw (x, y)
θw (x) = arctan = arctan
∂x ∂x


 πhf πx

 arctan − sin ,

 2Lx Lx


 (0 ≤ x ≤ Lx , 0 ≤ y ≤ lw /2),
= (5.22)

 πhf πx

 arctan sin ,

 2Lx Lx



(0 ≤ x ≤ Lx , Ly − lw /2 ≤ y ≤ Ly ),
and the local angle θf (y) between the fill fibre bundle and the y-axis is

∂ψfL (x, y) ∂ψfU (x, y)
θf (y) = arctan = arctan
∂y ∂y


 πhw πy

 arctan sin ,

 2Ly Ly


 (0 ≤ x ≤ lf /2, 0 ≤ y ≤ Ly ),
= (5.23)

 πhw πy

 arctan − sin ,

 2Ly Ly



(Lx − lf /2 ≤ x ≤ Lx , 0 ≤ y ≤ Ly ).
Suppose that the two-layer woven laminate is subjected to the thermal

∗
load of Φ − Φs and the mechanical mean stress σxx in the x-direction. The
geometrical parameters used in the three-dimensional finite element anlaysis
are the actual dimensions of NEMA grade G-11 woven GFRP laminates,
i.e., Lx = 0.81 mm, Ly = 0.63 mm, H = 0.19 mm, lw = 0.63 mm, lf =
0.65 mm, hw = 0.11 mm and hf = 0.08 mm. It is assumed that the fill fibre
bundle cracks are of the through-thickness type (c/hf = 1.0) for the cracked
woven laminate.
Table 5.3 shows the Young’s moduli Ex of the uncracked (c/hf = 0.0)
and cracked (c/hf = 1.0) woven laminates at the temperatures of Φ =
293, 77, 4 K from the three-dimensional finite element analysis (3-D FEA).
The experimental results [36] are also listed in the table. The predicted
values of Young’s modulus for c/hf = 0.0 are in good agreement with
the experimental values. Also, the presence of fill fibre bundle cracks leads
to decrease in Young’s modulus. Figure 5.6 shows the normalised stresses
Table 5.3. Young’s moduli Ex of two-layer woven laminates.
293 K 77 K 4K
Ex (c/hf = 0.0) 3-D FEA 26.75 GPa 33.10 GPa 36.04 GPa
Experimental 27.9 GPa 32.7 GPa 36.9 GPa
Ex (c/hf = 1.0) 3-D FEA 25.88 GPa 31.89 GPa 34.69 GPa
2.8
σ*xx = 350 MPa
x=0
y=0
2.4
Φ = 293 K
77 K
σxxw/σ*xx
4K
2.0
1.6
3-D FEA
2-D FEA
1.2
0 0.2 0.4 0.6 0.8 1.0
(z - hf)/hw
Fig. 5.6. ∗ .
Distributions of normalised stress σxxw /σxx
∗
σxxw /σxx in the warp fibre bundle at x = 0 and y = 0 (hf ≤ z ≤ hw + hf )
∗
as a function of the distance from the crack front (z − hf )/hw under σxx =
350 MPa at Φ = 293, 77, 4 K. For comparison purposes, the normalised
∗
stress σxxw /σxx from the two-dimensional finite element analysis (2-D FEA)
of the two-layer G-11 woven GFRP laminate (N = 1, h = (hw + hf )/2)
for Case A-1 and a/h = 1.0 at Φ = 4 K is also presented in the figure.
Similar to the two-dimensional finite element result (see Figure 5.2), the
stresses σxxw at 77 K and 4 K increase more sharply than that at 293 K when
approaching the crack front. The use of the three-dimensional finite element
method permits a realistic prediction of the internal stress distributions.
It should be noted that, although the two-dimensional model provides a
conservative prediction of the stress distributions near the crack tip, it
predicts the correct trend of the thermo-mechanical behaviour of the woven
laminates.
5.4 Crack Problem in CFRP Laminates

Candidate CFRP materials for RLV cryotanks must be characterised in
a cryogenic environment. With regard to the cryogenic response of woven
CFRP composite laminates, Kumagai et al. [38] performed cryogenic tensile
tests and evaluated the elastic and strength properties of the composites.
Recently, Takeda et al. [39] investigated the interlaminar shear behaviour
and electrical resistance-based damage self-sensing abilities of woven CFRP
laminates at cryogenic temperatures. In addition, microcracks in CFRP
composite laminates used for RLV cryotank applications provide a pathway
for the cryogenic fluids to leak through the composite laminates. There-
fore, an understanding of the crack behaviour in composite laminates at
cryogenic temperatures is a key issue for the reliable design of the storage
systems.
This section presents a brief description of our studies on the crack
problem in woven CFRP laminates. The T800H/3633 woven CFRP com-
posite laminates (Toray Industries, Inc., Japan) for RLV cryotank appli-
cations are considered here. The fibre reinforcement is a five harness satin
(5HS) weave T800H carbon fibre fabric, and the matrix is the 3633 tough-
ened epoxy resin system. A top view of the 5HS weave fabric architecture is
shown in Figure 5.7. In the 5HS weave, one warp fibre bundle is woven over
Fill
Warp
Fig. 5.7. Top view of 5HS weave fabric architecture.

four fill fibre bundles, and then under one fill fibre bundle. The woven CFRP
laminates are subjected to a thermal load and a tensile stress in the warp
direction. Also, cracks are assumed to be located in the fill fibre bundles.
A series of finite element analyses of the woven CFRP laminates with
cracks under combined mechanical and thermal loads were conducted. The
elastic properties and the stress distributions near the crack tip were stud-
ied for the two-layer [40] and multi-layer [41] woven laminates using two-
dimensinoal finite element analysis. The SIFs for the two-layer [42] and
multi-layer [43] woven laminates from the two-dimensional analysis were
also examined. In addition, three-dimensional finite elements were employed
to model the 5HS weave fabric architecture, and the numerical results for
the elastic properties and stress distributions near the crack front of the
woven laminates were discussed [44].
5.5 Summary
To derive the maximum benefit of woven composite materials for use in a
wide variety of cryogenic systems, basic research on the crack behaviour
in such composites at cryogenic temperatures is required. This chapter
summarises our cryomechanics studies on the crack problems in woven
GFRP/CFRP composite laminates subjected to tensile loading. The two-
and three-dimensional finite element models deal with transverse fibre
bundle cracking, and the elastic properties, the stress distributions near
the cracks and the SIFs are computed. Accurate predictions are possible
when the three-dimensional finite element model is used. However, the two-
dimensional model helps to identify basic aspects of the cryogenic response
for cracked woven composites. Furthermore, these studies would serve as
guidelines for understanding failure mechanisms of woven composites at
cryogenic temperatures.
References
[1] Chou, T.-W. and Ishikawa, T. (1989). ‘Analysis and modeling of two-
dimensional fabric composites’, in Chou, T.-W. and Ko, F.K. (eds), Textile
Structural Composites, Elsevier, New York, pp. 209–264.
[2] Mitchell, N., Bauer, P., Bessette, D., Devred, A., Gallix, R., Jong, C.,
Knaster, J., Libeyre, P., Lim, B., Sahu, A. and Simon, F. (2009). Status
of the ITER Magnets, Fusion Eng. Des., 84, 113–121.
[3] Choi, S. and Sankar, B.V. (2008). Gas Permeability of Various Graphite/
Epoxy Composite Laminates for Cryogenic Storage Systems, Compos. Part
B, 39, 782–791.
[4] Hartwig, G. (1988). Overview of Advanced Fibre Composites, Cryogenics,

28, 216–219.
[5] Reed, R.P. and Golda, M. (1994). Cryogenic Properties of Unidirectional
Composites, Cryogenics, 34, 909–928.
[6] Shindo, Y., Wang, R., Horiguchi, K. and Ueda, S. (1999). Theoretical and
Experimental Evaluation of Double-notch Shear Strength of G-10CR Glass-
cloth/Epoxy Laminates at Cryogenic Temperatures, ASME J. Eng. Mater.
Technol., 121, 367–373.
[7] Shindo, Y., Wang, R. and Horiguchi, K. (2001). Analytical and Experimen-
tal Studies of Short-beam Interlaminar Shear Strength of G-10CR Glass-
cloth/Epoxy Laminates at Cryogenic Temperatures, ASME J. Eng. Mater.
Technol., 123, 112–118.
[8] Kumagai, S., Shindo, Y., Horiguchi, K. and Narita, F. (2004). Experimental
and Finite-element Analysis of Woven Glass-cloth/Epoxy Laminate Tensile
Specimen at Room and Low Temperatures, Mech. Adv. Mater. Struct., 11,
51–66.
[9] Takeda, T., Takano, S., Shindo, Y. and Narita, F. (2005). Deformation and
Progressive Failure Behavior of Woven-fabric-reinforced Glass/Epoxy Com-
posite Laminates under Tensile Loading at Cryogenic Temperatures, Com-
pos. Sci. Technol., 65, 1691–1702.
[10] Shindo, Y., Tokairin, H., Sanada, K., Horiguchi, K. and Kudo, H. (1999).
Compression Behavior of Glass-cloth/Epoxy Laminates at Cryogenic Tem-
peratures, Cryogenics, 39, 821–827.
[11] Shindo, Y., Sumikawa, M., Narita, F. and Sanada, K. (2005). Acoustic Emis-
sion and Fracture Behavior of GFRP Woven Laminates at Cryogenic Tem-
peratures, Cryogenics, 45, 439–449.
[12] Shindo, Y., Horiguchi, K., Wang, R. and Kudo, H. (2001). Double Cantilever
Beam Measurement and Finite Element Analysis of Cryogenic Mode I Inter-
laminar Fracture Toughness of Glass-cloth/Epoxy Laminates, ASME J. Eng.
Mater. Technol., 123, 191–197.
[13] Sumikawa, M., Shindo, Y., Takeda, T., Narita, F., Takano, S. and Sanada, K.
(2005). Analysis of Mode I Interlaminar Fracture and Damage Behavior of
GFRP Woven Laminates at Cryogenic Temperatures, J. Compos. Mater.,
39, 2053–2066.
[14] Horiguchi, K., Shindo, Y., Kudo, H. and Kumagai, S. (2002). End-notched
Flexure Testing and Analysis of Mode II Interlaminar Fracture Behavior of
Glass-cloth/Epoxy Laminates at Cryogenic Temperatures, ASTM J. Com-
pos. Technol. Res., 24, 239–245.
[15] Shindo, Y., Narita, F. and Sato, T. (2006). Analysis of Mode II Interlaminar
Fracture and Damage Behavior in End Notched Flexure Testing of GFRP
Woven Laminates at Cryogenic Temperatures, Acta Mech., 187, 231–240.
[16] Shindo, Y., Sato, T., Narita, F. and Sanada, K. (2008). Mode II Interlaminar
Fracture and Damage Evaluation of GFRP Woven Laminates at Cryogenic
Temperatures Using the 4ENF Specimen, J. Compos. Mater., 42, 1089–1101.
[17] Rizov, V., Shindo, Y., Horiguchi, K. and Narita, F. (2006). Mode III
Interlaminar Fracture Behavior of Glass Fiber Reinforced Polymer Woven
Laminates at 293 to 4 K, Appl. Compos. Mater., 13, 287–304.
[18] Shindo, Y., Shinohe, D., Kumagai, S. and Horiguchi, K. (2005). Analysis and
Testing of Mixed-mode Interlaminar Fracture Behavior of Glass-cloth/Epoxy
Laminates at Cryogenic Temperatures, ASME J. Eng. Mater. Technol., 127,
468–475.
[19] Shindo, Y., Takahashi, S., Takeda, T., Narita, F. and Watanabe, S. (2008).
Mixed-mode Interlaminar Fracture and Damage Characterization in Woven
Fabric-reinforced Glass/Epoxy Composite Laminates at Cryogenic Temper-
atures Using the Finite Element and Improved Test Methods, Eng. Fract.
Mech., 75, 5101–5112.
[20] Miura, M., Shindo, Y., Takeda, T. and Narita, F. (2012). Cryogenic Interlam-
inar Fracture Properties of Woven Glass/Epoxy Composite Laminates under
Mixed-mode I/III Loading Conditions, Appl. Compos. Mater., 20, 587–599.
[21] Miura, M., Shindo, Y., Takeda, T. and Narita, F. (2012). Interlaminar Frac-
ture Characterization of Woven Glass/Epoxy Composites under Mixed-mode
II/III Loading Conditions at Cryogenic Temperatures, Eng. Fract. Mech., 96,
615–625.
[22] Kumagai, S., Shindo, Y. and Inamoto, A. (2005). Tension-tension Fatigue
Behavior of GFRP Woven Laminates at Low Temperatures, Cryogenics, 45,
123–128.
[23] Shindo, Y., Inamoto, A. and Narita, F. (2005). Characterization of Mode I
Fatigue Crack Growth in GFRP Woven Laminates at Low Temperatures,
Acta Mater., 53, 1389–1396.
[24] Shindo, Y., Inamoto, A., Narita, F. and Horiguchi, K. (2006). Mode I Fatigue
Delamination Growth in GFRP Woven Laminates at Low Temperatures,
Eng. Fract. Mech., 73, 2080–2090.
[25] Shindo, Y., Takeda, T., Narita, F., Saito, N., Watanabe, S. and Sanada, K.
(2009). Delamination Growth Mechanisms in Woven Glass Fiber Reinforced
Polymer Composites under Mode II Fatigue Loading at Cryogenic Temper-
atures, Compos. Sci. Technol., 69, 1904–1911.
[26] Miura, M., Shindo, Y., Narita, F., Watanabe, S. and Suzuki, M. (2009).
Mode III Fatigue Delamination Growth of Glass Fiber Reinforced Polymer
Woven Laminates at Cryogenic Temperatures, Cryogenics, 49, 407–412.
[27] Shindo, Y., Miura, M., Takeda, T., Saito, N. and Narita, F. (2011). Cryogenic
Delamination Growth in Woven Glass/Epoxy Composite Laminates under
Mixed-mode I/II Fatigue Loading, Compos. Sci. Technol., 71, 647–652.
[28] Kriz, R.D. (1985). Influence of Damage on Mechanical Properties of Woven
Composites at Low Temperatures, ASTM J. Compos. Technol. Res., 7,
55–58.
[29] Hyer, M.W. (1998). Stress Analysis of Fiber-reinforced Composite Materials,
McGraw-Hill, New York.
[30] Takeda, T., Shindo, Y., Kumagai, S., Narita, F. and Horiguchi, K. (2001).
Thermal-mechanical Response of Cracked Woven Glass/Epoxy Laminates in
a Tension Field at Low Temperature, Phys. Mesomech., 4, 97–104.
[31] Takeda, T., Shindo, Y., Narita, F. and Kumagai, S. (2002). Thermoelastic
Analysis of Cracked Woven GFRP Laminates at Cryogenic Temperatures,
Cryogenics, 42, 451–462.
[32] Takeda, T., Shindo, Y., Narita, F. and Sanada, K. (2004). Stress Intensity
Factors for Woven Glass/Epoxy Laminates with Cracks at Cryogenic Tem-
peratures, Mech. Adv. Mater. Struct., 11, 109–132.
[33] Takeda, T., Shindo, Y. and Narita, F. (2004). Fracture Mechanics Analy-
sis of Multi-layer Plain Weave Fabric Laminates with Transverse Cracks at
Cryogenic Temperatures, Int. J. Comput. Methods, 1, 151–169.
[34] Dahlerup-Petersen, K. and Perrot, A. (1979). Properties of Organic Compos-
ite Materials at Cryogenic Temperatures, ISR-BOM/79-39, CERN (Geneva,
Switzerland).
[35] Hashin, Z. (1979). Analysis of Properties of Fiber Composites with
Anisotropic Constituents, ASME J. Appl. Mech., 46, 543–550.
[36] Sanada, K., Shindo, Y. and Kudo, H. (1999). Damage and Fracture Mechan-
ics Analysis of G-11 Woven Glass-epoxy Laminates at Cryogenic Tempera-
tures, J. Japan Inst. Metals, 63, 221–229.
[37] Takeda, T., Shindo, Y. and Narita, F. (2004). Three-dimensional Ther-
moelastic Analysis of Cracked Plain Weave Glass/Epoxy Composites at
Cryogenic Temperatures, Compos. Sci. Technol., 64, 2353–2362.
[38] Kumagai, S., Shindo, Y., Horiguchi, K. and Takeda, T. (2003). Mechanical
Characterization of CFRP Woven Laminates between Room Temperature
and 4 K, JSME Int. J. Ser. A, 46, 359–364.
[39] Takeda, T., Shindo, Y., Fukuzaki, T. and Narita, F. (2012). Short Beam
Interlaminar Shear Behavior and Electrical Resistance-based Damage Self-
sensing of Woven Carbon/Epoxy Composite Laminates in a Cryogenic Envi-
ronment, J. Compos. Mater., 48, 119–128.
[40] Watanabe, S., Shindo, Y., Narita, F. and Takeda, T. (2009). Thermal-
mechanical Analysis of Satin Weave CFRP Composites with Cracks at
Cryogenic Temperatures, J. Reinf. Plast. Compos., 28, 1319–1337.
[41] Watanabe, S., Shindo, Y., Takeda, T. and Narita, F. (2008). Cryogenic
Mechanical Response of Multilayer Satin Weave CFRP Composites with
Cracks, Mech. Compos. Mater., 44, 331–340.
[42] Watanabe, S., Shindo, Y., Takeda, T., Natira, F. and Sanada, K. (2009).
Cryogenic Behavior of Cracks in Satin Woven CFRP Laminates under Tensile
Loading, J. Solid Mech. Mater. Eng., 3, 22–37.
[43] Watanabe, S., Shindo, Y., Takeda, T. and Narita, F. (2010). Singular Stress
Fields for Transverse Cracks in Multi-layer Satin Woven CFRP Compos-
ite Laminates under Cryogenic Conditions, Adv. Theor. Appl. Mech., 3,
349–367.
[44] Takeda, T., Shindo, Y., Watanabe, S. and Narita, F. (2012). Three-
dimensional Stress Analysis of Cracked Satin Woven Carbon Fiber
Reinforced/Polymer Composites under Tension at Cryogenic Temperatures,
Cryogenics, 52, 784–792.
Chapter 6
EFFECTS OF FABRICATION PROCESSES AND

NANOCLAY INCLUSION ON THE PERFORMANCE OF
WOVEN CARBON/EPOXY COMPOSITE LAMINATES
Mahesh Hosur∗,† , Farhan Chowdhury‡ and Shaik Jeelani∗

∗ Materials Science and Engineering Department, Tuskegee University,
Tuskegee, AL 36088, USA
† hosur@mytu.tuskegee.edu
‡ IGB Fellow, University of Illinois at Urbana-Champaign,

Urbana, IL 61801, USA
6.1 Introduction
Carbon fibre-reinforced polymer composites (CFRP) are revolutionising
industries such as aerospace, automobile, marine, civil infrastructures and
so on. A big swing is taking place in the weight-sensitive aerospace indus-
try in particular. Boeing has launched the 787 Dreamliner, a super-efficient
airplane, where as much as 50% of the primary structure, including the
fuselage and wing, is made of composite materials. The 787 provides air-
lines with unmatched fuel efficiency, lower maintenance costs, resulting in
exceptional environmental performance. The 787 will use 20% less fuel for
comparable missions than any similarly sized airplane. It will also travel at
speeds similar to today’s fastest wide bodies, Mach 0.85. Passengers will
also see improvements with the new airplane, from an interior environment
with higher humidity to increased comfort and convenience.
Several supercars (McLaren F1, Ferrari Enzo and Porsche Carrera GT)
over the past few decades have incorporated CFRP extensively in their man-
ufacture, using it for their monocoque chassis as well as other components
such as roofs (in the BMW M3), hoods, door panels and drive shafts. In civil
179
180 M. Hosur, F. Chowdhury and S. Jeelani
infrastructures, CFRP is used underneath spans to help prevent excessive

deflections, or wrapped around beams to limit shear stresses. Westgate
Bridge in Melbourne is the largest bridge in the world to be reinforced
with carbon fibre laminates. CFRP composites are also used in the hulls of
racing yachts, boat decks, structural pipelines, bicycle frames etc.
With the advent and application of nanotechnology, polymer compos-
ites are showing increasing promise for the next generation of material sys-
tems for structural applications. However, most of the work carried out in
the polymer nanotechnology area is concentrated on improving the thermal,
mechanical, optical and barrier properties of thermosetting and thermo-
plastic polymer systems. Inorganic nanoparticles have gained acceptance
as potential reinforcing materials because of their low cost and ease of
fabrication [1, 2]. Many researchers have attempted to enhance the prop-
erties of different types of polymers by the addition of nanoclays. Over
the past decade, a significant amount of work has focused on improving the
properties of epoxies using nanoclays. Some of the earliest work on inorganic
strengthening was performed in 1987 at Toyota. Researchers demonstrated
the possibility of functionalising the surface of layered inorganic silicates to
make it compatible with a polymeric matrix [3, 4].
Pinnavaia and co-workers [5–7] showed that by adding organophilic
montmorillonite into diglycidylether of bisphenol A (DGEBA), it was pos-
sible to significantly increase the tensile strength and modulus of the epoxy
compared with the control samples. This increase is more distinct if the
strength and modulus values of the starting epoxy are low. Park and Jana [8]
reported that the degree of exfoliation of nanoclay material in an epoxy
matrix is thought to be responsible for an increase in the properties of
the materials. The hypothesis of the exfoliation of nanoclay suggests that
elastic forces developed in clay galleries during epoxy curing are responsible
for the exfoliation of clay structures.
Over the last two decades, some studies have shown potential improve-
ment in the properties and performances of fibre-reinforced polymer matrix
materials in which nano- and micro-scale particles were incorporated. It
was also shown that the mechanical properties of CFRP were improved at
a small loading of wt. % clay. Schmidt [9], Novak [10], Kojima et al. [4],
Mark [11] and Hussain et al. [12] demonstrated the possible technology of
dispersing Al2 O3 particles in a matrix and investigated its effect on the
mechanical properties of CFRP. Gilbert et al. [13, 14] and Timmerman [15]
have demonstrated that fracture toughness and mechanical properties are
increased upon the incorporation of metal and inorganic particles. In these
Effects of Fabrication Processes on the Performance of Composite Laminates 181
studies, they developed the concept of La PolynanoGrESS (layered poly-

nanomeric graphite epoxy scaled system), which utilises the nanoparticle
effect in an epoxy matrix, and scaled it to continuous carbon fibre-reinforced
composites systems. Karaki et al. [16] incorporated layered clay, alumina
and titanium dioxide into an epoxy matrix and fabricated continuous
carbon fibre-reinforced polynanomeric matrices to study tension–tension
fatigue behaviour. They found that the number of micro-cracks in each
layer depended on the type of particles and their concentration. Hack-
man and Hollaway [17] studied the potential applications of clay nanocom-
posite materials to civil engineering structures. They concluded that its
ability to increase the service life of materials, subjected to aggressive
environments, could be utilised to increase the durability of glass and
carbon fibre composites. Miyagawa et al. [18] studied the influence of
biobased epoxy clay nanocomposites as a new matrix for carbon fibre
composites. In their study, they found that the flexural strength and
modulus did not change with the addition of nanoclay. However, they
reported increased inter-laminar shear strength. Haque et al. [19] studied
the effects of nanoclay on improving the mechanical and thermal proper-
ties of S2-glass/epoxy composites. They showed that by dispersing 1% by
weight of nanosilicates, S2-glass/epoxy-clay nanocomposites exhibited an
improvement of 44, 24 and 23% in inter-laminar shear strength, flexural
strength and fracture toughness respectively. Mahfuz et al. [20] studied the
tensile response of carbon nanoparticles/whiskers reinforced composites and
observed a 15–17% improvement in tensile strength and modulus. Chisholm
et al. [21] investigated the response of carbon/silicon carbide (SiC)-
epoxy nanocomposites and showed a 20–30% improvement in mechanical
properties.
In the current research, plain weave carbon/epoxy panels dispersed
with 1–3% by weight of nanoclay were fabricated by hand lay-up followed
by vacuum bagging and vacuum-assisted resin infusion moulding (VARIM)
processes. Flexural strength and modulus were determined through a 3-
point bend flexural test. Thermo-mechanical characterisation was carried
out using a TA-instrument, DMA 2980, to determine the dynamic mechan-
ical properties such as storage modulus, loss modulus and tan delta. The
glass transition temperature (Tg ) was identified as the temperature cor-
responding to a peak of tan delta versus temperature curve. Microscopic
approaches were used to investigate fracture behaviour and the mechanism
of the material. The fibre volume content of composites was also determined
using the matrix digestion method.
6.2 Experimental
6.2.1 Materials and fabrication of composites
Manufacturing of laminated composites for this investigation was carried
out using two methods: the first one was vacuum-assisted resin transfer
moulding (VARTM) and the second was the hand lay-up process followed
by vacuum bagging. The materials used in this study are shown in Table 6.1.
Woven fabrics are typically produced by the interlacing of warp (0◦ )
fibres and weft (90◦ ) fibres in a regular pattern or weave style. The integrity
of the woven fabrics is maintained by the mechanical interlocking of fibres.
Drape (an ability of a fabric to conform to a complex surface), surface
smoothness and stability of the fabric are controlled primarily by the weave
style. In a laminated composite, the surface interaction between the fibres
and resin is controlled by the degree of bonding that exists between the
two. This is heavily influenced by the treatment given to the fibre surface.
Table 6.2 lists the properties of a typical carbon fibre.
The thermoset polymeric matrix used in this investigation is a commer-
cially available SC-15 epoxy resin, obtained from Applied Poleramic, Inc.
It has two parts, part A (epoxy) and part B (hardener, alkyl polyamine).
Part A itself is a mixture of three components, namely, diglycidylether of
bisphenol A, aliphatic diglycidylether and an epoxy toughner. SC-15 is a low
viscosity two-phase toughened epoxy resin system and well established in
composite manufacturing technology. Some of the physical and mechanical
properties of SC-15 are shown in Tables 6.3 and 6.4.
Table 6.1. Materials used in this investigation.
Composites Experiments
No. of layers Fibre type Resin

8 Plain weave carbon, 6k Epoxy SC-15 Flexural, matrix
fibre tow digestion, WAXD, SEM
3 Plain weave carbon, 6k Epoxy SC-15 DMA
fibre tow
Table 6.2. Mechanical and thermal properties of carbon fibres [25].
Density, Modulus, Tensile strength, Poisson’s

g/cc GPa MPa ratio Thermal coefficient, α
1.81 235 3700 0.2 −0.5 × 10−6 /◦ C

Table 6.3. Physical properties of SC-15 epoxy resin [26].
Viscosity, Cured density, Mix ratio,

(cps) (gm/cm3 ) (by weight)
Mixed: 300 1.09 Part A: 100

Resin: 590 Part B: 30
Hardener: 65
Table 6.4. Mechanical properties of SC-15 epoxy resin [23].
Glass Transition temp., Flexural modulus, Fracture toughness, Tensile strength,

Tg ,(◦ C) (GPa) GIC , (J/m2 ) MPa
104 2.7 990 62
Nanoclay, in its raw state, is naturally occurring montmorillonite; a

2-to-1 layered smectite clay mineral with a plate-like structure. Surface-
modified montmorillonite clays are treated with ammonium or phospho-
nium ions giving rise to an intercalated structure. These modified clays,
produced by Nanocor Inc., also referred to as Nanomer, are utilised
to make nanocomposites. Depending on the surface treatment, they are
assigned a number. The nanoclay used in this study is Nanomer I.28E;
the physical properties of these clays are listed in Table 6.5.
The ultrasonic cavitation technique is one of the most efficient means to
disperse nanoparticles into a polymer [25–27]. The Sonics Vibra Cell ultra-
sonic processor (Ti-horn, 20 kHz, 100 W/cm2 ), as shown in Figure 6.1a, was
used to obtain a homogeneous mixture of epoxy resin and Nanomer I-28E
nanoclay. SC-15 epoxy comes with two parts: part A and part B. At first,
part A of the SC-15 epoxy (mixture of 60–70% diglycidylether of bisphenol
A and 10–20% aliphatic diglycidylether) was sonicated with nanoclay at an
amplitude of 55% with a total mixing time of 30 min. Nanoclays pick up
Table 6.5. Physical properties of Nanomer I.28E [24].
Appearance White Powder
Mean dry particle size (microns) 8–10

+325 Mesh residue (%) 0.1
Specific gravity 1.9
Bulk density (kg/m3 ) 313
Moisture (%) 3 max
Mineral purity (% min) 98.5
Fig. 6.1. (a) Vibra cell ultrasonic processor, (b) Cooling part of sonicated mix,
(c) Mechanical mixing of part A with nanoparticles and part B of SC-15 Epoxy,
(d) Degasification of resin mixture.
moisture from the air very easily. To get rid of any moisture, the nanoclay
was heated to 100◦ C for 2 hours before sonication. A pulser cycle (with a
turning on and offtime ratio of 2:1) was used to control the mixture tem-
perature. It was recommended to keep the mixture temperature at around
40–50◦C to ensure good mixing, which was monitored using an infrared
thermometer; the pulser cycle helps to achieve this goal. In addition, a
water bath was also used.
After sonication, the homogeneous mixture was cooled down to room
temperature in a refrigerating cooler maintained at 5◦ C as shown in
Figure 6.1b. As it cools down, the resin viscosity also increases promoting
a suitable condition for a more intercalated/exfoliated nanoclay structure.
The sonicated part A with nanoclay was then mixed with part B of the
SC-15 epoxy (epoxy hardener, 70–90% cycloaliphatic amine and 10–30%

polyoxylalkylamine) at a ratio of 10:3 using a high-speed mechanical stirrer
for about 5 min. Figure 6.1c illustrates the mechanical mixing of part A
and part B.
Mechanical mixing introduces air bubbles into the resin. This war-
ranted the need of removing the trapped air and reaction volatiles from
the mixture. This was achieved using a vacuum for about 30 min. The
resin mixture was kept in a sealed desiccating chamber, as can be seen in
Figure 6.1d, and connected to a vacuum suction pump. The vacuum pulls
out trapped air from the mixture and the resin is degasified.
The VARTM process uses vacuum pressure to remove air from the
fabric lay-up before and while the matrix resin is introduced into fabrics.
The pressure difference between the atmosphere and vacuum is the driving
force for infusion of the resin into the lay-up. For fabricating a laminate,
Frekote (mould-releasing agent) was sprayed onto the mould. The required
number of layers was carefully placed on the mould. Then, a sealant tape
was tacked onto the surface of the mould about 25–50 mm from the perime-
ter of the fabric layers. The resin supply tubes were connected with the
mould end of the tube connected to a spiral wrap, along with a distribution
mesh that lies on the top and bottom of the preform. This facilitates easy
flow of the resin over the top and through-thickness of the laminate when
the vacuum is applied. Tubes linking the vacuum pump and spiral wrap
were also connected. The resin traps were placed between the vacuum pump
and mould to collect excess resin. Finally, the vacuum bag was placed on the
mould and pressed firmly against the sealant tape to provide a vacuum tight
system. The preform was left to debulk under vacuum. After debulking,
the nanophased SC-15 resin system was infused with impregnating fabric
as the resin flow advanced towards the vacuum side. The resin inlet valve
was closed when the resin reached the suction side and the infused laminate
was left to cure at room temperature. The vacuum was maintained until the
end of the cure to remove any volatiles generated during polymerisation, in
addition to maintaining a pressure of one atmosphere. The arrangement of
the fabrication process is detailed schematically in Figure 6.2.
The hand lay-up technique is illustrated in Figure 6.3. A sheet of non-
porous Teflon release film was provided at the bottom on which the carbon
fabric was laid up. On each layer some prepared resin mixture was poured
and spread out using a hand roller. The gentle rolling action of the hand
roller ensured proper wetting of the carbon fabric and excess resin was
squeezed out. Another sheet of non-porous Teflon release film was provided
Suction line
Distribution mesh
Porous Teflon release film Fabric lay-up
Vacuum bagging
Resin infusion line

Cleaned aluminium tool
Fig. 6.2. VARTM lay-up sequence.
on top of the fabric lay-up. A suction line was provided on one side of the
lay-up as shown in Figure 6.3. An aluminium plate, placed on top of the
non-porous Teflon release film, gave a good surface finish. Finally, the entire
lay-up was vacuumbagged and left to cure for 24 hours.
The room temperature cured material was removed from the vacuum
bag and the edges were trimmed. Test samples were machined for ther-
mal and mechanical characterisation. They were thermally post-cured at
100◦ C for 5 hours in a mechanical convection oven to study the effect of
post-curing.
6.2.2 Characterisations
6.2.2.1 X-ray diffraction (XRD) analysis
Changes in the d-spacing between nanoclay platelets in the composite
panels were investigated using wide angle X-ray scattering (WAXS)
(a) (b) (c)
(f) (e) (d)

Fig. 6.3. (a) Resin is poured onto the carbon fabric, (b) Fibres are wetted with the
rolling action, (c) Non-porous Teflon release film at the top after fabric lay-up, (d)
Aluminium plate on top of the non-porous Teflon release film, (e) Lay-up is vacuum
bagged, (f) Edge trimmed hand lay-up panel.
measurements. This is an established method to evaluate the degree of clay

exfoliation. WAXS measurements were carried out in a Rigaku D/MAX
2200X-ray diffractometer with CuK α radiation (λ = 1.54 Å) using a scan-
ning speed of 1◦ /min and operating at 40 kV and 30 mA. During the XRD
experiments, samples were analysed in reflection mode. The sample size
was maintained at 17.5 mm (length) × 13.5 mm (width) × 1.7 mm (thick-
ness). Characteristics of the XRD diffraction pattern were collected and
analysed. Different wt. % loading of carbon/nanoclay-epoxy composites
showed different characteristics.
6.2.2.2 Matrix digestion test

The fibre volume fraction and void content of VARTM and hand lay-up
panels were estimated using the matrix digestion test according to ASTM
D 3171-99(2004) [28]. Samples of dimensions 3 × 3 × 0.2 cm were cut from
the panels and sides were subjected to grinding to ensure smooth edges.
Sample dimensions were measured and weighed accurately. Samples were
then placed in a bath of 80% concentrated nitric acid maintained at 75◦ C for
about 5 hours. At that stage, the matrix was completely digested and only
the fibres remained. The fibres were filtered and then washed with acetone
and water repeatedly. The fibre was then dried in an oven maintained at
100◦ C for about an hour. The fibre weight was measured. The fibre volume
percentage in the composite was calculated as follows:
Fibre vol. %, VF = [(W/F )/(w/c)] × 100
The matrix volume was calculated as follows:

(w−W )/ρm
Matrix vol. %, Vm = Vc × 100
Void vol. %, Vv = 100 − (Vm + Vf )

where:
W = weight of fibre in the composite
w = weight of the initial composite specimen
F = fibre density
ρm = density of the matrix
Vc = volume of the composite
c = composite density.
6.2.2.3 Flexure test

A 3-point bend flexure test was performed to evaluate the flexural modu-
lus and strength of each of the material systems. The span length of the
samples was 52 mm and the nominal thickness was 3.25 mm; the width
was maintained at 25.4 mm. This test was carried out according to ASTM
D790-02 [29]. Tests were conducted in displacement control mode with
a crosshead speed of 1.2 mm/min. Load-deflection data for each sample
was collected; the flexural stress and strain were then computed. The
flexural modulus was calculated from the slope of the stress–strain plot.
Five samples made of 8-layers of plain weave carbon fabric of each type
were tested, and the average values of flexural strength and modulus were
determined.
6.2.2.4 Scanning electron microscopy (SEM)

In order to investigate the change in microstructure due to nanoclay
infusion, SEM studies were carried out. SEM also facilitates observing the
failure mode at the micro-level. Failure samples from 3-point flexure tests
were examined to see changes in the failure mode. SEM analyses were car-
ried out using a JEOL JSM 5800. Samples were placed on a sample holder
covered with silver paint and coated with gold to prevent charge build-up
by electrons absorbed by the specimen. A 15 kV accelerating voltage was
applied to accomplish the desired magnification.
6.2.2.5 Shear strength of cured resin

The apparent shear strength of the resin was determined using a short
beam shear test on a 3-point bending fixture. Specimens were cut to a size
of 42 mm (length) × 7 mm (width) × 7 mm (thickness). The span length
between the two supports was 28 mm. Samples were loaded in displace-
ment control mode with a crosshead speed of 1.3 mm/min. Apparent shear
strength was measured as follows:
3 Pb
SH =
4 bd
where, Pb is the breaking load, b is the width of the specimen and d is the
thickness of the specimen. A minimum of three specimens were used for
each type of sample.
6.2.2.6 Dynamic mechanical analysis

The dynamic mechanical analysis (DMA) of various specimens was car-
ried out using a DMA 2980 (Manufacturer: TA Instruments, Newark, DE).
Samples made of 3-layered plain weave carbon fabric were cut into small
pieces using a diamond cutter and machined using a mechanical grinder
to maintain the specified sample dimensions. The width of the samples
was 12 mm and the span length to thickness ratio was 10. The test was
carried out according to ASTM D4065-01 [30]. Tests were run on a single
cantilever beam mode with a frequency of 1Hz and amplitude of 15 µm.
The temperature was ramped from 35 to 150◦ C at a rate of 3◦ C/min. This
heating rate was maintained throughout the test runs so that there is a
minimum temperature lag between the sample and furnace environment.
At least five samples were tested for each kind. From the test data, the stor-
age modulus, which gives the dynamic elastic response of the samples, loss
modulus, which gives the dynamic plastic response of the samples and tan
delta, which is the ratio of loss modulus/storage modulus, were determined.
6.3 Results and Discussion

6.3.1 XRD analysis
It is generally believed that the improvement of properties of nanoclay
composites is directly related to complete exfoliation of silicate layers in the
polymer matrix. However, a processing technique that produces complete
exfoliation is still a technical challenge. This may be due to the high
Fig. 6.4. WAXD patterns of nanocomposites with varying clay contents.
viscosity of the resin and the strong tendency of nanoclay particles to

agglomerate [31]. Vaia et al. [32] have suggested that the degree of exfoli-
ation can be improved through the aid of conventional shear devices such
as extruders, mixers, ultrasonicators etc. Correlation between the degree of
exfoliation and mechanical properties of nanocomposites will also be dis-
cussed in this section. Figure 6.4 shows WAXD patterns of carbon/epoxy
nanocomposites with varying clay contents. The patterns reveal the effect
of different weight percentage loading of nanocomposite on intercalation.
Pure Nanomer I.28E shows a characteristic diffraction peak corre-
sponding to the (001) plane at 2.37 nm. There is also a second-order peak
at 5.88 (1.5 nm) in the WAXD pattern for a nanoclay powder sample. The
control sample, lacking any nanoclay loading, shows no peak within the
smaller angle range as expected. The nanocomposite with 1 wt. % clay
shows a shallow peak with a d-spacing of 3.64 nm, whereas, the nanocom-
posite with 2 wt. % clay shows an even weaker peak with a d-spacing of
3.63 nm. However, nanocomposites with 3 wt. % clay content show a peak
from silicate basal planes (d001 ) with a d-spacing slightly lower than the
composites containing 2 wt. % clay. It reduces to 2.94 nm which is still
higher than pure nanoclay powder. A stronger peak implies a more ordered
intercalated nanocomposite. But it is evident that up to 2 wt. % nanoclay
loading, the sample peaks are highly disordered as shown by the significant
loss of intensity and broadening of peaks.
6.3.2 Matrixdigestion test results

The fibre volume fraction and hence void content in the composites were
estimated using the matrix digestion method. Table 6.6 summarises the
results of the matrix digestion test. It is seen from Table 6.6 that the fibre
volume fraction of VARTM samples (control, 1%, 2%, 3% nanoclay samples)
is about 56%, with a void content of about 6%.
Table 6.7 summarises the results of the matrix digestion test of hand
lay-up samples. It is seen from Table 6.7 that the fibre volume fraction of
1 wt. % and 2 wt. % nanoclay samples is 55%, while it is about 56% for the
control and 3 wt. % nanoclay samples. The void content is found to be close
to 3% for all samples. The lower amount of void content may attribute to
better compaction of hand lay-up panels.
When Tables 6.6 and 6.7 are compared, it is seen that the VARTM and
hand lay-up samples have almost the same amount of reinforcement, i.e.,
fibre volume fraction; but there is a significant difference in void content.
Possibly, the two different processing routes have influenced this change. In
the hand lay-up process, the resins are spread into fibres using a roller. The
rolling action and gentle squeeze makes the layers more compact, whereas
in VARTM manufacturing, the applied pressure on lay-up is only −1 atm
which may not be sufficient to the make layers more compact. One more
thing to be noted is that the fibre volume fraction in the VARTM samples
is more consistent than that for the hand lay-up samples. A slight difference
in the fibre volume fraction in the hand lay-up samples may be due to the
non-mechanised technique. Furthermore, it can be deduced that the hand
Table 6.6. Matrix digestion test results for VARTM samples.
1% Nanoclay 2% Nanoclay 3% Nanoclay

Control system system system
Fibre volume fraction, vf (%) 56.45 56.48 56.37 56.4

Void content, vv (%) 6.2 6.12 6.02 5.98
Table 6.7. Matrix digestion test result for hand lay-up panels.
1% Nanoclay 2% Nanoclay 3% Nanoclay

Control system system system
Fibre volume fraction, vf (%) 56.56 55.34 55.14 56.17

Void content, vv (%) 2.78 2.76 2.89 2.63
lay-up process is very much operator sensitive and operator skill is one of
the criteria for getting a good composite part.
6.3.3 Flexure Test

6.3.3.1 VARTM samples
Flexural tests of VARTM samples (with or without thermal post-curing)
were performed to evaluate the bulk stiffness and strength of the carbon/
epoxy nanocomposites. Typical stress–strain behaviours from the flexural
tests are shown in Figures 6.5 and 6.6. The positive effect of thermal
post-curing is also evident if we compare Figure 6.5 with Figure 6.6. In
Figures 6.5 and 6.6, it is evident that the composite strength is improved
with up to 2 wt. % of nanoclay loading. An increase in modulus is also
seen for samples with up to 2 wt. % nanoclay loading. Strength as well as
modulus decreased in the 3 wt. % of nanoclay samples. The average prop-
erties obtained from tests are listed in Tables 6.8 and 6.9. Table 6.8 shows
the properties of samples without thermal post-curing, while Table 6.9
shows the properties of samples which underwent thermal post-curing. From
Table 6.8, it is seen that for 2 wt. % nanoclay loading, the flexural strength
and modulus increased by about 25 and 14% respectively, while Table 6.9
400
300
Flexural stress, MPa
200
3%nanoclay
2%nanoclay
1%nanoclay
Control sample
100
0
0 0.0025 0.0050 0.0075 0.0100
Strain, m/m
Fig. 6.5. Flexural stress-strain plot of VARTM samples without post-curing.

500
400
300
200
3%nanoclay
2%nanoclay
1%nanoclay
Control sample
100
0
0 0.004 0.008 0.012
Strain, m/m
Fig. 6.6. Flexural stress-strain plot of post-cured VARTM samples.
Table 6.8. Flexural test results of VARTM samples without thermal post-curing.
Flexural strength, % Gain/loss Flexural modulus, % Gain/loss

MPa in strength GPa in modulus
Control sample 306 ± 13.5 — 35.0 ± 0.63 —

1% Nanoclay 354 ± 6.11 15.68 37.0 ± 0.5 5.71
2% Nanoclay 383 ± 11.59 25.16 40.0 ± 1.0 14.29
3% Nanoclay 361 ± 10.06 17.97 38.0 ± 0.66 8.57
Table 6.9. Flexural test results of thermally post-cured VARTM samples.

Control sample 365 ± 7.57 — 35.3 ± 0.15 —

1% Nanoclay 399 ± 18.88 9.31 36.7 ± 0.43 3.97
2% Nanoclay 415 ± 17.21 13.70 38.6 ± 0.65 9.35
3% Nanoclay 373 ± 9.29 2.19 37.7 ± 0.45 6.80
indicates an increase of strength and modulus by about 14 and 9% respec-

tively for 2 wt. % nanoclay loading. It is very clear from these tables that the
absolute values of strength for thermally post-cured samples are better than
room temperature cured ones. It should be noted that thermal post-curing
of VARTM samples hardly influenced the modulus. The average values
of modulus for room temperature cured and thermally post-cured samples

remained almost the same. The strength values were greatly improved upon
thermal post-curing.
6.3.3.2 Hand lay-up samples

Flexural tests of hand lay-up samples (with or without thermal post-curing)
were performed to evaluate the bulk stiffness and strength of the carbon/
epoxy nanocomposites. Typical stress–strain behaviours from the flexural
tests are shown in Figures 6.7 and 6.8. Some degree of non-linearity is
seen before reaching maximum stress, especially in the case of 1 wt. % and
2 wt. % nanoclay samples. The positive effect of thermal post-curing is also
evident if we compare Figure 6.7 with Figure 6.8. In Figures 6.7 and 6.8,
it is evident that the composite strength is improved with up to 2 wt. %
of nanoclay loading. An increase in modulus is seen in samples with up to
3 wt. % of nanoclay loading.
Tables 6.10 and 6.11 show the average properties of room temperature
cured and thermal post-cured samples respectively. From Table 6.10, it is
seen that with 2 wt. % nanoclay loading the flexural strength and modulus
increased by about 25% and 22% respectively, while Table 6.11 indicates an
increase of the strength and modulus by about 31% and 21% respectively
500
400
300
200
3%nanoclay
2%nanoclay
1%nanoclay
Control sample
100
0
0 0.003 0.006 0.009 0.012
Strain, m/m
Fig. 6.7. Flexural stress-strain plot of hand lay-up samples without post-curing.
600
500
400
300
200 3%nanoclay
2%nanoclay
1%nanoclay
Control sample
100
0
0 0.003 0.006 0.009 0.012
Strain, m/m
Fig. 6.8. Flexural stress-strain plot of post-cured hand lay-up samples.
Table 6.10. Flexural test results of room temperature cured hand lay-up samples.

Control sample 350 ± 12.73 — 37.11 ± 1.16 —

1% Nanoclay 413 ± 6.45 18.0 43.5 ± 1.03 17.21
2% Nanoclay 439 ± 7.32 25.42 45.35 ± 1.68 22.2
3% Nanoclay 411 ± 10.56 17.42 47.1 ± 1.23 26.9
Table 6.11. Flexural test results of thermally post-cured hand lay-up samples..

Control sample 380 ± 3.3 — 37.57 ± 0.77 —

1% Nanoclay 426 ± 10.81 12.10 43.8 ± 2.13 16.58
2% Nanoclay 498 ± 12.81 31.05 45.6 ± 0.81 21.37
3% Nanoclay 446 ± 8.95 17.36 46.9 ± 1.22 24.8
for 2 wt. % of nanoclay loading compared with the control samples. It is very
clear from these tables that the absolute values of strength for thermally
post-cured hand lay-up samples are better than room temperature cured
samples. Thermal post-curing does not have any appreciable effect on the
average modulus values. It can be seen from Tables 6.10 and 6.11 that the
modulus values remained almost the same in both room temperature cured
samples and thermally post-cured samples.
6.3.3.3 Comparison of VARTM and hand lay-up properties

Table 6.12 shows a comparison between VARTM and hand lay-up samples.
It is evident from this table that hand lay-up samples give better results
than VARTM samples in terms of strength and modulus. Improvement in
strength for hand lay-up samples over VARTM samples may be attributed
to the lower void content and better compaction. The trend in modulus,
as shown in Figures 6.5 and 6.6, for VARTM samples reveals that with up
to 2 wt. % of nanoclay loading, the modulus increases and then decreases
for 3 wt. %. This may be due to fact that with a higher percentage of
nanoclay loading there is a lot of agglomeration of particles, which resulted
in a bulk system with less than 3 wt. % loading. It may be suggested to
investigate the effect of mixing parameters (e.g., mixing time, volume of
mixing, amplitude of oscillation etc.) which may aid to reduce agglomera-
tion. However, Figures 6.7 and 6.8 for hand lay-up samples show that the
modulus increased with up to 3 wt. % of nanoclay loading. Spreading of
the resin with a hand roller might be responsible for this trend. The gentle
squeezing action of the hand roller might have generated sufficient shear
force to disperse some of the agglomerated particles into the resin.
Table 6.12. Comparison of flexural properties of VARTM and hand lay-up samples.
Strength, % Modulus, %
MPa Improvement GPa Improvement
Control sample
VARTM 365 ± 7.57 — 35.3 ± 0.15 —
hand lay-up 380 ± 3.3 4.11 37.57 ± 0.77 6.43
1% Nanoclay
VARTM 399 ± 18.88 — 36.7 ± 0.43 —
hand lay-up 426 ± 10.81 6.77 43.8 ± 2.13 19.35
2% Nanoclay
VARTM 415 ± 17.21 — 38.6 ± 0.65 —
hand lay-up 498 ± 12.81 20.00 45.6 ± 0.81 18.13
3% Nanoclay
VARTM 373 ± 9.29 — 37.7 ± 0.45 —
hand lay-up 446 ± 8.95 19.57 46.9 ± 1.22 24.40
6.3.4 SEM studies

Scanning electron microscopy studies were carried out on pristine as well as
samples subjected to flexural testing. Figures 6.9(b)–(d)–(f) illustrate the
fracture surfaces while Figures 6.9(a)–(c)–(e) show the pristine samples of
(a) Pristine control sample (b) Failed control sample
(c) Pristine 2% nanoclay sample (d) Failed 2% nanoclay sample
(e) Pristine 3% nanoclay sample (f) Failed 3% nanoclay sample
Fig. 6.9. SEM studies of fracture surfaces and pristine samples.

the control, 2 wt. % and 3 wt. % nanoclay loading respectively. These micro-
graphs were taken at a magnification of 300. It can be seen in Figure 6.9(b)
that the control samples do not exhibit good adhesion between the fibre
and matrix. Fibre pullout followed by fibre breakage is very evident from
Figure 6.9(b). Figure 6.9(d) shows a fracture surface of the 2 wt. % nan-
oclay sample. It is very clear from this micrograph that good bonding exists
between the fibre and matrix for 2 wt. % nanoclay samples. There is no fibre
pullout seen in this case. Delamination failure is observed in Figure 6.9(f)
for 3 wt. % nanoclay samples. Higher particle loading nanoclays start to
agglomerate and these agglomerations may have begun to act as flaws and
crack initiation sites [15]. As the wt. % of nanoclay particles increases,
the free volume allowed for nanoclay particles to move around would be
decreased. Therefore, mechanical stirring and ultrasonic separation tech-
niques cannot be effectively used to separate agglomerations of nanoclays,
as the wt. % of nanoclay particles in the epoxy resin increases, there is less
free volume for each nanoclay particle [33]. It is evident from SEM micro-
graphs that 2 wt. % nanoclay systems promote good interfacial bonding
between the fibre and matrix.

Scanning electron microscopy studies were also carried out for failure sur-
faces. Figure 6.10 illustrates the fracture surfaces of hand lay-up samples
for the control and a different weight percentage of particle loading. It
can be seen from this figure that the control samples exhibit some fibre
pullout failure, while 1 wt. % and 2 wt. % nanoclay showed fibre breakage
failure indicating good bonding between the fibre and matrix. It is also evi-
dent that the control sample splits open more than the nanoclay samples.
For the 3 wt. % nanoclay samples, fibre pullout is seen in Figure 6.10(d)
in the tension side of the flexural specimen, while a characteristic failure
is seen in the compression side of the flexural specimen. A kink band, as
shown in Figure 6.10(e), is observed in the compression side of the 3 wt. %
nanoclay sample, which is a typical failure for brittle fibres like carbon.
6.3.5 Shear strength of cured resin

As evident from the SEM images there is good adhesion between the fibres
and matrix; the short beam test was carried out to evaluate the shear
strength of the cured resin itself. It was intended to investigate if there is
any improvement in apparent shear strength properties upon the addition
(a) (b)
(d)
(c)
Formation of kink
bands
(e)
Fig. 6.10. SEM micrographs of failed hand lay-up samples for: (a) Control sample, (b)
1% nanoclay sample, (c) 2 wt. % nanoclay sample (d) 3 wt. % nanoclay sample subjected
to flexural loading, (e) Micro-buckling leading to the formation of kink zones observed
in the compression side of the 3 wt. % nanoclay sample.
of different nanoclay loading. Figure 6.11 shows the load versus displace-
ment curve of resin (without fibre reinforcement) for a short beam test
and Table 6.13 summarises the apparent shear strength results of different
specimens. It is very clear, as evident from the SEM micrographs, that
up to 2 wt. % nanoclay loading, the apparent shear strength of the resin
1000
800
600
Load, N
Control sample
1% nanoclay
2% nanoclay
3% nanoclay
400
200
0
0 0.0005 0.0010 0.0015 0.0020 0.0025
Displacement, m
Fig. 6.11. Load vs displacement curve of the resin for a short beam test.
Table 6.13. Apparent shear strength of the resin using the short beam
method.
Apparent shear strength, τxz (MPa) % Improvement
Control 11.66 ± 0.49 —

1% Nanoclay 14.25 ± 0.16 22.21
2% Nanoclay 15.18 ± 0.21 30.19
3% Nanoclay 12.78 ± 0.38 9.61
increases. About 30% improvement in apparent shear strength is seen for

2 wt. % nanoclay loading. It is possible that this increase in resin prop-
erty has contributed to an increase in flexural strength of the laminated
composite and also altered the mode of failure.
6.3.6 Dynamic mechanical analysis

Variations of storage modulus with temperature for VARTM samples (con-
trol, 1–3 wt. % nanoclay samples) without thermal post-curing are shown in
Figure 6.12. These results indicate that the storage modulus improves with
addition of nanoclay content up to 2 wt. % and then decreases for 3 wt. %.
Increased proportions of nanoclay improve up to a point beyond which
25
20 20.37 GPa
17.98 GPa
Storage modulus, GPa
16.21 GPa
15 Control sample
13.57 GPa 3% nanoclay
2% nanoclay
1% nanoclay
10
0
0 40 80 120 160
ο
Temperature C
Fig. 6.12. Temperature dependence of the storage modulus for VARTM samples with-
out thermal post-curing.
Table 6.14. DMA results of VARTM samples without thermal post-curing.
Storage Loss Change

modulus, % modulus, % Tg , in Tg ,
(GPa) Change (GPa) Change (◦ C) (◦ C)
Control sample 13.522 ± 0.19 — 2.029 ± 0.039 — 64.15 ± 0.18 —

1% Nanoclay 17.995 ± 0.23 33.08 2.444 ± 0.036 20.45 69.79± 0.09 5.64
2% Nanoclay 20.164 ± 0.27 49.12 2.695 ± 0.043 32.82 72.74± 0.13 8.59
3% Nanoclay 16.233 ± 0.21 20.05 2.135 ± 0.035 5.22 67.43± 0.10 3.28
they form agglomerates which could increase the brittleness to an extent of

lowering the modulus [34]. Table 6.14 shows a maximum improvement in
storage modulus of about 49% for 2 wt. % nanoclay samples over the control
samples at about 35◦ C. In Figure 6.22, a sharp drop in storage modulus
indicates the Tg . This sharp drop in storage modulus divides the entire
temperature range into two segments — the below Tg zone and the above
Tg zone. The below Tg zone is also referred to as the operating region
while the above Tg zone is referred to as the rubbery plateau. A flatter
operating region is seen for nanoclay samples in the below Tg zone when
compared with control samples. The rubbery plateau of nanoclay samples
distinctly shows an increased storage modulus compared with the control

samples which are indications of small degrees of improved crosslinking. In
general, the crosslink density (derived from E in the rubbery plateau) is
higher for nanoclay samples than for control samples. The answer to this is
that it is possible that nanoclay has a catalytic effect as well as the epoxy
resin reacting with the nanoclay. This is further evident from the study of
Uhl et al. [35]. In their study, on organically modified montmorillonites in
UV curable urethane acrylate films, they have shown that films containing
nanoclays have a higher crosslink density, as evident from the higher values
of the rubbery plateau in dynamic mechanical thermal analysis (DMTA)
results, which is also evident in the current DMA study. Based on real-time
infrared spectroscopy (RTIR) and photo-differential scanning calorimetry
(DSC) data, they also conclude that the presence of nanoclay can facili-
tate a crosslinking reaction and therefore the crosslink density is increased.
Additionally, they also conclude that clay may act as a crosslink and that
the physical aggregation of polymer chains onto the surface of a particulate,
results in a rise in the effective degree of crosslinking. Figure 6.13 shows the
variation of loss modulus with temperature for VARTM samples without
thermal post-curing. The loss modulus also increases up to 2 wt. % and
3
2.68 GPa
2.45 GPa
2.12 GPa
2.03 GPa
2
Loss modulus, GPa
3% nanoclay
2% nanoclay
1% nancolay
Control sample
0
20 40 60 80 100 120 140 160
ο
Temperature C
Fig. 6.13. Temperature dependence of loss modulus for VARTM samples without ther-
mal post-curing.
0.4
ο
63.3 C, 0.32
ο
69.4 C, 0.275
0.3 ο
67.5 C, 0.27
ο
69.8 C, 0.267
Tan delta
0.2
3% nanoclay
2% nanoclay
1% nanoclay
Control sample
0.1
0
30 50 70 90 110 130 150
ο
Temperature, C
Fig. 6.14. Variation of loss tan delta with temperature for VARTM samples without
thermal post-curing.
Table 6.14 indicates an increase of about 33% over the control samples. A
shift in peak is also seen here indicating an improvement in Tg .
Variations of tan delta with temperature for VARTM samples with-
out thermal post-curing are shown in Figure 6.14. Table 6.14 indicates
that the control sample shows a peak at an average Tg of about 64◦ C.
For nanoclay samples, these peaks, as seen in Figure 6.14, are observed
at higher temperatures than those for the control samples. The maximum
shift in peak to a higher temperature is seen for 2 wt. % nanoclay samples.
These peaks generally decrease in intensity upon increasing the nanoclay
content. The temperature variation of the storage modulus for thermally
post-cured VARTM samples (control, 1%, 2%, 3% nanoclay samples) is
shown in Figure 6.15. The storage modulus improves with the addition
of nanoclay content up to 2 wt. % and then again decreases for 3 wt. %.
Table 6.15 shows a maximum improvement in storage modulus of about 52%
for 2 wt. % nanoclay samples over the control samples at about 35◦ C. The
effect of post-curing is readily seen when Figure 6.12 and Figure 6.15 are
compared. There is an increase in absolute values of the storage modulus.
The operating region also increased considerably for thermally post-cured
samples. The rubbery plateau of nanoclay samples is also higher in the case
of thermally post-cured samples.
25
22.14 GFPa
20
2% nanoclay
Control sample
15
14.4 GPa
10
0
0 40 80 120 160
ο
Temperature C
Fig. 6.15. Effect of thermal post curing on the variation of storage modulus with tem-
perature of VARTM samples.
Table 6.15. DMA results of VARTM samples with thermal post-curing.
Storage Loss Change

modulus, % modulus, % Tg , in Tg ,
(GPa) Change (GPa) Change (◦ C) (◦ C)
Control sample 14.480 ± 0.13 — 1.585 ± 0.017 — 82.52 ± 0.21 —

1% Nanoclay 18.494 ± 0.17 27.72 1.920 ± 0.019 21.14 85.51 ± 0.18 2.99
2% Nanoclay 22.065 ± 0.19 52.38 2.335 ± 0.020 47.32 95.76 ± 0.19 13.24
3% Nanoclay 17.140 ± 0.16 18.37 1.748 ± 0.016 10.28 86.01 ± 0.17 3.49
Figure 6.16 shows the variation of loss modulus with temperature for
thermally post-cured VARTM samples. The loss modulus also increased
upon increasing the nanoclay content up to 2 wt. %. Table 6.15 indicates
an increase of loss modulus of about 47% over the control samples. A very
prominent shift in the loss modulus peak to a higher temperature is seen
for 2 wt. % nanoclay samples indicating an improvement in Tg . Variations
of tan delta with temperature for thermally post-cured VARIM samples
are shown in Figure 6.17. The control sample has a Tg of about 83◦ C. The
peaks of nanoclay samples shifted toward a higher temperature giving rise
to higher than the control samples. Table 6.15 indicates an improvement in
the Tg of about 13◦ C for 2 wt. % nanoclay samples. These peaks generally
2.5
2.3 GPa
1.92 GPa
2.0
2% nanoclay
Loss modulus, GPa

Control sample
1.5
1.0
0.5
0
20 40 60 80 100 120 140 160
ο
Temperature C
Fig. 6.16. Effect of thermal post-curing on the variation of loss modulus with temper-
ature of VARTM samples.
0.25
ο
82.69 C, 0.22 ο
o
86.01 C. 0.19
85.34 C, 0.21 ο
95.3 C, 0.20
0.20
3% nanoclay
0.15 2% nanoclay
Tan delta
1% nanoclay
Control sample
0.10
0.05
0
40 60 80 100 120 140 160
ο
Temperature C
Fig. 6.17. Effect of thermal post-curing on the variation of tan delta with temperature
of VARTM samples.
decrease in intensity and broaden upon increasing the nanoclay content.

This broadening of glass to rubber transition is often assumed to be due to
a distribution in molecular weight between crosslinks or some other kind of
heterogeneity in the network structure [36].
30
25 24.832 GPa
22.656 GPa
21.779 GPa
20

2% nanoclay
1% nanoclay
Control sample
15
10
0
0 40 80 120 160
ο
Temperature C
Fig. 6.18. Temperature dependence of the storage modulus for room temperature cured
hand lay-up samples.

Variations of the storage modulus with temperature for room temperature
cured hand lay-up samples (control, 1%, 2%, 3% nanoclay samples) are
shown in Figure 6.18. The storage modulus improves with the addition of
nanoclay content up to 2 wt. % and then decreases for 3 wt. % nanoclay
loading. Table 6.16 shows a maximum improvement in storage modulus of
about 24 % for 2 wt. % nanoclay samples over the control samples at about
35◦ C. The operating region of nanoclay samples with up to 2 wt. % loading
remained almost the same as the control sample except for an increase in the
storage modulus in nanoclay samples in the below Tg zone. The rubbery
plateau of nanoclay samples seemed to remain the same as the control
samples. In general, the crosslink density of the nanoclay samples (derived
from E in the rubbery plateau) is the same as that of the control samples.
3 wt. % nanoclay samples showed poor thermal properties compared with
other nanoclay samples by having a smaller operating region and lower
storage modulus in the above Tg region.
room temperature cured hand lay-up samples. The loss modulus increases
Table 6.16. DMA results of room temperature cured hand lay-up samples.
Storage Loss
modulus, % modulus, % Tg ,
(MPa) Change (MPa) Change (◦ C)
Control sample 19683 ± 198 — 2780 ± 66 — 72.21 ± 0.53

1% Nanoclay 22833 ± 139 16.00 3289 ± 42 18.31 73.38 ± 0.40
2% Nanoclay 24457 ± 281 24.25 3640 ± 54 30.94 72.68 ± 0.63
3% Nanoclay 21780 ± 141 10.65 3495 ± 69 25.72 69.19 ± 0.30
4
3.618 GPa
3.480 GPa
3.286 GPa
3 2.809 GPa
Loss modulus, GPa
3% nanoclay
2% nanoclay
1% nanoclay
1 Control sample
0
20 40 60 80 100 120 140 160
ο
Temperature C
Fig. 6.19. Temperature dependence of the loss modulus for room temperature cured
hand lay-up samples.
up to 2 wt. % and Table 6.16 indicates an increase of about 31% over

the control samples. There was no apparent shift in peak for nanoclay
samples indicating hardly any improvement in Tg . The peak loss modulus of
3 wt. % nanoclay samples shifted slightly towards a lower temperature than
the control samples. Variations of tan delta with temperature for samples
without thermal post-curing are shown in Figure 6.20. Table 6.16 indicates
that the control sample shows a peak at an average Tg of about 72◦ C.
For nanoclay samples, these peaks are more or less the same. No changes
in laminate Tg , as determined by peak in tan delta were seen. Therefore,
0.6
ο
68.89 C, 0.490
ο
72.42 C, 0.379 ο
73.28 C, 0.378
0.4
ο
72.73 C, 0.32
Tan delta
3% nanoclay
0.2 2% nanocly
1% nanoclay
Control sample
0
20 40 60 80 100 120 140 160
ο
Temperature C
Fig. 6.20. Variation of tan delta with temperature for room temperature cured hand
lay-up samples.
any changes in mechanical properties were due to the physical presence of

nanoclay as opposed to changes in the polymeric network structure. These
peaks generally increase in intensity and broaden with an increase in the
amount of nanoclay content.
Variation of the storage modulus with temperature for thermally post-
cured hand lay-up samples (control, 1%, 2%, 3% nanoclay samples) is shown
in Figure 6.21. The storage modulus improves upon the addition of nanoclay
content up to 2 wt. % and then again decreases for 3 wt. %. Table 6.17 shows
a maximum improvement in storage modulus of about 36% for 2 wt. %
nanoclay samples over the control samples at about 35◦ C. The effect of post-
curing is readily seen when Figure 6.18 and Figure 6.21 are compared. There
is an increase in absolute values of the storage modulus. The operating
zone increased considerably in the below Tg zone for thermally post-cured
samples. The rubbery plateau of nanoclay samples is somewhat the same
or slightly higher than the control samples.
thermally post-cured hand lay-up samples. The loss modulus also increased
upon increasing the nanoclay content up to 2 wt. % and Table 6.17 indicates
35
30 29.444 GPa
26.526 GPa
25 3% nanoclay

1% nanoclay
21.365 GPa Control sample
20
15
10
0
0 40 80 120 160
ο
Temperature C
Fig. 6.21. Effect of thermal post-curing on the variation of storage modulus with tem-
perature of hand lay-up samples.
Table 6.17. DMA results of thermally post-cured hand lay-up samples.
Storage Loss
modulus, % modulus, % Tg ,
(MPa) Change (MPa) Change (◦ C)
Control sample 21593 ± 149 — 2483 ± 43 — 99.03 ± 0.31

1% Nanoclay 26560 ± 70 23.00 3279 ± 39 32.06 99.26 ± 0.25
2% Nanoclay 29440 ± 71 36.34 3632 ± 37 46.27 99.55 ± 0.21
3% Nanoclay 23400 ± 89 8.37 3080 ± 31 24.04 98.81 ± 0.03
an increase of about 46% over the control samples. There is hardly any shift
in the loss modulus peak to higher temperature indicating no change in Tg .
Variations of tan delta with temperature for thermally post-cured hand
lay-up samples are shown in Figure 6.23. The control sample has a Tg of
about 99◦ C. The peaks of the nanoclay samples did not show any shift. The
only change associated with the nanoclay peaks is broadening of the peaks
with an increase in nanoclay content. Table 6.17 shows the Tg for different
wt. % of nanoclay samples.
4
3.603 GPa
3.256 GPa
3.074 GPa
3
2.514 GPa
Loss modulus, GPa
2
3% nanoclay
2% nanoclay
1% nanoclay
Control sample
0
0 40 80 120 160
ο
Temperatue C
Fig. 6.22. Effect of thermal post curing on the variation of loss modulus with temper-
ature of hand lay-up samples.
0.35
ο
99.78 C, 0.31
ο
98.97 C, 0.30
0.30
ο
98.95 C, 0.27
ο
0.25 99.01 C, 0.24
0.20
Tan delta
3% nanoclay
2% nanoclay
1% nanoclay
0.15 Control sample
0.10
0.05
0
20 40 60 80 100 120 140 160
ο
temperature C
Fig. 6.23. Effect of thermal post-curing on the variation of tan delta with temperature
of hand lay-up samples.
6.4 Conclusions
Nanoclays were used as nanoparticle fillers in woven carbon fibre-reinforced
polymeric composites. Nanoclay infusion at quite low concentrations
increased the flexural properties as well as the thermal stability of the
system. At 3 wt. % nanoclay loading, the properties of samples showed a
declining trend, thus giving rise to an optimum nanoclay loading of 2 wt. %.
Overall, this work showed that nanoclays can be easily used to modify tra-
ditional fibre-reinforced composite materials. Based on the study carried
out so far, the following is a summary of the present investigation.
From the WAXD analysis, it is seen that an intercalated nanoclay
structure is found in laminates. Also, nanoclays were dispersed in a highly
disorderly manner in the up to 2 wt. % samples, which is indicated by a
significant loss of intensity and broadening of the peaks; the amount of
reinforcement of fibres in laminates can be found from this test. The fibre
volume fraction for VARTM and hand lay-up samples is found to be around
56%. The void content in VARTM samples is found to be about 6% while
hand lay-up samples had a void content of about 3%.
Two wt. % of nanoclay seems to be the optimum loading for carbon/
SC-15 epoxy composites in terms of mechanical and thermal properties.
The flexural test results of thermally post-cured VARTM samples indicate
a maximum improvement in strength and modulus of about 14% and 9%
respectively. Test results of thermally post-cured hand lay-up samples indi-
cate a maximum improvement in strength and modulus of about 31% and
21% respectively. Comparisons of VARTM and hand lay-up laminates show
that hand lay-up samples give improved strength and modulus over VARTM
samples. This improvement in strength may be attributed to the lower void
content and better compaction. Microstructural studies of VARTM sam-
ples and resin shear strength results revealed that nanoclay promotes good
interfacial bonding between the fibre and matrix, thereby increasing the
mechanical properties. Good bonding between the fibres and matrix is also
seen in hand lay-up samples with up to 2 wt. % nanoclay. Kink zones are
observed in the compression side of the 3 wt. % nanoclay sample leading to
micro-buckling of brittle carbon fibres.
DMA results of thermally post-cured VARTM samples show a maxi-
mum improvement in the storage modulus by about 52% and an increase
in Tg of about 13◦ C. In the case of hand lay-up samples, it is seen that the
storage modulus increased by a significant amount while the Tg remained
almost unchanged for different weight percentages of nanoclay samples, as
determined by the peak in tan delta. Therefore, the changes in mechanical
properties were due to the physical presence of nanoclay as opposed to

changes in the polymeric network structure.
From the present study, it is seen that higher percentages of nanoclay
loading, in this case 3 wt. %, do not offer promising results in most cases.
The secret of success might be mixing of the nanoclays; however, it becomes
really challenging when mixing at higher weight percentage loading. All
mixing parameters in the ultrasonic cavitation method were held constant
in all types of nanoclay loading. The mixing time, amount of mixing, volume
and amplitude of oscillation can be varied to see the effect of mixing param-
eters when mixing higher weight percentages of nanoclay loading. Elevated
temperature processing will lead to decreased viscosity of the nanophased
polymer, which will enhance the good wetting of fibres and may provide an
avenue for the inclusion of higher wt. % of nanoclay.
Acknowledgements
The authors would like to thank the support received by NSF (grant num-
ber HRD-0317741) and Alabama EPSCoR (grant number EPS-0447675,
01158862) for carrying out this work.
References
[1] LeBaron, P.C., Wang, Z. and Pinnavaia, T.J. (1999). Polymer-layered silicate
nanocomposites: an overview, Applied Clay Science, 15, 11–29.
[2] Alexandre, M. and Dubois, P. (2000). Polymer-layered silicate nanocompos-
ites: preparation, properties and uses of a new class of materials, Materials
Science and Engineering, 28, 1–63.
[3] Fukushima, Y. and Inagaki, S. (1987). Synthesis of an intercalated compound
of montmorillonite and 6-polyamide, Journal of Inclusion Phenomena, 5,
473–482.
[4] Usuki, A., Kojima, Y., Kawasumi, M., Okada, A., Fukushima, Y. and
Kurauchi, T. (1993). Synthesis of nylon 6-clay hybrid, J. Mater. Res., 8,
1179–1184.
[5] Wang, M.S. and Pinnavaia, T.J. (1994). Clay-Polymer nanocomposites
formed from acidic derivatives of montmorillonite and an epoxy resin, Chem
Mater., 6, 468–474.
[6] Lan, T. and Pinnavaia, T.J. (1994). Clay-reinforced epoxy nanocomposites,
Chem. Mater., 6, 2216–2219.
[7] Wang, Z., Massam, J. and Pinnavaia, T.J. (2001). In: Pinnavaia, T.J., Beall,
G.W. (eds.) Polymer-Clay Nanocomposites. New York: Wiley.
[8] Park, J.H. and Jana, S.C. (2003). Mechanism of exfoliation of nanoclay par-
ticles in epoxy-clay nanocomposites, Macromolecules, 36, 2758–2768.
[9] Schmidt, H. (1985). New type of non-crystalline solids between inorganic and
organic materials, Journal of Non-Crystal Solids, 73, 681–691.
[10] Novac, B.M. (1993). Hybrid nanocomposite materials — between inorganic
glasses and organic polymers, Advanced Materials, 5, 422–433.
[11] Mark, J.E. (1996). Ceramic-reinforced polymers and polymer-modified
ceramics, Polymer Engineering Science, 36, 2905–2920.
[12] Hussain, M., Nakahira, A. and Niihara, K. (1996). Effects of wet ball-milling
on particle dispersion and mechanical properties of particulate epoxy com-
posites, Materials Letters, 26, 85–191.
[13] Gilbert, E.N., Hayes, B.S. and Seferis, J.C. (2002). Variable density com-
posite systems Constructed by metal particle modified prepregs, Journal of
Composite Materials, 36, 2045–2060.
[14] Gilbert, E.N., Hayes, B.S. and Seferis, J.C. (2002). Metal particle modifi-
cation of composite matrices for customized density applications, Polymer
Composites, 23, 132–140.
[15] Timmerman, J.F., Hayes, B.S. and Seferis, J.C. (2002). Nanoclay reinforce-
ment effects on the cryogenic microcracking of carbon fiber/epoxy compos-
ites, Composites Science and Technology, 62, 1249–1258.
[16] Karaki, T., Killgore, J.P. and Seferis, J.C. (2004). Characterization of fatigue
behavior of polynanomeric matrix composites, Proceedings of 49th SAMPE
Symposium and Exposition, Long Beach, CA, Paper 263.
[17] Hackman, I. and Hollaway, L. (2006). Epoxy-layered silicate nanocomposites
in civil engineering, Composites Part A, 37, 1161–1170.
[18] Miyagawa, H., Jurek, R.J., Mohanty, A.K., Misra, M. and Drzal, L.T. (2006).
Biobased epoxy/clay nanocomposites as a new matrix for CFRP, Composites
[19] Haque, A., Shamsuzzoha, M., Hussain, F. and Dean, D. (2003). S2-glass/
epoxy polymer nanocomposites: Manufacturing, structures, thermal and
mechanical properties, Journal of Composite Materials, 37, 1821–1837.
[20] Mahfuz, H., Adnan, A., Rangari, V.K., Jeelani, S. and Jang, B.Z. (2004).
Carbon nanoparticles/whiskers reinforced composites and their tensile
response, Composites Part A: Applied Science and Engineering, 35, 519–527.
[21] Chisholm, N., Mahfuz, H., Rangari, V.K., Adnan, A. and Jeelani, S. (2005).
Fabrication and mechanical characterization of carbon/SiC-epoxy nanocom-
posites, Composite Structures, 67, 115–124.
[22] Daniel, I.M. and Ishai, O. (2006). Engineering Mechanics of Composite Mate-
rials, 2 nd Ed. New York: Oxford University Press.
[23] Applied Poleramic Inc., Material Data Sheet: Sc-15, USA.
[24] Nanocor Inc., Material Data Sheet: Nanomer I.28E, USA.
[25] Eskin, G.I. (2001). Broad prospects for commercial application of the ultra-
sonic (cavitation) melt treatment of light alloys, Ultrasonic Sonochemistry,
8, 319–325.
[26] Eskin, G.I. (1995). Cavitation mechanism of ultrasonic melt degassing, Ultra-
sonics Sonochemistry, 2, S137–141.
[27] Eskin, G.I. (1994). Influence of cavitation treatment of melts on the pro-
cesses of nucleation and growth of crystals during solidification of ingots and
castings from light alloys, Ultrasonics Sonochemistry, 1, S59–63.
[28] Annual Book of ASTM Standards, D 3171-99 (2004). Standard Test Methods
for Constituent Content of Composite Materials.
[29] Annual Book of ASTM Standards, D 790-02 (2002). Standard Test Methods
for Flexural Properties of Unreinforced and reinforced Plastics and Electrical
Insulating Materials.
[30] Annual Book of ASTM Standards, D 4065-01. (2002). Standard Practice for
Determining and Reporting Dynamic Mechanical Properties of Plastics.
[31] Wu, C.L., Zhang, M.Q., Rong, M.Z. and Friedrich, K. (2002). Tensile per-
formance improvement of low nanoparticles filled-polypropylene composites,
Comp. Sci. Tech., 62, 1327–1340.
[32] Vaia, R.A, Jandt, K.D., Kramer, E.J. and Giannelis, E.P. (1996). Microstruc-
tural evolution of melt intercalated polymer-organically modified layered sil-
icates nanocomposites, Chem. Mater., 8, 2628–2635.
[33] Lam, C.K., Cheung, H.Y., Lau, K.T., Zhou, L.M, Ho, M.W. and Hui, D.
(2005). Cluster size effect in hardness of nanoclay/epoxy composites, Com-
posites: Part B, 36, 263–269.
[34] Ahmad, Z., Sarwar, M.I. and Mark, J.E. (1997). Dynamic-mechanical ther-
mal analysis of aramid-silica hybrid composites prepared in a sol-gel process,
J. Appl. Poly. Sci., 63, 1345–1352.
[35] Uhl, F.M., Davuluri, S.P., Wong, S.C. and Webster, D.C. (2004). Organically
modified montmorillonites in UV curable urethane acrylate films, Polymer,
45, 6175–6187.
[36] Neilsen, L.E. and Landel, R.F. (2002). Mechanical Properties of Polymers
and Composites, 2nd Ed. New York: M. Dekker.
Chapter 7
TEXTILE COMPOSITE FORMING SIMULATIONS
Philippe Boisse
Université de Lyon, LaMCoS, INSA-Lyon, F-69621, France
7.1 Introduction
A textile composite material is obtained by the association of a continuous
textile reinforcement and a matrix. The role of the matrix is to avoid the
possible relative sliding of the fibres and the resulting deformation of the
reinforcement. During forming processes, the matrix does not play its role
because it is absent or soft. Consequently, deformation of the composite is
possible in order to form complex shapes often with double curvatures. This
relative sliding of the fibres is specific to composite forming. The deforma-
tion modes and consequently the forming processes are quite different from
metal forming.
In the case of liquid composite moulding (LCM) processes [1, 2], a
textile reinforcement is first formed in order to get the so-called preform.
The resin is injected within this preform and polymerised at high temper-
ature to obtain the final textile composite (Figure 7.1). In these processes,
the reinforcement is formed before the resin is injected. Deformation of the
textile reinforcement is possible, in particular, due to the in-plane shear
strains and bending deformation. Double curved shapes can be obtained.
In the case of thermoset prepreg, the matrix is present within the rein-
forcement but it is not hardened; it is soft because it has not yet been
polymerised. This renders the forming possible with the same deforma-
tion modes because the low stiffness of the matrix (not hardened) enables
in-plane shear and bending strains of the fabric (Figure 7.2). Thermoset
prepreg draping is frequently used during hand or automatic draping in
the aeronautic industry where the rates of production are low [3, 4]. The
215
216 P. Boisse
Preforming of the texƟle Resin injecƟon Final composite part

reinforcement
Fig. 7.1. Resin transfer moulding (RTM) process.
Fig. 7.2. Thermoset prepreg draping.
quality of the manufactured composite, in particular the lack of porosity

between the layers, can be insured by applying pressure in an autoclave.
In the case of thermoplastic prepregs (CFRTP, continuous fibres rein-
forced thermoplastic), the possibility of relative motion is obtained by
preparatory heating of the prepreg at a temperature higher or close to the
fusion of the matrix. Forming is then possible by bending and in-plane shear
deformation of the reinforcement. After a consolidation stage (transverse
compression) which aims to avoid porosity between the plies, the composite
recovers its stiffness after cooling. These stages as a whole (Figure 7.3) are
fast (one or several minutes). This renders the process interesting to the
automotive industry, in particular, for the production of large numbers of
composite parts [5].
Textile Composite Forming Simulations 217
1 2
3 4
Fig. 7.3. Different stages of thermoforming. 1: Heating of the CFRTP, 2: Forming with
punch and die, 3: Reconsolidation phase, 4: Final part.
In the three cases above, the lack of efficiency of the matrix (it is
soft or absent) allows relative sliding between the fibres. This is specific
to composite materials. In addition, the fibre extension is usually small
during the forming stage, while the extension of a metal blank can be large
in sheet metal forming processes. Finally, the forming is mainly guided by
the textile reinforcement, which is the case in LCM processes, but also
in thermoset and thermoplastic forming, where the matrix modifies the
rigidity of the composite during forming but doesn’t change the deformation
modes of the textile reinforcement. In the case of the LCM process, the resin
injection onto the preform is an important and difficult stage which has been
intensively studied [1, 2, 6]; it is not studied in the present chapter.
In this chapter, after a description of the specific deformation modes
of textile reinforcements, four different approaches will be presented. The
kinematic models are the fastest methods but they are based on strong
assumptions. A hypoelastic model is described within a continuous finite
element simulation of the forming process. Semi-discrete elements introduce
the unit woven cell and its mechanical behaviour in shell elements. Finally,
a mesoscopic approach is presented where each yarn is modelled as the
contact and friction between the yarns. In this way the tow sliding can be
modelled.
218 P. Boisse
7.2 Deformation Modes of a Textile Composite

Reinforcement During Forming
7.2.1 Single curved and double curved shapes. Consequence
on the reinforcement
It is necessary to distinguish the forming on single curved and double curved
shapes. In the first case, the forming does not require membrane strain, i.e.,
strains in the mid-plane of the reinforcement. Only the bending strains are
necessary. When forming on a double curved shape, membrane strains are
necessary as the extension of the fibres is very small, and in-plane shear
strains are the main deformation mode which enable obtaining the double
curved shape. Consequently, the unidirectional reinforcements (UD) made
of parallel fibres without transverse links are not appropriate. Their forming
on a double curved surface would lead to transverse separation. They can be
used for simply curved parts such as in Figure 7.4. This beam is made by a
stack of 10UD plies [7]. The advantage of the UD reinforcement mainly lies
in the large stiffness due to the straightness of the fibres. To manufacture
double curved preforms or prepregs, it is necessary to ensure cohesion of
the ply and consequently, woven reinforcements or non-crimp fabric (NCF)
should be used. The weaving provides good cohesion of the fabric and a
large shear angle can be reached (Figure 7.5). To some extent, the NCF
stitch brings such cohesion and double curved preforms made of NCF can
be manufactured (Figure 7.5).
7.2.2 No sliding between warp and weft yarns

Figure 7.6 shows the hemispherical forming of a woven reinforcement on
which straight lines have been drawn in the warp and weft directions prior
Fig. 7.4. Single curved and double curved shapes.

Fig. 7.5. Large shear angle obtained with a woven reinforcement [8]. Double curved
hemispherical preform obtained with NCF reinforcement [9].
Fig. 7.6. Hemispherical forming.
to forming. These lines become curved after forming but remain continuous.
This implies that there is no sliding between warp and weft yarns. This is
important for modelling because this absence of sliding between the fibres
permits consideration of textile reinforcement as a continuous media and to
use continuum mechanics and finite elements. On the contrary, the forming
shown in Figure 7.7 shows strong loss of cohesion of the textile reinforcement
and large sliding between the yarns. This is due to excessive blank holder
forces. In this case, the reinforcement cannot be considered as a continuum.
It can be modelled and the forming can be simulated, but each yarn must
be described as well as the contact and sliding between the yarns [10] (see
Section 7.6). Nevertheless, the last phenomenon is a forming failure and in
most cases the yarns do not slide and the textile reinforcement, or more
precisely a ply of the reinforcement, can be considered as a continuum.
220 P. Boisse
Fig. 7.7. Hemispherical forming with sliding between warp and weft yarns [10].
7.2.3 Yarn extension

The fibres, and consequently the yarn they constitute, are elastic media
in tension. If the reinforcement is loaded in tension in the yarn direction,
it is stretched in this direction (Figure 7.8). This tensile strain is due to
both fibre extension and to decrimping in the case of woven fabrics [11, 12].
These strains are small (the maximum tensile strain of a carbon fabric
can be about 1.5%). Nevertheless, the corresponding strain energy can be
large because the yarns are stiff in tension. During compression in the yarn
AŌer in-plane shear

AŌer extension
IniƟal
IniƟal
Fig. 7.8. Extension and in-plane shear of a textile reinforcement.

direction, the yarns and the fabric can easily buckle because of the very
small diameter of the fibres compared with their length.
7.2.4 In-plane shear

This deformation is due to the relative rotation of the warp and weft yarns
of woven fabric (or NCF). Large in-plane shear strains are possible in the
reinforcement which behaves as a four-bar system (Figure 7.8). This defor-
mation mode enables the formation of double curved shapes and is essential
in composite forming [13]. There is a limit to the shear angle which is
often called the locking angle [14]. In practice, this limit is progressive and
depends on the loads on the reinforcement [15].
7.2.5 Transverse compression

It is often necessary to compact the reinforcement during manufacturing
processes. This is the case in LCM processes, before resin injection, and
in thermoforming in order to consolidate the laminate after forming. The
textile reinforcement is first easily compacted at low forces, then its rigidity
increases when the voids between the fibres are filled [16, 17].
7.2.6 Bending deformation

The bending stiffness of a textile reinforcement (without resin or with fluid
resin) is weak or very weak. The section of a fibre is very small (usually
some microns) and relative sliding between fibres is possible. Consequently,
the resultant bending stiffness of the textile reinforcement is very small [18].
It is sometimes neglected and the reinforcement can be modelled by a mem-
brane. Nevertheless, some thick fabrics have a larger bending rigidity and
they must be modelled by shell elements taking the bending stiffness into
account. Above all, the bending rigidity is of main importance for any type
of reinforcement, in order to model the wrinkles that can appear during a
forming process. The bending stiffness mainly determines the size of the
wrinkles [15].
7.2.7 Goal of the simulation of deformation during

the forming process
Simulation of reinforcement or prepreg forming aims to obtain information
on the process and to optimise this forming process without time-consuming
trial-and-error techniques. The first point concerns the conditions for the
222 P. Boisse
feasibility of the process. Is the shape of the preform possible using the
given textile reinforcement? What are the conditions (load on the tools, on
the blank holder etc.) necessary for a correct preform or prepreg stack? This
often requires modelling of the defects that can occur during forming. The
most common defects are wrinkles but porosities during thermoforming,
fibre fractures and waviness can also be important defects. The simulation
will give the conditions to avoid these defects in the part. A second point
that is specific to composite forming is the determination of the position and
orientation of the fibres in the preform, and consequently in the composite
part. In the case of the LCM process, the permeability of the preform
depends on the shear angles [19, 20]. Consequently, it is necessary to take
into account the computed shear angles in the injection simulation. Finally,
the mechanical properties of the final composite part (obtained after resin
hardening) depend on the fibre orientations. These orientations are the
main result of the forming simulation and must be taken into account in
the constitutive model of the composite.
7.3 Kinematic Models

The different forming simulation approaches are more or less complex
depending on the phenomena they take into account. The simplest
approaches for textile reinforcement forming on a double curved surface
(often called draping) are the kinematic models (also called fishnet algo-
rithms) [21–23]. These approaches are based on strong assumptions but
they are very fast. The yarns are assumed to be inextensible; the warp and
weft yarns can rotate freely at their intersection but they cannot slide; the
fabric comes progressively into contact with the tool and when a point is
in contact it does not slide relative to the tool.
The position of a point C is determined from those of the two neigh-
bours A and B. The lengths of AC and BC are known. C is the intersection
of the two geodesics of the surface of the tool with this length. This is a small
scalar problem which can be solved very quickly. The tool surface is defined
by plane elements or curved panels. To initiate the scheme of Figure 7.9 it is
necessary to fix a first node and to give two initial draping directions (warp
and weft); there is a symmetry axis when they exist. The result of the fishnet
approach depends on these two directions. These kinematical models are
very fast and widely used at design stages of composite parts. Nevertheless,
the strong assumptions listed above have drawbacks. For a given shape of
tool, the result is the same for any textile material. In addition, the loads
Fig. 7.9. Fishnet algorithm.
on the reinforcement, for instance due to blank holders, are not taken into
account. As the material and loads on the fabric are important points, the
use of the kinematic model is limited to the first steps of the design.
7.4 Continuous Approaches

In order to take into account the material behaviour and exterior loads
on the reinforcement, the main alternatives to the kinematic models are
defined within the finite element method (FEM). This FEM is currently
largely used for polymer and metal forming processes. The difficulty in the
case of composite forming lies in the very specific mechanical behaviour of
the textile reinforcement. It is made of fibres with a small diameter which
can slip. As most textile reinforcements have a small thickness, they are
usually modelled by shell or membrane FE. Nevertheless, some thick 3D
reinforcements need 3D finite elements [24].
In these approaches the textile reinforcement is assumed to be contin-
uous during forming. A specific anisotropic constitutive model takes into
account its particular mechanical behaviour. This model can be used with
standard shell, membrane or 3D finite elements. Several models have been
proposed, based on fabric-reinforced viscous fluids [25, 26], non-orthogonal
equations [27, 28], updated Lagrangian method [29], hypoelastic [30, 31]
or hyperelastic approaches [32, 33]. Nevertheless, none of these models can
describe all aspects of the mechanical behaviour of textile reinforcement
224 P. Boisse
during forming and there is currently no widely accepted model. In the next
section a hypoelastic approach is presented and used for the simulation of
the “double dome” forming benchmark [34–36].
7.4.1 Hypoelastic model

Finite element analyses at large strains are often based on hypoelastic laws
(also called rate constitutive equations) [37, 38]. A constitutive tensor C
relates a stress rate σ ∇ to the strain rate D:
σ ∇ = C:D (7.1)
The stress derivative must be performed in a frame fixed with respect to

the material in order to be objective. There are several objective derivatives.
In the case of textile reinforcement, the rotational objective derivative must
be based on the rotation of the fibre Q [39]:

d T •
σ∇ = Q · Q ·σ·Q · QT = σ + σ · Ω − Ω · σ (7.2)
dt
Ω is the spin corresponding to Q, i.e.,
Ω = Q̇ · QT (7.3)
The unit vectors (e1 , e2 ) in the current configuration are obtained from
the initial orientation axes, (e10 , e02 ) using the proper rotation R:
e1 = R · e01 e2 = R · e02 (7.4)
The current fibre directions f 1 and f 2 are determined from the trans-
formation gradient F. Assuming that the initial position of the fibres are
f 01 = e01 , f 02 = e02 :
F · e01 F · e02
f1 = f2 = (7.5)
F · e01 F · e02
In a step from time tn to time tn+1 , the FE code computes the strain
increment dεε. The components of the strain increment in the two frames g
and h are considered (α and β are indexes equal to 1 or 2):
dεε = dεgαβ gα ⊗ gβ = dεhαβ hα ⊗ hβ (7.6)

The two orthonormal frames g(g1 , g2 ) and h(h1 , h2 ) are based on the
two fibre directions with g1 = f 1 , and h2 = f 2 . In these frames, the fibre
stretching strain and the shear strain are calculated:
dεg11 = g1 · dεε · g1 dεg12 = g1 · dεε · g2 (7.7)

dεh22 = h2 · dεε · h2 dεh12 = h1 · dεε · h2 (7.8)
Eg and E h denote the stiffness in the warp and weft fibre directions,
and G is the in-plane shear stiffness of the textile reinforcement. From these
strain components the axial stress component and shear stress components
are calculated in each frame g and h:
dσg11 = Eg dεg11 dσg12 = Gdεg12 (7.9)

dσh22 =E h
dεh22 dσh12 = Gdεh12 (7.10)
The stresses are integrated on the time increment from time tn to time
n+1
t [37, 39]:
n+1/2 n+1/2
(σg11 )n+1 = (σg11 )n + dσg11 (σg12 )n+1 = (σg12 )n + dσg12 (7.11)
n+1/2 n+1/2
(σh11 )n+1 = (σh11 )n + dσh11 (σh12 )n+1 = (σh12 )n + dσh12 (7.12)
The material parameters are not necessarily constant, especially G

which depends strongly on the in-plane shear [3, 13, 14].
The Cauchy stress at time tn+1 in the textile reinforcement is the addi-
tion of the stresses in the two fibre frames:
σ n+1 = (σσ g )n+1 + (σσ h )n+1 (7.13)
In the next section, this approach is used to simulate the double dome
benchmark [34].
7.4.2 Double dome benchmark

The benchmark textile reinforcement is a co-mingled glass/polypropylene
yarn with balanced plain weave. This fabric has been tested by several labs
in [11]. Experimental forming tests have been performed in order to vali-
date the simulations (Figure 7.11). The blanks are prepared with the yarn
orientations of 0◦ /90◦ and ±45◦ with respect to the sides of the rectangular
blank.
A constant binder force of 100 N was applied. A blank holder (binder)
is used which applies a force of 100 N during the forming of all experimental
tests. Figure 7.12 shows the double dome model set-up which consists of
226 P. Boisse
Fig. 7.11. Experimental and computed deformed shapes for 0◦ /90◦ orientation. Shear
angle measurement.
Fig. 7.12. Double dome model set-up.
die, punch and binder used as forming tools. The material characteristics
and process parameters used for numerical forming tests are as below:
• Warp and weft elastic tensile modulus, E1 = E2 = 35400 MPa

• In-plane shear rigidity G12 , as a function of shear angle γ (radians)
G12 (γ) = 8.48γ 4 − 12.0972γ 3 + 6.1275γ 2 − 0.83γ + 0.051 (7.14)

• Blank holder force = 100 N

• Coefficient of friction = 0.25 (fabric to tools).
The material draw-in is defined as the amount of material flow in the

draping process along the contour from the undeformed position to the
deformed one. It has been measured at the locations shown in Figure 7.11.
This material draw-in and the in-plane shear angles obtained in numerical
and experimental tests are compared in Figures 7.13 and 7.14 in the case of
a 0◦ /90◦ initial orientation of the reinforcement. The location of the shear
angle measurement points are shown in Figure 7.11 (points 1 to 10). The
results are in good agreement. Additional results such as ±45◦ orientation,
multiply forming and different friction coefficients can be found in [34].
This forming case has been defined as an international benchmark of woven
composites [40].
7.5 Semi-discrete Approach

The goal of this approach is to introduce the woven structure of the textile
reinforcement while performing computations at the macroscopic scale. The
textile composite reinforcement is a set of unit woven cells. Finite elements
made of unit cells are considered (Figure 7.15). One unit cell is submitted
to the loads of its neighbouring yarns (Figure 7.16). Tensions T1 and T2
Numerical Experimental
30
25
Material draw-in (mm)
20
15
10
0
A B C D Ex Ey F G H L
Material draw-in points
Fig. 7.13. Comparison of numerical and experimental material draw-in for 0◦ /90◦
orientations of fabric yarns.
228 P. Boisse
Numerical Experimental (OSM)

45
40
35
Shear angle (deg.)
30
25
20
15
10
5
0
1 2 3 4 5 6 7 8 9 10
Shear angle locations
Fig. 7.14. Comparison of numerical and experimental shear angles for 0◦ /90◦ orienta-
tions of fabric yarns.
6 3
5
d6
β3 d5
n2 t1
n1
t2
β1 β2
1 2
t3
n3
d4
Fig. 7.15. Triangular finite element made of a set of woven cells.
are resultant of the loads on the warp and weft yarns. The in-plane shear
moment Ms is the moment due to these loads in the normal direction to the
fabric at the centre of the woven cell. M1 and M2 are the bending moments
on the warp and weft yarns.
The principle of virtual work relates the internal, exterior and acceler-
ation virtual work. In any virtual displacement field η such as η = 0 on the
Fig. 7.16. (a) Load resultants on a woven cell, (b) Tensions, (c) In-plane shear moment,
(d) Bending moments.
boundary with prescribed displacement:
Wext (η) − Wint (η) = Wacc (η) (7.15)
where
t s b
Wint (η) = Wint (η) + Wint (η) + Wint (η) (7.16)
t s b
Wint (η), Wint (η), Wint (η) are the virtual internal work of tension, in-plane
shear and bending with:
ncell

Wtint (η) = p
ε11 (η)p T1 p L1 + p ε22 (η)p T2 p L2 (7.17)
p=1
ncell

Wsint (η) = p
γ(η)p Ms (7.18)
p=1
ncell

Wbint (η) = p
χ11 (η)p M1 p L1 + p χ22 (η)p M2 p L2 (7.19)
p=1
Any quantity X is denoted p X when it concerns the unit woven cell

number p. The number of woven cells is denoted by ncell. L1 and L2 are
the lengths of the unit woven cell in the warp and weft directions. ε11 (η)
and ε22 (η) are the virtual axial strains in the warp and weft directions.
γ(η) is the virtual in-plane shear angle. χ11 (η) and χ22 (η) are the virtual
curvatures of warp and weft directions.
Assumptions are made with regards to the resultant load. Tensions T1
and T2 are assumed to only depend on the warp and weft axial strains.
Ms is assumed to be a function of the shear angle γ. The bending moments
230 P. Boisse
M1 and M2 are a function of the curvatures. These assumptions are correct

in most cases and they correspond to the result obtained from the standard
mechanical test for textile reinforcements. Nevertheless, they can be ques-
tionable in some cases. In particular, it has been shown that tension in the
yarns modifies the in-plane shear stiffness [41–43], but the measurements
and the modelling of this phenomenon are lacking for most reinforcements.
Finite elements can be developed from the above form of internal
virtual work of each woven unit cell. For instance, the three node triangle
shown in Figure 7.15 is seen as a set of unit cells [44]. The interpolated
displacement field within the finite element relates the generalised strains
ε11 , ε22 , γ, χ11 , χ22 to the displacement of the nodes. The triangle presented
in Figure 7.15 uses a rotation free approach [45, 46]. The curvatures χ11 , χ22 ,
are calculated from the displacement of the nodes of the element and those
of the neighbouring triangles. This avoids rotational degrees of freedom and
is consequently good from the numerical efficiency point of view.
A forming simulation example using this semi-discrete finite element
is shown in Figure 7.17. The experimental hemispherical forming has been
performed at the University of Nottingham. A blank holder is used in order
to avoid wrinkling in the hemispherical zone. The textile reinforcement is
very unbalanced. The ratio of warp and weft rigidities is 250. Consequently,
the weft direction is strongly stretched while the warp direction is not. An
initial square becomes a rectangle with a 1.8 ratio of the warp and weft
sides. This ratio is obtained both in experiments and simulation. There
are wrinkles outside of the blank holder and their shapes are correctly
Fig. 7.17. Hemispherical forming of a very unbalanced textile reinforcement.

described by the simulation. This is possible because the bending stiffness

is taken into account in the approach. If this forming is analysed using a
fishnet algorithm, the deformation will be the same in the warp and weft
directions, and there will be no wrinkles. More simulations of wrinkle onset
and development are detailed in [15].
7.6 Mesoscopic Approach

In this approach, the textile composite reinforcement is not considered as a
continuous material. It is a set of yarns in contact with their neighbours with
interlacing due to the weaving [8, 10, 47, 48]. Each fibre bundle is modelled.
This modelling must be simple enough in order to render the reinforcement
forming simulation possible, while in general there are a large number of
yarns and contact between yarns in the preform. The contact with friction
and possible slipping between yarns are explicitly taken into account. This
mesoscopic modelling allows for possible relative displacement between the
yarns.
Some approaches consider each fibre that is modelled using a beam
[49, 50]. Nevertheless, the number of fibres in a yarn is large and there are
many yarns in a preform. Consequently, this approach is not possible for
the textile reinforcement forming simulation.
7.6.1 FE model of the woven unit cell

Each fibre bundle is modelled, as well as its contact and friction, with the
neighbouring yarns. As there are a large number of yarns in the preform, the
mesh must be simple in order to allow the forming simulation of the rein-
forcement. A mesh with 416 degrees of freedom is proposed in Figure 7.18
for the unit woven cell in the case of a plain weave. It is based on the
shell finite element. It can be compared with a finer 3D mesh of the woven
cell which has been used for analyses of the unit cell mechanical behaviour
at the mesoscopic scale [39]. This mesh, with 47214 degrees of freedom,
cannot be used for a forming simulation because there are a large number
of unit woven cells in a preform. It is shown that despite its simplicity,
the coarse mesh of Figure 7.18 can correctly model a mesoscopic forming
simulation.
Each yarn is a set of fibres and its mechanical behaviour is very specific.
It is modelled by a hypoelastic model such as described in Section 7.4.1.
Some geometrical parameters of the woven cell (Figure 7.18) are identified
in order to obtain the correct mechanical behaviour. In particular, the width
232 P. Boisse
Fig. 7.18. Mesoscopic models of a plain weave cell. Fine and coarse meshes.
of yarn plays an important role in the in-plane shear response of the textile
reinforcement [10]. The thickness of the shell must also be such that both
bending and membrane behaviour are correct, although the correspondence
is different than that for continuous materials.
7.6.2 Bias extension test

The bias extension test is a procedure used to determine the in-plane shear
behaviour of textile reinforcement. This property is of importance since
in-plane shear is the main deformation mode during forming on a double
curved surface [13, 51, 52].
A specimen with yarns initially oriented at ±45◦ is stretched in a tensile
machine. Three zones (a, b and c in Figure 7.19c) develop during the defor-
mation. Assuming no extension of the yarns and no sliding at the crossovers,
the extension of the specimen can be related to the shear angle in the
Fig. 7.19. Bias extension test: (a) Experiments, (b) Mesoscopic simulation, (c) Theo-
retical deformation, (d) Shear angle versus displacement.
central part. Measuring the load using a tensile machine gives a curve of
the in-plane shear stiffness as a function of the shear angle. Nevertheless,
it is known that for angles larger than 40◦ the test is not reliable because
some sliding occurs between the yarns. A mesoscopic simulation of the bias
extension test is performed and compared with experiments and the theo-
retical solution (Figure 7.19). For angles smaller than 40◦ the theoretical,
experimental and computed curves are close. Over 40◦ the experimental
shear angle is smaller than predicted by the theory. This is due to motion
between the yarns. The mesoscopic simulation which can describe this slid-
ing is in good agreement with experiments. It can be seen that bending of
the shell elements plays an important role. The simulation based on the
membrane element gives poor results because the friction loads are too
small.
7.6.3 Forming of a hemisphere. Loss of cohesion

of the fabric
Figure 7.20 presents a hemispherical forming with high blank holder loads.
These loads lead to large sliding between the warp and weft yarns, i.e., a
loss of cohesion of the woven reinforcement. This forming process has been
simulated using the mesoscopic approach based on the model of the woven
unit cell shown in Figure 7.18 [10]. The large sliding is correctly obtained
using this approach. These results cannot be achieved using a continuum
approach. This type of defect is a real problem for some reinforcements, in
particular, those made of natural fibres [53].
Fig. 7.20. Hemispherical forming simulation with loss of cohesion of the woven
reinforcement.
234 P. Boisse
7.7 Conclusion
Composite reinforcement forming simulations help to determine the feasi-
bility of a preforming process. It can detect defects such as wrinkles, fibre
fractures or tow sliding. In addition, it gives the direction of the fibres
in the composite which is important for the mechanical properties of the
composite part and for resin injection.
The kinematic approaches are fast and efficient however, their strong
assumptions can lead to using methods based on the finite element method.
In most of these approaches, the textile reinforcement is considered as a
continuous material. Consequently, standard shell or 3D finite elements can
be used. The difficulty lies in the definition of a continuous model which
describes the very specific mechanical behaviour of the fibrous reinforce-
ment. Several models have been proposed. It will be necessary in the future
to distinguish a model among them which would be simple enough and
would efficiently take into account the main specificities of the mechanical
behaviour.
The continuity of the textile reinforcement is usually verified during
the forming process. Nevertheless, sliding of tows can happen under some
conditions. In these cases, mesoscopic modelling is necessary in order to
describe the relative motion of the tows.
References
[1] Advani, S.G. (1994). Flow and Rheology in Polymeric Composites Manufac-
turing. Amsterdam: Elsevier.
[2] Rudd, C.D. and Long, A.C. (1997). Liquid Molding Technologies. Cambridge
UK: Woodhead Publishing.
[3] Wang, J., Page, J.R. and Paton, R. (1998). Experimental investigation of the
draping properties of reinforcement fabrics, Composites Science and Technol-
ogy, 58, 229–237.
[4] Yong, J.C. (2011). Composites: application and assessment of Market,
ICCM18 Conference, Jeju, South Korea.
[5] Fuchs, E.R.H., Field, F.R., Roth, R. and Kirchain, R.E. (2008). Strategic
materials selection in the automobile body: Economic opportunities for poly-
mer composite design, Composites Science and Technology, 68, 1989–2002.
[6] Park, C.H., Lebel, A., Saouab, A., Bréard, J. and Lee, W. (2011). Modeling
and simulation of voids and saturation in liquid composite molding processes,
Composites Part A, 42, 658–668.
[7] Cheruet, A., Soulat, D., Boisse, P., Soccard, E. and Maison-Le Poec, S.
(2002). Analysis of the interply porosities in thermoplastic composites form-
ing processes, International Journal of Forming Processes, 5, 247–258.
[8] Allaoui, S., Boisse, P., Chatel, S., Hamila, N., Hivet, G., Soulat, D. and
Vidal-Salle, E. (2011). Experimental and numerical analyses of textile rein-
forcement forming of a tetrahedral shape, Composites Part A, 42, 612–622.
[9] Bel, S., Hamila, N., Boisse, P. and Dumont, F. (2012). Finite element model
for NCF composite reinforcement preforming: Importance of inter-ply sliding,
Composites Part A, 43, 2269–2277.
[10] Gatouillat, S., Bareggi, A., Vidal-Sallé, E. and Boisse, P. (2013). Meso mod-
elling for composite preform shaping — Simulation of the loss of cohesion of
the woven fibre network, Composites Part A, 54, 135–144.
[11] Kawabata, S., Niwa, M. and Kawai, H. (1973). The finite deformation theory
of plain weave fabrics. Part I: the biaxial deformation theory, J. Text. Inst.,
64, 21–46.
[12] Buet-Gautier, K. and Boisse, P. (2001). Experimental analysis and modeling
of biaxial mechanical behavior of woven composite reinforcements, ExpMech.,
41, 260–269.
[13] Cao, J., Akkerman, R., Boisse, P., Chen, J., Cheng, H.S., de Graaf, E.F.,
Gorczyca, J.L., Harrison, P., Hivet, G., Launay, J., Lee, W., Liu, L., Lomov,
S.V., Long, A.C., de Luycker, E., Morestin, F., Padvoiskis, J., Peng, X.Q.,
Sherwood, J., Stoilova, T.Z., Tao, X.M., Verpoest, I., Willems, A., Wiggers,
J., Yu, T.X. and Zhu, B. (2008). Characterization of mechanical behavior
of woven fabrics: Experimental methods and benchmark results, Composites
Part A, 39, 1037–1053.
[14] Prodromou, A.G. and Chen, J. (1997). On the relationship between shear
angle and wrinkling of textile composite performs, Composites Part A, 28,
491–503.
[15] Boisse, P., Hamila, N., Vidal-Sallé, E. and Dumont, F. (2011). Simulation
of wrinkling during textile composite reinforcement forming. Influence of
tensile, in-plane shear and bending stiffnesses, Compos. Sci. Technol., 71,
683–692.
[16] Chen, B. and Chou, T.W. (1999). Compaction of woven-fabric preforms in
liquid composite molding processes: Single-layer deformation, Comp. Sci.
Technol., 59, 1519–1526.
[17] Kelly, P.A. (2011). “Transverse Compression Properties of Composite Rein-
forcements”, In: Composite Reinforcements for Optimum Performance.
Cambridge UK: Woodhead Publishing.
[18] de Bilbao, E., Soulat, D., Hivet, G. and Gasser, A. (2010). Experimental
study of bending behaviour of reinforcements, ExpMech., 50, 333–351.
[19] Bickerton, S., Simacek, P., Guglielmi, S.E. and Advani, S.G. (1997). Inves-
tigation of draping and its effects on the mold filling process during man-
ufacturing of a compound curved composite part, Composites Part A, 28,
801–816.
[20] Loix, F., Badel, P., Orgéas, L., Geindreau, C. and Boisse, P. (2008). Woven
fabric permeability: From textile deformation to fluid flow mesoscale simu-
lations, Composites Science and Technology, 68, 1624–1630.
[21] Mark, C. and Taylor, H.M. (1956). The fitting of woven cloth to surfaces,
Journal of Textile Institute, 47, 477–488.
236 P. Boisse
[22] Van Der Ween, F. (1991). Algorithms for draping fabrics on doubly curved
surfaces, International Journal of Numerical Method in Engineering, 31,
1414–1426.
[23] Long, A.C. and Rudd, C.D. (1994). A simulation of reinforcement deforma-
tion during the production of preform for liquid moulding processes, I. Mech.
E. J. Eng. Manuf., 208, 269–278.
[24] DeLuycker, E., Morestin, F., Boisse, P. and Marsal, D. (2009). Simulation of
3D interlock composite performing, Composite Structures, 88, 615–623.
[25] Rogers, T.G. (1989). Rheological characterisation of anisotropic materials,
Composites, 20, 21–27.
[26] Spencer, A.J.M. (2000). Theory of fabric-reinforced viscous fluids, Compos-
ites Part A, 31, 1311–1321.
[27] Yu, W.R., Pourboghrata, F., Chungb, K., Zampaloni, M, and Kang, T.J.
(2002). Non-orthogonal constitutive equation for woven fabric reinforced
thermoplastic composites, Composites Part A, 33, 1095–1105.
[28] Peng, X. and Cao, J. (2005). A continuum mechanics-based nonorthogo-
nal constitutive model for woven composite fabrics, Composites Part A, 36,
859–874.
[29] Ten Thije, R.H.W., Akkerman, R. and Huétink, J. (2007). Large deforma-
tion simulation of anisotropic material using an updated Lagrangian finite
element method, Comput. Methods Appl. Mech. Eng., 196, 3141–3150.
[30] Dong, L., Lekakou, C. and Bader, M.G. (2001). Processing of composites:
Simulations of the draping of fabrics with updated material behaviour law,
J. Compos. Mater., 35, 138–163.
[31] Badel, P., Gauthier, S., Vidal-Sallé, E. and Boisse, P. (2009). Rate constitu-
tive equations for computational analyses of textile composite reinforcement
mechanical behaviour during forming, Composites Part A, 40, 997–1007.
[32] Aimène, Y., Vidal-Sallé, E., Hagège, B., Sidoroff, F. and Boisse, P. (2009).
A hyperelastic approach for composite reinforcement large deformation
analysis, J. Compos. Mater., 44, 5–26.
[33] Peng, X., Guo, Z., Du, T. and Yu, W.R. (2013). A simple anisotropic hyper-
elastic constitutive model for textile fabrics with application to forming sim-
ulation, Composites Part B, 52, 275–281.
[34] Khan, M.A., Mabrouki, T., Vidal-Sallé, E. and Boisse, P. (2010). Numerical
and experimental analyses of woven composite reinforcement forming using
a hypoelastic behavior: Application to the double dome benchmark, Journal
of Materials Processing Technology, 210, 378–388.
[35] Peng, X., Guo, Z. and Zhao, N. (2013). An anisotropic hyperelastic consti-
tutive model with shear interaction for cord-rubber composites, Composites
[36] Harrison, P., Gomes, R. and Curado-Correia, N. (2013). Press forming a 0/90
cross-ply advanced thermoplastic composite using the double-dome bench-
mark geometry, Composites Part A, 54, 56–69.
[37] Hughes, T.J.R. and Winget, J. (1980). Finite rotation effects in numerical
integration of rate constitutive equations arising in large deformation analy-
sis, Int. J. Num. Meth. Eng., 15, 1862–1867.
[38] Belytschko, T., Wing, K.L. and Moran, B. (2000). Nonlinear Finite Elements
for Continua and Structures. Chichester: John Wiley.
[39] Badel, P., Vidal-Sallé, E. and Boisse, P. (2008). Large deformation analysis
of fibrous materials using rate constitutive equations, Computers and Struc-
tures, 86, 1164–1175.
[40] Sargent, J., Chen, J., Sherwood, J., Cao, J., Boisse, P., Willem, A., Van-
clooster, K., Lomov, S.V., Khan, M., Mabrouki, T., Fetfatsidis, K. and
Jauffrès, D. (2010). Benchmark study of finite element models for simulat-
ing the thermostamping of woven-fabric reinforced composites, International
Journal of Material Forming, 3, 683–686.
[41] Lomov, S.V. and Verpoest, I. (2006). Model of shear of woven fabric and
parametric description of shear resistance of glass woven reinforcements,
[42] Launay, J., Hivet, G., Duong, A.V. and Boisse, P. (2008). Experimental
analysis of the influence of tensions on in plane shear behaviour of woven
composite reinforcements, Composites Science and Technology, 68, 506–515.
[43] Harrison, P., Abdiwi, F., Guo, Z., Potluri, P. and Yu, W.R. (2012). Charac-
terising the shear-tension coupling and wrinkling behaviour of woven engi-
neering fabrics, Composites Part A, 43, 903–914.
[44] Hamila, N., Boisse, P., Sabourin, F. and Brunet, M. (2009). A semi-discrete
shell finite element for textile composite reinforcement forming simulation,
Int. J. Numer. Method Eng., 79, 1443–66.
[45] Sabourin, F. and Brunet, M. (1995). Analyses of plates and shells with a
simplified 3 node triangular element, Thin Walled Structures, 21, 238–251.
[46] Onate, E. and Zarate, F. (2000). Rotation-free triangular plate and shell
elements, Int. J. Num. Meth. Eng., 47, 557–603.
[47] Pickett, A.K., Creech, G. and de Luca, P. (2005). Simplified and advanced
simulation methods for prediction of fabric draping, European Journal of
Computational Mechanics, 14, 677–691.
[48] Boukaber, B., Haussy, G. and Ganghoffer, J.F. (2007). Discrete models of
woven structures — Macroscopic approach, Composites Part B, 38, 498–505.
[49] Durville, D. (2005). Numerical simulation of entangled materials mechanical
properties, Journal of Materials Science, 40, 5941–5948.
[50] Miao, Y., Zhou, E., Wang, Y. and Cheeseman, B.A. (2008). Mechanics of
textile composites: Micro-geometry, Composites Science and Technology, 68,
1671–1678.
[51] Potter, K. (2002). Bias extension measurements on cross-plied unidirectional
prepreg, Composites Part A, 33, 63–73.
[52] Lebrun, G., Bureau, M.N. and Denault, J. (2003). Evaluation of bias-
extension and picture frame test for the measurement of shear properties
of PP/glass commingled fabrics, Composite Structures, 61, 341–352.
[53] Ouagne, P., Soulat, D., Moothoo, J., Capelle, E. and Gueret, S. (2013).
Complex shape forming of a flax woven fabric; Analysis of the tow buckling
and misalignment defect, Composites Part A, 51, 1–10.

January 6, 2015 10:14 Woven Composites 9in x 6in b1954-index page 239
INDEX
5-harness satin weave, 31 CT scanning, 142

8-harness satin weave, 31 cycloaliphatic amine, 185
apparent shear strength, 189, 199, 200 damage, 110

ASTM, 187 damage localisation, 59
damage self-sensing, 174
beam element, 143, 144, 147 decrimping, 220
benchmark, 225 deformation, 141, 142, 155
bending, 78, 216 delamination, 198
bending deformation, 215 differential scanning calorimetry,
bending stiffness, 221 202
bias extension test, 232 digital element, 152
double curved shapes, 218
carbon fabric, 189 durability, 181
dynamic mechanical analysis, 189,
CFRP, 160, 174, 175
202
compaction, 142, 145, 151, 152
complex shapes, 141
edge crack, 162, 169
composite, 180–182, 188, 192, 200, embedded element, 131
211
epoxy, 181, 211
composite cylinder assemblage (CCA) exfoliation, 180, 190
theory, 163
extension, 220
computational homogenisation, 8
continuity conditions, 7 failure mechanisms, 175
continuum damage mechanics, 49 fibre, 198, 211
convex hull algorithms, 149 fibre breakage, 198
crack behaviour, 159, 160, 174, 175 fibre pullout, 198
crosslink density, 206 fibre volume fraction, 191
crosslinking, 202 fibres, 212
cryogenic temperatures, 159, 160, finite element analysis, 130, 142, 150,
174, 175 160, 172, 173, 175
cryomechanics, 175 finite element method, 223
239
240 Index
flexural, 192, 194, 195, 198, 211 microstructure, 1

flexural strength, 200 modelling, 110, 142, 150
forming processes, 155, 215 modulus, 181, 188, 192, 194–196, 201,
fracture toughness, 180 211
monocoque, 179
generalised plane strain, 160, 162, 163 montmorillonite, 180, 183
geometrical model, 111 multi-filament methods, 143, 144
geometrical modelling, 115
geometry, 142, 147, 151 non-crimp fabric, 218
glass fibre reinforced polymer non-crimp fabric composites, 4
(GFRP), 159, 160, 161, 167, 170, non-local continuum damage, 59
172, 173, 175 non-local damage theory, 59
nanoclay, 180, 183, 184, 188–190, 192,
hand lay-up, 181, 182, 191, 194–196, 195, 196, 198–204, 206–209, 211,
206, 211 212
Hill–Mandel condition, 12 nanocomposite, 181, 190, 192, 194
hypoelastic model, 217 nanoparticle, 180, 181, 211
nanotechnology, 180
infrastructures, 180
in-plane shear, 221 objective derivative, 224
in-plane shear deformation, 216 operating region, 201, 203, 206
in-plane shear strains, 215 operating zone, 208
integrity, 160 organically modified
intercalated/exfoliated, 184 montmorillonites, 202
inter-laminar shear, 181
internal architecture, 110, 143 periodic boundary conditions, 142,
internal crack, 162, 169 144, 152
International Thermonuclear permeability, 122
Experimental Reactor (ITER), 159 plain woven composite, 3, 5, 23, 49,
55
kinematic modelling, 143 plain woven composite structures, 54
kinematic models, 217, 222 polymer composites, 179, 180, 202
kinematics, 146, 152 polymerisation, 185
knitted fabric composites, 4 post-cured hand lay-up, 208
post-curing, 192–195, 200, 201, 203,
locking angle, 221 207
loss modulus, 181, 189, 205–207, 209 preform, 141, 215
preform deformations, 141
matrix digestion, 191 primary structure, 179
mechanical behaviour, 110 progressive damage, 49, 59
meso-level textile processor, 111 progressive failure, 3, 54, 55
mesoscopic approach, 231 progressive failure anlayisis, 52
mesoscopic forming simulation, 231
method of cells, 5 real-time infrared spectroscopy, 202
method of inclusions, 124 reliable design, 174
microcracks, 160, 174 repetitive unit cell, 5, 9
Index 241
representative volume element, 5, 9 thermoplastic prepregs, 216

residual thermal stresses, 160 thermoset, 182
resin, 185, 189, 196, 199 thermoset prepreg, 215
resin shear strength, 211 three node triangle, 230
reusable launch vehicles (RLVs), 159, through-thickness reinforcement, 141
174 toughner, 182
rubbery plateau, 201–203, 206, 208 transverse compression, 221
twill woven composites, 3, 28, 29
satin weaves, 30
scanning electron microscopy, 188, ultrasonic cavitation, 183, 212
197 unbalanced, 230
semi-analytical, 4, 5, 52 unidirectional reinforcements, 218
semi-discrete, 227 unit cell, 110, 142, 143
semi-discrete elements, 217 unit cell geometry, 143
shear strength, 198
shell elements, 151 vacuum-assisted resin infusion
short beam test, 198 moulding, 181
SIF, 161, 166, 169, 170, 175 vacuum-assisted resin transfer
simulation, 154 moulding, 182
sonication, 184 vacuum bag, 153
stiffness, 192, 194 viscosity, 182
storage modulus, 181, 200–203, 206, void content, 191, 196
208, 211 voxel, 150
strength, 181, 188, 192, 194–196, 211
stress, 160–163, 166, 168, 172, 173, WAXD patterns of nanocomposites
175 with varying clay contents, 190
stress intensity factors, 160 weight-sensitive, 179
stress singularity, 167, 169 Westgate Bridge, 180
stress–strain, 188, 192, 194 WiseTex, 111
supercars, 179 woven composites, 2, 4, 41, 96, 109,
superconducting magnets, 159 141, 143, 159, 175
woven reinforcements, 78, 90, 95, 218
tan delta, 181, 189, 203, 204, 207, 211 wrinkles, 222
tessellating, 153 wrinkling, 230
TexGen, 142–144, 147, 148
textile composite, 1, 3, 110, 215 X-ray, 142
textile geometry, 142 X-ray CT, 142, 147, 154
textile mechanics, 146 X-ray diffraction, 186
textile preform, 141 X-ray scattering, 186
textile reinforcement, 215
thermal load, 162, 172, 175 yarn cross-section, 144, 148, 149
thermally post-cured, 211 yarn paths, 148, 154
thermally post-cured hand lay-up, Young’s moduli, 100, 162, 167, 172
209

(Computational and Experimental Methods in Structures 6) Aliabadi, Mohammad H - Woven Composites (2015, Imperial College Press)

Uploaded by

Document Information

Original Title

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

(Computational and Experimental Methods in Structures 6) Aliabadi, Mohammad H - Woven Composites (2015, Imperial College Press)

Uploaded by

Copyright:

Available Formats

Woven Composites

P984_9781783266173_tp.indd 1 30/12/14 11:32 am

Series Editor: Ferri M. H. Aliabadi (Imperial College London, UK)

Vol. 1 Buckling and Postbuckling Structures: Experimental, Analytical and

Vol. 2 Advances in Multiphysics Simulation and Experimental Testing of MEMS

Vol. 3 Multiscale Modeling in Solid Mechanics: Computational Approaches

Vol. 4 Boundary Element Methods in Engineering and Sciences

Vol. 5 Mathematical Methods and Models in Composites

Vol. 6 Woven Composites

Catherine - Woven Composites.indd 1 11/12/2014 2:51:12 PM

Imperial College Press

P984_9781783266173_tp.indd 2 30/12/14 11:32 am

Library of Congress Cataloging-in-Publication Data

British Library Cataloguing-in-Publication Data

Copyright © 2015 by Imperial College Press

Typeset by Stallion Press

Catherine - Woven Composites.indd 2 11/12/2014 2:51:12 PM

Chapter 1 Micromechanical Modelling of Textile Composites 1

Chapter 2 Mechanical Response and Failure of 2D Woven

Chapter 3 From a Virtual Textile to a Virtual Woven

Chapter 4 Modelling 3D Woven Composite Preform

Chapter 5 Cryomechanics and Crack Behaviour of Woven

Chapter 6 Eﬀects of Fabrication Processes and Nanoclay

Chapter 7 Textile Composite Forming Simulations 215

This page intentionally left blank

Omar Bacarreza∗,† , Pihua Wen‡ and M.H. Aliabadi∗,§

‡ School of Engineering and Material Sciences,

Queen Mary, University of London, London, UK, E1 4NS,

2 O. Bacarreza, P. Wen and M.H. Aliabadi

Table 1.1. Comparison of weave type and properties.

Plain + + ++ ++ +++ ++ +++++ ++

+++++ = Excellent, ++++ = Good, +++ = Acceptable, ++ = Poor, + = Very Poor

Micromechanical Modelling of Textile Composites 3

4 O. Bacarreza, P. Wen and M.H. Aliabadi

theory to predict the mechanical behaviour of twill composites. Numeri-

Micromechanical Modelling of Textile Composites 5

In this chapter, semi-analytical and numerical methods for the

1.2 Calculation of Eﬀective Elastic Properties

6 O. Bacarreza, P. Wen and M.H. Aliabadi

(i) The matrix is isotropic

In order to homogenise the subcells, through the thickness, the mate-

Micromechanical Modelling of Textile Composites 7

8 O. Bacarreza, P. Wen and M.H. Aliabadi

where Va , Vb are the volumes of the constituents “F ” or “M ”.

1.2.2 Homogenisation using FE Analysis

1.2.2.1 Geometric modelling

Micromechanical Modelling of Textile Composites 9

description. Often the models rely on accurate information of the tow

Fig. 1.2. Repetitive unit cell.

10 O. Bacarreza, P. Wen and M.H. Aliabadi

Fig. 1.3. Representative volume element.

Micromechanical Modelling of Textile Composites 11

Fig. 1.4. Poor geometry after ﬁle transfer.

1.2.2.2 Mesh generation

12 O. Bacarreza, P. Wen and M.H. Aliabadi

matrix constituent is usually very complex and generating a good mesh is

1.2.2.3 Boundary conditions

(i) Prescribed displacements

and satisfy the macro-homogeneity condition a priori [62, 70].

Micromechanical Modelling of Textile Composites 13

Table 1.2. Boundary condition with each load.

1.2.2.4 Determination of eﬀective mechanical properties

14 O. Bacarreza, P. Wen and M.H. Aliabadi