Accepted Manuscript

A Modelling Framework for Composites containing 3D Reinforcement

Fredrik Stig, Stefan Hallström

PII: S0263-8223(12)00111-0
DOI: 10.1016/j.compstruct.2012.03.009
Reference: COST 4568

To appear in: Composite Structures

Please cite this article as: Stig, F., Hallström, S., A Modelling Framework for Composites containing 3D
Reinforcement, Composite Structures (2012), doi: 10.1016/j.compstruct.2012.03.009

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 A Modelling Framework for Composites containing 3D
Reinforcement
Fredrik Stig, Stefan Hallström∗
Royal Institute of Technology (KTH)
Dept. of Aeronautical and Vehicle Engineering
SE 100 44 Stockholm, Sweden

Abstract

Composite materials reinforced with three-dimensionally (3D) woven carbon fibre

textiles are investigated and the challenge and the driver for the work is to generate
numerical models to predict the mechanical behaviour of these composites. The
result of the final modelling stage is near authentic finite element (FE) models of
representative volume elements (RVE)s of the composites. They are created by using
only a small number of input parameters, such as the size of the RVE, the number
of yarns and their mutual interlacing, and the yarn crimp. The FE models may
then be utilised for various purposes but are here used to derive homogenised elastic
mechanical properties of 3D reinforced composite materials. The correlation between
the models and experiments is good, both in terms of details in the architecture and
mechanical properties. There are however some deviations that could be explained
by the models being more regular than the real material.
Keywords: 3D weave, Textile Composites, Elastic Properties, FEM

∗
Corresponding Author
Email address: [email protected] (Stefan Hallström)

Preprint submitted to Composite Structures 14th March 2012

1. Introduction

The emerging textile technique facilitating three-dimensional (3D) weaving is

now receiving great interest, particularly from the aerospace industry, for its poten-
tial utilisation in structural composites. It involves cross-directional shedding and
picking, enabling net shape production of 3D textile profiles with great flexibility in
cross-sectional shape. 3D weaving with carbon fibres has successfully been used to
produce reinforcement preforms for subsequent resin infusion into composite beams.
There are many potential advantages with this novel weaving technique, such as

• possibility to weave complex 3D shapes directly to net shape in a single pro-

duction step

• virtually no scrap material is generated

• huge flexibility in shape, textile architecture and yarn content

• an inherent 3D, orthogonal orientation of interlaced yarns.

3D textiles are characterised by complex internal geometry which is sometimes

difficult to model accurately. According to Larve et al. [1], the complex internal
geometry is the reason for the lack of "robust and widely accepted methodology
for their design and analysis". For this material class to be more widely used and
accepted, new and better analysis tools are needed [2].
Simulations are important for several reasons. They provide cost-effective ways
to perform parameter studies of mechanical properties and their relation to the rein-
forcement type and composition. There are also properties, such as the out-of-plane
stiffness and strength, that are difficult to extract experimentally, where simulations
of representative volume elements (RVE)s could offer routes for predictions [3].

2
 According to Chapman and Whitcomb [4] small changes in the strand architecture
can have a substantial effect on the local stress predictions but also on the predicted
moduli. In this work "strand" is used for a region occupied by an infiltrated fibre
bundle, as opposed to regions containing pure matrix, in between the strands. They
further state that strands with high waviness are more sensitive to variations in strand
architecture than strands that exhibit low waviness. This illustrates the importance
of accurately representing and predicting the internal geometry within modelling
frameworks, especially for materials containing 3D textile reinforcements.
There are a number of different approaches to analyse the RVE of 3D textile re-
inforced composites and their mechanical properties. The most common approaches
can be categorised into mosaic models based on classical laminate theory (CLT) [5],
analytical models where the stiffness matrix is transformed for each sub-element and
homogenised [6, 7], and finite element (FE) models [8]. The geometry description in
the analytical models may be discretised in different ways, one example is to use so
called voxels or 3D cells [9]. Sheng and Hoa [10] state that the FE approach seems
to be more accurate than other methods, and acknowledge the difficulties of defining
the fabric geometry for a complex 3D-textile preform. They further state that "a
generalized volumetric approach is not yet feasible".
Much effort has been spent on creating frameworks for the modelling of compos-
ites in general but also for modelling 3D reinforced composites. There are a few mod-
elling frameworks in which the internal geometry is modelled and the homogenised
composite properties are evaluated, e.g. WiseTex [11] and TexGen combined with a
commercial FE software [12, 13, 14]. These frameworks have been used successfully
for a number of reinforcement types, but the complex 3D-weave architecture presents
challenges that are difficult to overcome with existing frameworks. The 3D weave
architecture is truly three-dimensional and does for instance not admit projection

3
onto a single plane for subsequent analysis with some 2D analogy. The strand cross
section area both changes size and shape along the strand trajectories and needs to
be treated accordingly to avoid volume overlaps in the geometrical description.
The software package WiseTex [11] contains a textile pre-processor [15] that
utilises the principle of minimum energy to calculate strand trajectories and strand
cross section shapes. Modelling of a number of weave types have been reported
[8, 16, 15], but to the authors’ knowledge the 3D-woven textile architecture has not
yet been implemented. In WiseTex the strand cross section shape is assumed to be
constant, although its size is allowed to vary [16]. Geometric overlap is not auto-
matically prevented but can be handled in a separate step using an intermediate
FE-calculation [8].
In TexGen, the yarn trajectories are defined by splines through sets of master
nodes. Sherburn et. al [13] proposed a method to generate spatial textile models in
TexGen together with a commercial FE software, using an energy approach. However
the method utilises a procedure where the architecture is projected onto a single
plane. It is therefore not applicable for textiles with crimped out-of-plane yarns, as
in the case of the present 3D-weave [13].
A third example is a procedure presented by Mahadik and Hallett [17] where
each strand of a layer-to-layer angle interlock weave is represented by 19 circular
beam elements in a concentric pattern, The final geometry model is created by first
creating a "loose" weave, aiming to avoid yarn inter-penetration, and then applying
a tensile force in an explicit FE simulation with general contact conditions, until
the final RVE size is obtained. This scheme appears to be applicable for 3D-woven
fabrics but when the authors of the present paper tried a similar approach they failed
to create a "loose" 3D-weave without strand inter-penetration.

4
 This work aims at setting forth a framework for modelling 3D-reinforced com-
posites in general, and composites containing 3D-woven reinforcement in particular.
An automated meso scale FE-approach is used, starting with the non trivial task of
modelling the internal geometry of the RVE. The primary reason for using a FE-
approach is that it could enable sufficiently high accuracy and level of detail to obtain
reliable local stress and strain fields. The outcome is then not only homogenised elas-
tic properties but also possibilities to study damage initiation and progression. The
geometry modelling part, presented in [18], generates geometry models of the RVE
without first having to weave the preform, infiltrate it and finally extract the geome-
try from computer tomography (CT) or optical microscope images. Thus the amount
of guess work and/or tedious measuring is reduced to a minimum. The presented
approach is intended to enable reliable strength predictions since they are likely to
depend on local stress concentrations in and between different material phases [19].
During geometry modelling the shape of the strand cross sections is allowed to vary,
while strand volume overlap is avoided by assigning a clearance between the strands.
In a subsequent step a RVE is meshed and the elastic mechanical properties are ex-
tracted by applying sets of standardised loads in FE-simulations. A Python script is
the link between all steps in the analysis - from the starting point where only a few
characteristic textile parameters are known - to the final results. The script keeps
track of nodes and elements and generates material directions and local fibre volume
fractions et cetera, which are then all recorded as element attributes needed to create
the FE input files.

2. The modelling framework

The presented modelling framework consists of a sequence of steps illustrated in

Fig. 1 and explained in the following.

5
 input 1) Geometry
data modelling

3) Property and 2) Definition of

BC assignment unit cell + meshing

4) Static 5) Extraction
FE-analysis of elastic prop.

Figure 1: The sequence of the presented elastic modelling framework.

Step 1 – Geometry modelling

As a first step the internal geometry is simulated using the modelling scheme previ-
ously presented by Stig and Hallström [18]. The geometry modelling starts with a
nominal rudimentary strand geometry description with circular strand cross sections
created in the textile geometry software TexGen [12]. In TexGen each strand trajec-
tory is defined by a spline through a number of master nodes. Then the strands are
converted into strongly anisotropic inflatable tubes in a FE-model. They are then
inflated to a target strand volume fraction under general contact conditions in an
explicit FE simulation.
Meshing of matrix pockets in between fibre bundles in detailed models of com-
posite materials is known to be a challenge due to their often complex shapes [1].
In the geometry modelling part of the proposed framework, there is a possibility
to control the contact clearance during the simulation thus creating sufficient gaps

6
between the strands to allow for small volume elements in between. The gaps help
facilitate mesh generation in general, and also avoiding too small elements in the
models. The concept of introducing gaps between strands was also used previously
by Crookston et al. [14].
Each expanded tube is divided into a number of slices in order to manage local
properties such as local fibre volume fraction and fibre orientation in subsequent
steps. A slice is defined as the volume confined by a band of elements hooping a
tube, see Fig. 2. The centroid and the cross section area of every slice are determined
from the nodal positions of the elements hooping the slice. The slice cross section
area, Aslice , and the effective fibre cross section area, Af , are then used to calculate
the local fibre volume fraction for each slice as

Af
vfslice = . (1)
Aslice

The local fibre volume fractions are saved as attributes of the slices later to be used
for calculating the local stiffness properties of the strands. The trajectory of each
tube is defined by a spline through selected slice centroids. The local fibre orientation
of each slice is then defined as the tangent of the trajectory spline at the slice centroid,
see Fig. 2, and also saved as a slice attribute.

Area Slice

Local fibre
Centroid orientation

Figure 2: Two slices of a strand from the tube model in Step 1, where centroids, a slice cross section
area (Aslice ) and the local fibre orientation are illustrated.

7
Step 2 – Definition of unit cell and meshing
A periodically repetitive RVE of the inflated modelled geometry is exported to a
mesh generating software, where the RVE boundaries are defined, see Fig 3, and
a periodic mesh of the RVE surfaces is created. A volume mesh is then generated
within the RVE, using quadratic tetrahedral elements.

Warp
V-weft

H-weft

(a) (b)

Figure 3: The sub-volumes of the FE model, constituting a) pure matrix regions and b) strands
comprising of warp, horizontal (H-weft) and vertical (V-weft) weft.

An elastic and transversely isotropic material model is used for the strands while
the matrix material in-between the strands is modelled using an isotropic material
description. Each volume element within the strands is associated with its closest
slice. This is done by an algorithm in which the shortest distance between the
element’s centroid and the slice plane is calculated. The slice plane is perpendicular
to the spline that defines the strand trajectory, and corresponds to the plane labelled
"Area" in Fig. 2. The elements are then assigned the local fibre volume fraction and
fibre orientation associated with their respective slice.
Step 3 – Assignment of properties and boundary conditions
The material properties of each element are finally calculated using constitutive
material data, vfslice and Chamis [20] micro-mechanics relations. The 3D reinforced

8
unit cell is assumed to be orthotropic and hence six cases of prescribed displacements,
corresponding to principal unit strains

ij = 1 , (were i, j = x, y, z) (2)

are applied to extract a complete set of six independent constitutive constants [19].
A set of boundary conditions are defined for each load case according to Li and
Wongsto [21]. The boundary conditions are implemented as equation constraints
between node pairs located on opposite boundary sides of the RVE. The node pairs
are identified using a sorting algorithm. To avoid hierarchical equation constraints,
which are not permitted in the FE software, the nodes on the edges and corners of
the RVE are treated separately.
Step 4 – Static FE analysis
The FE software Abaqus [22] is used to perform the static FE analyses using the
above load cases.
Step 5 – Extraction of elastic properties
In this final step, the constitutive constants are calculated using the methodology
described in Wang et al. [19]. The constitutive relation of the homogenised RVE is
expressed as
 = S σ, (3)

where  is the average strain vector, S the compliance matrix and σ the average
stress vector. The average strain vector is given as input for the six unit load cases
(2). The corresponding average stress vector for each load case is calculated as the
volume average, according to

1 X
σ= σ i dVi , (4)
V i

9
where σ i is the stress in element i, Vi is the volume of element i, and V is the total
volume. The compliance matrix is obtained by combining the system of equations
given by the load cases according to
−1
.. .. .. .. .. .. .. .. .. .. .. ..
 
. . . . . . . . . . . .
  
S =  1 2 3 4 5 6   (5)
  
σ1 σ2 σ3 σ4 σ5 σ6 
.. .. .. .. .. .. .. .. .. .. .. ..
  
. . . . . . . . . . . .

where the subscripts represent the load case numbers. The constitutive constants
may finally be extracted from the compliance matrix.

3. Case Study

A case study of composite materials reinforced with various 3D weaves is per-

formed to validate the proposed modelling framework. The 3D-weaving process is
characterised by dual directional shedding operations that fully interlace the warp
yarns with both horizontal and vertical weft yarns, according to the principles set
forth by Khokar [23]. The architecture of the 3D-woven textile is schematically illus-
trated in Fig. 4. As can be seen, the architecture of the surface layer is different from
the interior since the weft yarns turn in the surface layer and partly travel along the
weaving direction between insertion points. The interior RVE is outlined by black
lines in Fig. 4 and it consists of four warp yarns, four horizontal and four vertical
weft yarns. The case study is limited to the interior RVE. Four models are created
of different weaves; a baseline plain 3D-weave (B), a sparser 3D-weave (S), a denser
3D-weave (D) and a baseline 3D-weave containing stuffer yarns (BS). The nominally
straight stuffer yarns in BS are positioned in the pockets occurring between the warp
yarns, visible in the upper right illustration in Fig. 4.

10
 z z
C x Interior RVE y
A

x y
A x
H-weft
y V-weft
Warp

Figure 4: A plain 3D weave sample, comprising of warp, horizontal (H-weft) and vertical (V-weft)
weft, depicted along its three principal axes and in an isometric view. The interior RVE unit cell
is outlined as a prismatic box.

To be able to compare the results from the models and experiments, the input
variables needed to create the geometry model in Step 1 is generated from either CT
or microscopy images of composite samples, see Table 1. Crimp is here defined as the
physical length of the yarn along its path (Ly ) over the corresponding wavelength
(λ) as
Ly
crimp = . (6)
λ
As mentioned, the termination criterion for the inflation in Step 1 is when the overall
strand fraction, vs , reaches the target value. It is calculated as
vf
vs = , (7)
vfs

11
where vf is the overall fibre volume fraction and vfs the average fibre volume fraction
within the strands. Despite the complex nature of the 3D weave architecture, the
required input data is relatively limited.

Table 1: Geometric input data. The RVE length and width, assuming square cross-section, are
denoted λwarp and λwef t , respectively. It should be noted that the horizontal and vertical weft
yarn crimps are very small and difficult to measure, and should be treated accordingly.

Plain 3D woven
Weave style
3D woven with stuffer
Name S B D BS
λwarp [mm] 17.72 11.74 7.47 11.69
λwef t [mm] 2.12 2.18 2.33 2.62
crimp warp 1.012 1.025 1.055 1.0316
crimp weft 1.0002 1.0002 1.0002 1.0002
vf 0.45 0.44 0.42 0.58
vfs 0.54 0.54 0.57 0.68

Geometry models of reinforcements type B and BS are compared with pictures

of corresponding composite samples in Fig. 5 to assess the quality of the geometry
modelling. (The strand boundaries in the CT and microscope images are traced with
white lines to enhance visibility in the presentation.) The view cut in all four images
are where the warp yarns interlace with the horizontal weft yarns. The view cut is
also indicated in Fig. 4, labelled A. As can be seen in Fig. 5 there is a good overall
correlation.
Two types of yarns are used in the composite samples; Toray T700 12k in the
warp and Toho Tenax HTA40 6k in the weft. The vinyl ester Reichhold Dion 9500 is

12
 Specimen Geometry in FE model

Figure 5: CT-scan / microscopy images and corresponding FE geometry model views for specimen
type B (top) and BS (bottom).

used in all test specimens. The constituent material properties, presented in Table
2, are taken from manufacturers’ data sheets, except for those marked with an (*)
which are assumed similar as reported for Toray T300 by Chamis [20]. The warp
yarn’s (T700) longitudinal Young’s modulus is identical to that of T300, believed to
justify the use of T300’s transverse elastic properties. The longitudinal modulus of
HTA40 is 3% higher than that of T700.
All test specimens had square cross sections, and since the case study was limited
to the interior RVE, the surface layers were milled off creating dog bone shaped
specimens. An Instron 100 kN test machine was used for the tests together with a
100 kN load cell and the strain was measured using a strain gauge with a 50 mm

13
Table 2: Fibre and matrix material properties. The sub and superscripts f , m, l and t denote fibre,
matrix, longitudinal and transverse properties, respectively.

El [GPa] Et [GPa] Glt [GPa] Gtt [GPa] νlt νtt

Fibre 230** 14* 9.0* 4.8* 0.20* 0.25*
Matrix 3.1 3.1 1.1 1.1 0.35 0.35
* Reported by Chamis [20] for T300
** T700

gauge length.
In parallel with experiments, a FE model was created for each of the reinforcement
types listed in Table 1, using the proposed methodology. Relatively fine meshes are
needed to resolve the detailed geometry. A mesh convergence study was performed
showing that further refinements of the meshes were not necessary. All models con-
tained about a million degrees of freedom.

4. Results

In Table 3 results from the models are compared with experimental results and
results obtained from another approach [24], using simplified geometries in WiseTex
[11]. Since the RVE is orthotropic, properties transverse to the weave direction (y
and z directions) are presented with the index t in Table 3.
When the models are subjected to a tensile load (in their x direction) the highest
shear stresses are found in the warp strands in between horizontal and vertical wefts.
The shear stress in a warp strand is illustrated in Fig. 6 where the marker C is used
to illustrate the location of the maximum stress in the RVE, see Fig. 4.
A parameter study was conducted of the effect on the obtained elastic properties

14
Table 3: Elastic properties extracted using the proposed FE-modelling scheme, a Mori-Tanaka
(MT) model implemented in WiseTex [11] and experimental results.

Type Model Ex [GPa] Et [GPa] Gxt [GPa] Gtt [GPa] νxt νtt
MT 86.0 12.6 3.43 2.00 0.22 0.22
S FE 83.2 12.0 3.76 1.93 0.31 0.31
Exp. 74.6± 0.3 - - - - -
MT 66.3 15.6 3.77 2.07 0.20 0.17
B FE 65.8 12.5 3.94 1.99 0.36 0.28
Exp. 58.8 ±3.7 - - - - -
MT 33.3 19.9 3.54 2.13 0.15 0.12
D FE 36.2 11.5 3.84 2.19 0.41 0.27
Exp. 31.9±2.1 - - - - -
MT 109.2 16.7 4.22 2.44 0.18 0.16
BS FE 105.9 16.2 5.20 3.60 0.25 0.25
Exp. 96.4 - - - - -

from the degree of inflation. The results are presented in Fig. 7. The volume fraction
of strand, vs , is used as a measure and its final value corresponds well to that of the
real specimens (see Fig. 5).

5. Discussion

The Young’s modulus coming from the proposed modelling methodology agrees
well with the experimental results. However, both the FE models and the WiseTex
models consistently overpredict the longitudinal stiffness from the experiments about
10%. This is in line with previous work on composites reinforced with 3D textiles,

15
 !
xz

x y

Figure 6: The shear stress distribution in a warp strand in a type B model. Red and blue colours
indicate high positive and negative shear stresses τxy , respectively. The marker C relates the
position to the RVE in Fig. 4.

e.g. [25, 14, 26, 27, 19], where stiffness differences between models and experiments
from a few per cent up to 30-40% were reported.
A plausible reason for the current discrepancy is the models being more regular
than the real material. First of all, the reinforcement in the specimens is not per-
fectly regular, since the manufacturing of preforms was developed in parallel with
manufacturing of the woven preforms. Secondly, the preforms were typically slightly
misaligned with respect to the specimens’ principal direction due to the relatively el-
ementary moulding technique at infiltration. Visual inspection of microscope images
confirmed that the preforms were locally misaligned. This is more pronounced for
denser weaves. E.g. for the weave type D local misalignment angles up to 5 degrees
were found. Weave type D also exhibits the highest discrepancy between FE model
and experiments, see Table 3. None of the models take such preform deformation
into account. Thirdly, the filaments in a yarn are nominally parallel, as assumed in
the FE models, but that is not always true for the real material. As can be seen in
Fig. 8, the filament paths may diverge locally, affecting the warp strand stiffness of

16
 80

60

Ex [GPa]
40 Inflation corresponding
to real specimen
20

0
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9
vs

Figure 7: Young’s modulus in the x-direction for the baseline (B) model after various levels of tube
inflation in the first modelling step.

the samples negatively.

Figure 8: Filament orientations in a warp yarn.

Another source of uncertainty is related to to the difficulty to experimentally

determine the yarn crimp in a composite sample. Stig and Hallström [24] estimated
the 3D warp yarn crimp from 2D images and relatively large standard deviations
were found, especially for highly crimped samples. If the input warp yarn crimp is
set too low in the modeling framework then the constraints from the warp strands

17
on the weft strands will be too high during the inflation phase (Step 1), resulting in
relatively higher weft strand crimp in the model. The consequence of underestimated
warp crimp is therefore not only an overly stiff model in the warp direction but also
a too compliant model in the weft directions. Both of these tendencies are seen in
Table 3. In the upper images in Fig. 5 the weft strand crimp appears to be higher
in the model than in the real material.
The proposed FE methodology and the Mori-Tanaka WiseTex model (MT) gener-
ate very similar results for all configurations in the case study. The largest differences
are found for the Poisson’s ratios, where the FE model predicts slightly higher val-
ues. For the denser weaves the wavelength is shorter and since the weft yarn content
is the same in all models, the transverse Young’s modulus is expected to increase
with decreasing wavelength. This is the case for the MT models that are based on
simplified geometries but not for the FE models. An explanation could be that the
induced weft yarn crimp, which appears to be absent in the real material, reduces
the stiffness more than the higher weft density increases it.
The geometry model of the dense (D) weave was at the limit of what the presented
geometry generating approach could handle, due to the highly crimped warp yarns.
During inflation very large deformations occur in the initial (Step 1) FE simulation,
where both the trajectories and the cross section size and shape of the tubes/strands
change dramatically. The approach could handle quite extreme deformations but
there are limitations related to the fact that the tube walls are modeled as linear
elastic while in the real material slip between individual filaments within the yarns
is possible and likely to occur. When the models were forced into too large deforma-
tions, some tendencies towards mesh distortion problems and abnormal expansion of
the tubes started to become prominent.
One of the benefits of using a detailed FE model is that stress and strain dis-

18
tributions are obtained together with a complete stiffness matrix. The high level
of detail in the stress and strain fields may be used to study damage initiation and
develop failure criteria, and do also allow for analysis of e.g. failure progresion. A
pronounced non-linear behaviour of 3D-textile reinforced composites loaded in ten-
sion has previously been reported [28, 29, 30]. The onset of damage was referred to
as plastic tow straightening [28] and was believed to be associated with matrix fail-
ure due to high shear stresses. Using the proposed model generation with sufficient
mesh refinement, potentially enables corresponding accuracy in calculated stress and
strain fields. The locations of highest shear stresses (e.g. point C in Figs. 4 and 6)
are potential sites for large plastic deformation and/or crack initiation, in the matrix
or at fibre-matrix interfaces.
The parameter study, where the degree of inflation from Step 1 was varied, shows
that the longitudinal Young’s modulus is affected slightly. The reason for this is
that the strand trajectories change shape (not length) from the assumed starting
trajectory during inflation. Note that the overall fibre volume fraction in all three
resulting models is the same, but the strand fibre volume fraction is higher in the
less inflated models.
As can be seen in Table 3 for types S, B & D, warp yarn crimp has a substantial
effect on the longitudinal Young’s modulus. However, the effect of warp yarn crimp
on the transverse modulus, the shear moduli and the Poisson’s ratios are relatively
small.

6. Conclusions

A general FE based method for modelling composite materials containing 3D tex-

tile reinforcement and predicting their elastic properties is presented. The method
enables creation of almost authentic geometry descriptions of three different weave

19
types, which is one of the major challenges when a FE approach is chosen for mod-
elling of composite materials with complex reinforcement architectures.
The correlation between results from the models and experiments is very good,
both in terms of details in the architecture and mechanical properties. The FE-
models are all about 10% stiffer than corresponding real samples, which is considered
satisfactory given the plausible sources of discrepancies. The real materials were not
as regular as the models and it is possible that the FE modelling scheme could be
improved even further if the irregularities were also taken into account. It must
however be considered more worthwhile to improve the quality of the real materials
than to put great emphasis on modelling flaws that are likely to disappear in the
future, as a result of improvements in manufacturing.
The proposed FE model generation methodology may be employed for subsequent
analysis of stresses and strains, strength and damage analysis as well as infusion and
permeability simulations.

Acknowledgement

The presented work has been financially supported by the European Commission
through the EU FP7 Contract No. ACPO-GA-2010-266026 Cost Effective Rein-
forcement of Fastener Areas in Composites (CERFAC). Biteam AB is acknowledge
for supplying the 3D-woven preforms. The authors would also like to express their
gratitude to Florian Weyrauch and his colleagues at DLR in Stuttgart for carrying
out the CT scanning.

References

[1] E. V. Larve, D. H. Mollenhauer, E. G. Zhou, T. Breitzman, T. J. Whitney,

Independent mesh method-based prediction of local and volume average fields

20
 in textile composites, Composites Part A: Applied Science and Manufacturing
40 (2009) 1880 – 1890. Special Issue: CompTest 2008.

[2] M. Nicosia, F. Vineis, J. Lawrence, S. Holmes, Microstructural modeling of

three-dimensional woven fiber composites, in: Proceedings of the 9th Interna-
tional Conference on Textile Composites - TEXCOMP9: Recent Advances in
Textile Composites.

[3] A. E. Bogdanovich, Progressive failure modeling and strength predictions of 3-d

woven composites, in: Proceedings of TEXCOMP9: Recent Advances in Textile
Composites.

[4] C. Chapman, J. Whitcomb, Effect of assumed tow architecture on predicted

moduli and stresses in plain weave composites, Journal of Composite Materials
29 (1995) 2134–2159.

[5] T. Ishikawa, T.-W. Chou, Stiffness and strength behaviour of woven fabric
composites, Journal of Materials Science 17 (1982) 3211 – 20.

[6] P. Vandeurzen, J. Ivens, I. Verpoest, A three-dimensional micromechanical

analysis of woven-fabric composites. i. geometric analysis, Composites Science
and Technology 56 (1996) 1303 – 1315.

[7] P. Vandeurzen, J. Ivens, I. Verpoest, A three-dimensional micromechanical

analysis of woven-fabric composites. ii. elastic analysis, Composites Science and
Technology 56 (1996) 1317 – 1327.

[8] S. Lomov, D. Ivanov, I. Verpoest, M. Zako, T. Kurashiki, H. Nakai, S. Hirosawa,

Meso-FE modelling of textile composites: Road map, data flow and algorithms,
Composites Science and Technology 67 (2007) 1870 – 1891.

21
 [9] E. Potter, S. T. Pinho, P. Robinson, L. Iannucci, A. J. McMillan, Unit cell mesh
generation for 3d woven composites, in: In proceeding of ECCM-14 Budapest.

[10] S. Z. Sheng, S. Hoa, Modeling of 3D angle interlock woven fabric composites,

Journal of Thermoplastic Composite Materials 16 (2003) 45 – 58.

[11] I. Verpoest, S. V. Lomov, Virtual textile composites software WiseTex: Integra-

tion with micro-mechanical, permeability and structural analysis, Composites
Science and Technology 65 (2005) 2563 – 2574.

[12] M. Sherburn, Geometric and Mechanical Modelling of Textiles, Ph.D. thesis,

The University of Nottingham, 2007.

[13] M. Sherburn, A. Long, A. Jones, Prediction of textile geometry using strain en-
ergy minimisation, in: Proceedings of the 1st World Conference On 3D Fabrics.

[14] J. J. Crookston, S. Kari, N. A. Warrior, I. A. Jones, A. C. Long, 3D textile

composite mechanical properties prediction using automated FEA of the unit
cell, in: Proceedings of The Sixteenth International Conference on Composite
Materials, July 8-13, 2007: Kyoto, Japan.

[15] S. Lomov, A. Gusakov, G. Huysmans, A. Prodromou, I. Verpoest, Textile geom-

etry preprocessor for meso-mechanical models of woven composites, Composites
Science and Technology 60 (2000) 2083 – 2095.

[16] S. Lomov, D. Ivanov, G. Perie, I. Verpoest, Modeling 3D fabrics and 3D-

reinforced composites: Challenges and solutions, in: Proceedings of the 1st
World Conference On 3D Fabrics (CD-ROM).

22
[17] Y. Mahadik, S. Hallett, Finite element modelling of tow geometry in 3D woven
fabrics, Composites Part A: Applied Science and Manufacturing 41 (2010) 1192–
1200.

[18] F. Stig, S. Hallström, Spatial modelling of 3D-woven textiles, Composite Struc-

tures DOI: 10.1016/j.compstruct.2011.12.003 (2011).

[19] X. Wang, X. Wang, G. Zhou, C. Zhou, Multi-scale analyses of 3D woven compos-

ite based on periodicity boundary conditions, Journal of Composite Materials
41 (2007) 1773 – 1788.

[20] C. C. Chamis, Simplified Composite Micromechanics Equations For Strength,

Fracture Toughness and Enviromental Effects, SAMPE Quarterly (1984).

[21] S. Li, A. Wongsto, Unit cells for micromechanical analyses of particle-reinforced

composites, Mechanics of Materials 36 (2004) 543 – 572.

[22] www.simulia.com

[23] N. Khokar, 3D-weaving: Theory and practice, Journal of the Textile Institute
92 (2001) 193–207.

[24] F. Stig, S. Hallström, Influence of crimp on 3D-woven fibre reinforced compos-

ites, Submitted for publication (2012).

[25] Z. Wu, Three-dimensional exact modeling of geometric and mechanical proper-

ties of woven composites, Acta Mechanica Solida Sinica 22 (2009) 479 – 486.

[26] D. Li, D. Fang, N. Jiang, Y. Xuefeng, Finite element modeling of mechanical

properties of 3D five-directional rectangular braided composites, Composites
Part B: Engineering 42 (2011) 1373 – 1385.

23
[27] P. Lapeyronnie, P. L. Grognec, C. Binétruy, F. Boussu, Homogenization of
the elastic behavior of a layer-to-layer angle-interlock composite, Composite
Structures 93 (2011) 2795 – 2807.

[28] B. Cox, M. Dadkhah, W. Morris, J. Flintoff, Failure mechanisms of 3D woven

composites in tension, compression, and bending, Acta Metallurgica et Materi-
alia 42 (1994) 3967 – 3984.

[29] B. Cox, M. Dadkhah, W. Morris, On the tensile failure of 3D woven composites,

Composites - Part A: Applied Science and Manufacturing 27 (1996) 447 – 458.

[30] P. Callus, A. Mouritz, M. Bannister, K. Leong, Tensile properties and fail-

ure mechanisms of 3D woven GRP composites, Composites - Part A: Applied
Science and Manufacturing 30 (1999) 1277 – 1287.

24