International Journal of Solids and Structures 50 (2013) 19061915

Contents lists available at SciVerse ScienceDirect

International Journal of Solids and Structures

journal homepage: www.elsevier.com/locate/ijsolstr

Micromechanical analysis of polymer composites reinforced by unidirectional

bres: Part II Micromechanical analyses
A.R. Melro a,, P.P. Camanho b, F.M. Andrade Pires b, S.T. Pinho c
a

INEGI, Rua Dr. Roberto Frias, 400, 4200-465 Porto, Portugal

DEMec, Faculdade de Engenharia, Universidade do Porto, Rua Dr. Roberto Frias, s/n 4200-465 Porto, Portugal
c
Dept. of Aeronautics, Imperial College London, South Kensington, London SW7 2AZ, UK
b

a r t i c l e

i n f o

Article history:
Available online 18 February 2013
Keywords:
FEA
Damage
Micromechanics
Epoxy
Composites

a b s t r a c t
This paper presents the application of a new constitutive damage model for an epoxy matrix on micromechanical analyses of polymer composite materials. Different representative volume elements (RVEs)
are developed with a random distribution of bres. Upon application of periodic boundary conditions
(PBCs) on the RVEs, different loading scenarios are applied and the mechanical response of the composite
studied. Focus is given to the inuence of the interface between bre and matrix, as well as to the inuence of the epoxy matrix, on the strength properties of the composite, damage initiation and propagation
under different loading conditions.
2013 Elsevier Ltd. All rights reserved.

1. Introduction
This article describes the application of a novel constitutive
model for epoxy materials presented in an accompanying paper
(Melro et al., 2013) to the prediction of inelastic deformation and
fracture of polymer composite materials. In recent years, micromechanical numerical analysis has made substantial advances.
Proof of that is the outstanding increase in recent publications
addressing the issue of micromechanical modelling of composite
materials making use of nite element analysis (FEA). The rst issue which requires attention when modelling the micromechanical
behaviour of a composite is the distribution of reinforcements in
the matrix material. Initially, a simplication was used considering
that the distribution of reinforcements followed a regular pattern,
for example, square or hexagonal (Li, 2001). For example,
Romanowicz (2012) and Hobbiebrunken et al. (2005) performed
micromechanical analysis on an hexagonal distribution of reinforcements. However, such distributions do not appropriately
reect the stress eld in the matrix, namely hydrostatic pressure
imposed by the stiffer bre material. To circumvent this problem,
algorithms and special techniques were developed to generate spatial distributions of reinforcements. For example, Vaughan and
McCarthy (2010) proposed an experimentalnumerical approach
to generate statistically equivalent distributions of reinforcements
in a composite, while Melro et al. (2008) developed an algorithm to

Corresponding author. Tel.: +351 22 508 1753; fax: +351 22 508 1584.
E-mail address: amelro@fe.up.pt (A.R. Melro).
0020-7683/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijsolstr.2013.02.007

quickly reproduce a statistically proven random distribution of

reinforcements.
The other important aspect for micromechanical analysis is the
constitutive behaviour of the constituents of the composite. The
matrix material is known to be pressure dependent and extremely
ductile under shear. To simulate these properties, many authors
gave use to standard non-linear models found in the literature.
Vaughan and McCarthy (2010) and Totry et al. (2010) have applied
the MohrCoulomb elasto-plastic model, Romanowicz (2012)
made use of the DruckerPrager elasto-plastic model, and Canal
et al. (2012) considered the matrix to follow a plastic coupled with
damage model which does not consider any hardening effect. Independent experimental analysis have shown that these constitutive
models are not suited for modelling the mechanical behaviour of
epoxy resins. Ghorbel (2008) compared some of these models concluding that only a paraboloidal criterion can properly capture the
non-linear behaviour of polymers under compression. Similar conclusions were achieved by Raghava et al. (1973). Fiedler et al.
(2001) performed extensive experimental characterisation of
epoxy resins, reaching identical results. The use of MohrCoulomb
or linear DruckerPrager models overestimates the mechanical
behaviour of the polymer under compressive loading conditions.
Based on these conclusions, this paper delivers a rst attempt at
implementing a thermodynamically sound constitutive model for
the material behaviour of an epoxy in composite materials, using
micromechanical analyses. The constitutive model consists of
an elasto-plastic stress evolution law that is capable of capturing the most signicant characteristics of an epoxy, such as
pressure dependency on its yield and failure behaviour, and shear

1907

A.R. Melro et al. / International Journal of Solids and Structures 50 (2013) 19061915
Table 1
Material properties.

Youngs modulus [GPa]

Poissons ratio
Plastic Poissons ratio
Tensile strength [MPa]
Compressive strength [MPa]
Mode I fracture toughness [N/mm]

Glass bres

Epoxy matrix

74
0.2

3.76
0.39
0.3
93
124
0.09

non-linearity with an almost perfect plastic behaviour (Fiedler

et al., 2001). An isotropic damage model has been included in the
constitutive model in order to provide more accurate predictions
on damage initiation and propagation in an epoxy matrix. The
implementation of the damage model took into account the dissipated energy release upon crack opening and propagation, according to the crack band model proposed by Bazant and Oh (1983)
which allows for mesh independence of the obtained results. To
the authors knowledge, it is the rst time such a complete integration is performed in a micromechanical numerical analysis, by
considering non-linear effects resulting from hardening and
stiffness reduction due to failure. The model is robustly built in
order to appropriately represent the constitutive response under
any loading condition.
Application of this constitutive model is made to a batch of representative volume elements (RVEs) of a unidirectional composite
material. The RVEs are developed at the micro-scale level of the
composite with a random distribution of reinforcements. A set of
periodic boundary conditions (PBCs) are applied to the RVEs allowing for the application of different loading conditions. The interface
between the matrix and the reinforcements is accounted for by
applying cohesive elements between the two constituents in RVEs
(Gonzlez and LLorca, 2007). Upon homogenisation of results, the
capability for the performed micromechanical analyses to capture
damage initiation and subsequent propagation with increasing
load is demonstrated. The inuence of interfacial damage in the
non-linear behaviour of the composite is also addressed. Attention
is also given to the dependency of results to the size of the representative volume elements used in the micromechanical analyses.
2. Constitutive modelling
In the generated RVEs, three different regions are considered:
the matrix material, the reinforcing bres, and the interface between matrix and bre. Since focus of this article is on the inuence of the interface and matrix material, the reinforcing bres
are considered to possess linear elastic isotropic constitutive
behaviour. The matrix is modelled using the elasto-plastic with
isotropic damage constitutive model described in Part I of this article (Melro et al., 2013). The interface between bres and matrix is
represented using the cohesive element formulation existent in the
commercial nite element analysis software ABAQUS (Hibbit et al.,
2006). The cohesive element behaviour is linear elastic up to damage onset. Initial stiffness of the cohesive element is set to
108 MPa=m to maintain continuity of the stress and strain elds
between bre and matrix. The strength of the cohesive elements
is dependent on the loading direction: tensile or shear. A tensile
strength of 50 MPa and a shear strength of 70 MPa are considered
for these analyses (Vaughan and McCarthy, 2011). Damage evolution is controlled by a linear softening law until complete failure of
the cohesive element. The rate of damage progression is controlled
by the fracture energy of the cohesive element under mode I, mode
II or a combination of both (Hibbit et al., 2006). Based on previous
experimental work (Varna et al., 1997), this value was set to
G IC 2 J=m2 and G IIC 6 J=m2 . These low values of toughness are
justied not only from the experimental work in Varna et al.

(1997), but also from the micromechanical numerical analyses performed by Vaughan and McCarthy (2011) who have demonstrated
that the brittle behaviour in transverse tension typical of composites is only captured for such low values of interfacial toughness.
The material properties for both reinforcements and epoxy matrix were taken from the literature and are summarised in Table 1.
The hardening behaviour of the epoxy, as well as the elastic and
strength properties were taken from the experimental work conducted by Fiedler et al. (2001), while for the fracture toughness
an average value from what can be found in the literature is used
for Bisphenol-A type epoxies (Hsieh et al., 2010, for example).
The properties for the reinforcing bres are taken from Soden
et al. (1998). The diameter of the bre is considered constant and
equal to 5 lm.
3. Finite element modelling
To demonstrate the validity of the presented constitutive models, several three-dimensional RVEs were generated and different
loading conditions were applied, always considering that a random
distribution of reinforcements in the transverse section exists. The
algorithm developed by Melro et al. (2008) was used for this purpose. periodic boundary conditions (PBCs) were applied to the
RVEs following Barbero (2008). The main focus of this work is
not the appropriate choice of boundary conditions in RVE-based
modelling. We recognise that the damage pattern is affected by
this type of boundary conditions and further studies have to be
conducted to assess its impact.
PBCs can be incorporated in a nite element analysis by using
linear multi-point constraints. These are nothing more than kinematic constraints imposed on the degrees of freedom of each pair
of nodes belonging to opposite faces, edges or vertices of the RVE.
Not only the degrees of freedom of these nodes are variables in
these equations but also the far-eld applied strains. Depending
on which position the nodes are faces, edges or vertices a different set of equations must be applied to their degrees of freedom
in order to solve compatibility issues between different kinematic
constraints. Fig. 1 shows the location and numbering used for the
faces (a), edges (b), and vertices (c) of the RVE to apply PBCs.
Equations for establishing PBCs are summarised in the following (Barbero, 2008):
 Faces

u1i  u3i  ce0i1 0

u2i  u4i  ae0i2 0
u6i

u5i

0
i3

 be 0

 Edges

u2i  u4i  ce0i1  ae0i2 0

u1i  u3i  ce0i1 ae0i2 0
u6i  u8i  ce0i1  be0i3 0
u5i  u7i  ce0i1 be0i3 0

9
0
0
u11
i  ui  aei2  bei3 0
12
0
0
u10
i  ui  aei2 bei3 0

 Vertices

u3i  u5i  ce0i1  ae0i2  be0i3 0

u2i  u8i  ce0i1  ae0i2 be0i3 0
u7i  u1i ce0i1  ae0i2  be0i3 0
u4i

u6i

0
i1

0
i2

0
i3

 ce ae  be 0:

1908

A.R. Melro et al. / International Journal of Solids and Structures 50 (2013) 19061915

Fig. 1. Numbering on RVEs for application of PBCs.

Fig. 2. Generated RVEs and respective meshes.

First order hexahedra elements with reduced integration were

chosen to generate the mesh of the RVEs. However, due to the randomness of the distribution of reinforcements and consequent difculty to mesh such geometry, a few wedge elements had to be
included in the mesh. After post-processing results, it was conrmed that these elements presented no inuence in the mesomechanical behaviour of the composite.

Five different bre distributions were generated with a bre

volume fraction of 60% CASES 15. The RVEs have a transverse
side-measure of 10 the bre radius and a thickness of 0:3 the
bre radius in the longitudinal direction. One more RVE was generated with a side-measure of 20 the bre radius CASE 6 with
the purpose of evaluating the independence of the results from the
volume element size. These dimensions are justied from a preli-

A.R. Melro et al. / International Journal of Solids and Structures 50 (2013) 19061915
Table 2
Far-eld strain tensor applied for each load case.
Load case

Far-eld strain tensor 102 

Longitudinal shear

e e2  e2
eo e1  e2 e2  e1

Transverse shear

eo e2  e3 e3  e2

Transverse compression

eo e2  e2
eo e2  e2 e2  e3 e3  e2

Transverse compression and

transverse shear

minary study on the dependency of estimated elastic properties

with different geometric parameters of the RVE (Melro et al.,
2012). Fig. 2 shows the generated RVEs and their respective
meshes. The smaller RVEs contain an approximate number of
39,200 elements while the larger RVE was built with approximately 155,000 elements.
4. Application to composite volume elements
Five different loading conditions are applied to the RVEs which
should induce different mechanical responses to the material:
transverse tension, longitudinal shear, transverse shear, transverse

SDV4
(Avg: 75%)
+1.257e02
+5.000e03
+4.583e03
+4.167e03
+3.750e03
+3.333e03
+2.917e03
+2.500e03
+2.083e03
+1.667e03
+1.250e03
+8.333e04
+4.167e04
+0.000e+00

SDV4
(Avg: 75%)
+2.248e02
+5.000e03
+4.583e03
+4.167e03
+3.750e03
+3.333e03
+2.917e03
+2.500e03
+2.083e03
+1.667e03
+1.250e03
+8.333e04
+4.167e04
+0.000e+00
1.893e07

SDV3
(Avg: 75%)
+0.000e+00
+0.000e+00
+0.000e+00
+0.000e+00
+0.000e+00
+0.000e+00
+0.000e+00
+0.000e+00
+0.000e+00
+0.000e+00
+0.000e+00
+0.000e+00
+0.000e+00

SDV3
(Avg: 75%)
+9.979e01
+9.148e01
+8.316e01
+7.484e01
+6.653e01
+5.821e01
+4.990e01
+4.158e01
+3.326e01
+2.495e01
+1.663e01
+8.316e02
+0.000e+00

80
70

CASE 6 With Cohesive Elements

CASE 6 Without Cohesive Elements
CASES 15 With Cohesive Elements
CASES 15 Without Cohesive Elements

60
Stress (22)[MPa]

Transverse tension

Diagram

50
40
30
20
10
0
0

1909

0.002

0.004

0.006
Strain (22)

0.008

Fig. 3. Results for transverse tension example.

0.01

0.012

1910

A.R. Melro et al. / International Journal of Solids and Structures 50 (2013) 19061915

compression, and a combination of transverse compression and

transverse shear. The applied far-eld strain tensors are dened
in Table 2.
For each loading condition, three different outputs are provided: the eld distributions of equivalent plastic strain and of
damage in the matrix, and the homogenised stressstrain curves
up to localisation of damage in the volume element. Although only
one CASE is presented for each loading condition, it is representative in terms of damage localisation pattern of all other CASES. The
stressstrain curves are obtained after performing volumetric
homogenisation dened by:

1
V

Z
V

rij dV

p
1X
rk V k ;
V k1 ij

SDV4
(Avg: 75%)
+2.113e02
+1.000e02
+9.149e03
+8.298e03
+7.448e03
+6.597e03
+5.746e03
+4.895e03
+4.045e03
+3.194e03
+2.343e03
+1.492e03
+6.414e04
2.094e04

SDV4
(Avg: 75%)
+3.618e02
+1.000e02
+9.114e03
+8.228e03
+7.343e03
+6.457e03
+5.571e03
+4.685e03
+3.799e03
+2.914e03
+2.028e03
+1.142e03
+2.563e04
6.295e04

SDV3
(Avg: 75%)
+9.240e01
+8.470e01
+7.700e01
+6.930e01
+6.160e01
+5.390e01
+4.620e01
+3.850e01
+3.080e01
+2.310e01
+1.540e01
+7.700e02
+0.000e+00

SDV3
(Avg: 75%)
+9.998e01
+9.165e01
+8.331e01
+7.498e01
+6.665e01
+5.831e01
+4.998e01
+4.165e01
+3.331e01
+2.498e01
+1.665e01
+8.313e02
2.023e04

55
50
45

CASE 6 With Cohesive Elements

CASE 6 Without Cohesive Elements
CASES 15 With Cohesive Elements
CASES 15 Without Cohesive Elements

40
Stress (12)[MPa]

roij

where roij represents the homogenised stress component, rkij and V k

are the stress component determined at integration point k and
associated volume, respectively, and N p represents the total number
of integration points in the RVE. After localisation of damage, the
homogenisation procedure becomes ill-posed due to its dependence
on the size of the RVE and applied periodic boundary conditions.
Hence, the stressstrain diagrams are interrupted shortly after
localisation (the softening part of the curve is still plotted in dotted
line for reference).
Plots of homogenised stressstrain curves obtained with and
without considering cohesive elements along the brematrix
interface are compared. This comparison will help to distinguish
the roll played by different failure mechanisms in the numerical

35
30
25
20
15
10
5
0
0

0.001

0.002

0.003

0.004

0.005 0.006
Strain ( )

0.007

0.008

12

Fig. 4. Results for longitudinal shear example.

0.009

0.01

A.R. Melro et al. / International Journal of Solids and Structures 50 (2013) 19061915

simulation, and the inuence of each in the homogenised behaviour of the composite under different loading conditions.
4.1. Transverse tension
Fig. 3 shows the results for the transverse tension loading condition for CASE 2 of the generated bre distributions. A crack develops along a direction transverse to the applied loading. Those
elements where the damage variable has reached a value of 1 have
failed completely. In the models with cohesive elements, crack formation begins with the decohesion of the matrix from the bre, i.e.

with the failure of cohesive elements in those regions where two

bres are aligned with the loading direction (marked with A in
Fig. 3(b)). If interfacial failure is not considered, then the transverse
tensile strength of the composite is overestimated. The same occurs for the ultimate failure strain. This is visible in Fig. 3(c).
Although initial stiffness is similar considering or not interfacial
failure, the evolution of the micro-mechanical damage is substantially different. Interfacial decohesion leads to a more pronounced
stiffness reduction, ultimately leading to an earlier transverse failure of the unidirectional composite. This stiffness reduction is a direct consequence of interfacial failure all over the RVE. A damage

SDV4
(Avg: 75%)
+2.524e02
+1.500e02
+1.375e02
+1.250e02
+1.125e02
+1.000e02
+8.750e03
+7.500e03
+6.250e03
+5.000e03
+3.750e03
+2.500e03
+1.250e03
+0.000e+00

SDV4
(Avg: 75%)
+3.072e02
+1.500e02
+1.375e02
+1.250e02
+1.125e02
+1.000e02
+8.750e03
+7.500e03
+6.250e03
+5.000e03
+3.750e03
+2.500e03
+1.250e03
+0.000e+00

SDV3
(Avg: 75%)
+9.811e01
+8.993e01
+8.176e01
+7.358e01
+6.541e01
+5.723e01
+4.905e01
+4.088e01
+3.270e01
+2.453e01
+1.635e01
+8.176e02
+0.000e+00

SDV3
(Avg: 75%)
+1.000e+00
+9.167e01
+8.333e01
+7.500e01
+6.667e01
+5.833e01
+5.000e01
+4.167e01
+3.333e01
+2.500e01
+1.667e01
+8.333e02
+0.000e+00

60

50

CASE 6 With Cohesive Elements

CASE 6 Without Cohesive Elements
CASES 15 With Cohesive Elements
CASES 15 Without Cohesive Elements

23

Stress ( )[MPa]

40

30

20

10

0
0

0.001

0.002

0.003

0.004

1911

0.005 0.006
Strain ( )

0.007

0.008

23

Fig. 5. Results for transverse shear example.

0.009

0.01

1912

A.R. Melro et al. / International Journal of Solids and Structures 50 (2013) 19061915

mechanism clearly visible in these results is the formation of thin

necks of matrix material around the regions where interfacial damage is more pronounced (marked with B in Fig. 3b). This is conrmed by micrographs taken during experimental testing
(Hobbiebrunken et al., 2006, page 2252). Both experimental observations and numerical results demonstrate that brittleness of the
composite under transverse tension is consequence of the brittleness of the interface and not of the epoxy matrix.
Also worthy of attention is the inuence of the size of the RVE. A
larger RVE leads to equal results in terms of ultimate strength, but
there is a clear tendency for a sharper decline in the stiffness

reduction. This is to be expected since, as it was demonstrated in

the past (Nguyen et al., 2010), the homogenised stressstrain response does not scale with the RVE size. This tendency occurs in
all loading conditions to be analysed in the following subsections.
4.2. Longitudinal shear
Fig. 4 shows the results for the longitudinal shear loading condition for CASE 3 of the generated bre distributions. An horizontal
band of damaged material is formed in the matrix parallel to the
bres. The inuence of the brematrix interface is almost null

SDV4
(Avg: 75%)
+2.276e02
+1.500e02
+1.374e02
+1.249e02
+1.123e02
+9.979e03
+8.724e03
+7.469e03
+6.214e03
+4.959e03
+3.704e03
+2.448e03
+1.193e03
6.190e05

SDV4
(Avg: 75%)
+3.807e02
+1.500e02
+1.375e02
+1.250e02
+1.125e02
+9.995e03
+8.744e03
+7.493e03
+6.242e03
+4.991e03
+3.740e03
+2.488e03
+1.237e03
1.383e05

SDV3
(Avg: 75%)
+9.949e01
+9.120e01
+8.291e01
+7.462e01
+6.633e01
+5.804e01
+4.975e01
+4.145e01
+3.316e01
+2.487e01
+1.658e01
+8.291e02
+0.000e+00

SDV3
(Avg: 75%)
+1.122e+00
+1.020e+00
+9.172e01
+8.148e01
+7.123e01
+6.099e01
+5.075e01
+4.050e01
+3.026e01
+2.001e01
+9.770e02
4.734e03
1.072e01

160

140

CASE 6 With Cohesive Elements

CASE 6 Without Cohesive Elements
CASES 15 With Cohesive Elements
CASES 15 Without Cohesive Elements

Stress (|22|)[MPa]

120

100

80

60

40

20

0
0

0.002

0.004

0.006

0.008
Strain (| |)

0.01

0.012

22

Fig. 6. Results for transverse compression example.

0.014

0.016

A.R. Melro et al. / International Journal of Solids and Structures 50 (2013) 19061915

SDV4
(Avg: 75%)
+1.375e02
+5.000e03
+4.583e03
+4.167e03
+3.750e03
+3.333e03
+2.917e03
+2.500e03
+2.083e03
+1.667e03
+1.250e03
+8.333e04
+4.167e04
+0.000e+00

SDV4
(Avg: 75%)
+2.568e02
+5.000e03
+4.583e03
+4.167e03
+3.750e03
+3.333e03
+2.917e03
+2.500e03
+2.083e03
+1.667e03
+1.250e03
+8.333e04
+4.167e04
+0.000e+00

SDV3
(Avg: 75%)
+9.639e01
+8.836e01
+8.032e01
+7.229e01
+6.426e01
+5.623e01
+4.819e01
+4.016e01
+3.213e01
+2.410e01
+1.606e01
+8.032e02
+0.000e+00

SDV3
(Avg: 75%)
+9.984e01
+9.152e01
+8.320e01
+7.488e01
+6.656e01
+5.824e01
+4.992e01
+4.160e01
+3.328e01
+2.496e01
+1.664e01
+8.320e02
+0.000e+00

1913

55
50
45

CASE 6 With Cohesive Elements

CASE 6 Without Cohesive Elements
CASES 15 With Cohesive Elements
CASES 15 Without Cohesive Elements

Stress (|22|)[MPa]

40
35
30
25
20
15
10
5
0
0

4
Strain (|22|)

8
3

x 10

Fig. 7. Results for combined transverse compression and transverse shear example.

which leads to the conclusion that under longitudinal loading it is

the matrix which controls ultimate strength of the unidirectional
composite. Damage is rst activated in several regions in between
close bres where non-linear behaviour of the matrix also occurs
(marked with A in Fig. 4(a) and with B in Fig. 4(b)). As loading progresses, damage localises along an horizontal band in the matrix,
i.e. swerving around the bres. Albeit there is no inuence of the
brematrix interface up to ultimate strength, in the softening region of the stressstrain curves (Fig. 4(c)) there is a tendency for a
sharper decrease of stiffness. In other words, by not considering

the inuence of the interface, a greater non-linear response in

the homogenised stress eld is obtained, leading to greater values
of failure strain of the unidirectional composite.
4.3. Transverse shear
Fig. 5 shows the results for the transverse shear loading condition for CASE 4 of the generated bre distributions. A band of localised damage is formed along a diagonal direction (marked with A
in Fig. 5(b)). The inclination angle of this band is difcult to judge

1914

A.R. Melro et al. / International Journal of Solids and Structures 50 (2013) 19061915

given the constant swerving around the bres. Nevertheless, it is

possible to conclude that failure in the matrix occurs in a plane
roughly perpendicular to the maximum principal macro-stress.
Visible in Fig. 5(c) is the inuence of the brematrix interfacial
failure. Interface failure provides a faster decrease of stiffness for
the unidirectional composite, thus inuencing the predicted
value of ultimate stress and ultimate strain. Also, there is a longer
non-linear behaviour of the unidirectional composite when the
interface is not considered in the modelling, similarly to the
longitudinal shear case.
4.4. Transverse compression
Fig. 6 shows the results for the transverse compression loading
condition for CASE 1 of the generated bre distributions. The localised band of damage (marked with A in Fig. 6(b)) follows a direction not aligned with the compressive load. The inclination of
this band is approximately 53 with a vertical line. This result
was recurring for all CASES. The measured angle of the localised
band of damage is in excellent agreement with observed experimental data (Puck and Schrmann, 2002; Pinho et al., 2006). This
inclination angle also demonstrates that it is not due to the compressive effort, but due to shear stresses developing along an inclined plane that failure is bound to occur along that inclined
plane. In Fig. 6(c) it is possible to visualise the inuence of the brematrix interfacial damage. Failure at the interface due to shear
efforts is bound to help propagate damage faster up to maximum
stress with consequent decrease in stiffness. The presence of interfacial damage also causes a lesser non-linear region, as was the situation for the two previous shear loading conditions.
4.5. Combined transverse compression and transverse shear
Fig. 7 shows the results for the combined effort of transverse
compression with transverse shear for CASE 5 of the generated bre distributions. A different direction of the localised band of
damaged material is visible in this load case when comparing with
the individual loading cases of Figs. 5(b) and 6(b). For this load
case, the magnitude of the compressive load is equal to the magnitude of the transverse shear load. While the transverse compression imposes a fracture plane where the normal tensile
component is maximum at 53 with the vertical, the transverse
shear load imposes a different inclination angle for the fracture
plane. The combination of the two different loadings has an impact
on the distribution of shear stresses in the matrix affecting the
plane at which these shear stresses are maximum. Clearly visible
in Fig. 7(c) is also the inuence of cohesive elements. Unlike in
the individual loading cases previously discussed, a combination
of transverse compression and transverse shear implies the
re-distribution of shear stresses in the matrix, especially in the
brematrix interface, where damage initiation is visible (marked
with B in Fig. 7(b)). The presence of interfacial damage also causes
a smaller non-linear region in the homogenised stressstrain
curves.
4.6. Discussion of results
Analysing the homogenised stressstrain diagrams in Figs. 37,
it can be concluded that the material response is not signicantly
affected by the choice of boundary conditions and size of RVEs in
the hardening stage. This is visible independently of the loading
condition that is applied to the RVEs. The independence of results
from the size of the RVEs is an important conclusion as it removes
the need for the use of huge RVEs in order to assure the representativeness of results, as had been done by Trias et al. (2006) or
Vaughan and McCarthy (2011).

The material response during softening is not so linear and

there can be a dependency on both boundary conditions and size
of RVE. Recently, a novel type of boundary conditions for strain
localisation in micro-structural analyses was proposed (Coenen
et al., 2012) that is able to capture the constraining effect of the
material surrounding the RVE upon developing a localisation band.
As for the dependency on the size of the RVE during softening, one
possible remedy is the use of a failure zone averaging mechanism
(Nguyen et al., 2010). These alternatives should be considered in
future developments of the work presented in this paper.
The inuence of brematrix interface must also be underlined.
Although under certain loading conditions their presence scarcely
changes the homogenised stressstrain response (such as under
longitudinal shear), in all other loading scenarios studied in this
contribution their inuence is strongly felt. In future studies
involving micromechanical analyses where diverse boundary conditions will be applied, the brematrix interface will be
considered.
5. Conclusions
Upon development and implementation of a novel constitutive
damage model for a typical epoxy matrix in Part I of this contribution (Melro et al., 2013), representative volume elements of unidirectional composites were studied. The brematrix interface has
been modelled assuming a bi-linear traction-separation damage
law. The RVEs contain a random distribution of reinforcements in
the matrix and periodic boundary conditions were implemented.
Different loading conditions were applied in order to study damage
initiation and propagation in the matrix and brematrix interface
transverse tension, longitudinal shear, transverse shear, transverse compression, and a combination of transverse compression
and shear. The objective of such analyses is to demonstrate the
capacity of the implemented material model for the matrix to capture the different responses visible in equivalent experimental
analyses:
 Under transverse tension and transverse shear loading, interfacial damage is responsible for damage initiation. The formation
of narrow necks of epoxy material around the failed interfaces
demonstrates that failure under transverse tension is controlled
by the brittleness of the brematrix interface.
 Epoxy matrix is the sole responsible for damage initiation and
propagation under longitudinal shear load, thus justifying the
high shear straining measured experimentally.
 Inclination of the fracture angle is perfectly captured by the
micromechanical analyses. This demonstrates the good agreement with experimental results obtained from using the proposed constitutive model.
 Upon combination of transverse compression with transverse
shear, a combination of failure mechanisms is observed. The
damage is initiated along the brematrix interface and, simultaneously, shear straining is observed in the epoxy matrix.
Combination of these two mechanisms leads to considerable
non-linear behaviour of the composite.
Another important conclusion is the independence of results
from the size of the RVE before damage localisation.
Acknowledgements
The rst author acknowledges the nancial support from Fundao para a Cincia e a Tecnologia (FCT) through the doctoral degree grant SFRH-BD-24045-2005, and from the research project
PDCTE-EME-65099-2003 funded by POPH QREN Tipologia 4.1

A.R. Melro et al. / International Journal of Solids and Structures 50 (2013) 19061915

Formao Avanada, sponsored by Fundo Social Europeu. The

second author would like to thank the nancial support from FCT
under the Project PDCT/EME_PME/64984/2006.
References
Barbero, E.J., 2008. Finite Element Analysis of Composite Materials, rst ed. CRC.
Bazant, Z.P., Oh, B., 1983. Crack band theory for fracture of concrete. Materials and
Structures 16, 55177.
Canal, L.P., Gonzlez, C., Segurado, J., LLorca, J., 2012. Intraply fracture of brereinforced composites: microscopic mechanisms and modelling. Composites
Science and Technology 72, 12231232.
Coenen, E., Kouznetsova, V., Geers, M., 2012. Novel boundary conditions for strain
localisation analyses in microstructural volume elements. International Journal
for Numerical Methods in Engineering 90, 121.
Fiedler, B., Hojo, M., Ochiai, S., Schulte, K., Ando, M., 2001. Failure behaviour of an
epoxy matrix under different kinds of static loading. Composites Science and
Technology 61, 16151624.
Ghorbel, E., 2008. A viscoplastic constitutive model for polymeric materials.
International Journal of Plasticity 24, 20322058.
Gonzlez, C., LLorca, J., 2007. Mechanical behaviour of unidirectional brereinforced polymers under transverse compression: microscopic mechanisms
and modelling. Composites Science and Technology 67, 27952806.
Hibbit, Karlsson, Sorensen, I., 2006. ABAQUS 6.6/Standard Users Manual. Hibbit,
Karlsson & Sorensen, Inc.
Hobbiebrunken, T., Fiedler, B., Hojo, M., Ochiai, S., Schulte, K., 2005. Microscopic
yielding of CF/epoxy composites and the effect on the formation of thermal
residual stresses. Composites Science and Technology 65, 16261635.
Hobbiebrunken, T., Hojo, M., Adachi, T., De Jong, C., Fiedler, B., 2006. Evaluation of
interfacial strength in CF/epoxies using fem and in-situ experiments.
Composites Part A: Applied Science and Manufacturing 37, 22482256.
Hsieh, T., Kinloch, A., Masania, K., Taylor, A., Sprenger, S., 2010. The mechanisms and
mechanics of the toughening of epoxy polymers modied with silica
nanoparticles. Polymer 51, 62846294.
Li, S., 2001. General unit cells for micromechanical analyses of unidirectional
composites. Composites: Part A 32, 815826.
Melro, A., Camanho, P., Pinho, S., 2008. Generation of random distribution of bres
in long-bre reinforced composites. Composites Science and Technology 68,
20922102.
Melro, A., Camanho, P., Pinho, S., 2012. Inuence of geometrical parameters on the
elastic response of unidirectional composite materials. Composite Structures
94, 32233231.

1915

Melro, A., Camanho, P., Pires, F., Pinho, S., 2013. Micromechanical analysis of
polymer composites reinforced by unidirectional bres: Part I Constitutive
modelling. International Journal of Solids and Structures. http://dx.doi.org/
10.1016/j.ijsolstr.2013.02.009.
Nguyen, V.P., Lloberas-Valls, O., Stroeven, M., Sluys, L.J., 2010. On the existence of
representative volumes for softening quasi-brittle materials a failure zone
averaging scheme. Computer Methods in Applied Mechanics and Engineering
199, 30283038.
Pinho, S., Iannucci, L., Robinson, P., 2006. Physically based failure models and
criteria for laminated bre-reinforced composites with emphasis on bre
kinking. Part II: Fe implementation. Composites Part A: Applied Science and
Manufacturing 37, 766777.
Puck, A., Schrmann, H., 2002. Failure analysis of FRP laminates by means of
physically based phenomenological models. Composites Science and
Technology 62, 16331662.
Raghava, R., Caddell, R.M., Yeh, G.S.Y., 1973. The macroscopic yield behaviour of
polymers. Journal of Materials Science 8, 225232.
Romanowicz, M., 2012. A numerical approach for predicting the failure locus of
bre reinforced composites under combined transverse compression and axial
tension. Computational Materials Science 51, 712.
Soden, P., Hinton, M., Kaddour, A., 1998. Lamina properties, lay-up congurations
and loading conditions for a range of bre-reinforced composites laminates.
Composites Science and Technology 58, 12251254.
Totry, E., Molina-Aldaregua, J.M., Gonzlez, C., LLorca, J., 2010. Effect of bre, matrix
and interface properties on the in-plane shear deformation of carbon-bre
reinforced composites. Composites Science and Technology 70, 970980.
Trias, D., Costa, J., Fiedler, B., Hobbiebrunken, T., Hurtado, J.E., 2006. A two-scale
method for matrix cracking probability in bre-reinforced composites based on
a statistical representative volume element. Composites Science and
Technology 66, 17661777.
Varna, J., Berglund, L., Ericson, M., 1997. Transverse single-bre test for interfacial
debonding in composites: 2. Modelling. Composites Part A: Applied Science and
Manufacturing 28, 317326.
Vaughan, T., McCarthy, C., 2010. A combined experimental-numerical approach for
generating statistically equivalent bre distributions for high strength
laminated composite materials. Composites Science and Technology 70, 291
297.
Vaughan, T., McCarthy, C., 2011. Micromechanical modelling of the transverse
damage behaviour in bre reinforced composites. Composites Science and
Technology 71, 388396.