You are on page 1of 8

COMPOSITES

SCIENCE AND
TECHNOLOGY
Composites Science and Technology 67 (2007) 32783285
www.elsevier.com/locate/compscitech

Micromechanical modelling and experimental investigation of


random discontinuous glass ber polymermatrix composites
A. Bouaziz a, F. Zari a,*
, M. Nat-Abdelaziz a, J.M. Gloaguen b, J.M. Lefebvre b

a
Laboratoire de Mecanique de Lille (UMR CNRS 8107), USTL, PolytechLille, Avenue P. Langevin, 59655 Villeneuve dAscq Cedex, France
b
Laboratoire de Structure et Proprietes de lEtat Solide (UMR CNRS 8008), USTL, Batiment C6, 59655 Villeneuve dAscq Cedex, France

Received 13 October 2006; received in revised form 28 February 2007; accepted 28 March 2007
Available online 11 April 2007

Abstract

Videomeasurements were used to estimate the damage in chopped random glass ber polymermatrix composites. In order to predict
the overall mechanical behaviour, voiding evolution induced by ber debonding is incorporated into a micromechanics-based constitu-
tive model. The comparison between the experimental data and the numerical predictions shows a very good agreement.
2007 Elsevier Ltd. All rights reserved.

Keywords: A. Polymermatrix composites (PMCs); B. Modelling; B. Microstructure; B. Debonding; D. Videoextensometry

1. Introduction mechanical analyses have been proposed in the literature to


simulate the damage response of discontinuously ber-rein-
Fiber-reinforced polymermatrix composites are widely forced composites. Meraghni and Benzeggagh [1] and Mer-
used in industry as structural materials and then there is an aghni et al. [2,3] identied damage mechanisms in a
evident need for the prediction of their mechanical proper- random discontinuous ber-reinforced composite based
ties. In this kind of materials, various damage phenomena on acoustic emission. The authors proposed a microme-
can occur such as matrix cracking, interfacial debonding, chanical analysis based on a modied version of the
ber pull-out and ber fracture. The predominant damage Mori-Tanaka model to investigate the eect of damage
mechanism may vary according to ber volume fraction mechanisms on the overall response. Similar analyses were
and aspect ratio, orientation and distribution in the com- made by Fitoussi et al. [4,5] and Derrien et al. [6] relying on
posite, strengths of the interface and each constituent, the ultrasonic method and the Mori-Tanaka approach.
and loading mode. Progressive damage accumulation in Considering bers randomly oriented in a ductile poly-
the composite is known to aect the overall mechanical mermatrix, Lee and Simunovic [79] recently proposed a
properties. In order to estimate the overall response of micromechanical model. In all these investigations, the
the material, the accumulated damage must therefore be damage evolution was introduced into the micromechani-
included in the constitutive relations. Furthermore, in cal approach from probabilistic considerations in the form
order to achieve a rigorous description of material, the con- of a Weibull statistical density function. This probability
stitutive equations must be derived from micromechanical function introduces additional parameters generally cali-
considerations. However, due to the complexity of the brated on an experimental stressstrain curve.
microstructure, the damage mechanisms in ber-reinforced In the present paper, the damage behaviour of glass mat
polymermatrix composites are not fully understood from ber-reinforced polymermatrix composite is investigated
the experimental point of view. Nevertheless, some micro- using a combined approach of micromechanical modelling
and experimental characterization. A representative com-
*
Corresponding author. Tel.: +33 328767460; fax: +33 328767301. posite, consisting of a polyester matrix incorporating vari-
E-mail address: fahmi.zairi@polytech-lille.fr (F. Zari). ous volume fractions of randomly oriented glass bers, was

0266-3538/$ - see front matter 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compscitech.2007.03.031
A. Bouaziz et al. / Composites Science and Technology 67 (2007) 32783285 3279

chosen. Rather than using the methods (acoustic emission 25


and ultrasonic) usually employed, an alternative technique
based on volume strain variation is retained to quantify the 1
progressive interfacial damage in the material. The pro- 3
gressive damage observed on the macroscopic level was
2
identied at the microscopic level by SEM observations.
Damage evolution is introduced into a micromechanical
model based on the Eshelby formulation and the ensem- A
B
ble-volume averaging homogenization process. The result- 23 F
C
G representative
140 volume element
ing micromechanical model is implemented numerically D
E
and used to simulate uniaxial loading for the sake of com-
parison with experimental results.
R=120
2. Experiments

2.1. Material

The composite under study consists of randomly ori- Fig. 1. Specimen dimensions (in mm) and conguration of the seven
markers for strains measurement.
ented chopped (E-glass) bers embedded in an unsaturated
polyester matrix. The material was fabricated by compres-
sion molding at room temperature from mat densities of
300 and 450 g/m2. Each ber is constituted by a bundle load
interface
of 50 laments of about 10 lm in diameter and 50 mm in
length. The bers, randomly distributed in the plane of
the material, exhibit an aspect ratio a of about 100. The load
cell
ber volume fraction, obtained by burning o the polyester
matrix, is found to be 13% and 21%. For convenience in videotraction
marked
the remaining of the paper, the glass ber reinforced poly- sample workstation
video
ester (GFRP) composites are coded as follows: GFRP-21R white neon
camera
and GFRP-13R for the 21% and 13% ber volume frac- light

tions, respectively.

2.2. Experimental method


actuator
interface
In the literature, the damage response in ber reinforced electromechanical Instron
machine (model 5800)
polymer composites is experimentally addressed from acous-
tic emission or ultrasonic techniques. In these methods, the Fig. 2. Schematic illustration of the optical extensometer system.
void nucleation at the interfaces is not explicitly quantied
and is completely ignored in the associated modelling. In this
paper, video-controlled tensile tests and scanning electron extensometer allows to avoid the potential damage initia-
microscope (SEM) observations were achieved in order to tion on the specimen surface. The technique consists in fol-
understand the basic mechanisms governing deformation lowing, in real time, the location of the center of gravity of
in the composite material. ink marks drawn on the sample surface (Fig. 1). Five round
Parallelepipedic plates were machined to the specimen markers (A, B, C, D and E) are aligned along the tensile
shape presented in Fig. 1. Rather than using a normalized axis 1, each located 1 mm apart (center-to-center). Two
specimen with constant cross-section, we have designed this others (F and G) are placed on the perpendicular axis 2
specic geometry of the specimen in order to localize the and are separated as much as possible. Their diameter is
damage in the central part. The large radius of curvature about of 0.5 mm. The objective of the method is to deter-
allows to be freed from the stress triaxiality dependence. mine the three principal strains in the representative vol-
The mechanical behaviour of the composite material ume element (RVE) dened by the three transversal
was determined at room temperature by tensile tests on markers [11]. The thickness of the RVE is about 0.2 mm.
an Instron (model 5800) machine equipped with video- The surface containing the seven markers is digitized and
traction extensometer. The videotraction set-up consists the coordinates of their gravity centers are provided 50
of a CCD videocamera and an unit for image digitizing and times per second with a precision of about 0.05 pixel, which
analyzing as schematically represented in Fig. 2 [10,11]. gives a precision on the deformation of about 0.0002.
Contrary to standard strain measurement methods such The axial strain is determined from the partial axial
as gauge which impose a measure with contact, the optical strains induced by the relative displacement of markers
3280 A. Bouaziz et al. / Composites Science and Technology 67 (2007) 32783285

disposed along the axis 1. The partial axial strains are given 2,5E-05 5,E-05
by:
AB  A0 B0 2,0E-05 4,E-05
E11 AB
A0 B0

global strain rate (s-1)


local strain rate (s-1)
BC  B0 C 0 1,5E-05 3,E-05
E11 BC
B0 C 0
1
CD  C 0 D0
E11 CD 1,0E-05 2,E-05
C 0 D0
DE  D0 E0
E11 DE 5,0E-06 1,E-05
D0 E 0
where (A0B0, B0C0, C0D0, D0E0) and (AB, BC, CD, DE) are
initial and extended distances dened in Fig. 1. The axial 0,0E+00 0,E+00
0 50 100 150 200 250
strain in the RVE is obtained by a nonlinear polynomial
time (s)
interpolation from the four partial axial strains [11]:
E11 nonlinear interpolation from E11 AB; 1,E-04 2,0E-04

E11 BC; E11 CD; E11 DEFCG 2


8,E-05
The transversal strain is given by: 1,5E-04

global strain rate (s-1)


local strain rate (s-1)
FG  F 0 G0
E22 E22 FG 3 6,E-05
F 0 G0
1,0E-04
where F0G0 and FG are initial and extended distances of the 4,E-05
transversal markers. Note that since only one camera is
used, E22 and E33 are assumed equal. 5,0E-05
2,E-05
Once the three principal strains are measured in the
RVE, the local volume variation measuring the dilatation
in the material can be calculated from the following 0,E+00 0,0E+00
relation: 0 50 100 150 200 250
time (s)
V V0
Ev 1 E11 1 E22 1 E33  1 4
V0 Fig. 3. Local and global strain rates for GFRP-13R: (a) non strain-
controlled and (b) strain-controlled tests.
where V0 and V are the initial and current volume of the
RVE.
The damaged local volume variation is given by:
constant during deformation. Indeed, it shows a decrease
Edv Ev  Eev 5 followed by an increase due to an heterogeneous deforma-
where Eev is the elastic part of the volume variation dened tion. In order to model the local behaviour of the damaged
by: composites for a strain driven test, it is fundamental to
keep the local strain rate constant. The system used in this
R11
Eev 1  2m 6 study is able to control the local strain rate. Fig. 3b shows
E result of a test at a local constant strain rate of 4.105 s1.
In (6) m is the Poissons ratio, E is the Youngs modulus and Monitoring of the cross-head speed by the videotraction
R11 is the stress given by: system is clearly seen. This control is achieved in real time
F with the simultaneous data measurements.
R11 7 The stressstrain curves of composites conducted at a
S 0 1 E22 1 E33
local constant strain rate of 104 s1 are given in Fig. 4.
where F is the applied load on the specimen and S0 is the The strain at break of these composite specimens is of
initial cross-section. the order of 0.01. Therefore, these chopped random ber
composites are brittle in nature.
2.3. Experimental results As shown in Fig. 5, a whitening phenomenon is observed
in the central part of the specimen. This phenomenon is gen-
The local axial strain rate, during a tensile test at a con- erally associated in the literature with voids occurring in the
stant cross-head speed of 0.2 mm/min, is displayed in material. The whitening appears because the void size is lar-
Fig. 3a as a function of time. The global strain rate is deter- ger than the light wavelength (about 0.6 lm). It provides an
mined from the cross-head speed and the initial grip to grip optical verication of nucleation and propagation of voids in
distance (80 mm). It is clear that the local strain rate is not the material.
A. Bouaziz et al. / Composites Science and Technology 67 (2007) 32783285 3281

80 0.012 GFRP-21R, there is a small increase and decrease of the


damaged volume strain at the beginning of the loading; this
measure has however no meaning since it is in the scatter-

damaged volume variation


60 0.009 ing range of the measurement system. The incubation
strain depends on ber volume fraction since it is equal
stress (MPa)

to 0.004 for GFRP-21R and 0.0055 for GFRP-13R. In this


40 0.006 incubation period, the volume variation is associated to the
purely elastic dilatation due to the Poisson eect. Subse-
quently, an exponential damage increase is observed and
20 0.003 the behaviour is approximately the same for the two mate-
rials. The scaling of the volume strain depends on ber vol-
ume fraction. For an axial strain of 0.01, a damaged
0 0 volume variation of 0.007 for GFRP-21R and 0.005 for
0 0.003 0.006 0.009 0.012 GFRP-13R is found.
strain In order to identify the local damage mechanisms and
80 0.012
the location of voids, microscopic observations on fracture
surfaces at the RVE were achieved by SEM. Fig. 6 shows
that the material contains numerous voids. These voids
damaged volume variation

60 0.009
are localised around the bers. The origin of these voids
is attributed to the interfacial damage which seems to be
stress (MPa)

the predominant microscopic damage mechanism in the


composite. However, ber fracture was also seen. Due to
40 0.006
high hydrostatic stress, the region of broken bers is also
the site of void nucleation in the matrix. Furthermore, no
signicant matrix cracking was observed.
20 0.003

3. Modelling
0 0
0 0.003 0.006 0.009 0.012
3.1. Constitutive relationships
strain
Traditional continuum mechanics cannot directly pre-
Fig. 4. Experimental stressstrain and damaged volume strain curves for: dict the correct behaviour of the composite since such an
(a) GFRP-21R and (b) GFRP-13R.
approach is based on continuity and homogeneity of the
material. Indeed the concept of micromechanics is required
to perform a rigorous analysis of the material and to pre-
dict the eect of microstructure on its overall mechanical
properties. The constitutive law of the homogeneous med-
ium equivalent to the heterogeneous material is established
from a micromechanical modelling, taking into account

Fig. 5. Photograph of a fractured GFRP-21R specimen.

The damaged volume variation is presented as a func-


tion of axial strain in Fig. 4. The nucleation and growth
stages of microvoiding are both revealed in this gure. It
is evident that there exists an incubation strain where
no damaged volume strain is observed. Note that for Fig. 6. SEM photograph of the fractured surface for GFRP-21R.
3282 A. Bouaziz et al. / Composites Science and Technology 67 (2007) 32783285

both the existence of discontinuous randomly dispersed 1


and oriented glass bers, and damage evolution.
Chopped glass bers are assumed to be elastic spheroids
(prolate: a > 1) randomly dispersed and oriented in the
elastic polymermatrix. As the spheroidal bers are in the 1
12 plane, the composite as a whole is transversely
isotropic.
The constitutive equations of the equivalent homoge-
neous medium is dened by: 2

RC:E 8
where R and E are the macroscopic stress and macroscopic
strain tensors, respectively, and C is the fourth-rank elastic
stiness tensor of the composite polymer (:signies the
tensor contraction).
The volume-averaged stress R and strain E tensors are 2
given respectively by: Fig. 7. Local and global coordinates for a composite containing randomly
Z Z located and oriented spheroidal bers.
1 1
R rxdV and E exdV 9
V V V V
where r and e are the microscopic stress and microscopic where the transformation matrix Qij is given by:
strain tensors, respectively and V is the volume of a RVE. 2 3
The ber-reinforced composite presents initially perfect cos h sin h 0
interfacial bonding between the matrix (phase 0) and bers 6 7
Qij 4  sin h cos h 0 5 13
(phase 1). After its nucleation, damage (phase 2) is taken 0 0 1
into account as an added phase in the composite. There-
fore, the initial two-phase composite becomes a three-phase In (13), h (0 6 h 6 p) is the angle between x1 and x01 .
composite from the onset of damage nucleation. According to the studies of Tandon and Weng [14] and
According to the explicit formulation derived by Ju and Lee and Simunovic [7], the orientational averaging process
Chen [12] for unidirectionally aligned spheroid-reinforced for all orientations for the macroscopic stiness tensor can
composites, the macroscopic stiness tensor of the equiva- be written as:
lent homogeneous medium is dened by the following Z p
expression: hCi Qmi Qnj C mnpq Qpk Qql ph sin hdh 14
0
" #
X 2
1 where p(h) is the probability density function of the ran-
C C0 I Br I  Sr Br 10
dom orientation which is equal to 1/p for uniformly ran-
r1
dom orientation.
where I is the fourth-rank identity tensor, C0 is the elastic- In order to achieve more realistic behaviour predictions,
ity tensor of the matrix material, Sr is the Eshelbys tensor the damage experimentally identied must be included in
for phase r and Br is a fourth-rank tensor dened by: the modelling. Based on experimental evidence, the deb-
1 1 onding along the interface between the matrix and bers
Br /r Sr Cr  C0 C0  11
is the major failure mechanism in this material. The inter-
where Cr and /r are respectively, the elasticity tensor and facial debonding under increasing deformation leads to the
the volume fraction of the phase r. creation of new surfaces inducing void volume variation in
The Eshelbys tensor Sr is a function of the Poissons the material. The increase of the void volume fraction is
ratio of the matrix and the aspect ratio of the phase r. Its assumed to be a coupled eect of nucleation of voids and
formulation for a spheroidal inclusion embedded in an iso- growth of these voids. The rate of void volume fraction
tropic linear elastic medium can be found in the literature is governed by the following evolution relation:
[8,13]. /_ 2 /_ nucl /_ grow 15
Since for the studied material, the bers are randomly
oriented in the 12 plane, it is necessary to introduce global where /_ nucl is the nucleation rate of voids and /_ grow is the
axes denoted by the unprimed axes and local axes of each growth rate of existing voids.
bers denoted by the primed axes (Fig. 7). The random ori- The nucleation rate, linked to the local hydrostatic stress
entation of glass bers can then be described by introduc- applied on the ber, is expressed in a simple empirical way.
ing the angle h between the primed and unprimed axes: The progressive partial debonding between bers and the
matrix can be taken into account with the Weibull statisti-
xi Qij x0j 12 cal function [15]:
A. Bouaziz et al. / Composites Science and Technology 67 (2007) 32783285 3283
  m 
r1h where:
P d r1h 1  exp  16
ru Dt _
D/2 /2t /_ 2tDt  23
where r1h is the average internal hydrostatic stress of the 2
bers, m and ru are the Weibull parameters to be estimated The hydrostatic stress inside the bers is computed accord-
from experiments. m controls the shape of the Weibull ing to (17) and the current volume fraction of bonded bers
function and ru its scale. is updated in the following way:
From the Eshelby inclusion theory, the stress inside
/1tDt /  /dtDt 24
bers is assumed uniform and is given by [7]:
!1 The volume fractions of damage (22) and bonded bers
X
2
1
r1 C1 I  S1 B1 /1 I  Sr Br :E 17 (24) are updated at each time increment step and are used
r1 to compute the progressive evolution of the stiness tensor.
After computation of the current stiness tensor, the stres-
The cumulative volume fraction of debonded bers is then ses are updated.
given by: Considering the width (23 mm) of the specimens used in
/d /P d r1h 18 the mechanical tests (Fig. 1), the geometrical distribution
of the bers (nominal length of bers is 50 mm giving
where / is the original volume fraction of bonded bers.
a = 100) is expected to be inhomogeneous. Fig. 8 shows
After partial debonding, the debonded portion of the
the inuence of the aspect ratio a of the bers on the pre-
ber does not transmit any more stresses to the matrix.
dicted Youngs modulus of the composite. It is clear that a
The debonding process is mechanically complex because
inuences strongly the Youngs modulus for smaller values
it introduces a local induced anisotropy which is not easy
of a and its eect saturates when considering higher values
to account for. In order to simplify the problem, the deb-
(about a > 30). Therefore, it is worth assuming that all
onded part of the ber is substituted by an equivalent vol-
bers have identical geometry.
ume of matrix material. By this way, replacing elastic
mechanical properties of the debonded bers by that of
3.2. Comparison between predicted and measured behaviour
the matrix leads to a global weakening of the composite.
Moreover, the void volume fraction which is introduced
The formulation given in the previous section is now
as a third-phase, also acts as supplementary weakening fac-
applied to uniaxial loading condition. The material proper-
tor. The nucleation rate of interfacial voids by the debond-
ties of the unsaturated polyester matrix and bers are:
ing of bers and the matrix is assumed to be controlled by
E0 = 4400 MPa, m0 = 0.35, E1 = 72000 MPa, m1 = 0.25.
the stress in the bers:
The Youngs modulus and the Poissons ratio of the matrix
oP d r1h are measured from video-controlled tensile tests. The
/_ nucl c/_ d c/ r_ 1h 19
or1h aspect ratio a of bers is 100. The parameters of the voids
nucleation rate (19) are numerically determined using a
where c is a scaling coecient for void nucleation rate and least squares regression tting on the accumulated damage
is assumed to be constant. curves observed in the experiments. They are found to be:
The growth rate of the previously nucleated microvoids m = 5, ru = 118 MPa and c = 0.08. These values are taken
at the interface is given by [16]: independent of the ber volume fraction.
3
/_ grow /nucl R_ h 20
4l 8000

where l is the shear modulus of the composite. Void 7500


growth is governed by the macroscopic hydrostatic stress.
The voids are assumed to be spheroidal and to remain of
7000
modulus (MPa)

uniform size during deformation. Furthermore, the interac-


tions between the dierent phases are not taken into con-
6500
sideration in the present model.
The model is introduced into a computer algorithm
based on the strain driven scheme; the unknown macro- 6000
scopic stress history is determined by a given macroscopic GFRP-21R

strain: 5500 GFRP-13R

RtDt Rt C : DEtDt 21
5000
0 20 40 60 80 100
The incremental nonlinear equations for damage are
solved using the Trapezoidal scheme:
Fig. 8. Inuence of the shape parameter a on the predicted macroscopic
/2tDt /2t D/2 22 Youngs modulus.
3284 A. Bouaziz et al. / Composites Science and Technology 67 (2007) 32783285

The simulated stressstrain responses and damaged vol- 0.15


ume strain are given in Fig. 9. The model predictions are in

volume fraction of debonded fibers


very good agreement with experiments. However, the nal
0.12 GFRP-21R
stage of damage, leading to brutal fracture, is not well repre-
GFRP-13R
sented. In this nal stage, the simulated damage evolution
underestimates the experimental data since no threshold 0.09
stage is taken into account in the present modelling. Indeed,
during the rupture various damage mechanisms (debonding,
0.06
ber pull-out and breakage) simultaneously accumulate
leading to a sudden coalescence of voids which produce a
drastic increase of the damaged volume strain. 0.03
The volume fraction evolution of debonded bers pre-
dicted by the model is displayed in Fig. 10 for the two volume
0
fractions of bers. The case without damage representing an 0 0.003 0.006 0.009 0.012 0.015
upper bound and the damaged case are shown in Fig. 11. The strain
weakening eect due to damage can be clearly seen. This
Fig. 10. Predicted volume fraction of debonded bers.
weakening is more emphasized for GFRP-21R than
GFRP-13R.

4. Conclusion 140
GFRP-21R without damage
GFRP-21R with damage
120
The damage behaviour of chopped random glass ber GFRP-13R without damage
unsaturated polyester matrix composites was investigated. 100
GFRP-13R with damage
stress (MPa)

80 0.012 80

experiment 60
damaged volume variation

60 modelling 0.009
40
stress (MPa)

20
40 0.006
0
0 0.003 0.006 0.009 0.012 0.015
strain
20 0.003
Fig. 11. Eect of damage on the overall behaviour.

0 0
The experimental results have demonstrated the impor-
0 0.003 0.006 0.009 0.012
strain tance of taking into consideration the volume strain in
the accurate characterization of composites damage, espe-
80 0.012 cially for the composites having a weak interfacial bond
experiment strength. The experimental results were used for the valida-
modelling tion of a micromechanical model based on the ensemble-
damaged volume variation

60 0.009 volume averaged homogenization procedure introduced


by Lee and Simunovic [7]. The damage evolution law was
stress (MPa)

incorporated into the modelling in terms of nucleation


40 0.006 and growth of voids. The predictions of the micromechan-
ical damage model were found in good agreement with the
measured behaviour.
20 0.003 The modelling is clearly able to connect the complex
microstructure and the dissipative mechanisms with the
macroscopic behaviour of the material. Comparison with
0 0 experiments was done only for in-plane mechanical behav-
0 0.003 0.006 0.009 0.012 iour. Therefore, data collection with other mechanical tests
strain than uniaxial tension is required. Furthermore, the micro-
Fig. 9. Predicted stressstrain and damaged volume strain curves mechanical constitutive model is going to be implemented
compared to experimental data for: (a) GFRP-21R and (b) GFRP-13R. into a nite element code.
A. Bouaziz et al. / Composites Science and Technology 67 (2007) 32783285 3285

References [8] Lee HK, Simunovic S. A damage constitutive model of progressive


debonding in aligned discontinuous ber composites. Int J Solids
[1] Meraghni F, Benzeggagh ML. Micromechanical modeling of matrix Struct 2001;38:87595.
degradation in randomly discontinuous-bre composites. Comp Sci [9] Lee HK, Simunovic S. A damage mechanics model of crack-
Tech 1995;55:17186. weakened, chopped ber composites under impact loading. Comp
[2] Meraghni F, Blakeman CJ, Benzeggagh ML. Eect of interfacial PartB Eng 2002;33:2534.
decohesion on stiness reduction in a random discontinuous-bre [10] Francois P, Gloaguen JM, Hue B, Lefebvre JM. Volume strain
composite containing matrix microcracks. Comp Sci Tech measurements by optical extensometry: application to the tensile
1996;56:54155. behaviour of RT-PMMA. J Phys III 1994;4:3219.
[3] Meraghni F, Desrumaux F, Benzeggagh ML. Implementation of a [11] GSell C, Hiver JM, Dahoun A. Experimental characterization
constitutive micromechanical model for damage analysis in glass mat of deformation damage in solid polymers under tension, and
reinforced composite structures. Comp Sci Tech 2002;62:208797. its interrelation with necking. Int J Solids Struct 2002;39:
[4] Fitoussi J, Bourgeois N, Guo G, Baptiste D. Prediction of the 385772.
anisotropic damaged behavior of composite materials: introduction [12] Ju JW, Chen TM. Micromechanics and eective moduli of elastic
of multilocal failure criteria in a micro-macro relationship. Comp composites containing randomly dispersed ellipsoidal inhomogenei-
Mater Sci 1996;5:87100. ties. Acta Mech 1994;103:10321.
[5] Fitoussi J, Guo G, Baptiste D. A statistical micromechanical model of [13] Ju JW, Sun LZ. A novel formulation for the exterior-point
anisotropic damage for SMC composites. Comp Sci Tech Eshelbys tensor of an ellipsoidal inclusion. J Appl Mech 1999;66:
1998;58:75963. 5704.
[6] Derrien K, Fitoussi J, Guo G, Baptiste D. Prediction of the eective [14] Tandon GP, Weng GJ. Average stress in the matrix and eective
damage properties and failure properties of nonlinear anisotropic moduli of randomly oriented composites. Comp Sci Tech 1986;27:
discontinuous reinforced composites. Comp Meth Appl Mech Eng 11132.
2000;185:93107. [15] Weibull W. A statistical distribution function of wide applicability. J
[7] Lee HK, Simunovic S. Modeling of progressive damage in aligned Appl Mech 1951;18:2937.
and randomly oriented discontinuous ber polymermatrix compos- [16] Steenbrink AC, Van der Giessen E, Wu PD. Void growth in glassy
ites. Comp PartB Eng 2000;31:7786. polymers. J Mech Phys Solids 1997;45:40537.