Vous êtes sur la page 1sur 35

325

CHAPTER 4.1: MODELING OF DAMAGE IN FIBER AND PARTICLE REINFORCED


COMPOSITES
Wolfgang Lutz, Ming Dong, Ke Zhu, and Siegfried Schmauder
CONTENTS
1 Introduction 325
2 Damage Phenomena in Short Fiber rein-
forced Composites 326
3 Overview of Damage Modeling in Com-
posites 327
4 Modeling 329
4.1 Self-Consistent Model (SCM) . . . 329
4.2 Combined Cell Model (CCM) . . . 331
4.3 Statistical Combined Cell Models . 333
4.3.1 Static Loading Conditions . 333
4.3.2 Quasi-Static Cyclic Load-
ing Conditions . . . . . . . 334
4.4 Plastic-Damage Model . . . . . . . 335
4.4.1 Modications of the Plastic-
Damage Model . . . . . . . 335
5 Results and Application 336
5.1 Metal Matrix Composites (MMC) . 336
5.1.1 Material . . . . . . . . . . . 337
5.1.2 Results: Self-Consistent
Model . . . . . . . . . . . . 337
5.1.3 Results: Combined Cell
Model . . . . . . . . . . . . 339
5.1.4 Results: Statistical Com-
bined Cell Model . . . . . . 341
5.1.5 Conclusions . . . . . . . . . 342
5.2 Polymer Matrix Composites (PMCs) 343
5.2.1 Material . . . . . . . . . . . 343
5.2.2 Results: Combined Cell
Model (CCM) . . . . . . . 343
5.2.3 Results: Statistical Com-
bined Cell Model . . . . . . 345
5.2.4 Conclusions . . . . . . . . . 349
5.3 Gypsum Fiber Composites . . . . . 349
5.3.1 Material . . . . . . . . . . . 349
5.3.2 Results: Plastic-Damage
Model . . . . . . . . . . . . 350
5.3.3 Conclusions . . . . . . . . . 355
6 Summary 355
1. INTRODUCTION
Composites such as Metal Matrix Composites
(MMCs) and Polymer Matrix Composites (PMCs)
are frequently reinforced with strong continuous
or short bers. In the case of short ber rein-
forced MMCs and PMCs, random arrangements of
bers are observed. Their mechanical properties are
highly dependent on their composition, on the ma-
trix properties as well as on the type and volume
fraction of reinforcements. The complexity of such
affecting parameters makes a complete theoretical
description of the behavior and the failure proper-
ties of reinforced composite with ductile and brittle
matrix difcult. In this respect, a micromechanical
analysis of the local composite failure process opens
a possibility to predict the failure properties of these
composites [13]. Micromechanical models shown
in this Chapter can be applied to describe most tech-
nical relevant composites varying from simple in-
clusion type and interpenetrating microstructures to
functionally graded materials.
In this Chapter, models to simulate damage in
ber and particle reinforced composites with duc-
tile (metal or polymer matrix) or brittle (gypsum)
matrix are summarized on the basis of the works
of Dong et al. [46], Zhu et al. [7, 8], Kabir et
al. [9], Lutz et al. [10] and Rahman et al. [11].
They are the Self-consistent Model, the Combined
Cell Model, the Statistical Combined Cell Model
and the plastic-damage model. The embedded cell
of the self-consistent model represents a composite
where, instead of using xed or symmetric bound-
ary conditions around the ber-matrix or particle-
matrix cell, the inclusion-matrix cell is embedded in
326 DAMAGE SIMULATION
an equivalent composite material with the mechan-
ical behavior to be determined iteratively in a self-
consistent manner. Using the Combined Cell Model
in conjunction with the nite element method, the
mechanical behavior of composites with a certain
orientation of the bers can be simulated numeri-
cally by averaging results from different 2D and 3D
cell models representing a single ber in three prin-
cipal orthogonal planes in the composite. Apply-
ing an appropriate integration of the results of all
ber orientations, stress-strain curves in tension and
compression of the global material can be simulated
including the effects of residual stresses. As an ad-
vancement, the Statistical Combined Cell Model has
been developed to consider ber-cracks and ber-
matrix debonding using a Weibull statistical ap-
proach and the rule of mixture. The parameters of
the Weibull damage law have been determined us-
ing inverse modeling by comparing simulation and
experiment. The Statistical Combined Cell Model
was applied to static and cyclic loading conditions.
The plastic-damage model includes stiffness degra-
dation due to damage in the plasticity part by two
independent scalar damage parameters, for tension
and compression respectively. The effect of damage
is introduced by replacing all stress denitions (true
stress) by the reduced effective stress. Further on,
the plastic-damage model is based on a stiffness re-
covery scheme to simulate the effect of micro-crack
opening and closing. After introducing the models
and the methods to simulate damage evolution, the
presented models will be applied to different com-
posites: MMCs, PMCs, and cellulose ber rein-
forced gypsum composites. There, investigated ma-
terials will be explained. Moreover, an approach is
presented to apply the Combined Cell Model to re-
alistic microstructures of an injection molded PMC
sample.
2. DAMAGE PHENOMENA IN SHORT
FIBER REINFORCED COMPOSITES
Failure of ber or particle reinforced composites is
generally preceded by an accumulation of differ-
ent types of internal damages. During the damage
process in composite materials, the following phe-
nomena are known: formation and growth of micro-
cracks or voids, clustering, coalescence, formation
and growth of initial cracks, and propagation of one
of the cracks up to the failure of the specimen [12].
These steps depend on the type of reinforcing bers
or particles (inclusion), and on the interface between
inclusion and matrix. They vary with the type of
loading. For instance, in an early stage of load-
ing debonding appears if there is a weak inclusion-
matrix interface. Stress transfer from the matrix to
the inclusion in a composite takes place by shear
at the inclusion-matrix interface. An important as-
pect is the loading direction relative to the orienta-
tion of the inclusion. For example, in a PMC with a
glass ber, debonding occurs preferentially if load-
ing is applied perpendicular to the ber orientation.
If loading is applied in ber direction, bers will fail
or will be pulled out after reaching a certain criti-
cal load. Further on, in ber reinforced materials,
bers can exhibit a stitching action on the micro-
cracks preventing them from propagating. This in-
tervention retards the creation of micro-cracks lead-
ing to an overall improvement of the fracture resis-
tance [13].
Strong interfaces result in high strength and stiff-
ness, but low fracture toughness. On the other hand,
weak interfaces promote deection of matrix cracks
along the interface and lead to high fracture tough-
ness, but low strength and stiffness of composites.
The process of transfer of load between bers and
matrix in the neighborhood of a ber break or a
matrix crack depends on the strength of the inter-
face. Although ber and matrix can be character-
ized by conducting simple tests, interface proper-
ties are most difcult to determine. Interfacial shear
strength is an important parameter that controls the
inclusion-matrix debonding process [14].
Mechanical fatigue is the most common type of fail-
ure of structures. It is dened as the failure of
a component under the repeated application of a
stress smaller than that required to cause failure in
a single application. In fatigue, a crack is initiated
and slowly grows under the action of the uctuat-
ing stress until, eventually, failure occurs in a catas-
trophic manner with no great distortion preceding
the event. To understand fatigue damage of ber re-
inforced composites, a simple unidirectional com-
posite loaded in tension parallel to the bers is dis-
cussed in the following. If ber breakage occurs
when the local stress exceeds the strength of the
weakest ber, this causes shear stresses concentra-
tion at the ber-matrix interface near the broken
ber tip. The interface area acts as a stress con-
centrator for the longitudinal tensile stress, which
may exceed the fracture stress of the matrix, lead-
ing to transverse cracks in the matrix. These cracks
can be randomly distributed [15]. With the devel-
opment of the fatigue process, the local strains ex-
ceed a certain threshold, resulting eventually in ber
breakage and propagation of matrix cracks. During
matrix crack propagation, the ber-matrix interface
will also fail due to severe shear stress at the crack
tip. The nal failure occurs when a sufciently large
crack has developed. The lower strain limit for the
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 327
matrix is the threshold strain below which the ma-
trix cracks remain arrested by the bers. This strain
is observed to be approximately the fatigue strain
limit of the unreinforced matrix material. The upper
limit is given by the strain to failure of the compos-
ite, which is the strain to failure of the reinforcing
bers. The progressive damage mechanism is ma-
trix cracking with associated interfacial shear failure
and this governs the fatigue life. When bers are ar-
ranged perpendicular to the loading direction, dam-
age mechanisms are slightly different (similar to the
static case). Here, the lowest failure limit is given by
transverse ber-matrix debonding which is strongly
connected to the ber orientation angle . This is
reected in the fatigue limit strain [16].
A study of the fatigue damage mechanisms gives
indications of the weakest microstructural element,
which is a useful information in the selection of ma-
terials for improvement in service properties. Gen-
erally, PMCs possess weak interfaces, and fatigue
failure occurs by distributed debonding or longitudi-
nal matrix cracking followed by further ber break-
age. Macroscopically, the weak interface compos-
ites show shorter fatigue lives and more rapid fa-
tigue degradation. Fatigue damage can be studied
on a macroscopic and microscopic scale.
There are several differences between the fatigue be-
havior of metals and of ber reinforced composites.
In metals, the stage of gradual and invisible dete-
rioration spans nearly the complete lifetime during
service conditions. No signicant reduction of stiff-
ness is observed during the fatigue process. The
nal stage of the process starts with the formation
of small cracks, which are the only form of macro-
scopically observable damage. Gradual growth and
coalescence of these cracks quickly produce large
cracks and nal failure of the structural component.
As the stiffness of a metal remains quasi unaffected,
the linear relation between stress and strain remains
valid, and the fatigue process can be simulated in
most common cases by a linear elastic analysis and
linear fracture mechanics. In a ber reinforced com-
posite, damage starts very early and the extent of
the damage zones grows steadily, while the damage
type in these zones change (e.g., small matrix cracks
leading to large size delaminations). The radial dete-
rioration of a ber reinforced composite with a loss
of stiffness in the damaged zones leads to a contin-
uous redistribution of stresses and to a reduction of
stress concentrations inside a structural component.
As a consequence, an estimation of the actual state
or a prediction of the nal state (when and where -
nal failure is to be expected) requires the simulation
of the complete path of successive damage states
[16].
3. OVERVIEW OF DAMAGE MODELING IN
COMPOSITES
Many of the established models only consider one
or two of the above mentioned damage mechanisms.
Using nite element method (FEM) to model dam-
age, requires specic approaches to solve the dis-
crepancies between the quasi-continuum statement
of a problem and the random and discontinuous na-
ture of crack growth [12].
The unit cell approach is often used to simulate the
initiation of damage. Bao [17] uses a three phase
damage cell model taking into account the failure of
particles and particle-matrix debonding to simulate
strength and creep resistance of metals such as Al
and Ti reinforced with Al
2
O
3
. The deformation of
particle and whisker reinforced MMCs was investi-
gated by Llorca et al. applying cylindrical unit cells
to obtain the overall stress-strain behavior [18]. Ax-
isymmetric unit cells were used by Walter to model
damage initiation in ber reinforced composites by
cohesive elements [19]. The stress triaxiality and the
shape of voids were taken into account by Brocks et
al. to simulate effective stress vs. strain curves [20].
Thereafter, the relevant parameters of the Gurson-
Tvergaard-Needleman damage model were deter-
mined. 3D hexagonal cells were used by Sun et
al. to model the inuence of micro-crack densities
on the creep behavior of ferritic steels [21]. Also,
weak interfaces of polymer specimens were simu-
lated by cylindric unit cells containing a rigid par-
ticle [22, 23]. It is possible to include the effect
of particle and ber failure, the particle/ber-matrix
debonding by unit cell models. One basic assump-
tion of many (non self-consistent) unit cells is the
uniformdistribution of the inclusions and, therefore,
of the damage [12].
After a general introduction of unit cell models, the
Self-consistent Model, the (Statistical) Combined
Cell Model and the plastic-damage model will be
presented to simulate the mechanical behavior of
different ber and particle reinforced composites.
Initially, the mechanical behavior of a unidirec-
tionally continuous ber reinforced composite with
bers of circular cross-section was studied by
Adams [24] adopting nite element cell models
under plane strain conditions: a simple geometri-
cal cell composed of matrix and inclusion material
is repeated by appropriate boundary conditions to
represent a composite with a periodic microstruc-
ture. The inuence of different regular ber ar-
rangements on the strength of transversely loaded
boron ber reinforced Aluminum was analyzed in
[2527]. It was found that the square arrangement
328 DAMAGE SIMULATION
of bers represents two extremes of the strengthen-
ing: high strength levels are achieved if the com-
posite is loaded in a 0

direction of nearest neigh-


bors while the 45

loading direction is found to be


very weak for the same ber arrangement. A regular
hexagonal ber arrangement lies between these lim-
its [2529]. The transverse mechanical behavior of a
realistic ber reinforced composite containing about
thirty randomly arranged bers was found to be best
described but signicantly underestimated by the
hexagonal ber model [26]. Dietrich [30] found a
transversely isotropic square ber reinforced Ag/Ni
composite material using bers of different diame-
ters. A systematic study in which the ber volume
fraction and the ber arrangement effects have been
investigated, was founded into a simple model in
[29].
The inuence of ber shape and clustering was nu-
merically examined by Llorca et al. [18], Diet-
rich [30], and Sautter [31]. It was observed that
facetted ber cross-sections lead to higher strengths
compared to circular cross-sections except for bers
which possess predominantly facets with an angle of
45

with respect to the loading axis in close agree-


ment with ndings in particle reinforced MMCs
[32]. Thus, hindering of shear band formation
within the matrix was found to be responsible for
strengthening with respect to ber arrangement and
ber shape [18]. In [29, 3335] local distribu-
tions of stresses and strains within the microstruc-
ture have been identied to be also strongly inu-
enced by the arrangement of bers. However, no
agreement was found between the mechanical be-
havior of composites based on cell models with dif-
ferently arranged bers and experiments with ran-
domly arranged bers loaded in transverse direction.
The overall mechanical behavior of a particle rein-
forced composites was studied with axisymmetric -
nite element cell models by Bao et al. [36] to repre-
sent a uniformparticle distribution within an elastic-
plastic matrix. Tvergaard [37] introduced a modi-
ed cylindrical unit cell containing one half of a sin-
gle ber to model the axial performance of a peri-
odic square arrangement of staggered short bers.
Hom [38] and Weissenbek [39] have used three-
dimensional nite elements to model different regu-
lar arrangements of short bers and spherical as well
as cylindrical particles with relatively small volume
fractions ( f < 0.2). It was generally found that the
arrangement of bers strongly inuences the dif-
ferent overall behavior of composites. When short
bers are arranged in a side-by-side manner, they
constrain the plastic ow in the matrix and the com-
puted stress-strain response of the composite in the
ber direction is stiffer than that observed in exper-
iments. If the bers in the model are overlapping,
between neighboring bers strong plastic shearing
can develop in the ligament and the predicted load
carrying capacity of the composite is closer to the
experimental measurements.
The inuence of thermal residual stresses in ber
reinforced MMCs under transverse tension is stud-
ied in [27] and found to lead to signicant strength-
ening elevations in contrast to ndings in partic-
ulate reinforced MMCs where strength reductions
were calculated [40]. A limited study on the over-
all limit ow stress for composites with randomly
oriented disk-like or needle-like particles arranged
in a packet-like morphology is reported by Bao et al.
[36]. In [41, 42] a modied Oldroyd model has been
proposed to investigate analytically-numerically the
overall behavior of MMCs with randomly arranged
brittle particles. Duva [43] has introduced an ana-
lytical self-consistent model to represent a random
distribution of non-interacting rigid spherical parti-
cles perfectly bonded in a power law matrix in the
dilute regime of volume fractions of f < 0.2.
Composites with randomly arranged inclusions can
be modeled by a self-consistent procedure with em-
bedded cell models. This method of surrounding
a simulation cell by additional equivalent compos-
ite material was introduced in [30, 33] for struc-
tures which are periodical in loading direction, and
recently extended to non-periodic two-dimensional
[4, 44, 45] and three-dimensional composites [31,
42]. One reason for the discrepancy between exper-
iments and calculations based on simple cell mod-
els is believed to be the unnatural constraint gov-
erning the matrix material between inclusion and
simulation cell border [26, 34, 3638, 4648] re-
sulting in an unrealistic strength increase. Embed-
ded cell models are known to remove the unrealistic
constraints of the simple models described above.
An initial comparison of two- and three-dimensional
embedded cell models in case of perfectly plastic
matrix material depicts elevated strength levels for
the three-dimensional case [42] as it happened for
composites with regularly arranged bers [29].
For aligned short ber reinforced composites, some
approaches to determine the mechanical behavior
have been introduced in [49, 50] by considering two
geometrical aspects: cross-section along ber and
cross-section in transverse plane, which lead to dif-
ferent cell models for arrays of end-to-end aligned
short bers, axially clustered short bers, trans-
versely clustered arrays of short bers or misaligned
short bers. In case of short bers with small as-
pect ratio, periodic cell models can be used [49, 50]
to follow morphological effects, especially to de-
scribe in-plane misaligned short ceramic bers (SiC-
whisker) with small misorientation angles in a single
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 329
cell model. A further analysis is presented in [51],
where the Duvas model is applied to calculate the
overall ow behavior of short ber reinforced com-
posites. The effects of ber orientation, which inu-
ences the mechanical deformation behavior and the
ber damage behavior, have also been discussed in
[51]. Unit cell models have been applied in [52] to
analyze residual stress effects on uniaxial deforma-
tion of whisker reinforced Metal Matrix Compos-
ites, where models with different ber aspect ratios
are employed to predict the overall ow behavior of
short ber reinforced MMCs. Comparisons between
experimental and numerical results on two compos-
ites demonstrate that it is not possible to use a single
unit cell to predict the mechanical behavior of ran-
domly oriented short ber reinforced MMCs. A rep-
resentative volume with different ber orientations
is described in a two-dimensional model in [53] to
determine the mechanical behavior of composites
with randomly oriented bers, where the bers are
again very short (aspect ratio 2/1). A microme-
chanical model has been introduced in [54, 55] for
short ber reinforced aluminum alloys. There, three
elementary microstructural mathematical processes
were taken into account to investigate the creep be-
havior of these MMCs, without considering the mor-
phological aspect of the bers.
4. MODELING
In this section several models will be presented,
to simulate the mechanical behavior of composites.
The self-consistent and the (Statistical) Combined
Cell Models are indicative of unit cell approaches.
As a last example the plastic-damage model is intro-
duced, which can describe the homogenized consti-
tutive behavior of ber reinforced quasi-brittle ma-
terials. In Sec. 5, these four models will be then
applied to ber and particle reinforced composites
with metal or polymer matrix and a cellulose ber
gypsum composite. There, also the potential and
limitations of these models to simulate the specic
mechanical behavior will be discussed.
4.1. Self-Consistent Model (SCM)
For composites reinforced with aligned continu-
ous bers or with spherical particles, simple unit
cells with single inclusions can be taken from a
representative cross-section in a transverse plane
[4, 24, 26, 30, 34, 56] or from a cross-section along
the loading axis [5, 36, 40]. In the present section,
2D and 3D self-consistent embedded cell models
will be applied to model the mechanical behavior of
composites with random continuous ber and par-
ticle arrangements. Fig. 1 describes schematically
a typical plane strain (2D) embedded cell model
with a volume fraction f = (r/R)
2
or axisymmetric
(3D) embedded cell model with a volume fraction
f = (r/R)
3
. Here, instead of using xed or symme-
Figure 1. Embedded cell model with nite element
mesh [4].
try boundary conditions around the ber-matrix or
particle-matrix cell, the inclusion-matrix cell is em-
bedded in an equivalent composite material with the
mechanical behavior to be determined iteratively in
a self-consistent manner. If the dimension of the em-
bedding composite is sufciently large compared to
that of the embedded cell, e.g., L/R = 5, the exter-
nal geometrical boundary conditions introduced for
the embedding composite are almost without inu-
ence on the composite behavior of the inner embed-
ded cell. A typical FE mesh and corresponding sym-
metry and boundary conditions are given in Fig. 1,
where a circular ber or a spherical particle is sur-
rounded by a circular (for 2D) or spherical (for 3D)
shaped matrix, which is again embedded in the com-
posite material with the mechanical behavior to be
determined.
The ow stresses for transverse loading of Metal
Matrix Composites reinforced with continuous
bers and for uniaxial loading of spherical parti-
cle reinforced metal-matrix, composites were in-
vestigated in previous studies using embedded cell
models [4, 5]. These works describe a ber or a
spherical particle as surrounded by a metal-matrix,
which is again embedded in the composite material
330 DAMAGE SIMULATION
with the mechanical behavior to be determined it-
eratively in a self-consistent manner. It has been
veried in [4, 5] that such a self-consistent embed-
ded cell method is appropriate to represent Metal
Matrix Composites with randomly arranged parti-
cles or aligned continuous bers. The inclusion be-
haves elastically and its stiffness is much higher than
that of the matrix. In addition, the continuous bers
of circular cross-section and spherical particles are
assumed to be well bonded to the matrix so that
debonding or sliding at the inclusion-matrix inter-
face is not permitted. The uniaxial matrix stress-
strain behavior is described by a Ramberg-Osgood
type of power law [5].
The global mechanical response of the composite
under external loading is characterized by the over-
all stress as a function of the overall strain .
Moreover, to describe the results in a consistent way,
the reference axial yield stress
0
and yield strain

0
of the matrix (as dened in Eq. 1 for the 3D
case) will be taken to normalize the overall stress
and strain of the composite, respectively.
Following Bao et al. [36], the composite contain-
ing hard inclusions will necessarily harden with the
same strain hardening exponent N, as the matrix for
the case of hard inclusions when strains are in the
regime of fully developed plastic ow. At suf-
ciently large strains the composite behavior is then
described by
=
N
_

0
_
N
, (1)
where
N
is the asymptotic reference stress of the
composite, which can be determined by normalizing
the composite stress by the stress in the matrix at the
same overall strain , as indicated in Eq. 2 and Fig. 2:

N
=
0
_
( )
( )
_
for >>
0
. (2)
For a matrix of strain hardening capability N, the
limit value
N
_

0
is dened as composite strength-
ening level, which is an important value to describe
the mechanical behavior of composites. This value
depends only on ber and particle arrangement, in-
clusion volume fraction and matrix strain-hardening
exponent. Under axial deformation at the external
boundary, the overall response of the inner embed-
ded cell can be obtained by averaging the stresses
and strains at the boundary between the embedded
cell and the surrounding volume.
The embedding method is a self-consistent proce-
dure, which requires several iterations, as shown in
Figure 2. Composite strengthening [5].
Fig. 3. An initially assumed stress-strain curve (it-
eration 0 in Fig. 3) is rst assigned to the embed-
ding composite, in order to performthe rst iteration
step. An improved stress-strain curve of the com-
posite (iteration 1) will be obtained by analyzing
the average mechanical response of the embedded
cell. This procedure is repeated until the calculated
stress-strain curve from the embedded cell is almost
identical to that of the previous iteration. The con-
vergence of the iteration occurs typically at the fth
iteration step, as illustrated in Fig. 3. It has been
Figure 3. Iterative modeling procedure: stress-
strain curves for different iteration steps[5].
found from systematic studies that convergence of
the iteration to the nal stress-strain curve of the
composite is independent of the initial mechanical
behavior of the embedding composite (iteration 0).
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 331
4.2. Combined Cell Model (CCM)
In this section, CCMs are presented to simulate the
overall ow behavior of composites reinforced with
discontinuous short bers. These cell models in-
volve two 2D models and two 3D models repre-
senting a single ber in three principal orthogonal
planes of the local system in a composite. The over-
all ow behavior of the composites will be predicted
with in-plane randomly oriented and 3D randomly
arranged short bers by an appropriate integration
over all ber orientations.
Two different kinds of ber orientations are mod-
eled: in-plane random(2D random) and 3D random.
In Sec. 5.2 also composites with aligned and layered
ber orientations are considered. In the case of in-
plane randomorientation the bers are distributed in
preferred parallel planes as illustrated schematically
in Fig. 4a, whereas, in the other case, the bers lie
randomly in all directions of space without any pre-
ferred direction and plane, as shown in Fig. 4b. In
Figure 4. Schematics of the composites with a) in-
plane 2D random and b) 3D random ber orienta-
tions [6].
Fig. 5a, a global (X,Y, Z) and a local (x, y, z) coordi-
nate system are introduced. The Z and z-axis are
oriented in the loading direction. The local (x, y, z)
coordinate system for each single short ber in the
composites is dened in such a way that the ber of
length L lies in the yz-plane. The orientation of the
ber is then dened as the angle between the ber
direction and the applied loading direction. The lo-
cation of a ber can then be described as a vector L
in both coordinate systems:
L = L(X,Y, Z) = (Lsinsin,
Lcossin, Lcos ) (3)
L = L(x, y, z) = (0, Lsin, Lcos ).
In the local coordinate system, the ber orienta-
tion with respect to the loading direction is simply
Figure 5. (a) Global and local coordinate systems
for a ber with an orientation angle ; (b) three
principal orthogonal cross-section planes for a ber
in local coordinate system [6].
described by only one characteristic parameter ,
which denes the orientation of the ber with re-
spect to the external loading direction.
Because of the high aspect ratio and the different
orientations of the bers, it is difcult to use a con-
ventional FE unit cell model approach to represent
the deformation and plastic ow behavior of com-
posites. In order to get an appropriate cell model ap-
proach, a single ber of an orientation angle with
respect to the applied loading axis is considered in
three principal orthogonal cross-sections (plane A,
B and C in Figs. 5b and 6) in the local xyz-system of
a composite. These three planes are parallel and per-
Figure 6. Construction of cell models from the pro-
jections of a ber with an orientation angle on
three principal orthogonal cross-section planes A,
B and C [6].
pendicular to the loading direction and in the local
xyz-system, so they can characterize the ber ori-
entation in a simple and direct way. On the plane
A in Fig. 6, which is built up by the ber direc-
tion and the applied loading direction, the orienta-
tion and the geometrical size (length L and diame-
ter d) of the ber can be represented by a rectangle
(L x d) with an orientation angle . In the planes B
and C in Fig. 6, the ber is represented by its cuts of
332 DAMAGE SIMULATION
two ellipses, one with a minor axis of d (diameter of
the ber) and major axis of d/sin on plane B, and
another one with a minor axis of d and major axis
of d/cos on plane C. The major axes vary with
the ber orientation in the local xyz-system. The
rectangular- and ellipse-shaped cross-section of dis-
continuous bers in MMCs can be seen, e.g., in the
optical micrograph of a polished section. For com-
posites reinforced with in-plane randomly oriented
bers (Fig. 4a), the global XYZ-system is identical
to the local xyz-system. In this case, all the ellipses
of bers in cross-sections B and C possess the same
orientation (all major axes are parallel, see Fig. 4a),
whereas in the case of 3D random ber orientations
they depict different directions of major axes, see
Fig. 4b, in the global cross-section.
Due to different ber orientations, which lead to dif-
ferent shapes of rectangles and ellipses in the cross-
sections, as seen in Fig. 4, a single unit cell is not
sufcient to represent the complicated geometrical
situation and the mechanical behavior of the short
ber reinforced MMCs. More computational cells
should be taken into account in order to obtain the
mechanical behavior of the composites by simple
cell models.
On the basis of the geometrical description outlined
above, it is possible to dene an approach which
uses simple cell models to calculate and predict the
mechanical deformation and ow behavior. From
the geometrical shape of the single ber on the three
local principal orthogonal cross-sections (plane A,
B and C) four unit cell models (see Fig. 6) can be
constructed, which represent the local stress state of
a single ber with an orientation .
The cross-section of the single ber on the local
plane A is a rectangle with the orientation angle .
Unit cells can only be applied for the orientation of
= 0

or = 90

, as shown in Fig. 6. The ori-


ented ber is then separated into two essential parts,
i.e. model A2 and A3, where A2 is a 2D model for
bers under transverse loading and A3 is an axi-
symmetrical model for bers under axial loading.
The cross-sections of the single ber on the local
planes B and C are two ellipses with the minor axis
d and the major axes d/sin and d/cos, where d
is the diameter of the ber. The representative ap-
proaches to these cross-sections are given by con-
structing two cell models, one is B2 and the other
is C3. B2 is a two-dimensional model with an el-
liptical inclusion of a minor axis d and a major axis
d/sin, whereas C3 is a three-dimensional model
with an axisymmetrical ellipsoidal inclusion of the
same minor and major axes as those of B2. There
exist four unit models, i.e. A2, A3, B2 and C3 for
each ber orientation . For =0

and = 90

one
model, i.e. A3 and A2, respectively, is sufcient.
From the ber volume fraction f , the ber aspect ra-
tio L/d and the ber orientation , the geometric size
of four cell models can be determined and the local
stress-strain curves
A2
(, ),
A3
(, ),
B2
(, ) and

C3
(, ) can be calculated fromthe cell models with
the conventional unit cell technique [36, 56].
Four stress-strain curves,
A3
(, ),
A2
(, ),

B2
(, ), and
C3
(, ) from four unit cells for a
given single ber orientation , must be connected
in an appropriate way to get the stress-strain curve
of a composite with bers oriented in a direction .
The following heuristic procedure has been used to
establish the desired connection: at rst, the stress-
strain curve
A
(, ) can be calculated by averag-
ing the stresses
A3
(, ) and
A2
(, ) with the help
of the volume relationship between the two separate
parts A2 and A3 of the cross-section on the plane A,
V
A2
=V sin and V
A3
=V cos:

A
(, ) =

A2
(, ) V
A2
+
A3
(, ) V
A3
V
A2
+V
A3
(4)
=

A2
(, ) sin+
A3
(, ) cos
sin+cos
.
On the cross-sections of the plane B and C there
exists the same volume relationship between the
two models B2 and C3, so that an average stress

BC
(, ), associated with
B2
(, ) and
C3
(, )
can be written as

BC
(, ) =

B2
(, )sin+
C3
(, )cos
sin+cos
. (5)
The three stress-strain curves,
A
(, ),
B
(, ) and

C
(, ) must be averaged to get an overall ow be-
havior of ber reinforced composites with a single
ber orientation, in such a way that each of the three
stresses
A
,
B
and
C
contributes to the overall ow
behavior (, ):
(, ) =
_

A
(, ) +2
BC
(, )
_
3. (6)
It is assumed that Eq. 6 provides the mechanical be-
havior of a single ber as a function of the ber ori-
entation with respect to the applied loading in a
composite.
To obtain the overall ow behavior of short ber
reinforced composites, a weighted integration of
stress-strain curves for all the ber orientations, as
dened in Eq. 7, can be carried out by introducing a
weighting function f (), which describes the distri-
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 333
bution density of short bers in a composite:
() =

/
2
_
0
(, ) f () d

/
2
_
0
f ()d
. (7)
If the short bers are randomly distributed in MMCs
in a plane (2D random), they possess the same dis-
tribution density in all directions of the plane, as
schematically illustrated in Fig. 4a. In this case, the
weighting function f () has the constant value 1. If
the short bers are distributed randomly in MMCs
in all directions of space (3D random), the distri-
bution density changes with the orientation angle
in the same way as the change of the latitude in a
spherical coordinate, which is considered by intro-
ducing a weighting function f () = sin. For com-
posites with preferred ber orientation the weight-
ing function f () must be determined in correspon-
dence with the preferred ber orientations.
4.3. Statistical Combined Cell Models
4.3.1. Static Loading Conditions
Statistical Combined Cell Models (SCCMs) for
short ber reinforced composites with different ber
volume fractions have been developed on the basis
of the Combined Cell Models of the previous section
[6, 57, 58] and a Weibull statistical approach [59],
originally developed for ber fracture in composites.
The SCCM takes into consideration ber-cracks and
ber-matrix debonding. This allows to calculate the
two types of unit cells separately, i.e. unit cells with
unbroken and with broken bers. Then, the global
mechanical behavior of composites reinforced with
short bers is calculated on the basis of the rule of
mixture.
When loading is parallel to the ber orientation or if
no debonding occurs between ber and matrix, it is
found that ber failure is the main source of damage
in the composite (Fig. 7). The fracture probability
of each ber is a function of its volume and of the
maximumprincipal stress
U
F
in the ber. Therefore,
the Weibull law, from Eq. 8, can be written in terms
of ber failure as follows:
P
brk
(
F
) = 1exp
_

_

U
F

0F
_
m
F
_
. (8)
In this equation, P
brk
(
F
) is the failure probability
of ber fracture and m
F
is the shape parameter of
Weibulls law which corresponds to the scatter of
the ber breaking in the composite.
0F
is a scale
parameter and equivalent to the mean value of the
ber strength, which gives a cumulative breaking
probability of 63% and it corresponds to fraction of
broken bers for a given ber reinforced composite.
This parameter is strongly related to the reinforce-
ment material. In this manner, we can obtain the
mechanical behavior of composites for the unit cells
A3, A2, B and C with Eqs. 4 and 5. The damage
behavior of composites can be calculated according
to Eqs. 8 and 9:

brk
( , ) =
_
1P
brk
_

F
_
_

brk
UD
( , )
+P
brk
_

F
_

brk
FD
( , ). (9)
A further principal source of damage is the failure
Figure 7. Schematics of the Statistical Combined
Cell Models (SCCM) with fracture of brittle bers
in short ber strengthened composites.
of the ber-matrix interface (Fig. 8). This failure is
governed by a local criterion that is dominated by in-
terfacial normal stress. Because the interfacial dam-
age is distributed statistically as a function of the
spatial distribution of the microstructure, the local
interface failure criterion must be written in a statis-
tical form following Weibulls law:
P
deb
(
L
) = 1 (10)
exp

_

U
L

0L
_
2
+
_

U
L

0L
_
2

m
L

,
where P
deb
(
L
) denotes the ber-matrix interfacial
debonding probability relative to a given interfacial
state
U
L
, which is a function of the microscopic
stress
L
,
0L
denotes the interfacial stress, and m
L
334 DAMAGE SIMULATION
is the statistical parameter. The parameter
U
L
de-
notes the interfacial shear stress and
0L
is the char-
acteristic shear stress. If the ber is perpendicular
to the loading direction (90

), there is no signicant
inuence of shear stresses and the equation can be
written as
P
deb
(
L
) = 1exp
_

_

U
L

0L
_
m
L
_
. (11)
The stress state of a cell can be predicted by the mix-
ing rule [7, 60] in which undebonding stresses and
debonding stresses are taken into account:
( ) =
_
1P
deb
(
L
)
_

ud
( ) +
P
deb
(
L
)
db
( ), (12)
where
ud
( ) is the stress in an undamaged unit cell
and
db
( ) is the stress in a damaged unit cell due
to ber-matrix interfacial debonding. In this equa-
tion, ( ) is the stress behavior of a composite cell
with ber perpendicular to the loading direction that
includes the debonding damage behavior. The rst
term on the right-hand side indicates the stress be-
havior of the undamaged interface (ud) and the sec-
ond term indicates the stress behavior of damaged
interface (db) in a composite. Thus, the arithmetic
sum in Eq. 12 implies the stress behavior of a com-
posite cell with debonding failure. The mechanical
Figure 8. Schematics of the Statistical Combined
Cell Models with damage in the boundary layer be-
tween bers and matrix.
behavior derived from the unit cells A3, A2, B and
C with consideration of the damage between bers
and matrix follows in an analogous manner.
For the numerical investigation with consideration
of the ber-matrix adhesion effect, bers were ar-
ranged in the tensile specimen perpendicularly to the
loading direction. In this case there was damage at
the boundary layer between bers and matrix, but
no ber fractures took place. When using the CCM,
we have in this case = 0

, so that a description
of the model through the model part A3 is sufcient
(Fig. 8). As a rst application of the SCCM, the pa-
rameters in Eqs. 11 and 11 can be calculated by a
comparison between the computation and the exper-
iment.
From the experiments it can be seen that in case of
parallel loading there is a combined effect of ber
breaking and debonding on the composite failure.
Both effects can be combined in a composite unit
cell using the mixing rule [7, 60]
( ) =
_
1P
deb
(
L
) P
brk
(
L
)
_

ud
( ) +
P
deb
(
L
)
db
( ) +P
brk
(
L
)
brk
( ), (13)
where
brk
( ) is the stress in a damaged unit cell due
to broken bers.
The two Weibull parameters for interface failure and
ber failure are numerically identied by using the
data from micromechanical models and the calcu-
lated nite element results to compare them with the
experimental curves.
4.3.2. Quasi-Static Cyclic Loading Conditions
The micromechanical fatigue damage model in this
section is based on a statistical microscopic damage
law. Predictions of these types of failure have been
applied to determine damages in each loading cycle.
By comparing the simulation with the experimental
stress-strain curves for tension, the Weibull damage
parameters are determined. Using these damage pa-
rameters a mesoscopic model (Sec. 5.2.3) including
the effect of ber-clusters is developed and the dam-
age during cyclic loading is predicted.
To study the behavior of ber reinforced compos-
ites, 3D unit cell models are used to analyze the
microscopic failure. The statistical analysis of ber
breaking and ber-matrix interfacial debonding will
be predicted by Weibulls law [61, 62] as described
above. It is based on the assumption that the com-
posite fails as a result of accumulation of statistically
distributed ber aws. The equations of Weibulls
damage law for ber failure [7, 63] were taken from
Eqs. 8 and 13. The ber failure can be supplemented
by ber-matrix interfacial debonding.
Evolution of damage in a composite under cyclic
loading is calculated on the basis of the statistical
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 335
evolution of damage in the ber-matrix interfaces
and in the broken bers. Debonding failure and fail-
ure due to broken bers are considered mutually de-
pendent on each other. That means that, if debond-
ing occurs around the ber-matrix interface, ber
failure will not occur. On the other hand, where
the bers break, there is a negligible inuence of
debonding failure. The damage stress for each cycle
is calculated according to the total failure probability
due to ber failure and interface debonding. The ef-
fect of damage is embedded in the model by replac-
ing the stress of the previous cycle (the true stress in
the rst cycle) with the effective stress in the present
cycle. Any strain constitutive equation for the dam-
aged material is derived in the same way as for the
virgin material, except that the true stress is replaced
by the current effective stress [64]. Accordingly,
material properties are changed during the cycle due
to ber failure and interface debonding. Applying
the mixing rule, the stress after the k-th loading cy-
cle can be expressed as follows:

k+1
i j
(
i j
) =
k
i j
(
i j
)
P
brk
(
k
i j,unbr
(
i j
)
k
i j,brk
(
i j
))
P
deb
(
k
i j,unde
(
i j
)
k
i j,deb
(
i j
)), (14)
where i is the Element index, and j is the loading
step index in one cycle. Therefore, the new mate-
rial properties of the composite are calculated for
each loading cycle, which is then included into the
ABAQUS input le [65] for the next loading cycle
calculation.
4.4. Plastic-Damage Model
In this section, the homogenized constitutive frac-
ture behavior of materials will be described for
static and quasi-static cyclic loading with a plastic-
damage model proposed by Lubliner et al. [66] and
Lee and Fenves [67]. In this model, stiffness degra-
dation due to damage is embedded in the plasticity
part of the model. Damage is represented by two
independent scalar damage parameters, one for ten-
sion (d
t
) and another one for compression (d
c
). This
is necessary because many materials show different
damage mechanisms in tension and compression.
In tension, the damage is associated with cracking,
while in compression, it is associated with crush-
ing. The initial undamaged state and complete dam-
aged state of the material under tension and com-
pression are indicated by d
t
, d
c
= 0 and d
t
, d
c
= 1,
respectively. Apart from this, a stiffness recovery
scheme is used for simulating the effect of micro-
crack opening and closing. The effect of damage is
embedded in the plasticity theory and all stress de-
nitions (true stress) are reduced to the effective stress
[64]. This enables the decoupling of the constitutive
relations for the elastic-plastic response from stiff-
ness degradation (damage) response.
In the following equations, underlined symbols indi-
cate vector or tensor quantities, overlined stress ex-
pressions indicate effective stresses. Symbols with-
out underline are to be understood as scalar quan-
tities. All strain symbols with a tilde are equiva-
lent strains. In Eq. 15 Macaulay brackets have
been used, which are dened as x = x if x > 0,
otherwise x = 0. For the plasticity part, a non-
associated plasticity scheme is used. The yield sur-
face proposed by Lubliner et al. [66] is based on
modications of the classical Mohr-Coulomb plas-
ticity (Eq. 15):
F( ,
pl
) =
1
1
( q3 p+(
pl
)


max
_


max
_
)
c
(
pl
c
), (15)
where corresponds to the stress tensor,
c
is the
uniaxial compressive stress, p corresponds to the
effective hydrostatic pressure, and are material
constants, q corresponds to the equivalent effective
deviatoric stress,


max
is the max. principal stress,
and
pl
corresponds to the equivalent plastic strain.
A separate ow potential is used to determine the di-
rection of plastic ow in the principal stress space.
The ow potential chosen for this model is the
Drucker-Prager hyperbolic function G (Eq. 22 in
Sec. 5.3.2). At high conning pressure stress,
the function asymptotically approaches the linear
Drucker-Prager owpotential in the deviatoric plane
and intersects the hydrostatic pressure axis at 90

[68]. In Fig. 9 the yield surface and the ow po-


tential function are illustrated in the 2D principal
stress space. The material modeling has been per-
formed based on an existing implementation of the
plastic-damage model in ABAQUS. The details of
the mathematical formulation of the model are given
in [66, 6870].
4.4.1. Modications of the Plastic-Damage
Model
The simulation results of the static behavior of
the material presented in Sec. 5.3.1, obtained by
using the implemented plastic-damage model in
ABAQUS, is close to that of the experiment (see
Sec. 5.3.2). However, when applying the imple-
mented model to cyclic loading, considerable model
limitations are observed. The implemented model
336 DAMAGE SIMULATION
Figure 9. Illustration of (a) yield surface, and ow
potentials, (b) dilation angle [11].
reaches up to the point where unloading starts. Be-
yond this point the material behavior is complex,
showing different stiffnesses at different stages of
unloading and reloading. It is not possible to han-
dle the varying unloading and reloading stiffnesses
with the available stiffness recovery effects imple-
mented in the plastic-damage model provided by
ABAQUS. Therefore, the plastic-damage model has
been re-implemented with the necessary modica-
tions in a user dened material subroutine UMAT
in ABAQUS to improve the simulation of the quasi-
static cyclic experiments.
If the yield point in compression is not reached,
compression damage is absent (d
c
= 0). Then, dam-
age occurs only due to tensile loading, which is
represented by the scalar tension damage parame-
ter d
t
. The total damage parameter d in the modi-
ed plastic-damage model is correlated with tension
damage parameter d
t
as
d = d
t
s, (16)
s = 1w, (17)
where s is the stiffness recovery factor and w is
the weight factor that controls the stiffness recov-
ery. w = 1 means complete stiffness recovery cor-
responding to d = 0, whereas w = 0 means no stiff-
ness recovery corresponding to d = d
t
. On the yield
surface, d is obtained from Eq. 16 and 17. The evo-
lution of yield stress, tension damage, d
t
and weight
factor, w are functions of plastic strain in tension,
p
t
.
The material subroutine UMAT requires the evolu-
tion information as strain softening, damage evolu-
tion and stiffness recovery curves. During unload-
ing and reloading in the elastic domain, d is rede-
ned in the material subroutine UMAT based on
rules derived by observing the unloading/reloading
slope (E) in the uniaxial quasi-static cyclic stress-
strain curves. The corresponding damage parameter
d is obtained from the varying slope (E) and the ini-
tial stiffness (E
0
) as
d = 1
E
E
0
. (18)
The determination process of the rules controlling
d in the elastic domain and strain softening, damage
evolution and stiffness recovery curves are discussed
in Sec. 5.3.2.
5. RESULTS AND APPLICATION
After introduction of the investigated materials, in
this section the above presented models will be
applied to different composites which are MMCs,
PMCs, and cellulose ber reinforced gypsum mate-
rials. In the case of MMCs the self-consistent and
the (Statistical) Combined Cell Models are used.
The PMCs are studied applying the Statistical and
Combined Cell Model. Finally, the plastic-damage
model suit especially for quasi-brittle materials such
as the investigated cellulose ber reinforced gypsum
composite.
5.1. Metal Matrix Composites (MMC)
As a rst example the self-consistent and the (Sta-
tistical) Combined Cell Model will be applied to
Metal Matrix Composites. MMCs with strong in-
clusions are a relatively newclass of materials (com-
pare Chapter 2.1.4). Due to high strength and light
weight, they are potentially valuable in aerospace
and transportation applications [71].
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 337
5.1.1. Material
The metal-matrix is an Al/12% vol. Si cast alloy
(M124). The composite considered here has been
produced by Mahle GmbH, Stuttgart, via pressure
inltration of a ber preform with randomly ori-
ented short Al
2
O
3
-bers (Safl). These bers be-
have elastically (Youngs modulus E = 300 GPa,
and Poissons Ratio = 0.23), the ber content in
the composite is 15% vol., and the ber aspect ra-
tio is approximately 200m/3m. Fig. 10 shows an
optical micrograph of a polished section of the short
ber reinforced composite [72].
As a further example an Al/46% vol. B compos-
ite with random ber packing taken from [26] has
been selected to verify the embedded cell model.
This MMC is a 6061-O aluminum alloy reinforced
with unidirectional cylindrical boron ber of 46%
volume fraction. The room temperature elastic
properties of the bers are a Youngs modulus
of E
(B)
= 410 GPa, and a Poissons Ratio of

(B)
= 0.2. The experimentally determined me-
chanical properties of the 6061-O aluminum ma-
trix are Youngs modulus, E
(Al)
= 69 GPa, Pois-
sons Ratio,
(Al)
= 0.33, 0.2% offset tensile yield
strength,
0
= 43 MPa, and strain-hardening expo-
nent N = 1/n = 1/3.
Furthermore, the composite Ag/58% vol. Ni [42]
with random particle arrangement (Youngs modu-
lus, E
(Ni)
= 199.5 GPa, E
(Ag)
= 82.7 GPa, Pois-
sons Ratio,
(Ni)
= 0.312,
(Ag)
= 0.367, and
yield strength
(Ni)
0
= 193 MPa,
(Ag)
0
= 64 MPa)
has been investigated.
Figure 10. Optical micrograph of a polished sec-
tion of the discontinuous short ber Al alloy/15%
vol. Al
2
O
3
composite with 3D random ber orienta-
tion [6].
5.1.2. Results: Self-Consistent Model
In this section, self-consistent embedded cell mod-
els, which are described in Sec. 4.1, are applied to
simulate the transverse behavior of MMCs contain-
ing bers in a regular square or hexagonal arrange-
ment as well as the mechanical behavior of MMCs
containing particles in a regular arrangement. Two
aims are pursued: one is to investigate the me-
chanical behavior of MMCs reinforced with regular
or random arranged continuous bers under trans-
verse loading and particles under uniaxial loading.
The other one is to systematically study compos-
ite strengthening as a function of inclusion volume
fraction and matrix hardening ability. The Finite
Element Method (FEM) is employed to carry out
the calculations. The overall response of MMCs is
elastic-plastic. As regular ber spacings are dif-
cult to achieve in practice, most of the present ber
reinforced MMCs contain aligned but randomly ar-
ranged continuous bers.
The LARSTRAN nite element program [73] was
employed using 8 noded plane strain elements (for
2D) as well as axisymmetric biquadrilateral ele-
ments (for 3D) generated with the help of the pre-
and post-processing program PATRAN [74].
Fig. 11a shows a comparison of the stress-strain
curves of the composite Al/46% vol. B under trans-
verse loading from simulations of a real microstruc-
ture together with results from different cell mod-
els. The stress-strain curve from the embedded cell
model employed in this Chapter shows close agree-
ment with the curve from the calculated random
ber packing in the elastic and plastic regime, which
lies between the curves from square unit cell model-
ing under 0

loading and hexagonal unit cell model-


ing.
Furthermore, the stress-strain curve from another
experiment [42] on the composite Ag/ 58% vol. Ni
with random particle arrangement has been com-
pared with that from the self-consistent embedded
cell model (Fig. 11b). Close agreement in the regime
of plastic response is obtained, although the Ni-
particles in the experiment were not perfectly spher-
ical. These results indicate that the embedded cell
model can be used to successfully simulate compos-
ites with random inclusion arrangements and to pre-
dict the elastic-plastic composite behavior. A com-
parison of the stress-strain curves for the composite
Al/46% vol. B in Fig. 11a shows that the stress-
strain curve from randomber packing given in [26]
lies also between the curves from square unit cell
modeling under 0

loading and hexagonal unit cell


modeling.
338 DAMAGE SIMULATION
Figure 11. Comparison of the mechanical behavior
of (a) an Al-46% vol. B ber reinforced compos-
ite (N=1/3, f=0.46) under transverse loading from
different models, and (b) an Ag-58% vol. Ni partic-
ulate composite from embedded cell model and ex-
periment [5].
Geometrical shape of the embedded cell
As mentioned above, different shapes of cross-
section of the embedded cell model with a circular
shaped ber, as shown in Fig. 12a, are also taken
into account to investigate the inuence of the geo-
metrical shape of the embedded cells on the overall
behavior of the composite. The stress-strain curves
of all embedded cell models with different geomet-
rical shapes are plotted in Fig. 12b. With an ex-
ception of square - 45

embedded cell model, the


stress-strain curves are very close for all embedded
cell shapes, namely, square - 0

, circular, rectangu-
lar - 0

, rectangular - 90

, elliptic - 0

and elliptic
- 90

. From the calculated results of the embedded


Figure 12. Embedded cell models: inuence of (a)
different matrix shapes on (b) stress-strain curves
for an Al-46% vol. B (N=1/3, f=0.46) composite
[5].
cell models, localized ows have been found around
the hard ber with preferred yielding at 45

. Be-
cause of the special geometry of the square - 45

em-
bedded cell model with the cell boundary parallel to
the preferred yielding at 45

, the overall stresses of


the composite with such a geometrical cell shape are
therefore reduced, such that a relative lower stress-
strain curve has been obtained from the modeling.
The almost identical responses of the other embed-
ded cell models indicate that, besides of the special
shape of matrix with 45

cell boundaries, the pre-


dicted mechanical behavior of ber reinforced com-
posites under transverse loading is independent of
the modeling shape of the embedded composite cell.
Strengthening model
The strength of MMCs reinforced by hard inclu-
sions under external mechanical loading has been
shown to increase with inclusion volume fraction
and strain-hardening ability of the matrix for all in-
clusion arrangements investigated. From the pre-
sented numerical predictions, a strengthening model
for aligned continuous ber reinforced MMCs with
random, square (0

) and hexagonal arrangements


4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 339
as well as for spherical particle reinforced MMCs
with random, primitive cubic and hexagonal ar-
rangements can be derived as a function of the in-
clusion volume fraction f , and the strain-hardening
exponent N, of the matrix:

N
=
0

_
1
f
c
1
(2+N)
_
(c
2
N+c
3
)

0
c
4
_
f +
N
5
_
, (19)
where
0
is the matrix yield stress, and c
1
, c
2
, c
3
and c
4
are constants summarized in Tab. 1. Eq. 19
2D c
1
c
2
c
3
c
4
SCM 0.361 1.59 0.29 0.1
SM (0

) 0.405 2.35 0.65 0.22


HM 0.305 1.3 0.05 0.0
3D
SCAM 0.45 2.19 0.84 0.53
AM 0.38 2.5 0.7 0.66
PCM 0.34 2.3 0.65 0.5
Table 1. Constants for strengthening models: self-
consistent embedded cell model with random ber
arrangement (SCM), square unit cell model (SM),
hexagonal unit cell model (HM), Self-consistent
axisymmetric embedded cell model (SCAM), axi-
symmetric unit cell model (AM), primitive cubic unit
cell model (PCM) [5].
represents the closest approximation to the calcu-
lated composite strengthening values
N
for matrix
strain-hardening exponents N in the limit of 0.0 <
N < 0.5 for square 0

, hexagonal and random ber


arrangements, (practical ber volume fractions f in
the range of 0.0 < f <0.7), respectively. A compar-
ison of this strengthening model (Eq. 19) for random
ber arrangements with the values calculated by
using self-consistent embedded cell models shows
close agreement with an average error of 1.25% and
maximum error of 6.95%. Eq. 19 is also available
for matrix strain-hardening exponents N in the lim-
its of 0.0 < N < 0.5 for self-consistent axisymmet-
ric embedded cell models (particle volume fraction
f in the range of 0.05 < f < 0.65 with an average
error of 1.59% and a maximum error of 6.68% for
the extreme case f = 0.05, N = 0.5), axisymmet-
ric unit cell models (particle volume fraction f in
the range of 0.05 < f < 0.55 with an average er-
ror of 1.22% and a maximum error of 6.18% for the
extreme case f = 0.55, N = 0.5) and for primitive
cubic unit cell models (particle volume fraction f in
the range of 0.05 < f < 0.45 with an average error
of 1.43% and a maximum error of 6.38% for the ex-
treme case f = 0.05, N = 0.5).
5.1.3. Results: Combined Cell Model
The purpose of the present section is to investigate
the mechanical and thermo-mechanical behavior of
MMCs (Al/15% vol. Al
2
O
3
aluminum matrix com-
posite (Fig. 10) reinforced with randomly oriented
short bers by applying the Combined Cell Model
described in Sec. 4.2. The bers are well bonded to
the matrix so that debonding or sliding at the ber-
matrix interface is not permitted. The nite element
method (FEM) is employed within the framework of
continuum mechanics to carry out the calculations.
The uniaxial matrix elasto-plastic stress-strain be-
havior measured from experiments at room temper-
ature can be described by an exponential hardening
law:
= E
0
, (20)
=
0
_

0
_
N
>
0
,
where and are the uniaxial stress and strain of the
matrix, respectively,
0
is the ow stress, the ma-
trix yield strain is given as
0
= E/
0
, E is Youngs
modulus, and N is the strain hardening exponent. J
2
ow theory of plasticity with isotropic hardening is
employed with a von Mises yield criterion to char-
acterize the rateindependent matrix material. The
ow behavior is different in tension and compres-
sion and can be described using the following pa-
rameters: E = 76000 MPa, = 0.33, N = 0.2, and

tension
0.2
= 225 MPa,
compression
0.2
= 234 MPa.
Figs. 13a and 13b show the numerically obtained
stress-strain curves of the composite (M124/15%
vol. Al
2
O
3
) in uniaxial compression (a) and ten-
sion (b), respectively. The orientation angles consid-
ered here are 0

, 5

, 10

, 15

, 30

, 45

, 60

and 90

.
The experimental stress-strain curves of elastic ber
(Al
2
O
3
) and elastic-plastic matrix (Al/12% vol. Si-
alloy) are also shown in these gures. Composite-
strengthening increases with decreasing the ber
orientation angle from 90

to 0

. From 90

to 30

the increase is very small, but it becomes larger and


larger from 30

to 0

. After the integration by ap-


plying Eq. 7 (Sec. 4.2) for both cases of 2D and
3D random orientations with weighting functions
f () = 1 and f () = sin, respectively, we obtain
the two stress-strain curves (bold continuous and
dashed lines in Fig. 13) for the overall ow behavior
of these composites. For all the cases analyzed, the
averaged stress-strain curves of composites with 2D
random as well as 3D random ber reinforcements
lie in the neighboring of the stress-strain curve of
composites with 30

and 60

ber orientation, re-


spectively. The fact that the stress-strain curves of
340 DAMAGE SIMULATION
(a) compression
(b) tension
Figure 13. Numerical results of (a) compression
and (b) tension ow behavior of the M124/15% vol.
Al
2
O
3
(3D randomber orientation, ber aspect ra-
tio: 200m/3m) with different ber orientations
[6].
in-plane randomly oriented ber reinforced compos-
ites coincides with those of approximately 30

ori-
ented ber reinforced composites, has been also re-
ported in [51].
In Figs. 14a and 14b the numerical results obtained
for 3D random ber orientation are compared to the
experimental data obtained by uniaxial compression
and tension tests, respectively. In the case of com-
pression loading close agreement exists between ex-
periments and simulation in the elastic and plastic
regimes. However, at strains above 1.5% the numer-
ical simulation predicts higher strain hardening than
observed in the experiments. In the case of tensile
loading, close agreement between the experimen-
tal measurement and the numerical prediction is ob-
tained only for the elastic regime (see Fig. 14b). The
observed deviations between experimental and nu-
merical results can be attributed to the onset of mi-
(a) compression
(b) tension
Figure 14. Effects of residual stresses on the over-
all ow behavior of the M124/ 15% vol. Al
2
O
3
(3D random ber orientation, ber aspect ratio:
200mm/3mm) and comparison with experiments.
[6]
crodamage such as fracture of the brittle constituents
of the composite. Such damaging processes have
been observed both in metallographic studies and in
acoustic emission measurements (see Chapter 3.1).
The different deviation in tension and compression
may be attributed to the fact that the damaging pro-
cesses mentioned above are sensitive to the direction
of loading [75].
These results indicate that the Combined Cell Model
used in this study can be applied successfully to
composites with random ber orientation as long as
effects from micro-damage can be neglected. In or-
der to predict the macroscopic stress-strain curve of
short ber reinforced MMCs in tension, a more ac-
curate model including microscopic damage events
must be developed (see Sec. 4.3).
In a second step, the effects of residual stresses
have been estimated using the model. The internal
stresses and strains that form during cooling from
400

C to room temperature were calculated for
each cell under the simplifying assumptions that the
thermal expansion coefcients of the constituents as
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 341
well as the ow behavior of the matrix alloy are
identical in the whole temperature range.
Figs. 14a and 14b show, besides the comparison
with experiments, the comparisons between com-
puter predictions of the overall ow behavior of
MMC randomly reinforced with short bers, with
and without considering the initial thermal stresses.
Signicant inuences of residual stresses on the
overall mechanical behavior of short ber reinforced
MMCs were found: in both tension and compres-
sion the effective Youngs moduli are found to be
lower while the yield stresses are increased. Because
of the plastic deformation in the local area near the
ber under thermal loading, in these areas the lo-
cal stress states under mechanical loading are differ-
ent compared to the case without thermal loading.
Under mechanical loading, further ow in some lo-
cal areas directly after thermal loading reduces the
overall stress response at the small strain state. Be-
cause metal-matrix hardening takes place in some
local areas under thermal loading and the harden-
ing is isotropic, the material in this area is harder
than it would be without undergoing thermal load-
ing. With increasing the overall strain, i.e. when
higher ow stresses in the local area are reached,
the overall yield stresses of the composites will be
higher compared to the case when thermal stresses
are absent. The composite strengthening including
thermal loading naturally depends on temperature
change T, difference of thermal expansion coef-
cients of matrix and ber and on the value of the
yield stress of the matrix. The composite strength-
ening with thermal loading increases with increasing
value of T, and the yield stress of the matrix.
5.1.4. Results: Statistical Combined Cell Model
The SCCM model is applied to MMC materials
in static and quasi-static cyclic loading conditions.
These results are presented in the following two
paragraphs.
Static loading
In [58, 59] it has been experimentally established
that prior to the failure of the composite, frac-
ture of brittle bers takes place in short ber rein-
forced metal matrix composite M124-Safl under
tensile stress. Fig. 15 shows calculated stress-strain
curves of the ber composite M124-Safl (15% vol.
Al
2
O
3
-Safl bers) under consideration of ber fail-
ure in dependence of global strain. In Fig. 15a, the
Weibull modulus m was varied from 1 to 3, and in
Fig. 15b, the characteristic stress
0
of bers, from
500 MPa to innite. It can be observed that when
(a)
(b)
s
Figure 15. Comparison of stress-strain curves, for
a) different Weibull moduli m with
0
= 1000 MPa,
and b) different characteristic stresses
0
with
Weibull modulus m = 1 for metal matrix composite
with 3D random short bers [7].
the global strains are lower than 0.15%, the differ-
ence among the numerical and experimental results
are very small, because in this area there is hardly
any damage in bers. Close agreement of the calcu-
lated stress with the experimental result is found for
m = 1 and
0
= 1000 MPa. The Weibull modulus
m is usually found between 3 and 8 [58]. However,
the calculated stress-strain curve for m = 1 deviates
from the experimental results. As reported in [58],
strong ber clusters exist in the analyzed ber com-
posite M124-Safl (15% vol. Al
2
O
3
-Safl bers),
but are not considered in the present model. To con-
sider the inuence of ber clusters on the simulation
results, a mesoscopic concept has to be established
which takes into account accidental changes of ber
volume content and which allows to calculate statis-
tical ber failure in different ber cluster areas (see
Sec. 5.2).
Quasi-static cycling loading
The presented Statistical Combined Cell Model is
based on the reduction of the effective Youngs mod-
342 DAMAGE SIMULATION
ulus and damage as introduced by the evolution of
failure probability of bers and ber-matrix inter-
faces determined by the Weibull damage law. The
predescribed procedures are performed on the Al-
Al
2
O
3
short ber composite and compared with test
results from [63, 76, 77]. The calculations are per-
formed up to 10 loading cycles and the evolution of
damage on the Youngs modulus is calculated after
each loading cycle. FE simulation results and test
results are plotted for comparison in Fig. 16. Close
Figure 16. Comparison of experiment and simula-
tion with the reduction of the effective Youngs mod-
ulus in fatigue of an Al-Al
2
O
3
composite, for 0.15%
strain [9].
agreement is found between the experiment and the
simulation using the proposed damage model. It is
seen that the simulated model shows a slightly lower
decrease in the effective Youngs modulus. It is as-
sumed that, in reality, matrix cracks inuence the
composite failure, and hence reduce the effective
Youngs modulus. Taking also into account matrix
cracks would minimize the differences in the reduc-
tion of the effective Youngs modulus between ex-
periment and calculation.
5.1.5. Conclusions
The transverse elastic-plastic response of MMCs re-
inforced with unidirectional continuous bers and
the overall elastic-plastic response of Metal Matrix
Composites reinforced with spherical particles have
been shown to depend on the arrangement of rein-
forcing inclusions as well as on the inclusion vol-
ume fraction f , and the matrix strain-hardening ex-
ponent, N. Self-consistent axisymmetric embedded
cell models have been employed to predict the over-
all mechanical behavior of Metal Matrix Compos-
ites reinforced with randomly arranged continuous
bers and spherical particles perfectly bonded in
a power law matrix. Experimental ndings on an
aluminum matrix reinforced with aligned but ran-
domly arranged boron bers (Al/46% vol. B) as
well as a silver matrix reinforced with randomly ar-
ranged nickel inclusions (Ag/58% vol. Ni) and the
overall response of the same composites predicted
by embedded cell models are found to be in close
agreement. The strength of composites with aligned
but randomly arranged bers cannot be properly de-
scribed by conventional ber-matrix unit cell mod-
els, which simulate the strength of composites with
regular ber arrangements.
Systematic studies were carried out for predicting
composite limit ow stresses for a wide range of
parameters f and N. The results for random 3D
particle arrangements were then compared to regu-
lar 3D particle arrangements by using axisymmet-
ric unit cell models as well as primitive cubic unit
cell models. The strength of composites at low par-
ticle volume fractions were in close agreement ex-
cept for the modied Oldroyd model. With increas-
ing particle volume fractions f , and strain hard-
ening of the matrix N, the strength of composites
with randomly arranged particles cannot be properly
described by conventional particle-matrix unit cell
models, as those are only able to predict the strength
of composites with regular particle arrangements.
Finally, a strengthening model for randomly or reg-
ularly arranged continuous bers and particle rein-
forced composites under axial loading is derived,
providing a simple guidance for designing the me-
chanical properties of Metal Matrix Composites.
For any required strength level, Eq. 19 will provide
the possible combinations of particle volume frac-
tion f , and matrix hardening ability, N.
The ow behavior for Metal Matrix Composites re-
inforced with 2D (planar) and 3D randomly oriented
short Al
2
O
3
-bers is investigated by Combined Cell
Models in conjunction with the FEM. The mechan-
ical behavior of short ber reinforced Metal Matrix
Composites (MMCs) with a given ber orientation
can be simulated numerically by averaging results
derived from different cell models. These cell mod-
els involve two 2D models and two 3D models rep-
resenting a single ber in three principal orthogo-
nal planes in the composite. Stress-strain curves
have been calculated for MMCs reinforced with 2D
randomly planar and 3D randomly oriented short
bers by an appropriate integration of results of all
ber orientations. The numerical results are com-
pared with experimental data of a ber reinforced
aluminum alloy composite obtained in uniaxial ten-
sion and compression tests. Close agreement is ob-
tained between experimental results and the predic-
tions of the model in the regimes where no micro-
damage is observed experimentally. Finally, the ef-
fects of residual stresses have been estimated us-
ing the model. Both in tension and in compres-
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 343
sion Youngs modulus is found to be lower while
the yield stresses are increased compared to the case
when residual stresses are absent.
Applying the Statistical Combined Cell Model,
which includes damage effects in form of a statis-
tical Weibull approach, also the quasi-static cyclic
behavior of the MMC composite could be investi-
gated.
5.2. Polymer Matrix Composites (PMCs)
5.2.1. Material
PMCs are frequently reinforced with strong contin-
uous or short bers (Chapter 2.1.3). In the case of
short ber reinforced PMCs, a specic orientation
distribution is observed. Their mechanical proper-
ties are highly dependent on their structure. The
complexity of such affecting parameters impedes a
complete theoretical description of the behavior and
the failure properties of these composites. In this re-
spect, a micromechanical analysis of the local fail-
ure process opens a possibility to predict the macro-
scopic failure property of composites [13].
In this section, on the basis of [7], the Combined and
the Statistical Combined Cell Model [6] are applied
to describe the overall ow behavior of composites
reinforced with short bers (polypropylene matrix
with 8.1% vol. glass bers) with good and sparse
adhesive strength (Fig. 17). The failure of such com-
posites with different ber volume fractions is in-
vestigated using Statistical Combined Cell Models
based on Combined Cell Models [6] and Weibull
statistical approach [58, 59]. For this purpose, an
injection molded polypropylene was used. Injection
molded specimens usually show a complex layered
morphology with bers mainly oriented in process-
ing direction at the skin layer and normal to it in the
center of the specimen (core layer) due to shear and
elongation ow. Applying a push-pull processing
the melt can be pushed through the cavity several
times forth and back using a two component injec-
tion molding machine. Push-pull processing leads
to highly oriented bers (Fig. 17a) also in the cen-
ter of the specimen while the thickness of the core
layer is considerably reduced. This fact is expressed
in a high value of the effective Youngs modulus of
the composite in push-pull direction (||) compared
to the effective Youngs modulus perpendicular ()
to it [78]. The properties of the matrix and ber as
well as the composite are given in Tab. 2.
(a)
(b)
Figure 17. Micrograph of a polymer matrix with
8.1% vol. glass bers with (a) good adhesive
strength [7], and (b) sparse adhesive strength.
Properties Value
Youngs modulus of matrix 1.9 GPa
Youngs modulus of bers 72.0 GPa
Aspect ratio of bers 25
Diameter of bers 10 m
Number of push-pull cycles 4
Fiber content 8.1% vol.
Youngs modulus || (composite) 5.5 GPa
Youngs modulus (composite) 2.5 GPa
Table 2. Properties of the matrix, the glass bers
and the push-pull processed composite [9, 78].
5.2.2. Results: Combined Cell Model (CCM)
In this Chapter, CCMs [6, 57, 58] are applied to
describe the overall ow behavior of composites
reinforced with short bers and polymer matrix
(Fig. 17). As described in Sec. 5.1.3, the overall
ow behavior of composites with a certain ber ori-
entation can be calculated by an appropriate inte-
gration over all ber orientations. The numerical
results are compared to experimental data of short
ber reinforced Polymer Matrix Composites under
344 DAMAGE SIMULATION
tension. Close agreement has been obtained at small
strains between experiments and numerical predic-
tions by using these models. The larger the strain,
the stronger the deviation between experiments and
numerical predictions (Fig. 18). In order to predict
Figure 18. Comparison of experiments and FE pre-
dictions for polypropylene matrix composite with
planar random short bers [7].
the ow behavior of short ber reinforced compos-
ites in tension to a higher accuracy, ber cracking
and ber-matrix debonding can be taken into ac-
count [58], which is done in Sec. 5.2.3.
Consideration of complex ber orientations
The injection molding process leads to a complex
arrangement of the bers in the cavity due to shear
ow and elongational ow. The different orienta-
tions of the bers result in anisotropy of the com-
ponent properties. Using inserts to fabricate plates
containing a hole, leads to the splitting of the melt
front and nally to the formation of a weldline as a
result of the joining of the two melt fronts (Fig. 19).
Weldlines are known to be mechanically weak re-
gions of the component (compare Chapter 2.1.3). In
Figure 19. Successive patterns of lltime of the melt
(polyamide 6 reinforced with 30 weight-% glass-
bers - PA6GF30) at different stages of the process:
0.25 s, 0.96 s, 1.02 s and 1.17 s (end of lling).
this region, the ber orientations show great varia-
tions (Fig. 20). The ber orientation distribution de-
Figure 20. Averaged ber orientation over thickness
of a PA6GF30-specimen (simulation).
termined via microwave anisotropy measurements
(details of the method can be found in Chapter 1.2.3)
has been measured in the region near the hole within
a measurement eld of 60 x 45 mm
2
and a raster of
1,25 x 1,25 mm
2
. The experimental results are dis-
Figure 21. Experimental microwave orientation of a
PA6GF30-specimen (see Chapter 2.1.3).
played in Fig. 21 and compared to the simulation re-
sult in Fig. 20. Horizontal orientation in the left part
of the microwave orientation image, the ow around
the hole and the coalescence to the weldline with
a horizontal orientation of the bers can be identi-
ed. Since the measurement eld is larger than the
raster distance, artefacts appear near the free edges
(here: hole). The comparison of the simulated ber
orientation (Fig. 20) with experimental results of
microwave anisotropy investigations shows a good
correlation (Fig. 21). The ber orientation was
simulated for several layers (Fig. 22) in each ele-
ment with the use of an injection molding simula-
tion software (MoldowPlastic Insight). The linear-
elastic results were transferred (Fig. 23) to a strength
analysis FE code (ABAQUS). The simulation pro-
cedure is shown in Fig. 23. As a result, the effect
of ber orientations on the local mechanical behav-
ior as well as the macroscopic properties of a model
plate containing a hole and a weldline were investi-
gated by applying the CCM and compared to the re-
sults of the Tandon-Weng model [79]. Fig. 24 shows
the result of the linear-elastic simulation. According
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 345
Figure 22. Multilayer model.
Figure 23. Procedure of simulation.
to the Tandon-Weng model, the lowest values of the
stiffness E
11
appear near the injection point, behind
the hole and at the ow end. The maximal stiffness
is reported with 9.9 GPa (red regions). The result-
ing macroscopic stiffness of the global component is
calculated to be 7.1 GPa.
In a second step, the CCM is used to calculate the
global mechanical tensile properties of the compo-
nent (polyamide 6, 30% glass bers) taking into
account the local ber orientation, as shown in
Fig. 20. Elastic and elastic-plastic properties are
considered. The results of these simulations are
shown in Fig. 25. The solid line, which repre-
sents the results of the Combined Cell Model, is
compared to the isotropic elastic-plastic properties
(dashed line) and to the stiffness prediction of the
Tandon-Weng model (straight line). The anisotropic
simulation exhibits a stiffness of 6.9 GPa compared
to 6.7 GPa of the isotropic model. These two results
are in the same order of magnitude as the results of
the Tandon-Weng model (7.1 GPa). The differences
Figure 24. Local stiffness E
11
on the basis of the
simulated ber orientation distribution.
Figure 25. /-graph of the PA6-component with
isotropic and anisotropic properties (CCM model)
compared to the Tandon-Weng model.
between isotropic and anisotropic simulations are at-
tributed to the fact that the CCM model does not take
into account ber-matrix debonding and ber fail-
ure.
5.2.3. Results: Statistical Combined Cell Model
The two Weibull parameters, for interface and ber
failure for the polypropylene (Sec. 5.2.1), are nu-
merically identied by using the data from mi-
cromechanical models and the calculated nite el-
ement results to t the experimental curves. For
this purpose, unit cell models with bers reinforced
composites and the stress-strain behavior due to
debonding and ber breaking are calculated. Tensile
test data are taken from experimental tests at IKP,
University of Stuttgart [78]. The simulated Weibull
Figure 26. The Weibull curve is compared to the
experimental curve to determine Weibull parameters
[9].
curves are calculated using Eqs. 12 and 13. Then,
346 DAMAGE SIMULATION
the values of m and are determined. In this exam-
ple the Weibull parameters for interface failure are
m
L
= 1.4 and
0L
= 170 MPa and the Weibull pa-
rameters for ber failure are determined as m
F
=3.5
and
0F
= 250 MPa (Fig. 26).
Consideration of ber fractures
In [58, 59] it has been experimentally established
that prior to the failure of the composite, fracture
of brittle bers takes place in short ber reinforced
PMCs with glass bers under tensile stress. Figs. 27
and 28 show the stress-strain curves using the Statis-
tical Combined Cell Model. In Fig. 27 the Weibull
modulus m was varied from 8 to innite for
0
=
450 MPa and in Fig. 28 the characteristic stress
0
of bers was varied from 500 MPa to innite for
m = 12 (8.1% vol. bers). It was found that an
increase in ber volume fraction from 8.1% vol. to
13.1% vol. results in an increase of the stress-strain
curve [7]. With decreasing Weibull modulus and de-
creasing characteristic stress
0
of bers, the stress-
strain curve of the composite decreases. The curve
for m = (Fig. 27) and
0
= shows that there is
hardly any damage in the bers. The same result can
be obtained by applying the CCM (Fig. 18). The de-
viation between numerical and experimental results
with the SCCM is smaller than with the CCM. Close
Figure 27. Comparison of stress-strain curves for
different Weibull modules m and
0
= 450 MPa for
polymer matrix composite (8.1% vol. bers) with
planar random short bers [7].
agreement is found when m = 12 and
0
= 450 MPa
for the polymer matrix composite with 8.1% vol.
short glass bers. The probabilities of ber frac-
ture for PMCs depending on total strain are shown in
Fig. 29. The characteristic stress
0
of the bers was
varied from400 MPa to 500 MPa. It can be seen that
the fracture probability of the bers increases with
decreasing characteristic stress
0
. The zero area
of the fracture probability of bers is extended with
increasing characteristic stress
0
. This means that
Figure 28. Comparison of stress-strain diagrams
for different characteristic stresses
0
with a differ-
ent Weibull modulus m = 12 (8.1% vol. bers) for
polymer matrix composite with planar randomshort
bers [7].
Figure 29. Fracture probabilities for different char-
acteristic stresses
0
for polymer matrix composite
with planar random short bers (8.1% vol.) [7].
the stronger the ber in the composite, the smaller
the probability for the ber to break.
Consideration of the adhesion effect between
bers and matrix
The Weibull moduli and the characteristic stresses
(Eq. 8) are varied and the results are depicted in
Fig. 30. Here, it can be recognized that at different
characteristic stresses and at different Weibull mod-
uli the global stress at strains under 0.5% follows
the experimental curve. Increasing Weibull modu-
lus (Fig. 30a) and characteristic stresses (Fig. 30b),
the global stress increases. Through a compari-
son between the simulation and the experiment, the
Weibull modulus and the characteristic stress were
derived as m
G
= 5 and
G
0
= 450 MPa.
Consideration of ber fractures and ber-matrix
debonding
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 347
(a)
(b)
Figure 30. Comparison of the stress-strain curves
with the experimental results: a) with different
Weibull moduli at
G
0
= 450 MPa, and b) with differ-
ent characteristic stresses
G
0
and a Weibull modulus
of m
G
= 5.
The parameters of Weibulls law for short ber re-
inforced thermoplastics with different ber volume
fractions, gathered from the simulation, are recapit-
ulated in Tab. 3. Through the insertion of the param-
eters of Weibulls law in Eq. 14, and in Eqs. 4, 5
and 7, the stress-strain curve of short ber rein-
forced thermoplastics with sparse ber-matrix adhe-
sion can be described. The calculated stress-strain
Parameters 8.1% vol.

G
0
(MPa) 450
m
F
12

G
0
(MPa) 450
m
G
5
Table 3. Parameters of Weibulls law for ber rein-
forced PMCs.
curve of the PMCs with different ber-matrix ad-
hesion and consideration of the ber fractions and
ber-matrix debonding are compared to the experi-
mental results in Fig. 31. The simulation reproduces
Figure 31. Comparison of the calculated stress-
strain curves of ber reinforced composites (8.1%
vol. bers) with good or sparse ber-matrix ad-
hesion with the experimental results under tensile
loading.
well the curves obtained experimentally when the
global strain is smaller than 1.5%. At strains larger
than 1.5%, the simulation results deviate from the
experimental ones. This deviation permits to guess
that crack propagation in this area plays an impor-
tant role.
Cyclic simulations
A simple mesoscopic model of randomly distributed
8.1% vol. ber reinforced polypropylene matrix
composite with 3D continuum cubic elements is
simulated. Each element of the model is character-
ized by the properties of unit cells of different ber
volume fractions. In this model bers are assumed
to be scattered randomly throughout the composite
and to be aligned parallel to the loading direction
(Fig. 17a). A Gaussian distribution is used to de-
scribe the random distribution of bers in the com-
posite. In conjunction with the available test data,
the hypothesis is made that the ber volume frac-
tions vary between 1.8%, 3.8%, 8.1% and 13.1%
throughout the whole composite (Figs. 32, 33). In
order to introduce different damage conditions in the
above composite, different loading conditions with
different ber volume fractions are studied and dam-
age behaviors are introduced into each element of
the mesoscopic model. In strain-controlled simu-
lations strain is kept constant at 3%. Following the
algorithmand the described methodology of the Sta-
tistical Combined Cell Model for the cyclic simu-
lations, the cyclic loading process is continued for
two heterogeneous mesoscopic models of different
ber arrangements (model 1 and model 2). It is
348 DAMAGE SIMULATION
Figure 32. Randomdistribution of ber volume frac-
tions in a mesoscopic composite model [9].
Figure 33. Gaussian distribution of ber volume
fractions of the elements of a mesomodel [9].
seen from the stress-strain curve that after 8-9 cy-
cles, stabilized cycles are found in the case of 3%
strain (Fig. 34). Results correspond to the predic-
tion of material behavior of the polypropylene ma-
trix composite under cyclic loading including dam-
age effects and plasticity. As Fig. 34 shows, model
2 has a stronger damage effect than model 1. In
the elastic region there is no signicant change of
material property for either model. However, in the
plastic region only model 2 shows decreased elasto-
plastic behavior in comparison with the model 1 due
to signicant damage in the composite. Since the
evolution of damage depends on the heterogeneous
arrangements of bers, model 1 shows higher stiff-
ness than model 2 and good fatigue behavior under
cyclic loading.
Inuence of damage on the effective Youngs
modulus
The effect of damage due to debonding and ber
failure can be seen in the reduction of total stiff-
ness. The effective Youngs modulus is reduced con-
siderably due to damage after each cycle (Fig. 35).
Mechanisms of fatigue damage in composites re-
sult in cracks of various orientation, size and ge-
ometry as described above. These cracks are orig-
%
Figure 34. a) Two different heterogeneous models,
and b) cyclic stress-strain behavior of PMC under
damage for 8.1% vol. Polypropylene/glass ber [9].
Figure 35. Effect of damage (ber failure and ber-
matrix debonding) under cyclic loading on the re-
duction of the effective Youngs modulus without
consideration of the overall failure [9].
inated from different microscopic damage mecha-
nisms, which leads to the degradation of the overall
material properties, including stiffness and strength
in various directions.
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 349
From Fig. 35 effects of the heterogeneity of the
mesoscopic models on the degradation of the effec-
tive Youngs modulus with the increase in number of
cycles can be also predicted. In model 2 bers are
arranged more heterogeneously than in model 1. For
this reason, the evolution of damage is more visible
in model 2 than in model 1 and the reduction of the
effective Youngs modul is more pronounced.
5.2.4. Conclusions
The statistical strength of short ber reinforced
PMCs with different volume fractions was inves-
tigated in this section. The Combined Cell Mod-
els (CCM) were limited to simulate the linear me-
chanical behavior of short ber reinforced compos-
ites. Particularly, for the linear behavior of the stud-
ied material good numerical results can be obtained.
Statistical Combined Cell Models (SCCM) can be
used to predict additionally the failure properties
of short ber reinforced composites with consider-
ation of ber fractures and/or damage in the bound-
ary layer at the interface between bers and ma-
trix. With the developed SCCM, it is possible to
simulate the mechanical behavior of short ber rein-
forced composites with consideration of the damage
between matrix and ber. From numerical and ex-
perimental results the Weibull parameters were ob-
tained for the studied PMC injection molded mate-
rials.
A micromechanical fatigue damage model based
on the statistical microscopic damage law was pre-
sented. Statistical damage was described by the
Weibull damage law taking into consideration dif-
ferent ber volume fractions in the composite. Fiber
failure as well as ber-matrix interfacial debonding
in the composite are considered as damage mech-
anisms, which were introduced in heterogeneous
mesoscopic models where the random distributions
of bers were determined by Gaussian distribution.
Here, different heterogeneous arrangements inu-
encing the fatigue behavior are possible. Neverthe-
less, the proposed fatigue model can be applied to
different ber reinforced composites, as long as the
Weibull failure parameters for each composite are
determined. It was also found that the proposed
model can predict the experimental behaviors, but
as matrix-cracks are not included in the model, the
overall behavior can be varied accordingly. Further
modications of the model can be made by includ-
ing matrix cracking.
5.3. Gypsum Fiber Composites
5.3.1. Material
Cellulose ber reinforced gypsum based materials
are gaining increasing importance in the building in-
dustry. The material belongs to the short ber com-
posite material class (see Chapter 2.1.1). The non-
combustible panel material is produced in thick-
nesses of 10 to 40 mm and with a ber content of
about 20 % vol. The ber orientation in the compos-
ite is predominantly random planar. A major appli-
cation of the panels consists in sheathing and brac-
ing wall elements in a timber frame. The material
shows a macroscopic response which resembles that
of a ductile material with pronounced strain soften-
ing and high energy dissipation. Damage is local-
ized in a softening zone (crack band) perpendicular
to the loading direction. The key macroscopic fea-
tures of the material are the development of a spe-
cic yield surface with strain softening nature, per-
manent or plastic deformation, stiffness degradation,
and recovery.
The displacement-controlled experiments were per-
formed with unnotched specimens in uniaxial static
and quasi-static cyclic loading conditions. Contact-
free optical strain was measured by applying laser
extensometry and using an optical grid (Chapter
1.2.2). The same test setup (Fig. 36) with different
specimen dimensions (Figs. 36 and 37a) was used
for static and quasi-static cyclic experiments [67].
In the present experiments, the chosen gage lengths
Figure 36. Specimen for the static and quasi-static
cyclic investigations [11].
for the strain measurement are 50 mm (the whole
optical grid length), 20 mm, 7 mm (the softening
350 DAMAGE SIMULATION
zone itself) and 2 mm (the minimum gage length).
The static experiments were monotonic uniaxial ten-
Figure 37. Load scheme for (a) the tension threshold
test, and (b) for the tension-compression test [11].
sion tests with a displacement rate of 0.2 mm/min.
The quasi-static cyclic tests were performed with
two different load schemes. In the rst experiment,
called tension threshold test (Fig. 37a), the applied
displacement was varied between zero and a cer-
tain positive value. In the second experiment, called
alternating tension compression test (Fig. 37b), the
applied displacement was varied between the same
magnitude of positive and negative within the indi-
vidual displacement levels.
The displacement amplitudes of the quasi-static
cyclic tests were determined as multiples of the ulti-
mate force (F
u
) which was investigated in the static
experiments. The yield point was evaluated with
the 0.1 0.4 0.9 F
u
method [80]. In both quasi-
static cyclic experiments the applied displacement
levels were 0.025, 0.050, 0.075, 0.100, 0.125, 0.150,
0.175, 0.200 and 0.250 mm. The displacement lev-
els comprised three cycles each, except for the rst
two, which levels comprised one cycle each.
5.3.2. Results: Plastic-Damage Model
For a successful use of this material in building in-
dustry, a fundamental understanding and numerical
investigations of the material behavior are neces-
sary. The main focus of this section is the model-
ing of this specic material behavior based on av-
eraged material properties obtained from uniaxial
loading experiments. Simulations are carried out for
static and quasi-static cyclic loading conditions as
expected in regular and earthquake situations. The
determination process of material parameters from
the experiments is also discussed. The modeling of
the material behavior has been performed with the
nite element software ABAQUS [68]. The sim-
ulation of the material behavior under static load-
ing has been successfully carried out using the im-
plemented model. The simulation of the hystere-
sis loops occurring under quasi-static cyclic loading
has been achieved with modications of the imple-
mented model concerning the varying elastic stiff-
ness degradation and recovery inside the elastic do-
main.
Static loading conditions
In this section, the original plastic-damage model,
which is available in ABAQUS, is used. The elastic-
ity parameters Youngs modulus E = 3870 MPa and
Poissons Ratio = 0.19 have been used as obtained
from the static experiments. For plasticity and dam-
age, strain softening and damage evolution curves
are required. Apart from these parameters, a further
material constant called dilation angle is neces-
sary. In the following, the determination process of
the above mentioned curves and the dilation angle
are discussed.
The strain softening curve is basically given as yield
stress vs. inelastic strain relation. In order to avoid
mesh sensitivity effects, this curve is further con-
verted to a yield stress vs. displacement relation (ac-
cording to [68]). The inelastic strain is identied
by subtracting the elastic strain from total strain, as
given in Eq. 21:

in
=
el

in
=

E
. (21)
Therefore, using Eq. 21, the inelastic strain
in
can
be computed from the experimental stress strain
curve. The stress-strain curve in the softening
zone of the specimen is used for this purpose.
Furthermore, the inelastic strain
in
is converted to
displacement by multiplying with the length of the
softening zone. In this way, the yield stress vs. dis-
placement relation shown in Fig. 38a is determined.
The evolution of damage has been assumed as it ap-
pears in quasi-brittle materials such as concrete. The
damage evolution curve is given as damage vs. dis-
placement relation (Fig. 38b). From this curve pro-
nounced damage increase is found at the beginning,
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 351
Figure 38. Material input curves for the static sim-
ulation: (a) strain softening, and (b) damage evolu-
tion curve [11].
but later it gradually approaches 1. This is consis-
tent with the experimental observation that the ma-
trix in the present ber reinforced material cracks
near the tension yield point and most of the load car-
rying area is lost. At a later state of loading, in the
post-peak regime, ber breaking and pull-out occur
gradually and the rate of damage development is de-
creased.
The dilation angle controls the orientation of the
ow potential function G (see Fig. 9a), which is de-
ned as
G =
_
(
t0
tan)
2
+ q
2
ptan, (22)
where is the eccentricity (see Fig. 9b and [68] for
further explanations). = 0.1, which is a typical
value for quasi-brittle materials [66] and
t0
= 3.09
MPa is the yield strength in tension derived fromthe
experiment, Fig. 39a. The equivalent effective de-
viatoric stress q and the hydrostatic stress p are de-
ned as
p =
1
3
: I , q =
_
3
2

S :

S , (23)
where I is the unit matrix and S the effective stress
deviator. The material constants and in Eq. 15
have been assumed to take values of = 0.1 and
= 3, which are typical of quasi-brittle materials ac-
cording to [66]. Applying inverse modeling by com-
0 1 2 3 4
0
1
2
3
experiment
simulation
L=20mm
L=50mm
0
1
2
3
0 0.4 0.8 1.2 1.6
Figure 39. Static simulation results: (a) at length
scales of 20 mm and 50 mm, (b) comparison of dila-
tion angle and mesh size, and (c) contour plot of
damage variable d [11].
paring the stress-strain curves of simulation and ex-
periment, the dilation angle is determined.
The 2D simulation has been performed with four-
noded quadrilateral plane stress elements (CPS4R).
From Fig. 36a it can be seen that the length of the
minimum cross-section area of the specimen with
constant width is 50 mm. Assuming stochastically
distributed micro-defects an equal likelihood of fail-
ure occurrence exists throughout this length. There-
fore, to enforce strain softening in that area a weak
section, consisting in a row of nite elements, has
been inserted in the center of the specimen. The
thickness of the weak section corresponds to the
width of the crack band occurring during the exper-
iment. A yield stress of 3.09 MPa is assigned to the
weak section whereas, in the rest of the specimen,
352 DAMAGE SIMULATION
the yield stress is 3.10 MPa. The static simulation
results are presented in Figs. 39a, b and c.
The simulation results for the stress strain relation-
ship at the gage lengths of 20 and 50 mm show close
agreement with the experiment (Fig. 39a). Material
models are usually subjected to mesh sensitivity ef-
fects in modeling strain softening behavior [81]. The
plastic-damage model takes care of the mesh sensi-
tivity, based on the fracture energy criterion of Hille-
borg [81]. The effect of the mesh size (2 mm vs. 4
mm) on the simulated stress-strain behavior is very
small (Fig. 39b).
Fig. 39b also shows the effect of the dilation angle
on the stress-strain behavior. A dilation angle of
53

is determined applying inverse modeling. The


stress-strain curve associated to it is compared to the
result obtained with a dilation angle of 40

. It is
observed that the latter stress-strain curve obtained
with a dilation angle of 40

remains much higher


than with 53

. The dilation angle controls the shape


and the orientation of the plastic ow potential and,
consequently, the direction of plastic ow (Eq. 22).
Referring to Fig. 9a, it is observed that in the case
of a dilation angle of 40

the plastic ow possesses


components in

1
and

2
directions, where

1
and

2
are the directions of principal stresses, although
the direction of loading is in the direction of

2
.
The component of plastic ow in

1
direction con-
sumes additional energy and the resulting stress-
strain curve is, therefore, higher in case of = 40

.
For the dilation angle of 53

, the direction of plastic


ow is almost identical to the direction of loading.
Thus, the stress-strain curve shows close agreement
with experimental results.
The contour plot of Fig. 39c shows the appearance
of the almost fully damaged material state (d 1)
at 1.5% strain, related to a mean strain measurement
length of 50 mm in the center of the specimen. This
is expected because the softening zone is supposed
to appear at the weakest location. Thus the contour
plot of damage parameter d is capable to visualize
the damaged state of the material.
Quasi-static cyclic loading conditions
In this section, the modied plastic-damage model
is used. The material parameter determination pro-
cess, the nite element modeling and the simula-
tion results under quasi-static cyclic conditions as
well as the modications are discussed. The ex-
perimental stress-strain behavior obtained from the
softening zone is simulated using a single element
(a eight-noded, linear interpolating, hexahedral solid
element).
In order to explain the determination process of the
material parameters, as required by the modied ma-
terial model, three intermediate load cycles of the al-
ternating tension compression test have been chosen
(Fig. 40). In the following, the determination pro-
Figure 40. Illustration of the determination process
of the material parameters for the cyclic simulation
[11].
cess of the material input curves, namely, strain soft-
ening (Fig. 41a), damage evolution (Fig. 41b) and
stiffness recovery (Fig. 41c) are discussed. The de-
termination process of the rules governing the vary-
ing stiffness degradation-recovery, which control the
damage parameter d during unloading and reloading
are also discussed. Further details are available in
[13].
The strain softening curve of the material can be
extracted from the envelope of the stress-strain
curve obtained from quasi-static cyclic experiments
(Fig. 40), and it is given to the material model in the
form of a yield stress vs. plastic strain curve. Point 2
in Fig. 40 provides a yield stress of 2.4 MPa. From
there, the rst point A of the strain softening curve is
obtained as 2.4 MPa vs. 0% plastic strain (Fig. 41a).
Next, frompoints 3, 11 and 13 yield stresses of three
more points of the strain softening curve are gath-
ered. When full unloading is done fromthese points,
the points 5, 12 and 14 are reached which corre-
spond to the plastic strains of the points 3, 11 and
13. Thus points B, C and D of the strain softening
curve are obtained. The complete strain softening
curve shown in Fig. 41a is constructed in the same
manner from further cycles.
The damage evolution curve can be obtained by
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 353
Figure 41. Material input curves for cyclic simula-
tion: (a) strain softening, (b) damage evolution, and
(c) stiffness recovery [11].
computing the tension damage parameter d
t
and the
corresponding plastic strain at certain points of the
uniaxial stress-strain curve (Fig. 40). From the de-
graded unloading stiffness E, the tension damage d
t
can be computed using Eq. 18 (here, d is replaced
by d
t
).
Therefore, the rst d
t
value of the damage evolution
curve can be obtained from the unloading stiffness
beyond point 3. In this material, a varying stiff-
ness recovery is observed through the various un-
loading/reloading stages. The most damaged state
in this cycle is reached in segment 5-6, where the
stiffness is reduced to only 33.9 MPa. With the ini-
tial elastic stiffness, E
0
= 3870 MPa fromEq. 18, the
corresponding tension damage d
t
is 0.99124. The
corresponding plastic strain can be identied in the
same way as for the strain softening curve. Thus, a
plastic strain of 0.0254 can be associated with point
3. Hence, corresponding to point 3 in Fig. 40, point
B in Fig. 41b can be identied. Points C and D can
be identied in the same way from points 3 and 11.
From the next cycles it is possible to get additional
points of the damage evolution curve.
According to Eqs. 16 and 17, the damage parameter
d is controlled by stiffness recovery s, which is ex-
pressed in terms of the weight factor w. As opposed
to the implemented model in ABAQUS [68], w is
not assumed as a constant, but it varies with plas-
tic strain in tension
P
t
. This varying w with plastic
strain is given as a material input to the modied
model as a stiffness recovery curve. Corresponding
to point 3 in Fig. 40, the damage parameter d can
be found from the slope of the segment 3-5. Hence,
from Eqs. 16 and 17, w is identied as w=1d/d
t
.
From point 3 in Fig. 40 the point B in Fig. 41c is de-
rived. The corresponding plastic strains
t
are eval-
uated in the same way as described for the strain
softening curve. Similarly, points from the next cy-
cles can be derived and the stiffness recovery curve
can be constructed as shown in Fig. 41c. The dam-
age parameter d can also be directly identied. To
maintain the framework of the implemented plastic-
damage model, d is expressed in terms of d
t
and w.
In order to describe the determination process of
these rules, three intermediate cycles from the al-
ternating tension-compression test have been com-
pared to the simulated ones and the simulated curve
has been divided accordingly into several segments
(Fig. 40). Segment 1-2 belongs to the initial loading
at undamaged state (d = 0). In segment 2-3 (yield
surface), d is controlled by Eqs. 16 and 17 in accor-
dance with strain softening, damage evolution and
stiffness recovery curves. Segments starting from 3-
4-5 up to 8-9-10 lie in the elastic domain and show
varying stiffness degradation and recovery effects.
In the following, the rules governing the varying un-
loading/reloading stiffness (consequently, varying d
from Eq. 18) are derived by analyzing the experi-
mental data from the rst loading cycle. It will be
also shown that the presented rules are applicable
for the further cycles as well.
Segment 3-4-5 (Unloading in tension): the starting
point 3 of segment 3-4-5 (Fig. 40) is the point of
the rst unloading. From this point, the modied
model deviates from the implemented one. Accord-
ing to the implemented model, the unloading is sup-
posed to follow path 3-5, whereas the experimental
result shows an unloading path of higher stiffness.
Hence, in the modied material model path 3-4 is
followed instead of path 3-5. In the developed sub-
routine UMAT, the slope of segment 3-4 is set three
times higher than that of segment 3-5. In order to
determine the position of point 4, it is assumed that
this point stays at a stress level of 33% of that of
point 3 according to the observations in the exper-
iment. The position of point 5 is already known,
because the x-distance of point 5 from the origin is
the plastic strain accumulated along the path 2-3.
Segment 5-6-7 (Reloading in compression): the un-
loading path 5-6 (Fig. 40) is followed in accordance
354 DAMAGE SIMULATION
with the accumulated tension damage d
t
. Stiffness
recovery is not playing any role here and the slope
of 5-6 is related to d
t
through Eq. 18. In the modi-
ed model, point 6 is assumed as the position from
where the material starts to regain its stiffness due
to crack closure effects in compression. The strain
difference
67
between point 6 and point 7 is de-
rived from the experiment as 0.6% (alternating ten-
sion compression) and 1.5% (tension threshold) and
given as a material parameter to the material UMAT
subroutine. Since the stress-strain state of point 7 is
not known in advance, it is given as a material input.
In the present case, it is -4 MPa vs. 0.28% strain.
From point 7 the unloading in compression starts.
Once the position of this point is given as a material
input, this point and the subsequent unloading points
are stored as a state variable in the material UMAT
subroutine. Therefore, the position of the rst un-
loading point in compression (point 7) as a material
parameter is enough to predict the behavior of the
subsequent loading cycles.
Segment 7-8 (Unloading in compression): from
point 7 (Fig. 40), unloading is done in compression.
Since the material regains its initial stiffness dur-
ing the reloading step in compression, the unload-
ing path is very steep. According to the experiment,
the slope here is the same as for the initial elastic
stiffness (3870 MPa).
Segment 8-9-10 (Reloading in tension): at the begin-
ning of this segment, the cracks reopen. As a result,
the loading path 8-9 (Fig. 40) remains parallel to the
path 5-6. From point 6, the stiffness starts to regain
due to the reorientation of the bers along the ten-
sile loading direction. The strain difference
93
between point 9 and point 3 is observed as 1.5% and
taken as a material parameter. Since the UMAT ma-
terial subroutine stores the stress and strain level at
the previous unloading point 3 as a state variable, the
position of point 9 can be determined. As the posi-
tions of both points 9 and 3 are known, the slope of
path 9-3 can be determined. Furthermore, this slope
has to be reduced by 5% as derived fromexperimen-
tal results. Starting from point 9, the simulated load-
ing path ends up at point 10.
With the modied material model, the material re-
sponse in the softening zone of the specimen has
been modeled using a single element (Fig. 36). The
same level of strain as derived from the experimen-
tally obtained stress-strain response (Fig. 37a and
37b has been applied to the single element. All the
dimensions of the single element have been chosen
as unity. Therefore, the magnitude of strain to be ap-
plied is the same as the displacement applied in the
experiment. The geometry and the boundary con-
ditions, as applied to the single element, are shown
in Figs. 37a, b and c. The same elasticity parame-
ters for Youngs modulus E = 3870 MPa, and Pois-
sons Ratio = 0.19 have been used as in the simula-
tion under static loading conditions. Other necessary
material parameters and their determination process
have been visualized in Fig. 36. The simulation of
Figure 42. Cyclic simulation results of the tension
threshold test at all displacement levels [11].
the material response in the softening zone during
the tension threshold test has been performed at four
applied displacement levels (0.125 mm, 0.150 mm,
0.175 mm and 0.200 mm). In Fig. 42 the simula-
tion results at these displacement levels are shown.
The simulation results for the alternating tension-
compression tests have been presented in a similar
way (Fig. 43). Simulations are shown for three dis-
placement levels (0.150 mm, 0.175 mm and 0.200
mm). The basic reason for the modication of the
Figure 43. Cyclic simulation results of the compres-
sion test at all displacement levels [11].
implemented plastic-damage model was to incorpo-
rate the capability to reect the varying unloading-
reloading stiffnesses observed from the quasi-static
cyclic experiments. From the comparison of the
simulation results shown in Figs. 42 and 43 it is
found that the basic changes of stiffness during un-
loading and reloading have been captured. In par-
ticular, it is observed that the rules derived from the
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 355
rst load cycle governing the stiffness degradation
and recovery in the elastic domain can be maintained
for the subsequent load cycles.
The non-linear stress-strain response with contin-
uously varying unloading-reloading stiffnesses has
been approximated by a multi-linear stress-strain re-
sponse. In the modied model, emphasis has been
put on the close representation of the dissipated
energy inside the hysteresis loops. For instance,
the energy dissipation of the mid-cycle for the ten-
sion threshold test (displacement level of 0.150 mm,
Fig. 42) is 0.0216 MJ/m
3
(experiment) and 0.0180
MJ/m
3
(simulation), respectively. In the case of
alternating tension-compression test (displacement
level of 0.175 mm, Fig. 43) the mid-cycle energy
dissipations are found to be 0.0477 MJ/m
3
(exper-
iment) and 0.0455 MJ/m
3
(simulation). Therefore,
for these cycles, approximately 16% deviation is
observed regarding dissipated energy in the tension
threshold test while it is only 5% for the alternating
tension-compression test.
5.3.3. Conclusions
This section has described the nite element analy-
sis of the material behavior of a cellulose ber rein-
forced gypsummatrix composite. For the simulation
of the material behavior the plastic-damage model
of Lubliner et al. [66] and Lee and Fenves [69] has
been used for this specic material.
The simulations have been performed for static ten-
sile and quasi-static cycling loading of necked spec-
imens of a cellulose ber reinforced gypsum com-
posite. The introduction of a slightly pre-weakened
section in the center of the tensile bar, which re-
ects the heterogeneities of the material, has led to
a close representation of the material behavior of
the static experiment for different gage length with
nearly mesh independences and a unique set of pa-
rameters. The representation of the damage parame-
ter d as a contour plot reects well the damaged state
of the material.
Additionally, the dilation angle , which controls
the direction of the plastic ow as a parameter of
the plastic-damage model, has been determined to
= 53

using inverse modeling. Due to the complex


behavior of the material in the unloading-reloading
regime during quasi-static cyclic loading, the model
implemented in ABAQUS has been basically mod-
ied within the elastic domain. A material rou-
tine UMAT has been developed, which describes the
loading cycles in close agreement to the experiment.
In particular, the approximation of the energy dis-
sipation was improved considerably. On the basis
of these achievements, further studies can be per-
formed to adapt the so far developed material model
to the structural level. A further goal of future inves-
tigations will be the simulation of a macromechani-
cal seismically loaded component which is xed on
a timber frame by dowelled joints.
6. SUMMARY
In this Chapter an overview of damage modeling
in composites was given. Many approaches were
developed on the basis of simple unit cell models,
which, however, exhibit limitations which can be
overcome by applying modied models. As ex-
amples, the Self-consistent Model and the (Statis-
tical) Combined Cell, as well as the plastic-damage
model were described. These models have been ap-
plied to relevant technical materials such as Metal
Matrix Composites, Polymer Matrix Composites
and cellulose ber reinforced composites. It has
been possible to simulate the static and quasi-static
cyclic mechanical behavior of ber and particle re-
inforced composites as well as the damage process.
Depending on the specic material, certain mod-
els show advances to represent the particular be-
havior. The unit cell models presented here inte-
grate micromechanical damage phenomena, such as
inclusion-matrix debonding and inclusion cracks. In
future works macromechanical features (e.g., matrix
cracks) could be included to extend the applicability
of the simulations to large plastic deformations.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge support from
ENSAM, Paris, for supplying test data of Al/Al
2
O
3
,
as well as DAAD under contract no. D/0205761
for nancial supporting the co-operation with EN-
SAM. Additionally, the authors thank the Institute
for Polymer Testing and Polymer Science (IKP) for
providing the mechanical data and the SEM mi-
crographs. The following scientists contributed to
this Chapter (in alphabetical order): S. Aicher, O.
Bullinger, G. Busse, R. Finn, H. Gerhard, G. Lasko,
W. Lutz, T. Rahman, S. Schmauder, R. St oel and
K. Zhu. Further thanks go to A. Wanner, S. Predak
and R. Finn for the experiments and A. J ackel for
image editing.
356 DAMAGE SIMULATION
REFERENCES
[1] J. Bohse, K. Kroh: Micromechanics and acous-
tic emission analysis of the failure process of
thermoplastic composites, in Journal of Mate-
rials Science 27 (1992), p. 298-306.
[2] K. Guo-Zheng, Q. Gao: Tensile properties of
randomly oriented short Al
2
O
3
ber reinforced
aluminium alloy composites, in II. Finite ele-
ment analysis for stress transfer, elastic modu-
lus and stress-strain curve, in Composites A 33
(2002), p. 657-667.
[3] K. Derrien, D. Baptiste, D. Guedra-Degeorges,
J. Foulquier: Multiscale modeling of the dam-
aged plastic behavior and failure of Al/SiCp
composites, in International Journal of Plastic-
ity 15 (1999), p. 667-685.
[4] M. Dong, S. Schmauder: Transverse mechan-
ical behavior of ber reinforced composites -
FE Modeling with embedded cell models, in
Comp. Mat. Sci. 5 (1996), p. 53-66.
[5] Reprint from Acta Materialia, 44, M. Dong,
S. Schmauder: Modeling of metal matrix
composites by a self-consistent embedded cell
model, p. 2465-2478, Copyright (1996), with
permission from Elsevier.
[6] Reprint from Computational Materials Science
9, M. Dong, S. Schmauder, T. Bidlingmaier, A.
Wanner: Prediction of the mechanical behavior
of short ber reinforced MMCs by Combined
Cell Models, p. 121-133, Copyright (1997),
with permission from Elsevier.
[7] Reprint from Computational Materials Science
28, K. Zhu, S. Schmauder: Prediction of the
failure properties of short ber reinforced com-
posites with metal and polymer matrix, p. 743-
748, Copyright (2003), with permission from
Elsevier.
[8] K. Zhu, S. Schmauder, R. St oel, S. Predak,
G. Busse, W. Lutz: The failure properties of
short ber reinforced composites with polymer
matrix with consideration of the ber/matrix-
debonding, ICPNS 2004, Shanghai (2004).
[9] Reprint from Computational Materials Sci-
ence 36, M.R. Kabir, W. Lutz, K. Zhu, S.
Schmauder: Fatigue modeling of short ber re-
inforced composites with ductile matrix under
cyclic loading, p. 361-366, Copyright (2006),
with permission from Elsevier.
[10] W. Lutz, G. Lasko, S. Schmauder, S. Predak,
O. Bullinger, H. Gerhard, G. Busse: Injec-
tion molding simulation of a glass ber rein-
forced component with a weldline to calcu-
late ber orientation and resulting mechanical
properties, in 19. Stuttgarter Kunststoffkollo-
quium 4V5 (2005).
[11] Reprint fromComputational Materials Science
(in press), T. Rahman, W. Lutz, R. Finn, S.
Schmauder, S. Aicher: Simulation of the me-
chanical behavior and damage in components
made of strain softening cellulose ber rein-
forced gypsum materials, Copyright (2006),
with permission from Elsevier.
[12] L. Mishnaevsky: Numerical Experiments in
the Mesomechanics of Materials - Virtual Test-
ing of Microstructures as a Basis for the Opti-
mization of Materials, in Habilitation, Univer-
sity of Dortmund (2005).
[13] T. Rahman: Simulation of the damage devel-
opment in components made of cellulose ber
reinforced gypsum material, in Master Thesis,
University of Stuttgart (2005).
[14] E. M ader, H.J. Jakobasch, G. Grundke, T. Gi-
etzelt: Inuence of an optimised interphase
on the properties of polypropylene/glass bre
composites, in Composites A 27, (1996), p.
907-912
[15] W. Brostow, R.D. Corneliussen: Failure of
plastics, M unchen, Wien, New York, Hanser
(1986).
[16] R. Kabir: Fatigue modeling of short ber re-
inforced composites with ductile matrix under
cyclic loading, in Master Thesis, University of
Stuttgart (2003).
[17] G. Bao: A Micromechanical model for Dam-
age in Metal Matrix Composites, in AMD
142/MD-34 Damage Mechanics and Localiza-
tion, Ed. J.H.Ju, ASME 1992, p. 1-12.
[18] J. Llorca, A. Needleman, S. Suresh: An Anal-
ysis of The Effects of Matrix Void Growth
on Deformation and Ductility of Metal-
Ceramic Composites, in Acta Metall. Mater. 39
(10/1991), p. 2317-2335.
[19] M.E. Walter, G. Ravichandran, M. Ortiz:
Computational modeling of damage evolution
in unidirectional ber reinforced ceramic ma-
trix composites, in Comput. Mech. 20 (1997),
p. 192-198.
[20] W. Brocks, S. Hao, D. Steiglich: Microme-
chanical modeling of the Damage and Tough-
ness Behaviour of Nodular Cast Iron Materials,
in Proceedings of EUROMECH-MECAMAT
Conference Local Approaches to Fracture 86-
96, Fontainebleau (1996).
[21] D.Z. Sun, M. Sester, W. Schmitt: Development
and Application of Micromechanical Material
models for the Characterization of Materials,
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 357
in Proceedings of EUROMECH-MECAMAT
Conference Local Approaches to Fracture 86-
96, Fontainebleau (1996).
[22] V.V. Mozhev, L.L. Kozhevnikova: Unit
Cell Evolution in Structurally Damageable
Particulate-Filled Elastomeric Composites un-
der Simple Extension, in J. Adhesion 55
(1996), p. 209-219.
[23] V.V. Mozhev, L.L. Kozhevnikova: Highly Pre-
dictive Structural Cell for Particulate Poly-
meric Composites, in J. Adhesion 62 (1997),
p. 169-186.
[24] D. Adams: Inelastic Analysis of a Unidirec-
tional Composite Subjected to Transverse Nor-
mal Loading, in J. Compos. Mater. 4 (1970),
p. 310.
[25] J.R. Brockenbrough, S. Suresh: Plastic defor-
mation of continuous ber-reinforced metal-
matrix composites: effects of ber shape
and distribution, in Scripta Metall. Mater. 24
(1990), p. 325-330.
[26] J.R. Brockenbrough, S. Suresh, H.A. Wie-
necke: Deformation of metal-matrix compos-
ites with continuous bers: geometrical effects
of ber distribution and shape, in Acta Metall.
Mater. 39 (1991), p. 735-752.
[27] T. Nakamura, S. Suresh: Effects of thermal
residual stresses and ber packing on deforma-
tion of metal-matrix composites, in Acta Met-
all. Mater. 41 (1993), p. 1665-1681.
[28] S. Jansson: Homogenized nonlinear constitu-
tive properties and local stress concentrations
for composites with periodic internal structure,
in Int. J. Solids Structures 29 (1992), p. 2181-
2200.
[29] D.B. Zahl, S. Schmauder, R.M. McMeeking:
Transverse strength of metal matrix compos-
ites reinforced with strongly bonded contin-
uous bers in regular arrangements, in Acta
Metall. Mater. 42 (1994), p. 2983-2997.
[30] C. Dietrich: Mechanisches Verhalten von
Zweiphasengef ugen: Numerische und exper-
imentelle Untersuchungen zum Einuss der
Gef ugegeometrie, in VDI-Fortschrittsberichte
18 (128), VDI-Verlag, D usseldorf (1993).
[31] M. Sautter: Modellierung des Verfor-
mungsverhaltens mehrphasiger Werkstoffe mit
der Methode der Finiten Elemente, in PhD.
thesis, University of Stuttgart (1995).
[32] G.L. Povirk, M.G. Stout, M. Bourke, J.A.
Goldstone, A.C. Lawson, M. Lovato, S.R.
Macewen, S.R. Nutt, A. Needleman: Ther-
mally and mechanically induced residual
strains in Al-SiC composites, in Acta Metall.
Mater. 40 (1992), p. 2391-2412.
[33] C. Dietrich, M.H. Poech, S. Schmauder,
H.F. Fischmeister: Numerische Model-
lierung des mechanischen Verhaltens von
Faserverbundwerkstoffen unter transver-
saler Belastung, in Verbundwerkstoffe und
Werkstoffverbunde, Eds. G. Leonhardt et
al., DGM-Informationsgesellschaft mbH,
Oberursel (1993), p. 611-618.
[34] H. B ohm, F.G. Rammerstorfer, E. Weissenbek:
Some simple models for micromechanical in-
vestigations of ber arrangement effects in
MMCs, in Comput. Mater. Sci. 1 (1993),
p. 177-194.
[35] H.J. B ohm, F.G. Rammerstorfer, F.D. Fischer,
T. Siegmund: Microscale arrangement effects
on the thermomechanical behavior of advanced
two-phse materials, in ASME J. Engng. Mater.
Technol. 116 (1994), p.268-273.
[36] G. Bao, J.W. Hutchinson, R.M. McMeek-
ing: Particle reinforcement of ductile matrices
against plastic ow and creep, in Acta Metall.
Mater. 39 (1991), p. 1871-1882.
[37] V. Tvergaard: Analysis of tensile properties for
a whisker-reinforced metal-matrix composite,
in Acta Metall. Mater. 38 (1990), p. 185-194.
[38] C.L. Hom: Three-dimensional nite element
analysis of plastic deformation in a whisker-
reinforced metal matrix composite, in J. Mech.
Phys. Solids 40 (1992), p. 991-1008.
[39] E. Weissenbek: Finite Element modeling of
the discontinuously reinforced metal matrix
composites, in PhD. thesis, Technical Univer-
sity of Vienna (1993).
[40] D.B. Zahl, R.M. McMeeking: The inuence of
residual stress on the yielding of metal matrix
composites, in Acta Metall. Mater. 39 (1991),
p. 1171-1122.
[41] M.H. Poech: Deformation of two-phase mate-
rials: application of analytical elastic solutions
to plasticity Microstructure and ow stress of
cell forming metals, in Scripta Metall. Mater.
27 (1992), p. 1027-1031.
[42] L. Farrissey, S. Schmauder, M. Dong, E.
Soppa, M.H. Poech, P. McHugh: Investigation
of the strengthening of particulate reinforced
composites using different analytical and nite
element models, in Computational Materials
Science 15 (1999), p. 1-10.
[43] J.M. Duva, A self-consistent analysis of the
stiffening effect of rigid inclusions on a power
law material, in ASME J. Engng. Mater. Tech-
nol. 106 (1984), p. 317-321.
358 DAMAGE SIMULATION
[44] M. Sautter, C. Dietrich, M.H. Poech, S.
Schmauder, H.F. Fischmeister: Finite element
modeling of a transverse-loaded bre compos-
ite. Effects of section size and net density, in
Comput. Mater. Sci. 1 (1993), p. 225-233.
[45] D.B. Zahl, S. Schmauder: Transverse strength
of continuous ber metal matrix composites, in
Comput. Mater. Sci. 3 (1994), p. 293-299.
[46] T. Christman, A. Needleman, S. Suresh: An
experimental and numerical study of deforma-
tion in metalceramic composites, in Acta Met-
all. Mater. 37 (1989), p. 3029-3050.
[47] P.M. Suquet: MECAMAT 93, Int. Seminar on
Micromechan. Mater., Editions Eyrolles, Paris,
France (1993), p. 361.
[48] F. Th ebaud: PhD. Dissertation, Universit e de
Paris Sud (1993).
[49] N.J. Sfrensen: A planar model study of creep
in Metal Matrix Composites with misaligned
short bers, in Acta Metall. Mater. 41 (1993),
p. 2973-2983.
[50] N.J. Sfrensen, N. Hansen, Y.L. Liu: On the
inelastic behavior of Metal Matrix Compos-
ites, in Numerical predictions of deformation
processes and the behavior of real materials,
Eds.: S.I. Andersen et al., Roskilde, Denmark
(1994), p. 149-168.
[51] Y.L. Klipfel, M.Y. He, R.M. McMeeking, A.G.
Evans, R. Mehrabian: The processing and me-
chanical behavior of an aluminun matrix com-
posite reinforced with short bers, in Acta Met-
all. Mater. 38 (1990), p. 1063-1074.
[52] I. Dutta, J.D. Sims, D.M. Seigenthaler: An an-
alytical study of residual stress effects on uni-
axial deformation of whisker reinforced metal-
matrix composites, in Acta Metall. Mater. 41
(1993), p. 885-908.
[53] W.M.G. Courage, P.J.G. Schreurs: Effective
material parameters for composites with ran-
domly oriented short bers, in Computers &
Structures 44 (1992), p. 1179-1185.
[54] A. Dlouhy, N. Merk, G. Eggeler: A microme-
chanical study of creep in short ber reinforced
aluminium alloys, in Acta Metall. Mater. 41
(1993), p. 3245-3256.
[55] A. Dlouhy, G. Eggeler, N. Merk: A mi-
cromechanical model for creep in short ber
reinforced aluminium alloys, in Acta Metall.
Mater. 43 (1995), p. 535-550.
[56] D.B. Zahl, S. Schmauder: Transverse strength
of Metal Matrix Composites reinforced with
strongly bonded continuous bers in regular ar-
rangements, in Acta Metall. Mater. 42 (1994),
p. 2983-2997.
[57] H. Dietzhausen, M. Dong, S. Schmauder: Nu-
merical simulation of acoustic emission in ber
reinforced polymers, in Computational Materi-
als Science 9 (1997), p. 121-133.
[58] M. Dong, Characterization of failure prop-
agation in ber reinforced materials using
non-destructive methods, Cooperated Research
Center, Report 1994-1997 University Stuttgart
(1997).
[59] K. Zhu, M. Dong, K. Turan, N. Nguyen, Char-
acterization of failure propagation in ber re-
inforced materials using non-destructive meth-
ods, Cooperated Research Center, Report
1997-1999 University Stuttgart (1999)
[60] K. Zhu, S. Schmauder: Modelierung des
Sch adigungsverlaufs in Faserverbundwerkstof-
fen mit duktiler Matrix, SFB381, Ergebnis-
bericht (2000-2002).
[61] M.L. Dunn, H. Ledbetter: Elastic-Plastic be-
havior of textured Short Fiber composites, in
Acta Metall. Mater. 45 (8/1997), p. 3327-3340.
[62] J. Llorca: A numerical analysis of the damage
mechanisms in metal-matrix composites under
cyclic deformation, in Computational Materi-
als Science 7 (1996), p. 118-122.
[63] K. Derrien, J. Fitoussi, G. Guo, D. Baptiste:
Prediction of the effective damage properties
and failure properties of nonlinear anisotropic
discontinuous reinforced composites, in Re-
port, Laboratory LM3, ENSAM, CNRS URA
1219, Paris, France.
[64] J. Lemaitre: A course on damage mechanics,
Second Ed., Springer, Paris (1996).
[65] ABAQUS Version 5.8, Hibbit, Karlsson &
Sorensen.
[66] J. Lubliner, J. Oliver, S. Oller, E. O nate: A
plastic-damage model for concrete, in Int. J.
Solids and Structures 25 (3/1989), p. 229-326.
[67] S. Aicher, R. Finn: Fracture characterization
of cellulose ber gypsum composite subject to
in-plane tension loading, in Otto-Graf-Journal
15 (2004), p. 91-102.
[68] N. N., ABAQUS Theory manual, version 6.4,
3, Chapter 4.5.2 (2003), p. 1-13.
[69] J. Lee, G.L. Fenves: plastic-damage model for
cyclic loading of concrete structures, in J. Eng.
Mech. 124 (8/1998), p. 892-900.
[70] W. Lutz, T. Rahman, S. Schmauder, R. Finn,
S. Aicher: Simulation cellulosefaserverst arkter
Gips-Verbundwerkstoffe, 6. Leipziger Fachta-
gung Innovationen im Bauwesen - Faserver-
bundwerkstoffe, Leipzig, 1. - 2.12. (2005).
4.1 Modeling of Damage in Fiber and Particle Reinforced Composites 359
[71] W.C., Jr. Harrington: Mech. Properties Metall.
Compos., S. Ochiai, ed., Marcel Dekker Inc.,
NY, USA, 759(1993).
[72] T. Bidlingmaier, D. Vogt, A. Wanner:
Sch adigungsentwicklung bei der Kriechver-
formung einer kurzfaserverst arkten Alu-
miniumlegierung, in Verbundwerkstoffe und
Werkstoffverbunde, Ed.: G. Ziegler, DGM
Informationsgesellschaft Verlag (1995),
p. 245-248.
[73] LASSO Engineering Association, Markoman-
nenstr. 11, 70771 Leinfelden-Echterdingen,
Germany.
[74] PDAEngineering, 2975 Redhill Avenue, Costa
Mesa, California 92626, USA.
[75] T. Bidlingmaier, A. Wanner, E. Arzt: unpub-
lished results.
[76] E. Le Pen, D. Baptiste, G. Hug: Multi scale
fatigue behavior modeling of Al-Al
2
O
3
short
ber composites, in Int. J. of Fatigue 24 (2002),
p. 205-214.
[77] K. Derrien, D. Baptiste, D. Guedra-Degeorges,
J. Foulquier: Multiscale modeling of the dam-
aged plastic behavior and failure of Al/SiCp
composites, in Int. J. of Plasticity 15 (1999),
p. 667-685.
[78] G. R ub, N.C. Davidson, B. M oginger, P. Ey-
erer: Mechanical property simulation of par-
tially oriented composites using the Elemen-
tary Volume Concept, Conf. Proc. Polym.
Proc. Soc. Ann. Meet. (PPS 17), May 21-24,
Montr eal/Canada (2001).
[79] G.P. Tandon, G.T. Weng: The effect of aspect
ratio of inclusions on the elastic properties of
unidirectionally aligned composites, in Polym.
Comp. (1984), p. 327-333.
[80] B. Dujic, R.

Zarnic: Comparison of hystere-
sis responses of different sheathing to framing
joints, in Int. Council for Research and Inno-
vation in Building and Construction. Working
Commission W18. Proceedings Meeting 36,
Colorado, (2003) paper CIB W18/36 7 5.
[81] A. Hilleborg, M. Modeer, P. E. Petersson:
Analysis of crack formation and crack growth
in concrete by means of fracture mechanics
and nite elements, in Cement and concrete re-
search 6 (1976) p. 773-782.

Vous aimerez peut-être aussi