PERIODICA POLYTECHNICA SER. MECH. ENG. VOL. 46, NO. 2, PP.

139158 (2002)
FINITE ELEMENT MODELLING OF THE DAMAGE AND
FAILURE IN FIBER REINFORCED COMPOSITES (OVERVIEW)
Andrs SZEKRNYES

and Jzsef UJ

Research Group for Dynamics of Machines and Vehicles

Hungarian Academy of Sciences
H1051 Budapest, Hungary
Phone: +36 1 463 1170
e-mail: [email protected]

Department of Applied Mechanics

Budapest University of Technology and Economics
H1512 Budapest, Hungary
Phone: +36 1 463 2228
e-mail: [email protected]
Received: Mai 2, 2001
Abstract
The failure and the analysis of composite structures can be various. The failure modes can be inves-
tigated separately, but these modes generally arise together (e.g.: matrix cracking and delamination).
Due to this experimental observation the main problem is to construct models that can follow more
than one of the failure modes. To solve this problem the most proper method is the nite element
method (FEM), because of its simplicity. The aim of this paper is to summarize a few publications
about the damage and failure of composite materials and structures. The publications were grouped
according to the mode of failure, the aim is to give a short description about the nite element
modelling of the given failure mode. The models were reconstructed using the nite element code
COSMOS/M.
Keywords: composite, micromodel, macromodel, unit cell, specimen, FE analysis.
1. Introduction
Watching the structure of composites we can notice its heterogenous nature. The
composite material consists of bers and matrix. The bers are embedded into the
matrix. During the manufacturing process the matrix is warmed up to high tem-
perature, then the bers are added. Due to the high temperature and the mismatch
between the thermal properties of the components, there is interaction between the
bers and the matrix. The formation of interface is a consequence of this operation.
Hence the composite is a three-phased material. The length and the arrangement of
the bers are various. The most frequent ones are the long unidirectional continuous
and the short unidirectional arrangements. Another important feature of the com-
posites is the periodicity. This feature can make the FE analysis of the composite
materials simple. In literature there are three basic types of analysis [11, 19].
140 A. SZEKRNYES and J. UJ
Microscale analysis:
The modellingof the local constituents (ber, matrix, interface) usingthe periodicity
and the symmetry.
Macroscale analysis:
The modelling of the global composite structure using the homogenized material
properties of the composite.
Mesoscale analysis:
A scale between the macro- and microscale, which can be a micromodel that is not
periodic in all directions or, considering laminated structures, this may mean the
analysis of one elementary ply.
In fact, there is another scale, named multiscale analysis, coupling the three
ones above. In this case we need parallel computers with big capacity. Multiscale
approach was presented by FEYEL and CHABOCHE in [4] for composites.
2. Material Models for the Finite Element Analysis
There are signicant differences between the material properties of the ber and
matrix. The stiffness of ber is much higher than the matrix stiffness. Major part
of the loading is transferred onto the ber. Experimental data prove that the global
composite structure behaves as a linear elastic material [10]. If local analysis is
applied, then the considering of heterogeneity is necessary.
2.1. Materials of the Reinforcing Fibers
The material of bers has a wide range of type (e.g.: glass, carbon...etc.). It is as-
sumed that the glass ber is isotropic, the carbon ber is transversely isotropic linear
elastic material. The mechanical and other properties of the bers are available in
the literature [10].
2.2. Material Modelling of the Matrix
The modelling of the matrix is a far complex problem. In general the matrix is a
nonlinear elastic isotropic and hysteretic material. In case of polymer matrix (e.g.:
epoxy, vinylester, polyester...etc.) the applying of linear elastic material model
FINITE ELEMENT MODELLING 141
can give wrong results during the FE analysis. Due to this experience it is useful
to apply rheological material models for the matrix. These are the viscoelastic
models that consist of the well-known Maxwell and KelvinVoight elements and
the combinations of them [2, 6]. Metal matrix composites (MMC) require using
an elastoplastic, viscoplastic and elastoviscoplastic material model [3, 4]. Inelastic
material models are reachable in [14]. Inelastic behaviour can be simulated by
stiffness or strength degradation. Previous investigations describe that using a linear
material model damage occurs in higher loading than in the case of the nonlinear
model [19]. Because of that the nonlinear material law describes more accurately
the stress-strain relationship than the assumed linear. Commercial FE programs
usually contain nonlinear material models, but there is a chance to build in another
[19]. For that a user dened subroutine and the modication of it is needed [15].
2.3. Properties of the Interface
The modelling of interface also can cause problems. The interface has been widely
acknowledged, but the interface properties are not yet available [1]. The proper-
ties of the interface play critical role because the interface transfers the load onto
the bers. Experiments prove that the thickness of interface is not negligible.
Interface can be modelled as elastic isotropic material, but there are also other ap-
proximations. For example an elastic/perfectly plastic spring model can describe
the interface properties [3]. The spring model has different stiffness in the radial,
tangential and longitudinal directions. ANTIFANTIS described an interface model,
which assumes that the properties of interface (E, ) vary along the thickness of
interface [1]. Varying the adhesion parameters the perfect and imperfect interface
can be simulated. Other authors provided a cohesive interface model for metal
matrix, short ber composites [21]. Experiments show that stress peaks can occur
in the interface region [1]. In most cases the initiation of the damage depends on the
interface properties. Several papers neglect the interface; these models assume per-
fect contact between the ber and the matrix [3, 19]. The interface plays signicant
role in the damage process of composites.
3. Macro- and Micromodels, Unit Cells
3.1. Macroscale Finite Element Models
In the case of macroscale analysis we consider the composite structure as a homoge-
neous orthotropic continuum. Knowing the material properties of the constituents
and the ber-volume fraction we can compute the homogenized properties with
the law of mixture [10]. Most of the commercial FEM programs contain two- and
three-dimensional elements that are suitable for constructing global models. These
are shell, solid and multilayer shell elements. For the using of these elements the
142 A. SZEKRNYES and J. UJ
ber orientationis required. Applyingthese global models local microprocesses, in-
cluding local stress-strain distribution cannot be followed. These processes are very
important in the failure of composites (e.g.: interface damage, ber fracture...etc.).
3.2. Microscale Finite Element Models
Unit cells and RVE (Representative Volume Element) models are the most con-
venient forms in the micromechanical analysis. Making unit cells depends on the
geometry and the applied load. Depending on the structure and ber orientation
these models can be two- and three-dimensional ones. Due to the heterogeneity
different material models must be applied to describe the three-phased composite.
The construction and the mesh size of the nite element model have a great effect
on the results of the analysis. Constructing the FE model the prescription of proper
boundary conditions is required.
Fig. 1. 2D nite element model, square ber arrangement
Generally, the load of micromechanical models is displacement or strain con-
trolled. This means that one of the boundary curve or surface of the model is loaded
with displacement in a given direction [1, 19]. Fig. 1 shows the making of 2Dmodel,
the cut is perpendicular to the longitudinal direction of the bers. Assuming the
periodicity of composite a double symmetrical unit cell can be made. Using sym-
metrical conditions the FE model is only quarter of the former one. Fig. 1 also
shows the proper boundary conditions. In Fig. 1 the ber and the matrix are cov-
ered by triangular, the interface is covered by rectangular plane elements in plane
stress state. Depending on the aim of the analysis, there are several cases when the
cut must be longitudinal.
If a composite structure is built by plys with different orientation a 3D FE
model is required [3, 19]. In Fig. 2 a three-dimensional unit cell and nite element
FINITE ELEMENT MODELLING 143
model are shown. Using the periodicity a 3D multi-cell can be made, the nite
element model is just one eighth of the multi-cell. Proper boundary conditions can
be seen also in Fig. 2. Between ber and matrix a perfect bonding is assumed and
no debonding occurs. Three of the surfaces cannot move at the normal direction.
Remained surfaces can be loaded with given displacements.
Fig. 2. 3D nite element model
The ber arrangement is various. Most of the FEM programs have not got a
well-built in CAD system. These CAD systems do not guarantee smoothness and
accuracy that is always required. Because of that a proper CADsystem is applicable
[16]. Most of the FE codes can handle models drawn in CAD. Complicated ber
arrangements (e.g.: woven fabric, basket weave and triaxial braided unit cells) raise
the claim to use CAD system. More information can be found in [16]. The former
paper presents modelling of composite unit cells using CAD, and also describes the
construction of unit cells with logical (Boolean) operations.
4. Failure Analysis of Composites
Failure means that one of the components reaches the yield stress, then damage
occurrence and progression can be observed. Damage can progress on different
directions. Usuallymatrixcrackingis the rst damage process totake place since the
matrix has the lowest stress to failure. Afailure criterion is needed to establish initial
damage. This can be the maximum stress or strain criterion [19]. Damaged zones
can propagate and reach the border of each other. Finite element tests have a wide
application range. In most of the cases a simplied, loaded model is investigated
by FEM, the distribution of stress and strain elds gives enough information about
the damage process. Models can be analysed with holes and notches, considering
stress concentration effect. The following sections give a short review about the
nite element analysis of damage and failure processes incomposites using previous
publications.
144 A. SZEKRNYES and J. UJ
4.1. Modelling of Interface Damage
Experimental observations show that the damage process can be initiated by the
ber-matrix interface. The damage of the interface causes further problems (e.g.:
debonding, ber pullout). The interface strength plays an important role in the
damage process. Depending on the loading the interface failure be can caused by
the mismatch of the moduli and Poissons ratios. SCHLLER et al. presented a
very useful method for modelling the interface debonding [12]. A single ber is
embedded in a necked matrix block, as shown in Fig. 3. Mechanical loading can
be uniaxial compression or tension. Compression test has two disadvantages. First
the ber often breaks or buckles before interfacial debonding can occur. Another
drawback is the additional transverse stress components, which makes the stress
eld non-axisymmetric one. Under compression the transverse component is a
tensile one between the ber and the matrix.
Fig. 3. Necked single-ber composite specimens
The problem of ber buckling can be avoided by applying the tension test.
The transverse component is compressive; a sufciently small neck can compensate
it (see Fig. 3). The limit stress is determined by the matrix yield stress. The single-
ber specimen can be various, an axisymmetrical model is the most convenient
because the stress eld will be even axisymmetrical. The homogeneous tensile
loading of the interface allows the measurement of the real interface strength. For
the verifying of FE results real specimens are needed. The preparation of ax-
isymmetrical specimens is very difcult. This method cannot follow the effect of
adjacent bers. Interface damage plays an important role in most of the failure
modes.
FINITE ELEMENT MODELLING 145
4.2. The Effects of Thermal Residual Stresses
Opinions differ about thermal residual stresses. Ones say that major part of the
thermal residual stresses relax with the passage of time [19], others say that residual
stresses have signicant effects on the micromechanical behaviour of composites
[5, 7].
Manufacturing process induces thermal residual stresses in composites. After
curing and cooling the composite, the matrix is subjected to a triaxial stress state
[5]. Shrinkage during cooling and the mismatch between thermal properties of the
constituents are the main reasons of residual stresses. The thermal residual stress is a
compressive one in the ber and tensile one in the matrix. If the composite structure
is loaded the triaxial stress state in matrix will increase. FIEDLER et al. investigated
thermal residual stresses using three different models [5]. A concentric cylinder
model, a hexagonal model and a special composite model were used with long
continuous bers. All models consist of two-dimensional plane strain elements.
The models are depicted in Fig. 4. Thermal residual stresses are constants along
the circumference of ber in the cylinder model. The composite model contains
the hexagonal unit cell that is only 10% of the whole model. This local area is
surrounded by elements with homogeneous properties. The hexagonal arrangement
makes possible to investigate the effect of adjacent bers. Thermal residual stresses
were examined along the circumference of the central ber in radial, tangential and
longitudinal directions. The local ber-volume fraction was varied from 20% to
80%. Looking at the hexagonal and composite model residual stresses in all the
three directions show sinus-like distribution along the circumference. Increasing
ber-volume fraction the maximum and minimum values also increase.
Fig. 4. Cylinder (1.), hexagonal (2.) and special composite (3.) models
The hexagonal unit cell shows the results that neighbouring bers more or
less inuence the residual stress components. Results obtained by the composite
model show a decrease of thermal residual stresses with increasing local ber-
146 A. SZEKRNYES and J. UJ
volume fraction. The location and the value of the applied stress at which failure
occurs are inuenced by the local ber-volume fraction. Depending on the applied
matrix criterion the damaged area can be different. The three used criterions are the
parabolic, the strain energy density and the von Mises criterion. Varying the ber-
volume fraction these criterions give very distinct results. The von Mises criterion
predicts over the maximum load, so it is not suitable to predict initial matrix failure.
In fact thermal residual stresses reduce the maximum bearable load, but they do not
affect the location of initial failure.
The effect of thermal residual stresses in short-ber-reinforced composites
was investigated by ZHOU et al. [21] with Al matrix and SiC ber. The applied FE
model and the boundary conditions are shown in Fig.5.
Fig. 5. FE model for short-ber-reinforced Al/SiC composite
Cooling from high to room temperature the residual shear stress concentrates
near the corner of the ber end and induces initial damage along the interface.
Considering thermal cycling the radial, longitudinal and shear stress distributions
were investigated. The results show considerable stresses ahead of the ber end
and around the corner of the ber. Applying thermal cycling interface damage
is more dominant than matrix damage. Thermal cycling decreases the strength
and the toughness of metal matrix composites (MMC). The paper shows that the
thermal cycle range rather than the number of cycles dominate damage process.
Cooling and thermal cycling change the mode of failure. Considering only tensile
load failure arises as matrix damage around the ber end and no debonding occurs.
Considering cooling with tensile load debonding along the ber end and transverse
damage across the matrix occur. Finally applying thermal cycling and mechanical
load the MMC fails by debonding along the ber ends and ber sides, followed by
necking instability of the matrix ligament.
Microscopic behaviour of MMC is strongly inuenced by the inelastic de-
formation of the matrix. ISMAR and co-workers presented a study about MMCs
reinforced by different types of ber [7]. The damage of matrix is affected by the
properties of the bers. A non-symmetrical [0

/90

] ber arrangement model was

investigated, Fig. 6 depicts the 3D model. Boundary conditions are identical with
those shown in Fig. 2. The three different ber types: Al
2
O
3
, SiC and carbon
ber. All bers behave elastically, Al
2
O
3
and SiCare isotropic, the carbon ber
is described as transversely isotropic material. Interface between ber and matrix is
FINITE ELEMENT MODELLING 147
assumed to be very strong and no debonding occurs. The Al
2
O
3
has the highest
tensile modulus, SiC has the lower, and modulus of carbon ber is between the for-
mers. During fabrication process, the composite passes through several temperature
regions. Thermal residual stresses cause local inelastic deformation.
Fig. 6. 3D FE model for metal matrix composite
Acooling process is rstly examined. Seeing the three ber types the inelastic
strain shows highest values in the case of carbon ber. After cooling the equiva-
lent inelastic strain concentrates areas shown in Fig. 6. Inelastic strain distribution
reaches the maximum values where the bers are the closest to each other. Vary-
ing the ber-volume fraction does not signicantly alter the relative distribution
of equivalent inelastic strain but strongly affects the extent of it. Because of that
the effects of thermal cooling is considered in the following investigations of the
mechanical behaviour. Mechanical loading is displacement controlled with given
strain rate. Under monotonic loading Al
2
O
3
-reinforced composite shows maximum
inelastic strain owing to the larger difference between the tensile modulus of the
ber and matrix. The stress-strain curve of the three ber reinforced composites
shows that the slowest decline of the tangent modulus can be observed in the case
of carbon ber. On the other hand Al
2
O
3
ber has the fastest decline. Cyclic load-
ing was considered with given strain rate and temperature. Metal matrix behaves
inelastically applying the given loading amplitude. The density of inelastic perfor-
mance after forty cycles decreases depending on the ber type. Increasing loading
amplitude inelastic behaviour strongly increases. Cyclic loading indicates damage
progression and damage accumulation. Damage is distinctly concentrated in two
matrix areas; the Al
2
O
3
-reinforced composite exhibits the highest damage values.
Heightening the ber-volume fraction, the inelastic deformation slightly increases.
Fiber with large tensile modulus increases the strength of composite. On the other
hand the danger of premature failure during cyclic loading has to be considered.
Finally, the ber-volume fraction decisively affects the mechanical behaviour of
metal matrix composites.
148 A. SZEKRNYES and J. UJ
4.3. Matrix Damage and Matrix Cracking
Matrix crack occurs when the equivalent stress reaches the yield stress, after ber
end can be debonded and a local plastic zone in the matrix is the result of further
loading. In the case of short-ber reinforced composites around the ber end a
penny-shaped crack can arise. SIRIVEDIN et al. presented a study about matrix
cracking in short-ber composite [13]. Depending on the strength of the interface
this crack can progress on various ways. In the case of relatively weak interface, a
cylindrical interface crack propagates from the debonded ber end. If the interface
is strong then a conical matrix crack can be observed propagating fromthe debonded
ber end. This can be described by the angle of the conical crack. The applied
nite element model is depicted in Fig. 7. Using the symmetry of the unit cell a 2D
model can be presented.
Fig. 7. Unit cell and axisymmetrical FE model
Apenny-shaped crack resulting fromber-end debonding is assumed to exist.
The mechanical loading is tension. The matrix is described as an elastic-plastic ma-
terial. Under loading micro-void nucleation, growth and coalescence are observed.
Near the corner of ber a local plastic zone is formed. If the plastic strain reaches
the critical fracture strain value crack will be propagated. The value of the critical
strain can be determined approximately. The angle of crack affects the direction of
crack propagation; the angle is described by the location of the maximum stress.
The paper presents mechanical loading, thermal loading and the combination of the
two above. Results show those thermal residual stresses and strains strongly affect
the condition of the interface and change the location of the maximum equivalent
strain.
Laminated structures also require investigating the reason of matrix damage.
A [0

, 90

3
, 0

]
T
cross-ply structure is investigated in [19], in the case of epoxy
matrix and glass ber. The cross-ply contains long continuous bers. The FE
model is shown in Fig. 8. It is assumed that bers are uniformly distributed in
the matrix and have the same radii. The loading is strain controlled; the boundary
conditions are also depicted in Fig. 8. Attention is focused on the interior domain;
no free edge effect is included. Fibers are linear elastic; matrix behaves as nonlinear
FINITE ELEMENT MODELLING 149
viscoelastic material. Both constituents are homogeneous and isotropic.
p
Fig. 8. 3D FE model and location of the damage, [0

, 90

3
, 0

]
T
Criterion for the epoxy matrix describes if the maximum principal strain
reaches the critical strain value, damage progression will occur in matrix. FEresults
indicate that local damage zones arise in two areas, shown in Fig. 8. Increasing
loading both damage areas growand coalesce. Applying linear analysis the damage
occurs at higher loading. So that the nonlinear model describe more accurately the
real process. Experimental data support this fact. Nonlinearity plays an important
role in the analysis of composites.
4.4. Modelling of Composite Structures with Holes and Notches
Numerous applications of composites require the presence of holes and cutouts.
These can cause inhomogeneous stress distribution in the structure. Cracks and
aws act important role, because these affect the strength of the composite. Crack
can be caused by the manufacture process or loading. Concerning failure criteria,
three basic types of approaches can be found in literature [8, 10].
Fracture Mechanics Models:
These models assume a localised damage emanating from the hole, represented
by an equivalent crack, which causes failure when it reaches a critical size. The
strength of the laminate is related to the fracture toughness or the strain energy
release rate. The fracture toughness and the critical size seemed to be hole-size
and material dependent. This is the basic idea of LEFM (Linear Elastic Fracture
Mechanics).
150 A. SZEKRNYES and J. UJ
Point and Average Stress Criterion:
It is assumedthat failure occurs when the stress inone point at a certain characteristic
distance or the average stress along the characteristic distance from the hole attains
the unnotched strength of the laminate. Both criterions can be expressed in closed-
form equations. Experimental data show: those characteristic distances are also
hole-size and material dependent.
Progressive Damage Models:
Throughout the loading process damage and resulting changes occur in the stress
distribution. The appropriate combinations of failure modes, failure criterion and
property degradation law are required.
Open-hole tensile strength of quasi-isotropic laminates is investigated by
MORAIS in [8]. Stress analysis near edges and discontinuities can only be ac-
curately performed using 3D models. 2D models give non-negligible errors in
laminates. The central hole has the effect of stress concentration, the stress con-
centrates around the hole. Considering this fact MORAIS presented a mixed 2D/3D
model. The model is shown in Fig. 9. The specimen was modelled with 2D ele-
ments, except for a narrow zone around the hole, where 3D brick elements were
used. The connection between the 3D and 2D regions was done imposing identical
in-plane displacements on the boundary nodes of identical in-plane coordinates and
using interface elements. The longitudinal strength of laminated structures depends
on the size. Failure criterion does not follow this fact. First a [0

/45

/90

/ 45

]
s
laminate was investigated. The loading and boundary conditions are depicted in
Fig. 9. The laminate is transversely isotropic; properties can be obtained using
classical laminate theory. The inhomogeneous stress distribution can be obtained
along a central ligament. The 3Dstresses considered are layer average stresses. The
stress concentration depends on the hole-size. Comparing the 2D and the mixed
models the mixed model shows higher stress concentration values near the hole
than the 2D model. 2D models underpredict stress concentrations, further approx-
imately identical values were obtained by both models from the hole. The strength
of structure with central hole depends on the longitudinal strength of the laminate.
In situ ber strength can be estimated using experimental data and the mixed model.
Laminates with triple parallel-arranged and equally spaced notches can be
found in [20]. Glass- and carbon-ber/epoxy composites containing through or sur-
face notch were investigated. Fig. 10 shows the laminate and the FEmodel. The FE
model is onlya quarter of the structure, the stackingsequence is [0

/90

/0

/90

/0

]
s
.
The laminated composite behaves as orthotropic linear elastic continuum, the in-
terlaminar matrix is isotropic. The paper presents results taking account of the
interlaminar matrix and not, respectively. Notch can be surface or through notch.
The mesh was gradually rened to the notches. Stress concentration factors (SCF)
were introduced for the central and the deviate notch varying the distance between
FINITE ELEMENT MODELLING 151
Fig. 9. Mixed 2D/3D model with central hole
the two notches. The depth of layer describes the surface notch. Results show that
stress concentration factors near the notch tips change from high to low then revert
to high with spacing between both neighbouring notches.
Stress concentration factor is always larger at the deviate notch. Depth of the
notch increases the SCF. In order to validate the FE predictions a total of twenty
specimens withtriple parallel notches were examined. Failure tests showthe change
of the failure load exhibits an opposite trend with the notch stress concentration
factors. Tests also show that failure always occurs at one of the two deviate notches.
A notched laminated structure, including splitting and delamination was in-
vestigated by WISNOM and CHANG [18]. A cross-ply laminate of carbon/epoxy
with lay-up [90

/0

] with a center crack has shown that splitting occurs in the 0

plies perpendicular to the notch together with narrow areas of delamination. The
[90

/0

] laminate was modelled with four noded plane elements. The elements are
plane stress ones. Duplicate nodes were used on either side of any ply interfaces
where delamination is expected. The same interface element was also used to model
splitting in the 0

ply. The schematic illustration and the FE model are depicted

in Fig. 11. Separate elements were used to represent the 0

and 90

plies. Dupli-
cate coincident nodes were used for the two plies to allow relative displacement to
occur during delamination. These nodes were connected with interface elements.
Initially linear elasticity was assumed, but it became apparent that the nonlinear
in-plane shear response was very important. The interface was modelled with non-
linear springs between the coincident nodes. The springs are initially assumed to be
152 A. SZEKRNYES and J. UJ
Fig. 10. Laminated structure and FE model with triple parallel notches
elastic. When a critical relative displacement between plies is reached the interface
is assumed to fail. The tensile loading was applied in increments. Initial results
show splitting initiated from the end of the notch, and then progressively increased
in length.
It was accompanied by the formation of a narrow zone of delamination. The
extent of delamination is slightly less than the split length. It is the splitting that
drives the delamination. The analysis was repeated with rened FE mesh. Results
show that the analysis is not sensitive to the mesh size. The difference between
linear and nonlinear analysis was also investigated. The use of linear analysis
gives unreal results; hence nonlinearity acts an important role. The analysis was
also repeated with the spring stiffness halving to examine the effects of interface
parameters. In this case the damage progression is similar to the original case, but
the size of the yield zone has greatly increased.
4.5. Interlaminar Delamination and Transverse Cracking
Damage models can be constructed using proper combination of the global element
types. NISHIWAKI et al. developed quasi-three-dimensional model for the mod-
elling of interlaminar delamination and transverse cracking independently [9]. The
FINITE ELEMENT MODELLING 153
Fig. 11. Schematic illustration of damage and FE model
FE model is depicted in Fig. 12 and consists of shell and beam elements, which rep-
resents ber and resin respectively. The ber reinforcement is concentrated in the
shell elements with the maximum ber-volume fraction of 90.7%. Beam elements
represent the interlaminar matrix. The loading is lateral compression.
Fig. 12. Four layered quasi-three-dimensional cylinder model
The cylinder model was investigated with different stacking sequences. Re-
sults show different loading limits, transverse cracking occurs if a given shell ele-
ment reaches the yield stress and delamination can be observed if one of the beam
154 A. SZEKRNYES and J. UJ
elements yields. Increasing compressive load more and more shell and beam ele-
ments reach the yield stress. Interlaminar delamination is caused by shear stresses.
Finally, the estimation of the critical strength was performed. The critical strength
was the load level corresponding to initial yielding of beam elements along the axial
direction.
The interlaminar fracture is one of the major problems for ber composites.
Its occurrence greatly reduces the stiffness of a structure. Modelling of interlaminar
crack can be performed using end-notched exure specimen (ENF) under in-plane
shear deformation mode [17]. Applying the ENF the fracture toughness can be
measured. The aim is to provide quantitative information on interlaminar crack
growth in ber composites and understand the role of bers on the stress distribution
in the surrounding matrix. The used FE model and the meshed structure are shown
in Fig. 13. The model contains only a single ber on each side of the starting defect.
The mainconcernis the role of the blunt startingdefect, introducedbythe notch. The
in-plane shear loading was introduced by assigning a xed displacement to the top
and bottom surfaces. Boundary conditions are also shown in Fig. 13. Experimental
ENF tests were conducted to verify FE results using two different specimens. The
maximum principal stress shows that the highest stress occurs around the corners of
the starting defect, extending towards the interface. The stress distribution indicates
two mechanisms that are responsible for the fracture initiation. Fig. 14 depicts that
the fracture can be initiated either from the corner of the starting defect or at the
interface. The latter occurs when the interfacial strength is lower than a critical
percentage of the matrix strength. This percentage can be obtained from the stress
distribution around the notch.
g y g p
Fig. 13. End notched exure specimen and FE mesh
When the crack is initiated from the interface, its growth is expected in both
directions. The further crack can grow towards the starting defect and interfacial
debonding occurs. If the interface strength is sufciently high, the crack initiation
is caused by tensile failure of the matrix around the starting defect. The crack
FINITE ELEMENT MODELLING 155
Fig. 14. Local damage process
propagates towards the interface because of the shear failure. It is expected that the
interface strength can signicantly inuence the measured strain energy release rate.
4.6. Modelling of Fiber-Fragmentation
A nite element analysis was carried out by DAVIS and QU to simulate the ber-
fragmentation process in a single-ber composite (SFC) specimen [3]. A single
ber is embedded in a uniform matrix material. Under monotonic axial loading
the ber will fracture. As the load increases further, more ber fracture takes place
as a result of load transfer from the matrix to the ber. The ber-fragmentation
process will stop because either interfacial debonding or the fragments are so short
that not enough loads can be transferred to the ber for further fracture. This
saturated length is related to the interfacial strength. The SFC fragmentation test is
suitable to characterize the in situ interfacial strength and the understanding of the
micromechanics in SFCs can be transferred to real composites with higher ber-
volume fractions. The Al matrix is assumed to be elastic/plastic, the interface is
modelled by elastic/perfectly plastic spring layer. Fig. 15 illustrates the SFC and
the applied undamaged and damaged FE models. As the applied load is increased,
the ber will fracture when the fracture strain is reached. The ber fracture creates a
penny-shaped crack, creating a localized plastic zone in the matrix. The number of
fracture per unit length will increase; i.e. the ber-fragment length will get shorter
and shorter as load increases. The shear stress on interface will keep increasing
till the interface fails. There is a saturation length for the ber fragments, which
depends on the interface and the ber strength. The SFCspecimen containing many
fractures can be simplied into a unit cell model containing one fragment. Only one
quarter of the unit cell needs to be modelled for nite element analysis. A series of
nite element models of the unit cells with decreasing length was used to analyse the
fragmentation process. To simulate the successive ber fracture, a series of 21 SFC
unit cells with decreasing length were used, applying solid elements. The procedure
was the following. The ber strength was calculated and the load was increased
until the axial stress in the middle of the ber fragment exceeded the ber strength.
With the next specimen the calculation was repeated. This was done without and
with ber fracture on the model. It is found that although the radial and tangential
stresses in the matrix vary radially, the magnitude of the longitudinal stress is much
156 A. SZEKRNYES and J. UJ
higher. Therefore the latter dominates the von Mises stress. One notable difference
between the unfractured and fractured models is that the longitudinal shear stress
was not zero in the latter case.
Fig. 15. Finite element modelling of ber fracture
It seems that the stress-strain curve is not a good indicator of the interfacial
shear strength. The applied effective strain/normalized ber-fragment length is
much suitable to describe the effect of the interface strength. Finally, it is possible
to estimate the interface strength by inspecting the ber-fragment lengths from a
SFC test at a given effective strain.
5. Conclusions
The process of the nite element modelling of failure and damage is summarized in
Fig. 16. Due to the heterogeneous structure, the modelling of composites is a quite
complex problem. The macro-, meso- and micromechanical analyses are useful
ways to approach the failure and damage process of composite materials. The type
of the model depends on the failure mode and the aim of the FE analysis. A given
failure mode can be represented by more than one type of models. The composite
specimens play an important role; the results can be transferred into real composites.
Considering previous papers the damage process is always related to the interface
strength. The 3D models presented here consider perfect interface; the further
purpose can be the effects of varied interface parameters. Thermal residual stresses
strongly affect the strength of the composite. Increasing ber-volume fraction,
thermal residual stresses can increase. Holes and notches have the effect of stress
concentration. Fracture mechanic problems can be handled applying LEFM, point-
or average stress criterion and progressive damage models. Initial damage occurs
when considering the used criterion, stress or strain reaches a given critical value.
Using the nite element method the damage and failure can be predicted with
more or less accuracy, but to validate the results experimental tests and observations
are always required.
FINITE ELEMENT MODELLING 157
Results and conclusions
Location of the damage, effects of further loading,
Comparison with experimental data
?
FE analysis
Failure criterion for the matrix,
stress/strain relationship and distribution,
displacement eld
?
FE model
Material model for the ber, matrix and interface,
linear and nonlinear model, macro-, meso- and micromodel,
mesh generation, boundary conditions, applied load
?
CAD model
?
Failure mode
Interface damage, matrix cracking,
interfacial debonding, delamination, ber fracture, etc...
?
Composite Structure
Fiber, matrix, interface geometry,
material parameters
Fig. 16. Process of the nite element modelling of composites
158 A. SZEKRNYES and J. UJ
References
[1] ANTIFANTIS, N. K., Micromechanical Stress Analysis of Closely Packed Fibrous Composites.
Composites Science and Technology 60 (2000), pp. 12411248.
[2] BODOR, G. VAS, L. M., Polimer anyagszerkezettan, M uegyetemi Kiad; 1995.
[3] DAVIES, J. E. QU, J., Numerical Analysis of Fiber Fragmentation in a SiC/Al Single-Fiber
Composite Specimen, Composites Science and Technology, 60 (2000), pp. 22972307.
[4] FEYEL, F. CHABOCHE, J.-L., FE
2
Multiscale Approach for Modelling the Elastoviscoplastic
Behaviour of Long Fibre SiC/Ti Composite Materials, Computer Methods in Applied Mechanics
and Engineering, 183 (2000), pp. 309330.
[5] FIEDLER, B. HOJO, M. OCHIAI, S. SCHULTE, K. OCHI, M., Finite-Element Mod-
elling of Initial Matrix Failure in CFRP Under Static Transverse Tensile Load, Composites
Science and Technology, 61 (2001), pp. 95105.
[6] HOLMES, M. JUST, D. J., GRPin Structural Engineering, Applied Science Publishers, 1983.
[7] ISMAR, H. SCHRTER, F. STREICHER, F., Inelastic Behaviour of Metal-Matrix Compos-
ites Reinforced with Fiber of Silicon Carbide, Alumina or Carbon: a Finite-Element Analysis,
Composites Science and Technology, 60 (2000), pp. 21292136.
[8] MORAIS, de A. B., Open-Hole Tensile Strength of Quasi-Isotropic Laminates, Composites
Science and Technology, 60 (2000), pp. 19972004.
[9] NISHIWAKI, T. YOKOYAMA, A. MAEKAWA, Z. HAMADA, H. MORI, S., Quasi-
Three-Dimensional Analysis for Composite Cylinder under Lateral Compressive Loading, Com-
posites for the Pressure Vessels Industry, 302 (1995), pp. 8792.
[10] PHILLIPS, L. N., Design with Advanced Composite Materials, The Design Council, Springer-
Verlag 1989.
[11] ROSPARS, C. DANTEC, E. LE, LECUYER, F., AMicromechanical Model for Thermostruc-
tural Composites, Composites Science and Technology, 60 (2000), pp. 10951102.
[12] SCHLLER, T, BECKERT, LAUKE, B. FRIEDRICH, K., Single-Fibre Transverse Debond-
ing: Tensile Test of a Necked Specimen, Composites Science and Technology, 60 (2000),
pp. 20772082.
[13] SIRIVEDIN, S. FENNER, D. N. NATH, R. B. GALIOTIS, C., Matrix Crack Propagation
Criteria for Model Short-CarbonFibre/EpoxyComposites, Composites Science andTechnology,
60 (2000), pp. 28352847.
[14] SIVASELVEN, M. V. REINHORN, A. M., Hysteretic Models for Deteriorating Inelastic Struc-
tures, Journal of Engineering Mechanics, 126 (2000), pp. 633640.
[15] Structural Research and Analysis Corp. Cosmos/M 2.0 Complete Finite Element Analysis Sys-
tem Documentation 1996.
[16] SUN, W. LIN, F. HU, X., Computer-Aided Design and Modelling of Composite Unit Cells,
Composites Science and Technology, 60 (2001), pp. 289299.
[17] TODO, M. JAR, P.-Y.B. TAKAHASHI, K., Initiation of Mode-II Interlaminar Crack from
an Insert Film in the End-Notched Flexure Composite Specimen, Composites Science and
Technology 60 (2000), pp. 263272.
[18] WISNOM, M. R. CHANG, F., Modelling of Splitting and Delamination in Notched Cross-Ply
Laminates, Composites Science and Technology, 60 (2000), pp. 28492856.
[19] XIA, Z. CHEN, Y. ELLYIN, F., A Meso/Micro-Mechanical Model for Damage Progression
in Glass-Fiber/Epoxy Cross-Ply Laminates by Finite-Element Analysis, Composites Science
and Technology, 60 (2000), pp. 11711179.
[20] ZHAO, J. H. WANG, X. F. DHARANI, L. R. WEI, J., Notch Stress Concentrations
and Failure Characteristics in Laminates with Triple Parallel Notches, Composites Science and
Technology, 60 (2000), pp. 28652872.
[21] ZHOU, C. YANG, W. FANG, D., Damage of Short-Fiber-Reinforced Composites Consid-
ering Cooling and Thermal Cycling, Journal of Engineering Materials and Technology, 122
(2000), pp. 203208.