Available online at www.sciencedirect.

com
COMPOSITES
SCIENCE AND
TECHNOLOGY
Composites Science and Technology 68 (2008) 8089
www.elsevier.com/locate/compscitech

Modelling the interaction between matrix cracks

and delamination damage in scaled quasi-isotropic specimens
Stephen R. Hallett *, Wen-Guang Jiang, Bijoysri Khan, Michael R. Wisnom
Department of Aerospace Engineering, University of Bristol, University Walk, Bristol BS8 1TR, UK

Received 17 February 2006; received in revised form 22 May 2007; accepted 22 May 2007
Available online 7 June 2007

Abstract

A series of tensile tests on scaled quasi-isotropic laminates have been carried out and modelled using nite element analysis to predict
failure. Observations during testing and examination of the failed test specimens showed signicant inuence of matrix cracking and
delamination on the nal failure. Initially the virtual crack closure technique (VCCT) was used to determine the applied load that would
cause free edge delamination. Experimental results showed that failure occurred at loads lower than those predicted. Matrix cracks,
observed in the testing, were introduced into the model and interface elements were used to model the delamination development. This
approach gave good correlation to the behaviour observed in the tests. Even in the specimens apparently dominated by bre failure,
delamination was shown to be signicant.
 2007 Elsevier Ltd. All rights reserved.

Keywords: B. Strength; C. Delamination; C. Transverse cracking; C. Finite element analysis

1. Introduction in acceptance as shown for example by the implementation

The question of damage and failure in composites is one
which has received considerable attention in the literature.
The world wide failure exercise of Hinton, Kaddour and
Soden [1] has collated and benchmarked many of the fail-
ure criteria which are in common use. Stress based criteria
are however generally not able to predict failure taking into
account specimen layup and size. In addition to the many
laminate based criteria there exist other techniques for
the prediction of individual failure modes, taking account
of their progressive eects on the laminate behaviour.
One of the most signicant single failure modes is delami-
nation since its presence can cause a reduction in the com-
pressive strength of a laminate. Techniques for predicting
delamination failure include the virtual crack closure tech-
nique (VCCT) [2] and the cohesive zone or interface ele-
ment technique. The use of interface elements is growing
*
Corresponding author. Tel.: +44 117 33 17098; fax: +44 117 927 2771.
E-mail address: stephen.hallett@bristol.ac.uk (S.R. Hallett).
of the formulation of Camanho et al. [3] in version 6.5 of
the commercial nite element software ABAQUS and that
of Hellweg [4] in the LUSAS nite element software. Many
other formulations have also been presented in the litera-
ture (e.g. Allix and Ladeveze [5] and Borg et al. [6]). Both
VCCT and interface elements have successfully been
applied to the case of delaminations arising from a free
edge in composite materials [7,8].
The failure process is further complicated by the pres-
ence of matrix cracks which often occur at stress levels
below those which cause delamination. There is some con-
sideration of the eects of interaction between matrix
cracks and delamination in the literature but this is gener-
ally limited to that occurring at the micro or meso scale.
Ladeveze et al. [9] take account of the eect of matrix
cracking on delamination by modifying the respective dam-
age parameters in a model in which the matrix cracking is
considered as distributed throughout the continuum. Sal-
pekar et al. [10] have considered delamination growing at
the intersection of a matrix crack and a free edge in an

0266-3538/$ - see front matter  2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compscitech.2007.05.038
 S.R. Hallett et al. / Composites Science and Technology 68 (2008) 8089 81

angle ply laminate using VCCT. This more closely resem-
bles the situation observed experimentally here but concen-
trates on the growth of the delamination rather than its
overall eect on the laminate. Work by Noh and Whit-
comb [11] also examines the interaction between matrix
cracks and delaminations using VCCT. Another study
which combines the eect of matrix cracks and delamina-
tions is that of Akoi et al. [12] which uses interface elements
to predict delaminations arising from matrix cracks during
out of plane impact to at plates.
Here a series of tensile tests on carbon/epoxy specimens
manufactured from uni-directional pre-preg tape are pre-
sented. The layup was selected after analysis using VCCT
so as to minimise edge delamination. In order to examine
the already well documented size eect [13] for this partic-
ular layup and material, tensile tests on specimens with
dimensions scaled in all directions were carried out. In all
cases the specimens failed at strains below that predicted
by the VCCT analysis leading one to expect that the fail-
ures would not show any delaminations. This however
was not the case, especially for the larger specimens which
showed extensive delamination back to the grips before
bre failure. Observation of the specimens during the test
and examination of the failed specimens showed the exis-
tence of matrix cracks and progressive delamination devel-
opment. It was felt necessary to include this in the analysis
in order to achieve good correlation. The interface element
formulation presented by Jiang et al. [14] was adopted to
model the progressive delamination and investigate its
interaction with matrix cracks which were included in the
model from experimental observations.
2. Experimental programme
An experimental programme has been carried out on un-
notched carbon/epoxy composite specimens (Hexply IM7/
8552) scaled in all dimensions by up to a factor of 8 and tested
in tension at quasi-static rates of strain. Scaling in the thick-
ness direction has been achieved by increasing the ply thick-
ness as opposed to sublaminate level scaling which is perhaps
more commonly used. This gives layups of [45m/90m/45m/
0m]s where m is number of plies blocked together and also
the scaling factor. The stacking sequence was chosen such
that it gave the quasi-isotropic laminate least likely to delam-
inate from the free edge whilst still maintaining a 45 ply on
the surface as is consistent with industrial practice. This was
calculated prior to the experimental programme using the
virtual crack closure technique (VCCT) as described in the
following section. Table 1 gives specimen dimensions for
each of the scaling factors tested as well as mean stress levels
(calculated from nominal thickness) for the dierent failure
mechanisms observed.
In the baseline or m = 1 case no matrix cracking or
delamination could be observed during the loading of the
specimens before ultimate failure occurred by bre fracture
at a mean stress of 842 MPa for the 11 specimens tested.
Fig. 1a shows a typical failed specimen. Post failure exam-
ination by dye penetrant testing (Ardrox 9812) of the spec-
imen in a region away from the fracture location revealed
the existence of matrix cracks in the 45 surface ply and
delamination at the 45/90 interface (Fig. 1b) which is
believed to have occurred during specimen loading. It is
dicult to present this as incontrovertible evidence that

Table 1
Dierent failure events in ply level scaled specimens (failure stress shown in bold)
Case Lay-up No. of Gauge Width 45/90 delam. C.V. 45/0 delam C.V.(%) Fibre failure C.V.
tests length (mm) (mm) stress (MPa) (%) stress (MPa) stress (MPa) (%)
m=1 (45/90/45/0)S 11 30 8 842 7.6
m=2 (452/902/452/02)S 8 60 16 418 13.8 660 3.3
m=4 (454/904/454/04)S 11 120 32 316 11.4 458 5.8 541 5.2
m=8 (458/908/458/08)S 10 240 64 222 10.3 321 2.9 458 7.2

Fig. 1. (a) Failed specimen (scaling factor, m = 1) and (b) detail of failed specimen edge.
82 S.R. Hallett et al. / Composites Science and Technology 68 (2008) 8089

sub-critical damage occurred and contributed to the failure

process due to the sudden nature of the ultimate specimen
failure which may have caused such damage. The fractured
specimen in Fig. 1a however shows plies which have pulled
out from between each other by delamination with only the
central 0 plies exhibiting bre failure. This is consistent
with the presence of sub-critical damage with a more pla-
nar brittle type of fracture being expected in its absence.
In the case of the scaling factor of 2, matrix cracks were
observed in the surface 45 plies relatively early on in the
loading process at a stress level of about 343396 MPa.
This was followed by delamination at the 45/90 interface
in the form of triangular regions, which developed at the
intersection of a surface matrix crack with the specimen
free edge and grew progressively with increasing load.
These triangular regions were observed visually as darker Fig. 3. Sequence of events leading to failure observed at various stages of
loading for a scaling factor of 4.
areas on the specimen surface as the surface ply lifted up
and are shown in Fig. 2a with a schematic in Fig. 2b for
clarity. The ultimate failure is by bre fracture at and damage during delamination. The dierent failure
660 MPa and the pattern is similar to that of the specimens events have been highlighted in Fig. 3 on a plot of cross-
with a scaling factor of 1. In the case of a scaling factor of 2 head displacement vs. applied stress for the tests with a
however the delamination surfaces are more clearly dened
and the line of 0 bre failure follows that of the matrix
crack in the adjacent 45 ply.
Specimens with a scaling factor of 4 and 8 also exhibited
matrix cracking in the surface ply at relatively low load lev-
els. This was followed by crack development in the 90 ply
and progressive development of delamination at the 45/90
interface, detected from visual observation of the specimen
edges during the test. Delamination at the 90/45 interface
was accompanied by an audible sound and small drops in
load. Subsequent delamination at the 45/0 interface was
more signicant, extending right back to the grips and
causing a greater load drop. The stress corresponding to
the delamination at the 45/0 interface has been taken as
the failure stress since it causes the rst major load drop
and essentially leaves only a unidirectional specimen. This
corresponds to an average of 458 MPa and 321 MPa for
m = 4 and 8, respectively. Further loading did ultimately
cause bre failure which was initiated at the grips where
it will have been inuenced by local stress concentrations Fig. 4. Photograph of a typical specimen failure, m = 4.

Fig. 2. Photograph and schematic of delaminations in ply level scaled specimens (scaling factor, m = 2).
 S.R. Hallett et al. / Composites Science and Technology 68 (2008) 8089
scaling factor of 4. Fig. 4 shows a photograph of a speci-
men which is typical of the type of failure for both m = 4
and 8.
3. Numerical studies
3.1. Virtual crack closure technique (VCCT) analysis
Prior to commencing the testing programme all possible
stacking sequences for a 0, 90, +45 and 45 ply quasi-
isotropic layup were analysed using the virtual crack clo-
sure technique (VCCT) to determine the stress level at
which a free edge delamination would propagate. The
stacking sequence that was selected was that which had
the highest delamination stress whilst maintaining a 45
ply on the surface according to industrial practice.
A thin slice model of dimension Dh in the loading direc-
tion with one layer of 8 nodded solid elements (Fig. 5) was
used to calculate the strain energy release rate for a crack
growing from the free edge at a single interface. Fig. 5
shows four cracks due to the symmetry of the problem.
Accurate boundary conditions were used to take account
of the applied axial strain, e0 [15]. The relationship between
any pair of corresponding nodes n(x,y,z) and n 0 (x+Dh,y,z)
on the opposite cross-section of the slice can be expressed
as
u0 u e0 Dh
v0 v
w0 w;
where u 0 , v 0 and w 0 are the x, y and z displacement compo-
nents of node n 0 and u, v and w the x, y and z displacement
components of node n.
For each stacking sequence a pre-crack half a ply thick-
ness long (0.0625 mm) was inserted at each interface in
turn. The normalised strain energy release rates for each
of mode I, II and III GI =e20 ; GII =e20 and GIII =e20 were calcu-
lated from the crack closing nodal forces and relative nodal
displacements at the crack front by practicing conventional
two step VCCT procedures. These initially increased with
crack length and then reached constant values after the
delamination crack had reached a length approximately
equal to three ply thicknesses in each case. These constant
Fig. 5. Schematic and mesh of slice model showing application of boundary conditions.
83
values were used in the failure calculations. It is noted that
variation in Gi with respect to da/a (crack extension over
crack length) has been reported for very small values of
da/a [16]. In the case presented here values of da/a are large
enough such that constant values of Gi are obtained. The
critical strain for delamination propagation was deter-
mined using the calculated energy release rates and the fail-
ure criterion as
GI GII GIII
1 2
GIC GIIC GIIIC
Values of 0.2 and 1 N/mm were used for the critical strain
energy release rates, GIC and GIIC, respectively, as these, to-
gether with Eq. (2), gave the best t to single and mixed
mode experimental data [17,18]. GIIIC was taken to be the
same as GIIC since they are both shear modes and there
is a lack of good experimental data for this value. One
study by Ratclie [19] showed very similar values for GIIC
and GIIIC for IM7/8552 but expressed doubts about the
absolute magnitude due to concern about the validity of
the test method. The elastic properties of the plies were
taken from Ref. [20] and are shown in Table 2. Thermal
stresses were not seen to have a large eect on the free edge
delamination propagation in the absence of matrix cracks
and so have not been included in the analysis presented
here.
Using the results for the predicted rst delamination
failure for the various permutations of 45, 0 and 90
1 plies for a quasi-isotropic layup, the [45/90/45/0]s
stacking sequence was selected for the testing programme
as being the least likely to suer free edge delamination
whilst still retaining a 45 ply on the surface according
to industrial practice. Table 3 shows the predicted critical
Table 2
Elastic properties used in VCCT analysis
E1 161GPa
E2 = E3 11.38 GPa
G12 = G13 5.17 GPa
G23 3.98 GPa
m12 = m13 0.32
m23 0.436
84 S.R. Hallett et al. / Composites Science and Technology 68 (2008) 8089

strains for each scaling factor for this layup together 3.2. Interface element analysis
with the failure stress calculated from an equivalent
modulus of 61.64 GPa obtained from laminate analysis The interface element has been developed in the explicit
theory. nite element software LS-DYNA and uses a bilinear soft-
Comparing results in Table 3 with those in Table 1 it ening cohesivedecohesive constitutive law which relates
would clearly be expected that the experimental results the interfacial traction components to the relative displace-
should not be inuenced by free edge delamination. The ment components [14]. The elements are located between
predicted free edge delamination stress is higher than coincident nodes of adjacent laminae of the composite
the ultimate strength in the case of m = 1 and 2 and structure to simulate both initiation and non-self-similar
higher than the delamination stresses for all of the inter- growth of delamination cracks without the need to specify
faces in the case of m = 4 and 8. The detailed examination an initial delamination. The complete failure of the inter-
of the failure processes leading to the strengths given in face element under mixed mode conditions is based on
Table 1 however shows this not to be the case. It is clear Eq. (2) but mode II and mode III are not treated sepa-
that the models based on the VCCT analysis are not rately, a combined resultant shear is used instead. Fig. 6
correctly predicting the delamination events which appear shows the element traction relative displacement formula-
to be critical in determining the ultimate strength of the tion graphically (not to scale). The same values as for the
specimens. It was felt that the level of detail included by VCCT analysis were used for GIC and GIIC and the maxi-
assuming prismatic behaviour in the slice model was not mum mode I and II stresses were 60 and 90 MPa, respec-
capturing the physical process by which the delamination tively, based on strengths for typical epoxy matrix
was occurring. An interface element technique that has materials. The analysis is not very sensitive to these values
previously been used for prediction of delamination and since failure of the specimens depends on delamination
splitting in notched composites [21] was therefore further propagation which is energy controlled. The initial stiness
developed and applied to the current experimental is relatively high and is based on an assumed resin rich
conguration. region 0.005 mm thick between the plies, giving
1.0 106 MPa/mm.
A nite element model of each of the scaled specimens
was prepared for analysis with LS-DYNA and the interface
element approach. Each ply was modelled with a single
Table 3 layer of fully integrated eight nodded brick elements. The
Critical failure strains of the scaled ply-level scaled [45m/90m/45m/0m]s in-plane dimensions of the elements in the regions of inter-
specimens predicted by VCCT est were in the order of one ply thickness (0.125 mm). Anal-
Critical Delamination stress Delamination yses on the same layup but with a centrally located hole
strain (MPa) interface showed almost identical results between one element
m=1 1.61% 992 45/90 through the thickness of each ply and four elements
m=2 1.14% 703 45/90 through the thickness of each ply [14]. Interface elements
m=4 0.81% 497 45/90 were inserted between coincident nodes at the ply inter-
m=8 0.57% 351 45/90
faces. Within each ply a single matrix crack was included

Fig. 6. Traction -relative displacement behaviour of interface elements for: (a) mode II and (b) mixed mode.
 S.R. Hallett et al. / Composites Science and Technology 68 (2008) 8089 85

using coincident nodes along the line of the crack (shown in
Fig. 7). Observation of the failed specimens (e.g. Fig. 2c)
showed that failure occurred at a single critical matrix
crack in each ply and that the pattern used in the analysis
seemed reasonable. Interface elements were not used to
model the matrix cracking as experimental results showed
that the strain at which this occurred was suciently lower
than the delamination strain and that the cracks would be
fully developed before delamination propagation started.
Fig. 7 shows the location of the splits as well as the bound-
ary conditions and loading on the model. Prior to the
mechanical loading a thermal load of 160 C was applied
to model the eect of residual stresses from the laminate
manufacturing process in which the specimen is cooled
down from the cure temperature of 180 C to room temper-
ature (20 C). The elastic properties used were the same as
those shown in Table 2 and the coecients of thermal
expansion (CTE) were a1 = 0.0, a2 = a3 = 3.0 105 C1.
These thermal stresses have a marked eect since they
cause an opening up of the embedded matrix cracks which
are the initiation points for the delaminations.
The numerical prediction of delamination growth was
tracked during loading. In all cases the development of
damage was similar and the case of the scaling factor of
4 is shown in detail in Fig. 8 together with the applied
strain and recorded stress at which each image was taken.
The dark areas indicate failed interface elements. The
striped pattern on the lower right hand image is a result
of the size of the marker used to indicate each failed ele-
ment being smaller than the continuum elements to which
they are attached due to the coarse mesh used away from
the specimen centre. Small amounts of free edge delamina-

Fig. 7. Finite element model (mesh not shown) with embedded matrix cracks.

Fig. 8. Predicted failure of interface elements at the dierent interfaces for scaling factor of 4.
86 S.R. Hallett et al. / Composites Science and Technology 68 (2008) 8089
tion away from the matrix cracks can be seen at the higher
loads, particularly at the 45/90 interface. Delamination ini-
tiated at the 45/90 interface at the position of the matrix
crack. This developed into a triangular region between
the matrix crack and the specimen edge which then propa-
gated in the direction of loading. Once the delamination
had propagated a short distance the surface ply started to
lift up from the laminate. This behaviour was also observed
in the test as seen in Fig. 2a, which shows a series of matrix
cracks, each of which has a darker region at its intersection
with the specimen edge which is the surface ply lifting up
above a delamination.
In the analysis the delamination then reaches the 90
matrix crack and progresses through the thickness into
the 90/45 interface. A small amount of free edge delami-
nation can be seen occurring away from the matrix cracks
in the 45/90 interface at this point. The delamination at the
90/45 interface grows very quickly from here to cover the
triangle bounded by the 90 and 45 matrix cracks. It
then drops down to the 45/0 interface where it propagates
back to the grips. This last rapid phase of delamination
propagation is accompanied by a load drop on the applied
displacement vs. stress curve as measured from the nodal
reaction forces. This is compared to a typical experimental
result in Fig. 9. Since only crosshead displacements, which
include additional compliance, were available from the
experiments, the analytical displacement has been scaled
so that the initial elastic stinesses match. It can be seen
that the behaviour at the failure load when extensive
45/0 delamination occurs is very similar. The results
reported for this delamination stress are recorded in the
same way as experimentally, i.e. the peak load before the
rst signicant load drop. The non-linearity in the experi-
mental result may be due to end tab or crosshead eects
which were not included in the analysis or considered sig-
nicant. No bre failure was included in the model, there-
fore even those specimens which did not reach the point of
600
500
Applied Stress (MPa)
400
300
200
100
0
0 0.5
Fig. 9. Typical experimental crosshead displacement vs. applied stress and scaled analysis results for scaling factor of 4.
extensive delamination back to the grips experimentally
still had a predicted stress associated with the onset of
delamination in the nite element model. These stresses
taken from the analysis are compared to the experimental
failure stresses from Table 1 in Fig. 10. It can be seen that
a very good level of correlation is obtained for all but the
smallest specimen size.
The good correlation in the case of the scaling factor of
2 is perhaps surprising as the experimental failure stress of
660 MPa is determined by bre failure with extensive
delamination not being observed. If the maximum 0 bre
direction stress in the analysis is examined at the point at
which the 45/0 delamination starts to propagate it is
found to be 2537 MPa. Due to the well established size
eect in composite materials [13] it is necessary to compare
this to strength values from appropriate volumes of mate-
rial. The measured unidirectional strength for a volume
similar to that of the 0 plies only is 2687 MPa [22]. These
two values are suciently close to each other to suggest
that once the delamination has progressed through the
thickness of the laminate and reaches the 0 plies, the local
stress concentration is suciently high that the 45/0
delamination and bre failure occur simultaneously. Fur-
ther evidence that this is what has happened is given by
the location of the stress concentration in the 0 ply which
occurs in a line along the adjacent 45 ply matrix crack.
Fig. 11a shows the deformed shape (displacements magni-
ed) and Fig. 11b shows the 0 ply bre direction stress
contours. Referring back to Fig. 2c it can be seen that
the line of bre failure does indeed occur at the same
45 angle, following the matrix crack in the adjacent ply.
Correlation with test for the baseline case (m = 1) was
not as good as for the larger specimens. Here no delamina-
tion was observed during loading in the test and it might
therefore be expected that the failure is dominated by bre
strength. The total volume of 0 ply material present is obvi-
ously reduced from the case of the scaling factor of 2.
Analysis, m=4
Experiment, m=4
1 1.5 2 2.5
Crosshead Displacement (mm)
 S.R. Hallett et al. / Composites Science and Technology 68 (2008) 8089 87

1000

FE delamination
800
Experimental failure

Stress (MPa)
600

400

200

0
1 2 4 8

Scaling factor

Fig. 10. Test to analysis correlation.

Fig. 11. Scaling factor of 2 (a) deformed shaped (displacements magnied) and (b) stress concentration in 0 ply once delamination reaches the 45/0
interface.

Fig. 12. (a) Predicted delamination and (b) 0 ply bre direction stress at 842 MPa for scaling factor of 1.

Results from unidirectional specimens with a similarly quasi-isotropic layup calculated from laminate analysis
reduced volume of material give the measured UD strength theory using the maximum stress failure criterion and
as 2806 MPa [22] compared to 2687 MPa as used previ- neglecting the small contribution due to residual stresses
ously for the m = 2 case. The equivalent strength for a is 1074 MPa, which is somewhat higher than the
88 S.R. Hallett et al. / Composites Science and Technology 68 (2008) 8089
842 MPa measured experimentally for m = 1. Further con-
sideration was therefore given to the analysis to explain this
apparent reduction in strength from that calculated from
the ply properties.
It has already been seen experimentally that if the free
edge of a failed specimen away from the ultimate failure
location is examined, small areas of delamination at the
45/90 interface can be observed (Fig. 1b). In the nite ele-
ment analysis at an applied stress level of 842 MPa (the
experimental failure stress) a small amount of delamination
can be observed at the 45/90 interface at the intersection
between the matrix crack and specimen edge (Fig. 12a).
Also shown in Fig. 12b are the 0 ply bre direction stress
contours. It can be seen that there is a localised stress con-
centration caused by the delamination of the outer ply giv-
ing a maximum 0 ply bre direction stress of 2357 MPa. In
this case the local stress level is not high enough to cause
bre failure.
An examination of all the failed specimens from the
experimental programme revealed that there was somewhat
greater variation in the fracture pattern for the m = 1 case
than for m = 2, particularly the location of bre failure.
This is possibly as a result of the reduced sub-critical dam-
age (typical samples are shown in Fig. 13a and b). To
model this a second conguration of matrix cracks was
used in the nite element model of m = 1 in which the cen-
tral 90 crack was moved to half way between its previous
position and the point at which the 45 crack intersected
the specimen edge (Fig. 13c). The stress at which delamina-
tion propagated in this specimen was reduced to 825 MPa,
slightly below the mean experimental failure stress of
842 MPa. At this point the maximum stress in the 0 ply
is 2707 MPa which, although below the measured
2806 MPa for unidirectional material, is suciently close
to give an indication of a possible mechanism by which
the specimen might fail at a lower stress than that predicted
Fig. 13. (a) Sample specimens for m = 2 and (b) m = 1 showing variation in fracture pattern, (c) modied matrix crack pattern in m = 1 model and (d) 0
ply bre direction stress at delamination.
by laminate analysis from the strength of the unidirectional
material.
4. Discussion
It can be seen from the experimental results that the un-
notched tensile strength for a quasi-isotropic laminate
reduces by approximately 62% with an increase in speci-
men size by a factor of 8. This in itself was not an unex-
pected phenomenon but the variation in failure mode
with increasing size and the role of delamination makes
understanding the failure mechanisms and predicting fail-
ure a signicant task. Initially the specimen layup was cho-
sen in an attempt to avoid free edge delamination failure.
Experimental results showed failure at stress levels below
that predicted by VCCT but still dominated by delamina-
tion failure events. Even the baseline case, which showed
no obvious delamination during loading, failed at a stress
level below that calculated from the unidirectional ply
strength. The interface element technique used to model
the delamination and take account of the matrix cracks
that occurred in the specimens showed good agreement
with the tests in predicting the sub-critical damage. Good
results were also achieved when comparing the experimen-
tal ultimate failure stresses for the dierent size specimens
with the stress predicted to cause the onset of delamination
at the 45/0 interface. For the case of m = 4 and 8 this is
the mechanism by which the test laminates failed. For the
case of m = 1 and 2 the experimental failure was dominated
by bre failure but the analysis has shown that this is initi-
ated by the delaminations which cause local stress concen-
trations. The exact position of the matrix cracks does have
some eect on the results, in particular the 0 ply bre
direction stress. In the case of the larger specimens not
exhibiting bre failure it is expected that variation in the
relative position of the matrix cracks would contribute to
 S.R. Hallett et al. / Composites Science and Technology 68 (2008) 8089
the experimental scatter. The nite element results were
obtained using only independently measured material
parameters (GIC, GIIC, matrix damage initiation strengths
and UD material strength) without reference to the lami-
nate test results.
5. Conclusions
It has been shown that even for the relatively simple case
of a at quasi-isotropic laminate prediction of failure is not
straight forward. Strength is strongly aected by specimen
size. Experimental results show a strong inuence of matrix
cracking and delamination and the interaction between
them. Considering the free edge delamination only, as
was done in the VCCT analysis, was not sucient to pre-
dict failure accurately. In order to achieve good failure pre-
dictions it is necessary to consider in detail the interaction
between delamination and matrix cracks as has been done
here. The interface element analysis approach gave excel-
lent results for prediction of both failure mechanisms and
stresses.
Acknowledgements
The authors gratefully acknowledge the support of the
UK Engineering and Physical Sciences Research Council
(Grant No. GR/R89462), the UK Ministry of Defence
and Airbus UK as well as the supply of material by Hexcel
Composites.
References
[1] Hinton MJ, Kaddour AS, Soden PD. Failure criteria in bre
reinforced polymer composites: the world-wide failure exercise.
Amsterdam: Elsevier; 2004.
[2] Rybricki EF, Kanninen MF. A nite element calculation of stress
intensity factors by a modied crack closure integral. Eng Fract Mech
1977;9:9318.
[3] Camanho PP, Davila CG, De Moura MF. Numerical simulation of
mixed-mode progressive delamination in composite materials. J
Compos Mater 2003;37(16):141538.
[4] Hellweg HB. Non-linear failure simulation of thick composites. PhD
Thesis, Imperial College, Department of Aeronautics; 1995.
[5] Allix O, Ladeveze P. Interlaminar interface modelling for the
prediction of delamination. Compos Struct 1992;22(4):23542.
89
[6] Borg R, Nilsson L, Simonsson K. Simulation of delamination in ber
composites with discrete cohesive failure model. Compos Sci Technol
2001;61(5):66777.
[7] T.K. OBrien. Characterization of delamination onset and growth in
a composite laminate. In: Damage in composite materials, ASTM
STP 775; 1982: p. 14067.
[8] Schellekens JCJ, De Borst R. A non-linear nite element approach for
the analysis of mode I free edge delaminations in composites. Int J
Solids Struct 1993;30(9):123953.
[9] Ladeveze P, Lubineau G, Marsal D. Towards a bridge between the
micro- and mesomechanics of delamination for laminated compos-
ites. Compos Sci Technol 2006;66(6):698712.
[10] Salpekar SA, OBrien TK, Shivakumar KN. Analysis of local
delaminations caused by angle ply matrix cracks. J Compos Mater
1996;30(4):41840.
[11] Noh J, Whitcomb J. Prediction of delamination growth and opening
near intersection of transverse matrix cracks and delamination. J
Compos Mater 2005;39(15):133552.
[12] Aoki Y, Yamada K, Suemasu H, Ishikawa T. Damage propagation in
CFRP laminates due to lateral load. In: proceedings of ICCM15
conference, South Africa, 2005.
[13] Wisnom MR. Size eects in the testing of bre-composite materials.
Compos Sci Technol 1999;59(13):193757.
[14] Jiang WG, Hallett SR, Wisnom MR, Green B. A concise interface
constitutive law for analysis of delamination and splitting in
composite materials and its application to scaled notched tensile
specimens. Int J Numer Meth Eng 2007;69:198295.
[15] Jiang WG, Henshall JL. Analysis of composite laminate beams using
coupling cross-section nite element method. Appl Math Mech
2006;27(12):170918.
[16] Sun CT, Jih CJ. On strain energy release rates for interfacial cracks in
bi-material media. Eng Fract Mech 1987;28(1):1320.
[17] Jimenez MA, Miravete A. Application of the nite element method to
predict the onset of delamination growth. J Compos Mater
2004;38(15):130935.
[18] Hansen P, Martin R. DCB, 4ENF and MMB delamination charac-
terisation of S2/8552 and IM7/8552. Technical Report N68171-98-M-
5177. Hertford, U.K.: Materials Engineering Research Laboratory
Ltd. (MERL); 1999.
[19] Ratclie JG. Characterization of the edge crack torsion (ECT) test for
mode III fracture toughness measurement of laminated composites.
NASA/TM-2004-213269; 2004.
[20] OBrien TK, Krueger R. Analysis of ninety degree exure tests for
characterization of composite transverse tensile strength. NASATM-
2001-211227 ARL-TR-2568; 2001.
[21] Hallett SR, Wisnom MR. Numerical investigation of progressive
damage and the eect of layup in notched tensile tests. J Compo
Mater 2006;40(14):122945.
[22] Wisnom MR, Khan B, Hallett SR. Size eects in unnotched tensile
strength of unidirectional and quasi-isotropic carbon/epoxy compos-
ites. Compos Struct. doi:10.1016/j.compstruct.2007.06.002.