Professional Documents
Culture Documents
Pinho Et Al 2012 Material and Structural Response of Polymer Matrix Fibre Reinforced Composites
Pinho Et Al 2012 Material and Structural Response of Polymer Matrix Fibre Reinforced Composites
Pinho Et Al 2012 Material and Structural Response of Polymer Matrix Fibre Reinforced Composites
COMPOSITE
Article M AT E R I A L S
Journal of Composite Materials
46(19–20) 2313–2341
! The Author(s) 2012
Material and structural response Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
of polymer-matrix fibre-reinforced DOI: 10.1177/0021998312454478
jcm.sagepub.com
composites
Abstract
This paper presents a pressure-dependent three-dimensional constitutive law to predict failure for laminated composites.
The nonlinear constitutive response in shear and in the transverse and through-the-thickness directions, which is
measured experimentally, is incorporated directly into the model. In addition, secant stiffnesses are dependent on the
state of hydrostatic pressure and on the general state of strain. The failure criteria distinguish between matrix failure,
fibre kinking and fibre tensile failure. In-situ strengths are used for matrix failure. Propagation of failure takes into
consideration the fracture energy associated with each failure mode and, for matrix failure, the accumulation of
cracks in the plies. A detailed discussion is undertaken of the mismatch between the available experimental data and
the physical properties required to characterise the constitutive response up to final failure. The model is employed to
make blind predictions of the triaxial failure envelopes and stress–strain curves of all 12 test cases provided by the
organisers of the second World-Wide Failure Exercise.
Keywords
WWFE-II, 3D model, pressure dependent, progressive failure
γ II I II I ε
Figure 1. Typical response of a polymer; (a) three regions in a shear curve; (b) compressive loading and (c) brittle response
in tension.
is not available and has to be estimated. The compari- While the exact hardening curve depends on the
sons with experimental results will ultimately establish polymer, the response tends to be qualitatively simi-
whether the simplifications, assumptions and estima- lar for different data published in the literature.7-12
tions are appropriate, or whether they need to be reas- For shear and compressive loading, three different
sessed. Despite the current incomplete physical areas can be identified, Figure 1(a) and (b). In the
understanding of failure in composites, the physical first region, the polymer behaves linearly. In the
basis for the models provides a solid foundation second region, the entangled polymer chains
which not only allows for predicting the failure event straighten and move relative to each other, eventu-
but also the failure mode and the consequences of fail- ally breaking some cross-links in the process. In
ure in terms of residual materials properties. the third region, there is an added resistance to
This paper starts by discussing the nonlinear any further straightening, which might be caused
material response of composites, which is assumed by the cross links and entanglement of the polymer
to result exclusively from the nonlinear response of chains. For tensile loading, Figure 1(c), yielding is
the resin. The experimental findings show that the usually not as prominent because existing micro-
constitutive response of laminated composites is rela- cracks propagate and break the specimen before
tively complex, but the inclusion of this complexity in any substantial yielding takes place. Rabinowitz
a model is required to obtain an accurate stress field et al.13 and Schuecker et al.14 point out that there
for each ply of a laminate. In turn, an accurate stress are two different mechanisms competing for failure:
field is essential to accurately predict failure of a yielding and brittle fracture. For certain stress states,
laminate. Failure criteria are formulated at the ply failure results from excessive yielding, while in others
level and can distinguish between matrix cracking, brittle fracture occurs prior to any significant
fibre kinking and fibre tensile failure. The present yielding.
damage propagation model predicts the response of Under superimposed hydrostatic pressure, both
laminates from first ply failure until final laminate the Young’s modulus and the shear modulus
failure. In addition to any imperfections associated increase with pressure,12,15 see Figure 2(a). For
with the models presented, the ability of these most experimental data published, the dependence
models to predict the test cases of the WWFE-II of initial (linear elastic) moduli with pressure can
depends on the exact definition of the problems to be approximated reasonably with a linear function,
be simulated and on the knowledge of the relevant see e.g. Figure 2(b). However, many authors identify
material properties. The factors affecting the accuracy two separate regions with different slopes,13,15,17 with
of the models and respective predictions for each test the kink between the regions related to the depend-
case are presented. ence of the glass transition temperature on pressure.
The dependence of the elastic moduli on the hydro-
static pressure is probably related to the compress-
Nonlinear constitutive response ibility of the resin and the added resistance to
of unidirectional composite plies deformation of the entangled polymer chains when
hydrostatic pressure is superimposed. The linear
Nonlinear response of polymers
dependence of the moduli with hydrostatic pressure
Most polymers used in engineering respond linearly can be predicted with good accuracy for a wide
for low applied strains, after which yielding with range of polymers15 using a model proposed by
hardening occurs, as illustrated in Figure 1. Birch.18 Birch’s model is simple and only requires
Pinho et al. 2315
0.6
p = 690 MPa
200 550 MPa
415 MPa 0.5
150 0.4
140 MPa
0.2
50
0.1
0 0
0 5 10 15 20 25 0 200 400 600
Compressive strain (%) Pressure (MPa)
Figure 2. Effect of pressure on (a) the compressive response of a polymer, after reference [16], and (b) the initial shear modulus,
after reference [11].
the moduli at atmospheric pressure and the Poisson’s response of the composite. Experimental data for sev-
ratio of the polymer: eral composites15 show a pressure-dependence of the
Young’s and shear moduli (see Figure 3), similar to
E ¼ Eo þ E pr with E ¼ 2ð5 4Þð1 Þ, the one observed in pure polymer. Therefore, the
3ð3 4Þ ð1Þ dependence of the moduli with pressure can be
G ¼ Go þ G pr with G ¼ , described using expressions analogous to those in equa-
2ð1 þ Þ
tion (1). As with pure resin, the most significant effects
where E and G are the initial Young’s and shear of hydrostatic pressure appear to be to shift the stress
moduli, respectively, Eo and Go are the Young’s and vs. strain curves up, and to allow larger plastic deform-
shear moduli (in the elastic region) at a reference pres- ation to occur before fracture. Any distortion or shift-
sure, respectively, pr is the pressure (relative to the pres- ing in the strain axis appears to be much less significant.
sure at which Eo and Go are measured) and is the Just like for pure resin, these observations suggest that
Poisson’s ratio. Values of the slope parameters E and yielding is mainly controlled by the state of deviatoric
G measured experimentally for different polymer sys- strain rather than by the state of stress.
tems are shown in Table 1. The slope coefficients E and G for the composite
Figure 2(a) shows that, while the hydrostatic pres- can either be measured experimentally or deduced from
sure results in an increased yield stress, the yield strain the slope coefficients for the polymer using a suitable
is not significantly affected. In fact, the boundaries micromechanical model. Hine et al.17 compared the
between the three regions identified in Figure 1(a) and two approaches for a glass/epoxy using Wilczynski’s
(b) are almost unaffected by the superimposed hydro- model19 and obtained a good agreement. If the slope
static pressure. The effect of hydrostatic pressure is coefficients of the composite are obtained from the
essentially to shift the curves up, while any horizontal slope coefficients of the polymer, then those latter coef-
shifting appears to be less significant. These observa- ficients can be either measured experimentally or calcu-
tions suggest that the yielding in the resin is controlled lated using Birch’s model.18 Values of the slope
by the state of deviatoric strain, rather than by the state parameters E and G measured experimentally for dif-
of stress, which seems reasonable, as the yielding results ferent composites are shown in Table 1.
from the relative movement of the entangled polymer Yielding and matrix cracking are two different mech-
chains. anisms. While yielding is related to the deviatoric strain
state in the matrix, matrix cracking is a fracture mech-
anics problem. The amount of yielding that can take
Nonlinear response of composites
place before fracture initiation depends on the load case
For polymer-matrix fibre-reinforced composites, the and on the stacking sequence. For composites, unstable
nonlinear response of the matrix induces a nonlinear crack propagation can take place at early stages of
2316 Journal of Composite Materials 46(19–20)
Reference Material E G L
1.5
140
500 MPa
Shear stress (MPa)
120 1.4
300 MPa
100 200 MPa
100 MPa 1.3
80 0.1 MPa
60 1.2
40
1.1
20
0 1
0 5 10 15 20 0 200 400 600 800
Shear strain (%) Pressure (MPa)
Figure 3. Effect of pressure on (a) the whole shear response of a composite, and (b) the initial shear modulus, after reference [12].
deformation, due to existing microscopic defects in the stacking sequence. This explains for instance why a
composite. Therefore, the amount of yielding and the 45 ASTM D3518 standard specimen for the deter-
matrix-related strengths of the composite are very mination of shear properties25 exhibits more yielding
dependent on these initial defects, particularly when than a V-notched rail shear specimen.26 While the
the state of stress includes no significant compression small individual layer thickness in the 45 specimen
to close existing microcrack tips. This argument has limits the maximum defect size, and the bounding
two main implications. Firstly, because the magnitude layers restrict unstable crack propagation, the unidirec-
of the initial defects depends on the type of specimen tional rail shear specimen can accommodate large man-
and stacking sequence, the amount of yielding before ufacturing defects, and the stacking sequence is such
fracture also depends on the type of specimen and that cracks can propagate unstably throughout the
Pinho et al. 2317
thickness of the laminate. The second implication is and G are slope coefficients for the transverse
that, under conditions that tend to close the initial Young’s and shear moduli, respectively. Because the
microcracks, the amount of yielding before failure stress in the fibre direction is mainly supported by the
should increase. This explains why there is more yield- fibres, the variable p, which corresponds to the hydro-
ing at higher hydrostatic pressures, as demonstrated in static pressure pr in equation (1), is redefined here as the
Figure 3(a). As a result, it appears clear that these two transverse hydrostatic pressure
mechanisms – yielding and matrix cracking – have to be
represented by two separate models. The model for 2 þ 3
p¼ : ð5Þ
the nonlinear constitutive response with yielding and 2
that for the fracture of the matrix are presented below.
As discussed above, the secant Young’s and shear
Nonlinear constitutive model for unidirectional moduli for zero (or atmospheric) hydrostatic pressure,
Eo2 and Go12 , are functions of the strain state, since they
composite plies
are affected by yielding. Since the second invariant of
Pressure dependence and yielding. The strains e in a com- the deviatoric strain tensor includes terms in "1 , which
posite ply are related to the stresses p in the same ply is considered here not to contribute significantly to
through Hooke’s law for orthotropic materials yielding, a modified version of the invariant that
excludes terms in "1 is empirically chosen to define an
8 9 equivalent strain
8 9 2 3>
"1 1=E1 12 =E1 13 =E1 0 0 0 >> 1>>
>
> >
> > > > qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
>
>
>
>
> 6 7>>
> 2 >
>
>
>
>
> "2 >
>
>
6
6 1=E2 23 =E2 0 0 0 7
7>> >
> > "eq ¼ ð"2 "3 Þ2 þ12 2 þ 2 þ 2 ð6Þ
>
> >
> 6 7> >
>
23 31
>
> >
> 6 >
< >
=
< "3 = 6 1=E3 0 0 0 7
7 3
¼6 7 ,
>
> 12 >> 6
6
7>
7> > such that, under monotonic loading, Eo2 and Go12 are
>
>
>
>
>
> 6
1=G12 0 0
7>> 12 >
>
> defined as:
>
> >
> 6 7>> > >
>
> >
> 6 7> >
>
>
23 > 4 sym: 1=G23 0 5> >
> 23 >
>
> >
: >
; > >
>
31
>
1=G31 :
>
; Eo2 ¼ Eo2 "eq
31 ð7Þ
Go12 ¼ Go12 "eq :
ð2Þ
The nonlinear curves defined by equation (7) are
or, in short, determined experimentally under uniaxial loading con-
ditions. The experimental curves are assumed to remain
e ¼ Dp: ð3Þ valid for all combinations of loading conditions when
expressed in terms of
the
equivalent strain
"eq . Since the
Assuming transverse isotropy in equation (2), functions Eo2 ¼ Eo2 "eq and Go12 ¼ Go12 "eq are only
E2
E3 ¼ E2 , G13 ¼ G12 , 13 ¼ 12 and G23 ¼ 2ð1þ 23 Þ
. To known at points, and not as a mathematical function,
completely define the constitutive law, the dependence spline-interpolations of these curves were implemented
of five material properties with loading needs to be for the range of experimental data (in a stand-alone
established: E1 , E2 , G12 , 12 and 23 . However, the program for the unidirectional test cases and in a
variation of E1 , 12 and 23 is less significant than the finite element program for the laminate test cases).
variation of E2 and G12 . Therefore, only the depend- Because the composite can withstand for specific load
ence of E2 and G12 on the stress and strain state combinations higher strains than those verified by uni-
is considered here. By virtue of their dependence on axial testing, an extrapolation of the experimental data
E2 and G12 (due to the assumed transverse isotropy), is also necessary. The extrapolations of the curves for
the moduli E3 , G13 and G23 also depend on the stress the five material systems used in the exercise are shown
and strain states. in Figure 4. It should be noted that the extrapolation is
To represent the effect of pressure on the moduli subjective, particularly on what concerns the inflection
detailed above, the following expressions for E2 and points visible in Figure 4. Ideally, the extrapolation
G12 are proposed: ought to be guided by suitable multi-axial data for
each material system.
E2 ¼ Eo2 þ E p
ð4Þ
G12 ¼ Go12 þ G p ,
Unloading. The test cases proposed by the present exer-
where Eo2 and Go12 are the secant Young’s and shear cise do not require the capability to model unloading.
moduli at atmospheric pressure, respectively, and E However, material healing during unloading can be
2318 Journal of Composite Materials 46(19–20)
150 300
T300/ PR319 T300/ PR319
(MPa)
50 Experimental points 100
Interpolation and extrapolation
0 0
0 5 10 15 0 1 2 3 4
Shear strain (%) Compressive transverse strain (%)
150 300
E-Glass/ MY750 E-Glass/ MY750
100 200
(MPa)
50 100
0 0
0 5 10 15 0 1 2 3 4
Shear strain (%) Compressive transverse strain (%)
150 300
S Glass/ Epoxy Compressive transverse stress S Glass/ Epoxy
Shear stress (MPa)
100 200
(MPa)
50 100
0 0
0 5 10 15 0 1 2 3 4
Shear strain (%) Compressive transverse strain (%)
150 300
A-S Carbon/ Epoxy A-S Carbon/ Epoxy
Compressive transverse stress
Shear stress (MPa)
100 200
(MPa)
50 100
0 0
0 5 10 15 0 1 2 3 4
Shear strain (%) Compressive transverse strain (%)
150 300
IM7/ 8551-7 IM7/ 8551-7
Compressive transverse stress
Shear stress (MPa)
100 200
(MPa)
50 100
0 0
0 5 10 15 0 1 2 3 4
Shear strain (%) Compressive transverse strain (%)
Figure 4. Extrapolation of the experimental longitudinal shear and transverse compression curves.
Pinho et al. 2319
60
Table 2. Failure criteria for polymers
50 Criterion Equationa
k (MPa) (deviatoric component)
Figure 5. Experimental failure points for an epoxy resin, after Experimental data can be predicted with good accur-
reference [22], and predictions using Raghava and modified von acy using different pressure-dependent criteria, such as
Mises criteria. the Mohr-Coulomb, modified Tresca, modified von
Mises, Raghava,29 Li-Wu30 and Altenbach-Tushev28
criteria. Some of these are summarised in Table 2.
The criterion proposed by the authors for the failure
prevented by simply defining the history field variable exercise is the Raghava29 criterion:
at time t as
3 k2 ðT CÞh
ðtÞ ¼ max "eq ðt Þ ð8Þ FIp ¼ , ð10Þ
t t
TC
(a) (b)
2 a0 L h 2 a0
L
2 a0
(c) (d)
h 2 a0 L
h 2a
L 2 a0
0
Figure 6. Slit crack considered for in-situ effects. (a) Ply in a UD laminate; (b) thin outer ply; (c) thin embedded ply and (d) thick
embedded ply.
This observation is at the root of the in-situ effects on represent the area under the strain versus stress curve
the strength presented in this section. up to the point ð", Þ, and are thus defined as
Consider the different types of plies presented in
Figure 6, which include an equivalent slit crack. The Z "ð Þ
equivalent slit crack is an equivalent crack of a well- x2 ð Þ ¼ 2 d"2
defined shape and orientation which purposes to repre- 0
Z ð Þ
sent the existing microcracks in the ply, resulting from
xL ð Þ ¼ L dL ð14Þ
manufacturing. The conditions for crack propagation 0
can be related to the longitudinal and transverse critical Z ð Þ
energy release rates for mode I and mode II by (see also xT ð Þ ¼ T dT :
References [2,6,31,32]): 0
ao
GIc ¼ Y 4x2 YisT The function x2 ð Þ is obtained from a standard
m
ao is transverse tensile test. The function xL ð Þ is obtained
GLIIc ¼Y x S ð13Þ from a longitudinal shear test and the function xT ð Þ
m L L
ao is is obtained from a transverse compressive test.
GTIIc ¼Y x S
m T T However, the functions are not derived directly from
the tests but rather from the spline curves with inter-
where GIc , GLIIc and GTIIc are the mode I and mode II and extrapolation shown in Figure 4. The variables L
(longitudinal and transverse) fracture toughness, and T are the longitudinal and transverse shear stresses
respectively, Y is a geometry-dependent factor, m is (see Figure 7), and their significance is discussed in the
equal to either 2, for unstable propagation in the trans- next section. The fracture mechanics solutions for the
verse direction, or 4, for unstable propagation in the four types of plies are summarised in Table 3 (see also
longitudinal direction. The strengths YisT , SisL and SisT are References [1,2,6]).
the in-situ transverse tensile strength, longitudinal shear
strength and transverse shear strengths, respectively. Matrix failure criterion. Matrix-dominated failure in com-
These strengths are in-situ because they depend on posites has similarities to that of pure polymer. This
the thickness of the ply and on the location of the ply would indicate that criteria analogous to Raghava’s29
in the laminate (inner or outer ply). The functions xð Þ would be amongst the most suitable to predict matrix
Pinho et al. 2321
(a) (b)
τ
3 2 αo
αο
2
2.6mm 2α o ST
c
σ
−Y −Y
c
2
(c) 3 (d)
α τT τL
σn
2
Figure 7. Traction components acting on the matrix fracture plane, from reference [2].
UD
YT SL cosð o Þ
ST ¼ YC cosð o Þ sinð o Þ þ
tanð2 o Þ
a 1 221
with ¼ 2 E2 E1 and h0 ¼ h=cos o.
failure in a composite. However, to predict the conse- UD composite plies. The failure index for matrix failure
quences of failure in composites knowing the fracture is defined as:
angle becomes extremely important. Variations of the
2 2
Mohr-Coulomb criterion, first proposed for composites T L hN iþ 2
by Puck and co-workers,33-37 have proven to be capable FIM ¼ þ þ ,
SisT T N SisL L N YisT
of predicting matrix failure under multi-axial stress
ð15Þ
states and have the added advantage that the fracture
angle comes directly with the prediction. For these rea-
sons, the criterion proposed here for matrix failure is an with failure being predicted when FIM 1. The last
adaptation of the Mohr-Coulomb failure criterion for term in the criterion represents the contribution from
2322 Journal of Composite Materials 46(19–20)
the positive normal traction in opening the cracks. bands are triggered by localised matrix failure next to
Therefore, this criterion is intended to be applicable misaligned fibres.41,42 Both streams have in common
for both tensile and compressive matrix failure. that they do not predict directly how kink bands
In equation (15), L , T and N are the traction com- form but rather predict an event (microbuckling or
ponents in the (potential) fracture plane, see Figure 7, matrix failure) which triggers the formation of kink
and are obtained by stress transformation: bands.
The micrographs in Figure 8 show different stages of
2 þ 3 2 3 kink-band formation in a T300/913 specimen. Matrix
N ¼ þ cosð2 Þ þ 23 sinð2 Þ
2 2 splitting in between the fibres can be identified in
2 3 ð16Þ Figure 8(b) and is the result of the high shear stresses
T ¼ sinð2 Þ þ 23 cosð2 Þ
2 introduced by failure in the neighbouring plies. In gen-
L ¼ 12 cosð Þ þ 31 sinð Þ eral, the high localised shear stresses can also be intro-
duced by manufacturing defects, such as fibre
where is the angle that maximizes FIM and is obtained misalignments. The splitting promotes further bending
numerically by evaluating the function at selected of the fibres, which in turn results in more splitting,
angles in the interval 0 5 180 as per References Figure 8(c). The bent fibres eventually break due to
[2,5,38]. The particular value of for pure transverse the combination of bending and compressive stresses,
compression, o, is a material property that can be first at one end and then at the other, finally resulting in
measured experimentally. Several sources have a kink band, Figure 8(d). Experimental observations, as
observed that the fracture angle for either glass or shown in Figure 8, suggest that kink bands are pre-
carbon composites is typically 51 o 55 .2,4,36 ceded by matrix failure and that microbuckling is not
The slope or friction coefficients T and L in equa- necessarily the triggering factor for failure.
tion (15) are introduced to account for the effect of Following the previous observations, fibre kinking is
pressure on the failure response. Their effect is that of assumed to result from shear-dominated matrix failure
increasing the respective shear strengths in the presence in a misaligned frame, under significant longitudinal
of a compressive normal traction and reducing the compression. However, if the longitudinal compression
respective shear strengths in the presence of a tensile is not significant, the shear-dominated matrix failure on
normal traction. The slope or friction coefficient T is the misaligned frame results in fibre splitting but not
obtained from the pure transverse compression test as a necessarily in fibre kinking. Experimental data for com-
function of o,2 bined longitudinal compression and inplane shear for a
carbon/epoxy T300/LY556-HY917-DY07043 suggests
1 that fibre kinking only takes place for an absolute
T ¼ , ð17Þ
tanð2 o Þ value of longitudinal compression greater than XC/2.
However, for longitudinal compression combined
while the slope or friction coefficient L is an independ- with transverse tension, experimental results for T800/
ent material property that needs to be measured 92444 indicate that no kink bands are formed if the
experimentally, see Table 1. Finally, the McCauley magnitude of the longitudinal compression is lower
brackets hiþ in equation (15) are defined as than XC.
hxiþ ¼ maxf0,xg. The criteria proposed here for fibre kinking and
for splitting use the same failure index equation
written as
Fibre kinking failure m
2 m
2
23 12
The physics of axial compressive failure are arguably FIKINK ¼ FISPLIT ¼ is þ is
ST T 2m SL L 2m
less understood than for any other failure mode in com- !2
posites. The experimental fact is that kink bands are 2m þ
þ : ð18Þ
observed in the material after failure. Although similar YisT
kink bands appear in different materials and at different
scales, the reasons for their formation, at least for poly- The two failure modes are then distinguished based
mer-matrix fibre-reinforced composites, seems far from on the magnitude of longitudinal compression with
agreed upon by all researchers in the area. While a 1 XC =2 indicating fibre kinking and 1 XC =2
stream of researchers follows the hypothesis from signifying fibre splitting. This distinction is relevant for
Rosen39 according to which kink band failure is some- the propagation of failure.
how the final result of the microbuckling of the fibres, The rotated coordinate systems relevant for the
another stream follows Argon40 and argues that kink description of a kink band are shown in Figure 9.
Pinho et al. 2323
(a) (b)
(c) (d)
Figure 8. Sequence of events during kink-band formation. The laminate is being loaded in compression in the vertical direction.
(a) Misalignment introduced by a matrix crack in an adjacent layer. (b) Matrix-fibre splitting exists throughout (see zoom); the first fibre
failures are indicated. (c) Further fibre failure. (d) Final kink band. Legend. K Matrix cracking; L Fibre failure.
2324 Journal of Composite Materials 46(19–20)
Kink plane 3
1 m m
ϕ 1
3ψ n
2m
l 2
3ψ 2ψ
3 ψ 2
1
σ σo
d = 1−
σ
σo
U crack σ o ε f −ε
d = 1−
σ ε f −ε d
d =0
U delam
ε el
ε o
εd ε f
d =1
the predicted fracture plane are reduced to zero. The The McCauley brackets in the first term indicate that
reduction of tractions is performed in such a way that the normal component of the traction is degraded in
the energy dissipation predicted by the numerical model tension but not in compression. The nominal stresses
per unit area of created surface equals the fracture expressed in the coordinate system aligned with the
toughness of the material for the respective failure fracture plane are
mode. The stresses obtained from Hooke’s law are
n oT
designated effective stresses, as they are the stresses
p‘mn ¼ ‘ m 1 d hnniþ n ‘m ð1 dÞmn ð1 dÞn‘ :
that act on the effective resisting area of the material
once failure started propagating. The stresses which are ð27Þ
in fact applied to the material, dependent on damage,
are designated nominal stresses. This nominal stress vector is then rotated back to the
The effective stress tensor in the material coord- material axes.
inate system p 123 is rotated to a coordinate sys- The evolution of the damage variable d defines how
tem aligned with the fracture plane, as shown in much energy is dissipated by damage. To assess this
Figure 10, thus defining p ‘mn ¼ f ‘ m energy, the non-elastic component of the strain due to
n ‘m mn n‘ gT . The effective traction vector is yielding is first removed from the total strain. The elas-
given as t ¼ f n mn n‘ gT . The presence of the tic strain, in the material coordinate system, is given by
crack/damage is simulated by degrading the traction the expression
components, such that the nominal tractions are
defined as eel123 ¼ Del1 p 123 , ð28Þ
n oT
where Del is similar to D defined in Equation 3, but
t¼ 1 d hnniþ n ð1 dÞmn ð1 dÞn‘ : ð26Þ
using the initial moduli instead of the moduli affect
2326 Journal of Composite Materials 46(19–20)
σo σ o ε f −ε
d = 1−
σ ε f −ε d
d =0
d =1
εo εf ε el
Figure 12. Wedge effect for an angle 6¼ 0.
will be formed. During this period of crack accumula- Outer ply. For the purpose of accumulation of
tion, it is here assumed that the nominal stress remains matrix cracks, outer plies differ from embedded plies
constant with increased strain. Eventually, the number because shear from neighbouring plies are only trans-
of cracks will saturate as the distance between neigh- mitted from one side of the ply. Applying a shear lag
bouring cracks becomes too small to transmit stresses model for outer plies results in a crack saturation dens-
to the layer. The saturation crack density, n, can be ity half of that for embedded plies:
calculated from a shear lag model by assuming constant
shear stresses at the interface equal to the shear strength SisL
(see Figure 13), as n¼ : ð33Þ
2hYisT
200
–200
σy = σz (MPa)
–400
Predictions
–600
Data given
Modified von Mises
Modified Tresca
–800
–800 –600 –400 –200 0 200
σx (MPa)
Figure 14. Predicted envelope for test case 1 – shear failure of MY750 epoxy resin under superimposed hydrostatic pressure.
250
200
150
τ12 (MPa)
100
0
–800 –700 –600 –500 –400 –300 –200 –100 0 100
σ1 = σ2 = σ3 (MPa)
Figure 15. Predicted envelope for test case 2 – 12 vs. 2 failure envelope with 1 ¼ 2 ¼ 3 for T300/PR319.
The pressure-dependent strength of pure resin is and that it predicts no failure for pure hydrostatic
approximately linear, at least for low superimposed compression.
hydrostatic pressures, and needs to be characterised
experimentally. The proposed prediction for test
UD laminates
case 1 is shown in Figure 14. It can be seen
that using the Raghava criterion leads to a slightly non- Test case 2. In test case 2, the 12 vs. 2 failure enve-
linear influence of hydrostatic pressure on strength lope with 1 ¼ 2 ¼ 3 for T300/PR319 is requested.
2332 Journal of Composite Materials 46(19–20)
0.6
Predictions, matrix cracking with α = 0
0.3
0.2
0.1
0
–800 –700 –600 –500 –400 –300 –200 –100 0 100
σ1 = σ2 = σ3 (MPa)
Figure 16. Predicted envelope for test case 3 – 12 vs. 2 failure envelope with 1 ¼ 2 ¼ 3 for T300/PR319.
150
Prediction
100
τ12 (MPa)
50 σ1 = σ2 = σ3 = –600 MPa
0
0 0.05 0.1 0.15
γ12
Figure 17. Predicted stress vs. strain response for test case 4 – 12 vs. 12 with 1 ¼ 2 ¼ 3 ¼ 600 MPa for T300/PR319.
Figure 15 shows that the predicted envelope is open for exercise. Nevertheless, any deviation from linearity of
2 5 0, and the composite is thus not expected to fail the experimental data would indicate a limitation of the
under hydrostatic compression. The failure model current model.
(equation (15)) predicts for this case a constant slope Coincidentally, this loading condition and material
of the 12 vs. 2 failure envelope for 2 5 0. The mag- has been studied by Shin and Pae.3,11,12 They tested
nitude of the slope of the envelope is defined by the composite tubes under torsion and hydrostatic pres-
material property L , which is not provided for this sure, with fibres at 0 and at 90 to the axial direction.
Pinho et al. 2333
100
90°
0
0°
–100
α = 53°
σ1 = σ3 (MPa)
–200
°
37
=
3
-5
–300 90
–500
–450 –400 –350 –300 –250 –200 –150 –100 –50 0 50
σ2 (MPa) Data given
Figure 18. Predicted envelope for test case 5 – 2 vs. 3 failure envelope for E-glass/MY750 epoxy with 1 ¼ 3 .
200
0
σ2 = σ3 (MPa)
–200
–400
Fib
Pure hydrostatic
–600 pressure
–800
–2500 –2000 –1500 –1000 –500 0 500 1000 1500 2000
σ1 (MPa) Data given
Figure 19. Predicted envelope for test case 6 – 1 vs. 3 failure envelope for S-glass/epoxy with 2 ¼ 3 .
The experimental results are also shown in Figure 15. factors are responsible for the different strengths mea-
The in-plane shear strength is different for the 0 and sured. Another aspect worth mentioning is that, for
90 specimens, although it could be expected that it moderate compression, the slope for both types of spe-
should be one single property. It is likely that the dif- cimens appears to be the same. This coincidence might
ferent manufacturing required by the different stacking be because the compression closes any previously exist-
sequences results in different initial defects in the mater- ing micro-cracks (and therefore the slope is less sensi-
ial. Furthermore, the rotation of the fibres during load- tive to the magnitude of manufacturing imperfections)
ing is different for the different specimens. Both these and the nonlinear geometric effects are not yet very
2334 Journal of Composite Materials 46(19–20)
100
Splitting Matrix cracking
–200
g
k in Predictions
k in
–400
re
Fib
–600
Pure hydrostatic pressure
–800
–3000 –2000 –1000 0 1000 2000
σ1 (MPa) Data given
Figure 20. Predicted envelope for test case 7 – 1 vs. 3 failure envelope with 2 ¼ 3 for A-S carbon/epoxy.
200
0°
α=
g
ittin
Spl 90°
0
Initial failure
Final failure
–200
ing
ink
re k
–400 Fib
°
45
=
g ,α
–600 c k in
cra
trix
Hydrostatic Ma
pressure
–800
–900 –800 –700 –600 –500 –400 –300 –200 –100 0 100
σx = σz (MPa)
Figure 21. Predicted envelope for test case 8 – y vs. z (with x ¼ z ) envelope for a 35 laminate of E-glass/MY750 epoxy.
significant. For higher compressive loads, the strength Because the specimens with fibres at 90 should not
and thus strain to failure increases, and the geometric suffer fibre rotation during the application of shear, the
effects are therefore more significant. Geometric effects parameter L used in the predictions was obtained by
might explain why the strengths for 0 and 90 speci- best fitting the corresponding experimental data.
mens becomes different again. The different response Therefore, while the linear dependence of the shear
of specimens that are conceptually identical illustrates strength on the hydrostatic pressure is a characteristic
well the problem of predicting failure in composites. of the model, the good agreement with the experimental
Pinho et al. 2335
ε
–0.05 –0.04 –0.03 –0.02 –0.01 0 0.01 0.02 0.03
0
–100
σy (MPa)
–200
εy εx
8 plies model
–300
Initial failure
–400
50 plies model
–500
Figure 22. Predicted stress vs. strain response for test case 9 – y vs. "x and y vs. "y at z ¼ x ¼ 100 MPa for a 35 laminate of
E-glass/MY750 epoxy.
100
α = 60°
60
τyz (MPa)
20
α = 30°
α = 90°
0
-500 -400 -300 -200 -100 0 100
σz (MPa)
Figure 23. Predicted envelope for test case 10 – yz vs. z envelope for a (0 /90 /45 )s laminate of IM7/8551-7.
data presented is also due to the fact that L was previously, allowing for a 12 vs. 2 envelope to be
obtained from the same data set. drawn, as shown in Figure 16. This figure also shows
the predictions of the current model, requested for test
Test case 3. Shin and Pae11,12 also measured the strain case 3. It can be seen that the model predicts an almost
to failure for the T300/PR319 specimens mentioned linear dependence of the shear strain to failure on 2 .
2336 Journal of Composite Materials 46(19–20)
100
α = 60°
60
τyz (MPa)
40
20
α = 30° α = 90°
0
–600 –500 –400 –300 –200 –100 0 100
σz (MPa)
Figure 24. Predicted envelope for test case 11 – yz vs. z envelope for a (0 /90 )s laminate of IM7/8551-7.
No experimental failure strain-related information linear. The lack of interaction between 2 and 3 for
from Shin and Pae11,12 was used for the case of this positive values of both stress components is an interest-
envelope, and any agreement between experimental ing feature of the predicted envelope. The failure enve-
and numerical results is fully predictive. Given a pre- lope is symmetrical about the pure hydrostatic pressure
diction for the strength of the specimen, the aspects of line. For the load combinations in test case 5, the fail-
the model that determine the predicted strain to failure ure mode is always matrix failure. In particular, fibre
are those related to the nonlinear response and include kinking is never predicted as the first failure mode to
the pressure-dependence of G12 (equation (4)), the def- occur. However, it should be noted that, for matrix
inition of the equivalent strain used (equation (6)) and failure under considerable superposed hydrostatic pres-
the necessary extrapolation of the experimental shear sure, the stress state in the specimen after matrix failure
stress vs. strain curve (Figure 4). can in principle lead to subsequent fibre kinking,
depending on the specimen and on how the load is
Test case 4. The curve 12 vs. 12 for applied. The angle of matrix failure changes from
1 ¼ 2 ¼ 3 ¼ 600 MPa is requested in test case 4 ¼ 0 to ¼ 53 , the latter being an input value ( o ).
for the same material T300/PR319 which was also mea- In the first quadrant, the change in angle ¼ 90 to
sured by Shin and Pae,11,12 see Figure 17. This figure ¼ 0 occurs at a stress ratio 2 =3 ¼ 1. In the
also shows the predictions from the current model. As second quadrant, the change in angle ¼ 53 to
for the previous test case, the experimental curve from ¼ 90 occurs for stress ratios 2 =3 in the approxi-
Shin and Pae11,12 is not used in any way for the predic- mate range ½2:75, 1:73 . In the fourth quadrant, the
tions, and any agreement between experimental and change in angle ¼ 37 to ¼ 0 occurs for stress
numerical is fully predictive. The aspects of the model ratios 2 =3 in the approximate range ½0:58, 0:36 .
that are relevant for this prediction are the strength
prediction (equation (15)) and the nonlinear aspects Test case 6. The 1 vs. 3 failure envelope for S-glass/
of the model as explained in the discussion of test epoxy with 2 ¼ 3 is requested for test case 6. The
case 3. predicted failure envelope, shown in Figure 19, is
again open for pure hydrostatic pressure. On the left
Test case 5. Test case 5 requests the 2 vs. 3 failure part of the envelope, it can be seen that an increase in
envelope for E-glass/MY750 epoxy with 1 ¼ 3 . The fibre-kinking strength is predicted with hydrostatic
predictions, shown in Figure 18, indicate that the enve- pressure. The exact (non-constant) slope of the enve-
lope is open under pure hydrostatic pressure. The seg- lope in this region results from the combination of
ments adjacent to the open part of the envelope are many features of the model, including nonlinear
Pinho et al. 2337
ε
–0.04 –0.035 –0.03 –0.025 –0.02 –0.015 –0.01 –0.005 0 0.005 0.01
0
–100
–200
σz (MPa)
–300
–400
εy = εx
–500
εz
–600
Figure 25. Predicted stress vs. strain response for test case 12 – z vs. "z and z vs. "x ¼ "y for a (0 /90 )s laminate of IM7/8551-7.
response, slope coefficients on the elastic properties and failure modes are expected to provide a validation of
shear strengths, definition of equivalent strain, and the general model assumptions.
extrapolation of the nonlinear stress vs. strain curves.
Test case 8. Test case 8 requires the y vs. z (with
Test case 7. Test case 7 is the same as test case 6, but for x ¼ z ) envelope for a 35 laminate of E-glass/
A-S carbon/epoxy instead of S-glass/epoxy. The predic- MY750 epoxy, as shown in Figure 21. It can be
tions, shown in Figure 20, are qualitatively similar, and observed that the envelope is open for pure hydrostatic
the same discussion as for test case 6 applies, since the pressure. Depending on the load combination, the spe-
proposed model does not distinguish between carbon cimen can fail by matrix cracking, fibre splitting or fibre
and glass fibre composites. Although glass fibres tend kinking. The matrix fracture angle within each individ-
to exhibit some nonlinear response, this effect is not ual ply, ¼ 45 , is interesting, as it has not appeared in
considered to be significant and is not taken into previous envelopes; it is the result of shear-dominated
account. matrix failure, with some transverse compression.
Notice that, due to the multidirectional nature of the
laminate, the macroscopic fracture angle of the lamin-
Multidirectional laminates ate is not 45 . However, the 45 fracture on each ply
In the present model, the strength for the matrix- should be observable using microscopy.
dominated failure modes depends on the thickness of
the plies in the laminate, which is here assumed to be Test case 9. The stress vs. strain curves for compression
0.25 mm. Furthermore, their accurate evaluation in y-direction, y vs. "x and y vs. "y at
requires assuming the values of the respective fracture z ¼ x ¼ 100 MPa are requested in test case 9 for
energies and the extrapolation of experimental non- the same laminate as in test case 8, see Figure 22. The
linear constitutive curves. Finally, the effect of weaker pronounced nonlinearity for high strains can be
outer plies on the final failure of the laminate depends observed very clearly. This nonlinearity results from
on the number of plies, which is here unless otherwise the nonlinear matrix response, and hence depends on
stated assumed to be 16. As a result, there is consider- the extrapolation of the constitutive response
able scope for the properties used and laminate simu- (Figure 4). Two curves are shown in Figure 22; one
lated not corresponding exactly to the experimental corresponding to an 8-ply model and another corres-
tests; consequently, some deviation might be expected ponding to a 50-ply model. The initiation of failure is
regarding quantitative predictions. Qualitative results predicted in both cases for the same stress at the outer
such as the shape of failure envelopes and predicted plies. However, the effect of the failure of the outer plies
2338 Journal of Composite Materials 46(19–20)
in the subsequent response of the laminate can be complexity of a model and its predictive capability is
observed to be more pronounced for the model with 8 not always straightforward to define.
plies only. Both models predict a matrix fracture angle 4. In this paper, a nonlinear constitutive model for lami-
within each individual ply ¼ 45 . nated composites which includes failure prediction is
proposed. The application of the model to the test
Test case 10. Test case 10 requests the yz vs. z envelope cases allows predicting not only failure itself but
for a (0 /90 /45 )s laminate of IM7/8551-7. The speci- also the failure modes and orientation of fracture
men fails by matrix failure in the outer ply for all load planes. The model appears to have captured interest-
combinations, without clear separation between initial ing physical phenomena, such as open envelopes for
and final failure, see Figure 23. Moderate through-the- hydrostatic compression, and the influence of hydro-
thickness compression is seen to improve the strength of static pressure on compressive fibre failure. While it is
the laminate. The ply to fail first is the outer ply, which expected that the model captures the relevant phys-
restricts how further load can be applied to the speci- ical mechanisms and their effects appropriately, there
men, given the z component of the loading. The fracture are several issues that interfere with an accurate pre-
angles range from 30 to 90 , as indicated in Figure 23. diction of failure stresses and the amount of nonli-
nearity. First, material properties which are
Test case 11. Test case 11 is the same as test case 10, determinants for the initiation and propagation of
but for a (0 /90 )s laminate. The results, presented in failure, such as intralaminar fracture toughnesses,
Figure 24, are qualitatively similar to test case 10. are not always available or readily measurable.
Second, multidirectional laminates are structures on
Test case 12. The stress–strain curves z vs. "z and z vs. their own right, and their failure strength is deter-
"x ¼ "y for the same laminate as the one in test case 11 mined not only by the material properties of the indi-
are requested in test case 12. The nonlinearity that can vidual plies and generic stacking sequence but also by
be observed in Figure 25, in particular the stiffening at a the total number of plies in the laminate and the
stress of about 200 MPa, depends largely on the number of plies clustered together.
extrapolation of the constitutive shear curves. More
evolved processes for extrapolating the constitutive
curves for cases which exhibit more plasticity than in Funding
the tests used to measure material properties ought to This research received no specific grant from any
be investigated in the future. funding agency in the public, commercial, or not-for-profit
The predictions from this model are compared with sectors.
those involved in the WWFE-II in Reference [50].
Acknowledgements
ST Pinho would like to acknowledge the support from the
Conclusions Royal Society through the Research Grants Scheme, and of
the EPSRC through Overseas Travel Grant EP/F020872/1.
1. The mechanical response of fibre-reinforced lami- PP Camanho would like to acknowledge the support from
nated composites is nonlinear and pressure-depen- the Royal Society through the International Incoming Short
Visits programme. All authors would like to acknowledge Dr
dent. At the microscopic scale, different mechanisms
C Dávila’s invaluable contribution through many discussions
such as deformation of polymer chains and breakage and in proof-reading this manuscript.
of cross-links, crazing or void formation and propa-
gation of micro-cracks compete for damage and fail-
Conflict of interest
ure of the matrix. For fibre kinking, the respective
micromechanics are not yet clearly understood. None declared.
2. At a mesoscopic scale (ply level), the presence of
neighbouring plies influences the propagation of
microcracks, thus resulting in an in-situ strength
Nomenclature
which is different from the strength measured on a Lowercase Roman letters
unidirectional specimen.
3. The understanding of all of these phenomena is not
complete yet, but significant progress has been made ao equivalent slit crack dimension
in the past decades. The relation between complexity d damage variable
of a phenomenon and its relevance in terms of dinst instantaneous value of the damage variable
material response is not always easy to assess, and h total thickness of a group of 90 plies clus-
therefore the most appropriate compromise between tered together
Pinho et al. 2339
34. Puck A. Festigkeitsnachweis von FKV-Bauteilen 43. Mannigel M. Influence of shear stresses on the fibre failure
Strength proof of design for fibre-reinforced plastics com- behaviour in carbon fibre reinforced plastics, PhD Thesis,
ponents. Kunststoffe Plast Europe 1996; 86(6): 2. Von der Fakultät für Maschinenwesen der Rheinisch-
35. Puck A and Mannigel M. Physically based non-linear Westfälischen (RWTH), Technischen Hochschule
stress-strain relations for the inter-fibre fracture analysis Aachen, 2007.
of FRP laminates. Compos Sci Tech 2007; 67(9): 44. Li X. Comparative study on the failure behaviour of
1955–1964. advanced carbon-epoxy composites. MSc Thesis,
36. Puck A and Schürmann H. Failure analysis of Department of Aeronautics, Imperial College London,
FRP laminates by means of physically based 2007.
phenomenological models. Compos Sci Tech 1998; 45. Hinton MJ, Kaddour AS and Soden PD. A compari-
58(7): 1045–1067. son of the predictive capabilities of current failure the-
37. Puck A and Schürmann H. Failure analysis of FRP ories for composite laminates, judged against
laminates by means of physically based phenomeno- experimental evidence. Compos Sci Tech 2002;
logical models. Compos Sci Tech 2002; 62(12–13): 62(12–13): 1725–1797.
1633–1662. 46. Cox HL. Elasticity and strength of paper and other
38. Dávila CG and Camanho PP. Failure criteria for FRP fibrous materials. Br J Appl Phys 1952; 3(3): 72–79.
laminates in plane stress, NASA/TM-2003-212663. 47. Maimi P, Camanho PP, Mayugo JA, et al. A continuum
Hampton, VA: NASA Langley Research Center, 2003. damage model for composite laminates: Part I–
39. Rosen BW. Mechanics of composite strengthening. In: Constitutive model. Mech Mater 2007; 39(10): 897–908.
Fiber composite materials. Metals Park, OH: American 48. Maimi P, Camanho PP, Mayugo JA, et al. A continuum
Society of Metals, 1965, pp.37–75. damage model for composite laminates: Part II–
40. Argon AS. Fracture of composites. In: Treatise on mater- Computational implementation and validation. Mech
ials science and technology. Vol. 1, Academic Press pub- Mater 2007; 39(10): 909–919.
lishers, 1972, pp.79–114. 49. Kaddour AS and Hinton MJ. Input data for test cases
41. Schultheisz CR and Waas AM. Compressive failure of used in benchmarking triaxial failure theories of compos-
composites, Part I: testing and micromechanical theories. ites. J Compos Mater 2012; 46: 2295–2312.
Prog Aero Sci 1996; 32(1): 1–42. 50. Kaddour AS and Hinton MJ. Benchmarking of triaxial
42. Waas AM and Schultheisz CR. Compressive failure of failure criteria for composite laminates: comparison
composites, Part II: experimental studies. Prog Aero Sci between models of ‘Part (A)’ of ‘WWFE-II’. J Compos
1996; 32(1): 43–78. Mater 2012; 46: 2595–2634.