Composite Structures 62 (2003) 373–378

www.elsevier.com/locate/compstruct

Damage model development for SMC composites

E.V. Morozov *, K.E. Morozov, V. Selvarajalu
School of Mechanical Engineering, University of Natal, Durban 4041, South Africa

Abstract
This work deals with the development of an SMC composite material model, incorporating a homogenisation model and sta-
tistical damage criteria. The model theory and the basis for implementation of the model into a commercial finite element analysis
package are both presented here. Coarse model validation is performed by evaluating the model parameters for a set of sample data,
and applying these to the modelling of the progressive fracture of an SMC plate under tensile loading.

2003 Elsevier Ltd. All rights reserved.

Keywords: SMC composite; Progressive failure; Damage model

1. Introduction 2.1. Material constitution

An overview of the progressive damage models that The constitution of a transversely isotropic solid (in
have been developed for the analysis of SMC composite this case the fibre phase in local coordinates) may be
materials reveals two distinct methods of modelling. expressed as follows:
Firstly, the material may be modelled on the basis of



s k l e
microscopic characterisation of the material constitu- ¼
ents, viz. the fibres and matrix, with these two consti- r l n e ð1Þ
tuents then homogenised to result in a single material s12 ¼ 2pe12 ; s13 ¼ 2pe13 ; s23 ¼ 2me23
with the theoretically averaged properties of the two.
Secondly, the SMC material may be characterised by where
way of macroscopic testing on a sample of the actual s ¼ 12ðr22 þ r33 Þ; r ¼ r11 ;
material itself, with the properties obtained from simple ð2Þ
e ¼ e22 þ e33 ; e ¼ e11
testing used to derive the nature of the material under
more complex loading, through typical finite element Here r11 is the stress, in the local coordinate frame of
analysis. The model developed here is an example of the the fibre, experienced along the longitudinal axis of the
former, with the material constitution adopted from fibre, r22 is the stress transverse to the fibre, and r33 is
work carried by Chen et al. [1], and a newly proposed the stress along the orthogonal axis in the transverse
damaging model based on the work of Desrumaux et al. plane. The strains e11 , e22 and e33 are similarly assigned
[2]. to these respective axes. Stresses s12 , s13 and s23 are the
shear stresses acting in the planes ð1; 2Þ, ð1; 3Þ and ð2; 3Þ
respectively.
The parameters k, l, m, n and p are Hill’s elastic
2. Model description
moduli [3]. Specifically, k is the plane-strain bulk mod-
ulus for lateral dilation without longitudinal extension, n
The model makes use of the Mori–Tanaka homoge-
is the modulus for longitudinal uniaxial straining, l is
nization scheme (see [1]) to express the overall properties
the associated cross modulus, m is the shear modulus in
of a randomly reinforced short fibre composite in terms
any transverse direction and p is the shear modulus for
of the properties of its constituents.
longitudinal shearing.
For composites reinforced with randomly oriented
*
Corresponding author. Fax: +27-31-260-3217. fibres, with the matrix (denoted by subscript r ¼ 1) and
E-mail address: morozov@nu.ac.za (E.V. Morozov). fibre phases (denoted by subscripts r ¼ 2; 3; . . . ; N ) being
0263-8223/$ - see front matter
2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruct.2003.09.007
374 E.V. Morozov et al. / Composite Structures 62 (2003) 373–378

isotropic and transversely isotropic in the local frame rather test samples that take this impregnated effect into
respectively, the overall moduli may be given by the account. To thus replace our fibre phase in the above
following relations [1]: formulation by a unidirectional composite, which is
simpler to test and also takes into account the composite
1X N
ðdr
3K1 ar Þ nature of the very inclusion itself, would appear justifi-
K ¼ K1 þ cr  PN 
3 r¼2 c1 þ r¼2 cr ar able.
ð3Þ
1X N
ðg
2G1 b Þ
G ¼ G1 þ c r  r PN r 
2 r¼2 c1 þ r¼2 cr br 2.2. Statistical damage law

where K, G are the overall bulk and shear moduli, re- It is proposed that damage is introduced in the ma-
spectively, of the composite material, cr are the volu- terial defined by the above formulation by introducing
metric fractions of each phase r, with the matrix phase statistical theory based on that presented by Desrumaux
assigned r ¼ 1 and the inclusion (fibre) phases following et al. [2], which makes use of the Weibull distribution.
as r ¼ 2; 3; . . . ; N . Since usually only one fibre material The application of this distribution to characterise the
or geometry is used for reinforcement, r ¼ 2 usually failure of brittle materials particularly is fairly well es-
pertains to that single material. tablished (see [4]). The distribution may be imposed to
Thus statistically control the failure pattern of the compo-
1 ðd2
3K1 a2 Þ site material, by extrapolating the results of testing of
K ¼ K1 þ c2 standard samples to the analysis of structures under a
3 ½c1 þ c2 a2 
ð4Þ variety of loading, through certain statistical relations.
1 ðg2
2G1 b2 Þ
G ¼ G 1 þ c2 Statistical analysis also in some way accounts for the
2 ½c1 þ c2 b2  large scatter in experimental results that usually exists in
Parameters ar , br , dr , gr depend on the moduli and geo- the testing of randomly reinforced composites particu-
metry of the phases, and for cylindrical fibres are given larly, attributable to the material’s inherent structural
as follows: inhomogeneity, the random distribution of the inclusion
phase, and difficulty in maintaining manufacturing
3K1 þ 3G1 þ kr
lr
ar ¼ precision and repeatability.
3G1 þ 3kr The general statement for probability of failure of the


1 4G1 þ ð2kr þ lr Þ 4G1 2ðc1 þ G1 Þ material in question may be expressed as
br ¼ þ þ
5 3G1 þ 3kr pr þ G1 c1 þ mr

 P ¼ 1
½1
F ðrÞ ð7Þ
1 ð2kr þ lr Þð3K1 þ 2G1
lr Þ
dr ¼ nr þ 2lr þ
3 kr þ G1 where P is the distribution function for the probability

ð5Þ of failure of the material under the action of stress r,
1 2
gr ¼ ðnr
lr Þ which is defined by the specific strength criteria in place.
5 3
For example, for the three dimensional case, when
8mr G1 ð3K1 þ 4G1 Þ
þ applying the shear strain energy (Von Mises) criterion
mr ð3K1 þ 4G1 Þ þ G1 ð3K1 þ 3mr þ G1 Þ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 1
8pr G1 4kr G1
4lr G1
2l2r þ 2kr lr r ¼ pffiffiffi ðr1
r2 Þ2 þ ðr2
r3 Þ2 þ ðr3
r1 Þ2 ð8Þ
þ þ 2
pr þ G1 3kr þ 3G1
where for an isotropic matrix, where r1 , r2 and r3 are principal stresses. The function F
in Eq. (7) is the probability function, expressed in terms
3G1 K1 þ G21 of the Weibull distribution
c1 ¼ ð6Þ
3K1 þ 7G1

V  r m
Note that the elastic moduli of the fibre phases are de- F ðrÞ ¼ 1
exp
ð9Þ
V0 S
fined in the local coordinate frame of that fibre.
It is also possible using this formulation to incorpo- where m is the Weibull shape parameter, S is the Weibull
rate the possibility of the inclusion phases being so scale parameter, V0 is the original volume of the samples
called Ôcomposite fibres’, made up of cylindrical uni- used for testing to obtain the Weibull distribution pa-
directional fibre composite material. The advantages of rameters, and V is the volume of the structure to which
this modelling option are immediately evident. Typi- the distribution is extrapolated.
cally, characterisation of the individual component Whichever strength criterion is adopted, once a pre-
phases of a composite is inconsistent and sometimes determined limit for the overall probability, P , is at-
even unrealistic. The characteristics of the dry fibre are tained, the material in question is considered Ôfailed’ and
for example not necessarily reflective of the same fibre either holds no more load carrying capability or is
impregnated with matrix material, thus it makes sense to gradually degraded by the lowering of its stiffness
 E.V. Morozov et al. / Composite Structures 62 (2003) 373–378 375

properties, until the material is rendered completely Data derived from sample tensile testing is extrapo-
failed. lated to component analysis by applying the Weibull
distribution function expressed in Eqs. (7) and (9)
above. The purpose of the sample testing is to statisti-
3. Model implementation cally describe the distribution of the material strength,
by evaluation of the Weibull shape and scale parame-
Implementation in a finite element analysis environ- ters. The shape parameter m, may be found by plotting a
ment requires, as a first step, definitions of the inputs of graph of (lnð
lnð1
F ÞÞ) vs. lnðrÞ, where F corre-
the model. From the above description, the following sponds to the percentage of samples that fails at stress r.
inputs are revealed. The slope of the graph may be shown to be equal to m,
The bulk and shear moduli, K1 and G1 , of the matrix, and the scale parameter S, is then obtainable from
E
where K1 ¼ 3ð1
2mÞ E
and G1 ¼ 2ð1þmÞ respectively, and where functions relating the scale and shape parameters (see
E is the elastic modulus, and m the Poisson’s ratio of the Eq. (11) for example).
matrix material. These parameters may be obtained Depending on the number of samples tested, the
from standard specimen tensile testing, or, for com- range of failure for the samples should thus first be sub-
monly used materials such as polyester matrix, such divided into appropriate intervals. The cumulative
data is readily available. number of failures at each interval is then noted and
Parameters k, l, m, n, p for the fibre material can be plotted (Fig. 1).
found as follows. The model definition allows for the Each interval may then be characterised by its median
possibility that the reinforcement phase may be made up stress, rfm , and the cumulative percentage failures
of different materials, but since usually only one fibre (which corresponds to F for the sample distribution)
material is included, only one set of these constants is may be plotted against these in the form
required. Once again, these material properties may be (lnð
lnð1
F ÞÞ) vs. lnðrfm Þ (Fig. 2) so that the corre-
obtained from testing, however for traditionally used sponding value of m, the shape parameter, may be
materials, such as glass fibres, such material constants found.
are available from reference. The constants may be re- The Weibull shape parameter is thus found to be 3.37
lated to more commonly adopted stresses by manipu- (by suitably approximating the slope of the graph). The
lating Eq. (1), which may be rewritten as




1
k l s e
Cumulative Percentage Failed

¼ 120
l n r e
100

1 

1
ðr þ r33 Þ e22 þ e33 80
¼ 2 22
r11 e11 60

1 

1 40
r e22
¼ 2 22 ð10Þ
r11 e11 20
0
(Assuming no stress and strain contributions along third 28 -36 36 -44 44 -52 52 -60 60 -68
axis (two-dimensional case).) Failure Stress Range (MPa)
Thus we need the strain response of the material
Fig. 1. Cumulative distribution for failure stress of samples.
under loading (stress), which allows us to find the
moduli ðk; l; nÞ. This may be obtained, once again, from
standard specimen testing.
The only shear term that applies for our two-dimen- 2.5
sional case is s12 , and since s12 ¼ 2pe12 (Eq. (1)), this
implies that 2p corresponds with the material shear 2

modulus G12 , which is also obtainable from testing or 1.5

ln (-ln (1-FA))

reference. 1
The outputs of the material model, which will in fact 0.5
serve as inputs to a finite element analysis solver, are the
0
overall bulk and shear moduli, K and G respectively, of 3.4 3.6 3.8 4 4.2
-0.5
the SMC composite material in question. These moduli
-1
are defined in a similar manner to which the Young’s
modulus, E, and the shear modulus, G, are defined for -1.5
ln (σ fm )
an isotropic material, within the solver. The proposed
progressive failure model may then be applied. Fig. 2. Weibull parameter identification.
376 E.V. Morozov et al. / Composite Structures 62 (2003) 373–378

scale parameter, S, may then be evaluated, since it may 100

be shown (see [4]), that when lnðrfm Þ ¼ 0, 90
80
ðlnð
lnð1
F ÞÞÞ ¼
m lnðSÞ ð11Þ

Probability of Failure
70
60
The scale parameter S is thus evaluated for our set of
50
experimental data to be 43 MPa.
40
Alternative methods are also applicable, one such
30
method being an iterative process for finding suitable
20
shape and scale pairs for a given set of experimental
10
data. This involves selecting one of the two variables
0
and then calculating the second, and repeating until 0 100 200 300 400
convergence of the selected and calculated values occurs Stress (MPa)
(see [4]). Using this technique, values of m and S were
Fig. 3. Probability of failure vs. stress.
found to be 3.5 and 51 MPa respectively for the set of
data presented here.
It should also be noted that a so called small-sample The suitably unbiased shape parameter and the cal-
Ôunbiasing factor’, a function of the number of samples culated scale parameter may then be substituted into the
taken, N , is usually applied to the shape parameter for expression for F (Eq. (9)) to find the probability of
relatively small values of N i.e. a low number of obser- failure of the analysed component. The Von Mises
vations, typically less than 120. The unbiasing factor for stresses (for the shear strain strength criterion) in the
N ¼ 8 samples here is given as 0.82 (see [4]). material are calculated by the finite element analysis

Fig. 4. Progression of stress damage in tensile specimen. (a) Loading of 55 kN. Damage initiated. Only 1 element fails. (b) Load increased to 56 kN.
Damage progresses to incorporate second element. (c) Load increased to 57 kN. More elements fail. (d) Load increased to 57.5 kN. The two right
quarters now almost completely meet the criteriea and are both degraded by halving their stifness properties. (e) The load in the new Ôdamaged’
material is increased to 57.6 kN and the stresses redistribute accordingly. The right half meets degradation criteria again and its stiffness is again
degraded by half. The left half is degraded for the first time. (f)–(h) This pattern continues, with each 100 N increment yielding sucessive degradation
of the material. (i) At a load of 58 kN the material is degraded fully to some limit level (10% stiffness chosen here) and then enters the accelerated
fracture phase, characterised by catastrophic degradation of the material stiffness.
 E.V. Morozov et al. / Composite Structures 62 (2003) 373–378 377

solver, and we may then apply the probability of failure 70000

(P ), as governed by the above derived formulae, to de- 60000 A
B
termine whether an element fails (and therefore either 50000

Force (N)
does not contribute any load-carrying capability in the 40000
next solver iteration, or is degraded), or not.
30000
For the purposes of an illustrative example, we take a
20000 C
150 · 100 · 5 mm SMC plate, loaded in tension. The
associated volume of this specimen, 7500 mm3 repre- 10000

sents our value V0 , the original volume. We take our 0

0 10 20 30 40 50
analysis volume, V , here to be equal to the volume of Displacement (mm)
each element of the finite element analysis. Thus, failure
is attributable at the element level and we may then Fig. 5. Applied force vs. displacement curve for illustrative example.
observe the progressive degradation of the structure as
elements are progressively degraded. Since our analysis
uses an 8 · 8 mesh, the fraction V =V0 is thus 1/64 here. carrying ability. The region B–C is modelled by an in-
We use the values for the shape and scale parameter verse analysis, where the material stiffness is degraded by
yielded by the iterative technique described above and a factor of 50% at each increment, and the corre-
then draw up a table of Probability of Failure vs. Von sponding force calculated at regular displacement in-
Mises Stress, for any level of stress, and define that stress tervals.
level (according to the corresponding probability) that
denotes failure of the element. Setting 70% as our
probability limit here, we see that a stress level of 230 4. Conclusions
MPa corresponds to this limit (Fig. 3).
Thus, above a stressed level of 230 MPa, we define At present those material models used for the analysis
that the element in question is rendered failed, and Ôre- of SMC composite materials within commercially
moved’ (converted to so called Ônull’ elements which available finite element software packages take advan-
contribute to volume but not strength), or degraded for tage of the overall isotropic nature of these materials.
the next iteration. Once all the failed elements are Ôre- Though this approximation is indicative of the general
moved’ or degraded, the solver is to recalculate the response of SMC composite materials, it does not allow
stresses in the material at the next iteration step. We for any analysis in terms of the material constituents. A
choose to merely degrade the stiffness properties of the model developing a relation between the properties of
failed elements by a factor of 50%, which might be more the constituent phases allows for comparative analysis
applicable to more ductile materials. The stress levels in using various fibre materials, resin materials, and fibre
the specimen are computed at each step by finite element volume percentages.
analysis, as shown in Fig. 4. The next phase of implementation of the new model
For computational ease, the example shown here presented here is being performed, viz. integration into a
degrades the specimen by groups of 16 elements at a commercial finite element analysis package as a user-
time (quarters of the model), but of course, a more re- defined option for analysis, so that real-time degrada-
fined degradation (which would require individual ele- tion of each element may be performed as and when it
ments being assigned individual material properties) is meets the failure criteria.
simple to institute-especially in a tailor-made analysis
algorithm within a user-defined material in a suitable
finite element analysis package. The darkest areas of the Acknowledgements
specimen as shown in Fig. 4 correspond to the areas
where stress has exceeded the limit stress of 230 MPa, The authors gratefully acknowledge the support of a
and is thus degraded appropriately. For this coarse ex- Technology and Human Resources for Industry Pro-
ample, once more than 50% of the elements in any gramme (THRIP South Africa) Grant (GUN 1759) and
quarter exceeded the limit, the entire quarter was de- a National Research Foundation (South Africa) Grant
graded by halving the material stiffness properties of (NRF GUN 2050592/2053318) to enable this research to
that quarter. be undertaken.
The force vs. displacement curve shown in Fig. 5
shows the nature of the material behaviour as it un-
References
dergoes progressive degradation. Point A represents the
initiation of damage (initial reduction in stiffness in a [1] Chen T, Dvorak GJ, Benveniste Y. Mori–Tanaka estimates of the
portion of the structure, as described in Fig. 4). Point B overall elastic moduli of certain composite materials. J Appl Mech
represents the initiation of catastrophic loss of load 1992;59:539–46.
378 E.V. Morozov et al. / Composite Structures 62 (2003) 373–378

[2] Desrumaux F, Meraghni F, Benzeggagh ML. Micromechanical [3] Hill R. Theory of mechanical properties of fibre-strengthened
modelling of damage mechanisms in randomly orientated materials: I. Elastic behaviour. J Mech Phys Solids 1964;12:199–
discontinuous fibre composite. In: ICCM-10, Whistler, BC, 212.
Canada, Fracture and fatigue, vol. I, 1995. p. 487– [4] Bury KV. Statistical models in applied science. John Wiley & Sons;
94. 1975.