Eective properties of a viscoplastic constitutive model

obtained by homogenisation
O. van der Sluis
*
, P.J.G. Schreurs, H.E.H. Meijer
Eindhoven University of Technology, Faculty of Mechanical Engineering, P.O. Box 513, 5600 MB Eindhoven, The Netherlands
Received 17 November 1998; received in revised form 10 June 1999
Abstract
Heterogeneous materials are used more and more frequent due to their enhanced mechanical properties. If the re-
lation between the microscopic deformation and the macroscopic mechanical behaviour can be obtained, it can be used
to design new materials with desired properties such as high strength, high stiness or high toughness. A method for
obtaining this relation is called homogenisation, by which the heterogeneous material is replaced by an equivalent
homogeneous continuum. In this paper, a homogenisation method is proposed which oers the possibility to determine
eective material properties for the homogeneous equivalent continuum, modelled by Perzyna's viscoplastic consti-
tutive law. To this end, nite element calculations are performed on a representative volume element, the geometry of
which is dened by the microstructure of the considered material. The mechanical behaviour of this RVE will also be
described by a viscoplastic model, clearly with a given parameter set. The proposed homogenisation strategy provides a
way to acquire the constitutive parameters for the equivalent medium. To validate the results of the homogenisation,
nite element calculations of the deformation behaviour of a perforated plate that is subjected to dierent loading
histories are performed. The global mechanical behaviour of the homogenised simulations and direct calculations,
where the heterogeneous structure is completely discretised, will be compared. 1999 Elsevier Science Ltd. All rights
reserved.
Keywords: Homogenisation; Constitutive behaviour; Inhomogeneous material; Microstructures; Finite elements; Viscoplasticity;
Strain softening
1. Introduction
In the past few decades, tremendous eort has
been put into the development of new materials
with enhanced mechanical properties, such as high
stiness, high strength and toughness. These ma-
terials all have in common an heterogeneous na-
ture of the microstructure. Extensive experimental
research has been reported on this subject, clearly
proving the considerable inuence of the micro-
structure on the macroscopic mechanical behav-
iour. Nevertheless, it seems apparent that more
fundamental studies are needed for a better un-
derstanding of the deformation behaviour of these
advanced materials. Relations between the micro-
structural phenomena and the macroscopic de-
formation behaviour are indispensable when
predicting macroscopic properties from the mi-
crostructure of a material. In this respect, the
works of Hill (1965), Gurson (1977) and Hashin
(1983) were pioneering.
Mechanics of Materials 31 (1999) 743759
www.elsevier.com/locate/mechmat
*
Corresponding author. Tel.: +31-40-247-2827; fax: +31-40-
244-7355.
E-mail address: olafs@wfw.wtb.tue.nl (O. van der Sluis)
0167-6636/99/$ - see front matter 1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 6 7 - 6 6 3 6 ( 9 9 ) 0 0 0 2 8 - 9
Generally speaking, we can distinguish three
dierent routes to predict the macroscopic re-
sponse of a heterogeneous material subject to a
mechanical load: (1) choice of a macroscopic
constitutive model, of which the material param-
eters are tted onto experimental data, (e.g. Geers,
1997; Meuwissen, 1998); (2) the application of a
multilevel nite element method, where on the
dierent levels, simultaneous numerical simula-
tions are performed, (e.g. Ghosh et al., 1995;
Ghosh and Moorthy, 1995; Ghosh et al., 1996;
Smit et al., 1998; Kruch et al., 1998); (3) homog-
enisation of the heterogeneous material towards a
closed form macroscopic constitutive equation,
either by analytical (Ponte Casta~ neda, 1996;
Garajeu and Suquet, 1997), or by numerical ap-
proach (Vosbeek, 1994; Van der Sluis et al., 1999).
The rst possibility seems to be the most ob-
vious one and is most widely used. A macroscopic
constitutive equation is chosen and consequently,
by collecting sucient experimental data, values of
the used parameters in the constitutive equation
can be determined. Although no direct micro-
structural information is available in the thus de-
termined constitutive model, valuable insight can
be gained from it.
The most important benet of the multilevel
method appears to be the readily available infor-
mation of the evolution of the microstructure
during loading of the macroscopic structure and
therefore, the results of this method are quite ac-
curate (Ghosh et al., 1996; Smit et al., 1998; Kruch
et al., 1998). Nevertheless, this approach poses
severe limitations on practical applications with
complex microstructures as well as complex mac-
rostructures, since in that case, such simultaneous
simulations require an enormous amount of com-
puter memory and computational time.
Another possibility is to predict the macro-
scopic response from microscopic analyses fol-
lowed by an homogenisation technique that results
in a so-called Homogeneous Equivalent Continu-
um (HEC). According to Maugin (1992) and
Vosbeek (1994), this homogenisation process aims
at replacing the macroscopic heterogeneous
structure with a continuum model that `best' rep-
resents the structural model (Fig. 5). The funda-
mental assumption in these homogenisation
procedures is the statistical homogeneity of the
heterogeneous material (Hashin, 1983), that is, all
statistical properties of the state variables are the
same at any point in the material. In that case, it is
possible to identify an element whose mechanical
behaviour is representative for the heterogeneous
medium as a whole. Such an element is called a
representative volume element, or short, RVE.
Several analytical approaches have been proposed
for elastic, viscoelastic and elasto-plastic models
(Hashin, 1983; Mura, 1987; Nemat-Nasser and
Hori, 1993; Boutin, 1996; Ju and Tseng, 1996).
More recently, also for viscoplastic models, eec-
tive potentials have been studied (Ponte Cas-
ta~ neda, 1996; Li and Weng, 1997a,b; Garajeu and
Suquet, 1997). It is evident that simplifying as-
sumptions have to be made (e.g. concerning the
stiness of the microscopic inclusions or the con-
stitutive behaviour of the individual materials), to
be able to derive expressions for the overall be-
haviour.
The numerical homogenisation method pro-
posed in this paper, does not require any simpli-
fying assumptions on account of the
microstructure of the material. Instead, nite ele-
ment calculations are performed on the micro-
structural level, hereby circumventing the
diculties encountered in the analytical treatment.
This procedure can be summarised as follows
(Vosbeek, 1994; Van der Sluis et al., 1999). First,
relations between the microscopic and macro-
scopic state variables (e.g. stresses and strains)
have to be dened. Next, appropriate boundary
conditions are developed which follow from these
micromacro relations. These boundary condi-
tions are prescribed on the RVE to generate the
volume averaged state variables, necessary to de-
termine the constitutive equations of the equiva-
lent homogeneous continuum.
When considering strain softening material be-
haviour, attention should be focussed on the
constitutive modelling. When trying to capture this
phenomenon in nite element analyses by using
conventional constitutive models, the obtained
results will reveal a strong sensitivity with respect
to the employed spatial discretisation. In this case,
the obtained results will not converge to a unique
solution upon mesh renement (see for instance
744 O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759
Needleman, 1988; De Borst and M uhlhaus, 1991;
De Borst et al., 1993). The initial boundary value
problem of the underlying mechanical process
becomes ill-posed upon entering the strain soft-
ening regime. Locally, the character of the partial
dierential equations changes from elliptic to hy-
perbolic in quasi-static problems, and from hy-
perbolic to elliptic in dynamic analyses. In
literature, several methods have been proposed to
preserve well-posedness of the mechanical problem
and to obtain mesh-independent results in nu-
merical analyses. Examples of the mentioned
methods include non-local models (Bazant, 1991;
Brekelmans, 1993; Tvergaard and Needleman,
1995) and gradient models (Triantafyllidis and
Aifantis, 1986; Peerlings et al., 1996), Cosserat
continua (Toupin, 1962; De Borst, 1990), and rate-
dependent models (Wu and Freund, 1984; Nee-
dleman, 1988; Sluys and De Borst, 1992; De Borst
et al., 1993; Wang et al., 1996). For an overview,
see De Borst et al. (1993). In eect, these so-called
regularisation models introduce an internal length
scale parameter which regularises the localisation
process.
In order to be able to deal with softening ma-
terials, a rate-dependent model is adopted. More
specically, Perzyna's viscoplastic model (Perzyna,
1966, 1971; Owen and Hinton, 1980) has been se-
lected. The corresponding constitutive equations
will be presented, after which the nite element
formulation is discussed. To show the capability of
the model to describe strain softening behaviour in
a proper way, quasi-static two-dimensional plane
stress nite element simulations on an imperfect
specimen are performed. Next, the homogenisa-
tion strategy will be elucidated. At this stage, the
emphasis is on the ability to determine an eective
parameter set for the homogeneous equivalent
material. This implies that the choice of the mi-
crostructure is rather arbitrary. We have assumed
a regular cubic porous microstructure, which is
fairly easy and fast to model (see Fig. 6). This is by
no means a limitation of the method, as also the
most oddly shaped inclusions, elastic or visco-
plastic, can be taken into account. Interparticle
interactions also could be modelled by simply ex-
panding the mesh of the RVE with a number of
inclusions. Values for the eective material
parameters are obtained from our homogenisation
procedure. These values are used to simulate an
ordinary uniaxial tensile test, a shear test, a biaxial
tensile test and a cyclic (tension-compression) test
on a perforated plate. The obtained global be-
haviour is confronted with direct nite element
calculations on the complete heterogeneous struc-
ture.
2. Model formulation
In this section, Perzyna's viscoplastic model
formulation will be discussed. First, the constitu-
tive equations will be given, after which the full
NewtonRaphson procedure for the solution of
the non-linear discretised equilibrium equations
will be considered. Next, the capability of the
model to describe strain softening behaviour
properly will be shown by means of quasi-static
nite element simulations on an imperfect speci-
men.
2.1. The constitutive equations
Analogous to the elastoplastic models, the total
strain tensor e is decomposed into an elastic part
and a viscoplastic part (Zienkiewicz and Cormeau,
1974),
e = e
e
e
vp
; (1)
when initial strains are neglected. The relation
between the Cauchy stress tensor r and the elastic
part of the strain tensor is simply dened by
Hooke's law:
r =
4
D
e
: e
e
with
4
D
e
=
mE
(1 m)(1 2m)
II
_

1 2m
m
4
I
_
;
(2)
where E is Young's modulus, m Poisson's ratio, I
the second order unit tensor, and
4
I the fourth
order unit tensor. This relation completely de-
scribes the reversible part of the deformation. The
viscoplastic response of the material becomes
manifest as soon as some specied combination of
the stress components exceeds a characteristic
scalar value. The occurrence of plastic yielding
O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759 745
depends on a yield condition F (r; j), where j is a,
yet to be dened, history parameter. Here, we will
adopt the pressure-independent Von Mises yield
criterion, which is dened by
F (r; j) =

3J
2
_
1(j); (3)
with J
2
the second invariant of the deviatoric stress
tensor, J
2
= (1=2)r
d
: r
d
. The function 1(j) denes
the current yield stress and is taken to be linear in
j,
1(j) = r
Y
hj; (4)
with r
Y
the initial yield stress, h the hardening
modulus and j the equivalent viscoplastic strain
given by
j =
_
t
0
_ j ds with _ j =

2
3
_ e
vp
: _ e
vp
_
: (5)
Clearly, positive values of h correspond to hard-
ening behaviour, whereas negative values indicate
strain softening. The direction of the viscoplastic
strain rate is dened as the gradient of the ow
surface in stress space, (i.e. oF =or), as expressed by
(Perzyna, 1966, 1971; Zienkiewicz and Cormeau,
1974; Owen and Hinton, 1980),
_ e
vp
=
_
k
oF (r; j)
or
=
_
ka; where
_
k = c[[/(F )[[
(6)
with c a uidity parameter, /(F ) a ow function,
and [[[[ referring to [[x[[ = xH(x), with H(x) the
Heaviside step function. Hence, viscoplastic
strains are non-zero only if /(F ) P0, i.e. stress
states on or outside the yield surface. Notice that
this is in constrast with elastoplastic models, where
only stress states inside or on the yield surface are
legal. In this respect, the term `overstress' model is
evident. The amount of the viscoplastic strain rate
is equal to the scalar value c/(F ). A power law is
chosen for the ow function /(F ) (Perzyna, 1971)
/(F ) =
F
r
Y
_ _
N
; (7)
with N the rate-sensitivity parameter.
2.2. Finite element formulation
The partial dierential equations for the visco-
plastic model, combined with the equilibrium
equations, will be solved numerically, using the
nite element method. The method is based on the
weighted residual formulation, that is obtained by
multiplying the equilibrium equations with
weighting functions ~w, integrating over the domain
X, integrating by parts and applying Cauchy's
theorem, which states that the stress vector
t
i
= r
ij
n
j
:
_
X
ow
i
ox
j
r
ij
dX =
_
C
w
i
t
i
dC \w
i
: (8)
When applying the generally used matrix with
dierential operators L and introducing the col-
umns e
T
= [e
11
; e
22
; c
12
[, r
T
= [r
11
; r
22
; r
12
[ and
u
T
= [u
1
; u
2
[, appropriate for the two-dimensional
plane stress case, it follows that e = Lu. Using this,
accompanied with the denitions w
T
= [w
1
; w
2
[
and t
T
= [t
1
; t
2
[, we obtain (8) in matrix notation
_
X
(Lw)
T
r dX =
_
C
w
T
t dC \w: (9)
This equation is interpolated, making use of an
isoparametric formulation (Zienkiewicz and Tay-
lor, 1989). Following Galerkin's method, the in-
terpolation functions N for the displacements and
the weighting functions are chosen identically
u(n; g) = N
T
(n; g) u
e
and
w(n; g) = N
T
(n; g) w
e
; (10)
with u
e
and w
e
columns containing the values of
the degrees of freedom and the weighting functions
at the nodes, respectively. Dening the matrix B
according to Lw = Bw
e
and recalling that (9)
holds for all w, we obtain the discretised equilib-
rium equations
_
X
B
T
r dX =
_
C
N
T
t dC or f
int
= f
ext
; (11)
with f
int
the internal force vector, and f
ext
the
external load vector.
Because of the history and time dependent
material behaviour, the value of the state variables
at the end of the loading, cannot be determined
746 O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759
directly. Instead, an incremental-iterative proce-
dure is adopted. Here, the total loading history
will be applied in a number of small loading steps,
also called increments, in order to calculate the
deformation history accurately. Thus, the state of
the material is determined at a nite number of
discrete moments in time t
n
, with n = 0; 1; 2; . . .
From now on, the attention is focussed on the
current increment, which starts at t = t
n
and
ends at t = t
n1
= t
n
Dt. Furthermore, it is as-
sumed that the state at time t = t
n
satises the
equilibrium equations. These equations can now
be written as
_
X
n1
B
T
r
n1
dX = f
ext
n1
: (12)
The NewtonRaphson iteration procedure is used
to determine a sequence of approximate solutions
until (12) is satised to a certain degree of accu-
racy. The unknown current value of the stresses is
written as r
n1
= r
+
n1
dr, where r
+
n1
is the es-
timation of the stresses and dr the correction
thereof. Hence, (12) becomes
_
X
n1
B
T
dr
n1
dX = f
ext
n1

_
X
n1
B
T
r
+
n1
dX
= f
ext
n1
f
int
n1
= r
n1
; (13)
in which r
n1
is the residual force vector which has
to become zero to satisfy the equilibrium equa-
tions. This residual is dependent on r
n1
, which
can be calculated in an accurate way by a so-called
stress update algorithm, that will be developed
below.
The equilibrium equation (13) is used to deter-
mine an iterative displacement eld du. This is only
possible when the iterative stress dr is expressed in
du, or equivalently, in de,
dr = Mde; (14)
where M is a material matrix, which follows from
the chosen constitutive model. The obvious choice
then is to use the so-called modular matrix, which
relates the stress rate to the strain rate, _ r = C_ e,
and can be derived from the constitutive equa-
tions. However, Criseld (1991) has pointed out
that this matrix is inconsistent with the applied
backward Euler time integration scheme (or any
other integration scheme for that matter), which
will destroy the quadratic convergence of the
NewtonRaphson procedure. To restore this rapid
convergence rate, a Jacobian matrix D
vp
has to be
derived instead, which can be used to formulate
the consistent tangential stiness matrix K. This
can be accomplished by substituting dr
n1
=
D
vp
n1
de
n1
into (13). Henceforth, the iterative
equilibrium Eq. (13) can be used to determine du,
according to
K
n1
du = r
n1
: (15)
Stress update algorithm: As was already men-
tioned, an iterative stress update algorithm is re-
quired. To this end, the stresses at time t
n1
are
written as
r
n1
= r
n
D
e
(De De
vp
)
= r
tr
D
e
De
vp
; (16)
where r
tr
are the elastic trial stresses. To calculate
the incremental value of the viscoplastic strains,
De
vp
, a time integration scheme must be used.
Here, we adopt the implicit backward Euler algo-
rithm, from which (6) can be written as
De
vp
= Dka
n1
: (17)
The backward Euler time integration scheme in-
volves a vector a
n1
, that is dened at the end of
the increment, and cannot be computed directly.
Therefore, an inner-iterative loop at integration
point level is employed in order to calculate the
stresses in an accurate way. Following Criseld
(1997), we dene the residuals
s = r
n1
r
tr
DkD
e
a
n1
; (18)
g = Dk Dtc[[/(F )[[
n1
: (19)
Obviously, the iterative scheme should minimise
these residuals. The residual vector s
i1
= s
i
ds is
approximated by a rst order Taylor series (clearly
with xed r
tr
)
s
i1
= s
i
dr
n1
D
e
a
n1
dk
DkD
e
oa
or
_ _
n1
dr
n1
= 0; (20)
since it follows from (3) and (6) that oa=oj = 0.
Rearranging gives
O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759 747
dr
n1
= Q
1
s
i
Ra
n1
dk; (21)
with
Q = I
_
DkD
e
oa
or
_ _
n1
_
and
R = Q
1
D
e
: (22)
The scalar residual g
i1
= g
i
dg is also approxi-
mated by a truncated Taylor series
g
i1
= g
i
dk Dtc/
/
F
a
T
n1
dr
n1
Dtc/
/
F
F
/
j
dk; (23)
where /
/
F
= o/=oF and F
/
j
= oF =oj. Elimination
of dr
n1
from (23) using (21) results in
dk =
g
i
Dtc/
/
F
a
T
n1
Q
1
s
i
1 Dtc/
/
F
a
T
n1
Ra
n1
Dtc/
/
F
F
/
j
: (24)
This expression can then be used to update the
stresses r
n1
by substituting the value of dk in (21).
Tangential stiness matrix: The derivation of
the consistent tangential stiness matrix is based
on the variation of (16) combined with (17)
dr
n1
= D
e
de
n1
D
e
a
n1
dk
DkD
e
oa
or
_ _
n1
dr
n1
: (25)
Rearranging then gives
dr
n1
= R(de
n1
a
n1
dk): (26)
Elimination of dk can be achieved by writing (6) in
incremental form,
Dtc[[/(F )[[
n1
= Dk: (27)
Variation gives
Dtc/
/
F
a
T
n1
dr
n1
Dtc/
/
F
F
/
j
dk = dk: (28)
When substituting (26) into (28), we obtain a re-
lation between dk and de
dk =
Dtc/
/
F
1 Dtc/
/
F
a
T
n1
Ra
n1
Dtc/
/
F
F
/
j
_ _
a
T
n1
Rde
n1
: (29)
Elimination of dk from (26) using (29) gives the
relation for the Jacobian matrix D
vp
:
dr
n1
= R
_

1
(Dtc/
/
F
)
1
a
T
n1
Ra
n1
F
/
j
Ra
n1
a
T
n1
R
_
de
n1
= D
vp
n1
de
n1
: (30)
The stress-update algorithm and the consistent
tangential stiness matrix described above, was
also found by Peric (1993).
2.3. Strain softening simulations
Quasi-static nite element simulations on an
imperfect specimen will be carried out. To check
the mesh-objectivity required to properly model
strain softening, the response of the simulations
must indeed converge to a unique solution upon
rening the mesh. To this end, a tensile specimen is
used which is shaped in such a way that in the
center, a stress concentration is initiated, resulting
in localisation of the deformation. The specimen
and the applied boundary conditions are shown in
Fig. 1. The following values for the material pa-
rameters are chosen: E = 10 000 (MPa),
m = 0:3 (), r
Y
= 25 (MPa), h = 1000 (MPa),
c = 10 (1/s) and N = 1 (). The prescribed strain
rate is _ e = 1 s
1
. To check the inuence of the
spatial discretisation of the specimen on the results
of the simulations, several meshes will be used
which are shown in Fig. 2.
As can be inferred from this gure, due to
symmetry of the specimen and the loading condi-
tions, only a quarter of the specimen has to be
Fig. 1. The tensile specimen with boundary conditions.
748 O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759
modelled. The resulting forcedisplacement curves
for the dierent meshes are shown in Fig. 3. From
this, we may conclude that the viscoplastic model
can indeed be used to describe strain softening
material behaviour in a proper way. Of course,
when the viscosity is reduced to zero (i.e. c ),
a rate-independent plasticity model is recovered,
that is not able to describe strain softening.
To illustrate the typical characteristics of the
model, several local stressstrain curves are shown
in Fig. 4. The left picture shows the behaviour
during continuous loading, obviously caused by
the stress concentration in the center of the speci-
men. The picture in the center clearly illustrates the
loading and subsequent elastic unloading behav-
iour which appears to be properly captured by the
Fig. 3. The resulting forcedisplacement curves for the dierent
spatial discretisations.
Fig. 2. Four dierent discretisations.
Fig. 4. Local stressstrain curves during loading of the specimen.
O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759 749
model. The right curve shows the elastic response,
since in this part of the specimen, stresses simply
are not high enough to give rise to any viscoplastic
deformation.
3. Homogenisation of the viscoplastic model
The purpose of the homogenisation process is to
model a heterogeneous material by means of a
unique continuous medium(Maugin, 1992). In that
way, it is possible to establish a constitutive equa-
tion which can be used to describe the mechanical
behaviour of the homogeneous equivalent contin-
uum, which `best' represents the response of the
real heterogeneous material (Fig. 5). The eective
properties of a heterogeneous material dene the
relations between averages of eld variables such as
stress and strain when their spatial variation is
statistically homogeneous, that is, the eective
properties dene the constitutive equations of the
equivalent homogeneous continuum.
The homogenisation strategy which is discus-
sed in this section, can be outlined as follows.
The rst step is dening the relations between the
microscopic and macroscopic state variables, e.g.
stresses and strains. By using these denitions, it
is then possible to derive microscopic boundary
conditions which can be applied to the represen-
tative volume element (RVE). These boundary
conditions are formulated as displacements re-
lated to the macroscopic deformation quantities.
It is important to observe that the deformation of
the RVE does not follow from any periodicity
considerations. This is contradictory to the gen-
erally employed homogenisation techniques,
which are based on periodicity demands of the
microstructure (e.g. Auriault, 1991; Ghosh et al.,
1995, 1996; Kruch et al., 1998; Smit et al., 1998).
The response of the RVE is subsequently calcu-
lated by nite element simulations. Then, the
macroscopic state variables can be calculated,
using the micromacro relations. Since the
prescribed values of the macroscopic strains and
strain rates are known, and subsequently, the
macroscopic state quantities can be calculated, it
is possible to determine an eective parameter set
for the constitutive equations describing the me-
chanical behaviour of the homogeneous equiva-
lent continuum.
Fig. 5. Graphical representation of the homogenisation process.
750 O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759
3.1. Consistent boundary conditions
The macroscopic strain tensor is dened as the
average of the corresponding quantity dened on
the RVE,
E(x) = e)(x); x X; (31)
in which ) denotes the averaging operator. In the
deterministic theories (Hill, 1965; Maugin, 1992),
the average is taken over the region R, occupied by
the RVE,
e)(x) =
1
V
_
R
e(x; y) dy; with V =
_
R
dy: (32)
The macroscopic strain tensor will be applied to
the RVE through appropriate boundary condi-
tions. These boundary conditions have to be con-
sistent with the denition (32). The consistency
conditions are a direct consequence of the deni-
tion of the linear strain tensor and the divergence
theorem,
E
ij
(x) =
1
2V
_
oR
u
i
(x; y)n
j
(x; y)
_
u
j
(x; y)n
i
(x; y)
_
ds; (33)
where n(x; y) is the outward normal vector on oR.
When we assume a uniform strain eld as boun-
dary conditions,
u
i
(x; y) = u
0i
(x) E
ij
(x)y
j
on oR; (34)
it is easily veried that these boundary conditions
are consistent with (33).
3.2. Determination of the eective parameters
The eective parameters for the description of
the mechanical behaviour of the equivalent ho-
mogeneous continuum, now have to be tted onto
nite element calculations performed on the RVE.
For simplicity, the RVE which we will use and will
be discussed here, represents a regular cubic hole
distribution. The nite element mesh is shown in
Fig. 6. The characteristic size of the RVE is set to
a = 1 (mm). The radius of the hole is r = 0:25
(mm), which corresponds to an initial void volume
fraction of f = 20%. Notice from Fig. 6, that the
hole is also discretised into a number of nite el-
ements. This precludes the necessity of prescribing
additional boundary conditions on the inner edge,
caused by the hole. The inclusion is assumed iso-
tropic elastic with a modulus of E = 1 (MPa) and
m = 0:3. These values are based on a comparison of
the global mechanical response of the `lled' RVE
and an RVE with an actual hole. The forcedis-
placement curves of the two RVEs, obviously
obtained from identical boundary conditions, were
identical using these values. An additional ad-
vantage of discretising the hole is that also arbi-
trary inclusions with dierent mechanical
properties can be modelled using the same nite
element mesh.
The viscoplastic model is applied both on the
macrolevel and on the microlevel. However, on the
microlevel, the material parameters are known
(Table 1), and on the macrolevel, the parameters
must follow from our homogenisation procedure.
Furthermore, a thoroughly motivated selection of
a constitutive model at the macroscopic level re-
mains dicult. In literature, several macroscopic
models are suggested for the description of voided
materials (e.g. Gurson, 1977; Becker and Needle-
man, 1986; Haghi and Anand, 1992; Garajeu and
Suquet, 1997). In these models, pressure dependent
yield surfaces follow from microstructural con-
Fig. 6. The nite element mesh of the RVE representing a
regular cubic hole stacking.
O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759 751
siderations. In this paper, we will conne ourselves
to pressure independent yield at the macroscopic
level, hereby not providing plastic compressibility.
The strategy of determining the eective prop-
erties for the equivalent homogeneous continuum
is divided into three steps: (i) eective elastic
properties, resulting in values for

E and m; (ii)
eective plastic properties, giving values for r
Y
and

h; (iii) and eective viscous properties, yielding c

and

N. These three steps will be elucidated in the
following. Using the averaging operator intro-
duced in (32), the following micromacro relations
are dened
R(x) =
1
V
_
R
r(x; y) dy; R
eq
=

3
2
R
d
: R
d
_
; (35)
E(x) =
1
V
_
R
e(x; y) dy; (36)
K(x) =
1
V
_
R
j(x; y) dy: (37)
In the following paragraphs, the uniform boun-
dary conditions (34) are applied under tensile
conditions by prescribing displacements on
y
1
= a=2. Henceforth, the applied constant
macroscopic strain rate is
_
E
11
= 1 s
1
. Because of
geometric and loading symmetry, only one quarter
of the RVE has to be modelled. In addition, the
vertical displacements on the boundaries
y
2
= a=2 are prescribed in such a way that the
boundaries should remain straight during loading,
while the resulting force in y
2
-direction equals zero
(Idesman et al., 1995). This can be easily imple-
mented in the nite element program by applying
nodal constraints (e.g. Bathe, 1982).
3.2.1. Eective elastic properties
Since the boundary conditions represent a ten-
sile conguration, we assume that the following
holds: R
22
= 0; R
12
= 0 and E
e
12
= 0. From the
averaged elastic constitutive Eq. (2), the following
expressions for the eective elastic properties are
readily obtained
m =
E
e
22
E
e
11
and

E =
R
11
E
e
11
: (38)
Of course, the resulting equivalent stresses in the
RVE are not allowed to exceed the local yield limit
of the material. The eective values are obtained as

E = 6191 (MPa) and m =0.27 (). Ju and Tseng

(1996) found

E =6547 (MPa) and m =0.28 () for
a given volume fraction of 20%.
3.2.2. Eective plastic properties
For determining the eective initial yield stress
r
Y
, we assume that at the instant of time at which
local yielding occurs in any point y in the RVE, the
eective initial yield stress is reached. At this time
instant, it holds that
r
Y
= R
eq
: (39)
The time instant can be calculated accurately by
increment renement. Using the boundary condi-
tions (34) and prescribing only the macroscopic
normal strain in the y
1
-direction including the
specied nodal ties applied on the horizontal
edges, the converged value of the eective yield
stress was obtained as r
Y
= 8:1 (MPa). Further
increment renement did not alter this value.
When averaging Eqs. (3) and (4),
R
eq
r
Y

hK = 0; (40)
an expression can be formulated for determining
the eective hardening coecient

h =
oR
eq
oK
: (41)
For the purpose of obtaining the value for

h, we
again perform a tensile test on the RVE (Fig. 6).
Table 1
The constitutive parameters for the matrix material at the microlevel
Elastic Plastic Viscous
E (MPa) m () r
Y
(MPa) h (MPa) c (1/s) N ()
10 000 0.3 25 1000 10 1
752 O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759
Contradictory to the loading of the elastic simu-
lation, the value of the imposed strain E
11
is such
that also viscoplastic deformations are allowed. By
means of the proposed micromacro relations, the
averaged equivalent stress R
eq
in the RVE and the
averaged equivalent viscoplastic strain K can be
calculated. This is shown in Fig. 7(a). The eective
hardening modulus is obtained by using a stan-
dard MarquardtLevenberg least-squares tting
procedure:

h = 1026 (MPa). In Fig. 7(b), the
calculated and tted plots of R
eq
K are shown.
It should be remarked that after reaching the
initial yield point, the mechanical behaviour be-
comes rate dependent. This implies that the stress
level during viscoplastic loading is rate dependent.
To investigate the inuence of the imposed strain
rate on the value of the eective hardening modu-
lus, we performed tensile tests on the RVE with
three dierent imposed strain rates,
_
E
11
= 0:5 s
1
; 1:0 s
1
and 2:0 s
1
. The evolution
of the averaged equivalent stress as function of the
averaged equivalent viscoplastic strain for these
three cases is shown in Fig. 8. From this, we may
conclude that the value of

h is nearly independent of
the imposed strain rates. Actual tting revealed that
deviations were well within 10%. It will be obvious
that the ideal loading case for the determination of

h would be a very low value of the imposed strain

rate, since in that case, the stresses should be nearly
on the yield surface instead of outside.
3.2.3. Eective viscous properties
The two parameters that are not yet deter-
mined, are the eective viscosity parameter c and
the ow constant

N. To obtain these values, it is
proposed to use the averaged response of the
RVE obtained from a tensile test, using the
boundary conditions (34). The constitutive Eqs.
(1), (2) and (6) for the averaged state variables in
the viscoplastic regime under tensile conditions
read
Fig. 7. (a) The averaged equivalent stress as function of the averaged equivalent viscoplastic strain and (b) the tted and calculated
plot of R
eq
K.
Fig. 8. The averaged equivalent stress as function of the aver-
aged equivalent viscoplastic strain for dierent imposed strain
rates.
O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759 753
_
R
11
=

E
1 m
2
_
E
11
_
_
_
m
_
E
22
c
R
11
r
Y

hK
r
Y
_ _
N
1
_

m
2
_
_
_
_
; (42)
0 =

E
1 m
2
_
E
22
_
_
_
m
_
E
11
c
R
11
r
Y

hK
r
Y
_ _
N
m
_

1
2
_
_
_
_
: (43)
Observe that in these equations, we have only the
two parameters left as unknowns. These two
equations can be rearranged to give
c
R
11
r
Y

hK
r
Y
_ _
N
= 1
_

m
2
_
1
_
E
11
_
m
_
E
22

_
R
11

E
(1 m
2
)
_
;
(44)
c
R
11
r
Y

hK
r
Y
_ _
N
= m
_

1
2
_
1
_
E
22
_
m
_
E
11
_
:
(45)
Substituting the relation for the total strain rate
tensor,
_
E =
_
E
e

_
E
vp
into (44), and using the
elastic relation for the stress rate, we obtain after
rearranging
c
R
11
r
Y

hK
r
Y
_ _
N
= 1
_

m
2
_
1
_
E
vp
11
_
m
_
E
vp
22
_
:
(46)
Upon substitution of the decomposition relation
of the total strain rate into (45) and equating the
right-hand side of this equation with the right-
hand side of (46), we have
m
_

1
2
_
_
E
vp
11
_
m
_
E
vp
22
_
= 1
_

m
2
_
_
E
vp
22
_
m
_
E
vp
11
_
;
(47)
from which it follows that at the macroscopic level,
we also have incompressible viscoplastic ow,
_
E
vp
22
=
1
2
_
E
vp
11
: (48)
When we use the incompressibility (48) with (46),
we obtain a simplied expression
c
R
11
r
Y

hK
r
Y
_ _
N
=
_
E
vp
11
: (49)
This equation will be used to obtain the eective
values c and

N. Note that the corresponding values
only depend on the viscoplastic response of the
RVE, according to (49). Again, by applying a
standard MarquardtLevenberg least-squares
algorithm, the tted parameters are obtained:
c =0.65 (1/s) and

N =2.62 (). For convenience,
the eective properties are given in Table 2. In
Fig. 9, the right-hand side of Eq. (49) is depicted as
the calculated part, whereas the left-hand side of
Eq. (49) is plotted as the tted part. Also from this
gure, it can be seen that the tted curve does not
match the calculated curve exactly, indicating a
deciency in the macroscopic model. It is believed
that the assumption of macroscopic incompress-
ible plastic ow, as well as the constant visco-
plastic material parameters are the main cause for
this observation.
Another possibility of determining the eective
viscoplastic parameters could be to perform tensile
tests on the RVE with a wide range of imposed
strain rates, resulting in a large database of sample
points. This database could then be used with an
Table 2
The eective constitutive parameters for the homogeneous equivalent continuum
Elastic Plastic Viscous

E (MPa) m () r
Y
(MPa)

h (MPa) c (1/s)

N ()
6191 0.27 8.1 1026 0.65 2.62
754 O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759
estimation algorithm to obtain values for the ef-
fective parameters. Such strategies are commonly
used in mixed numericalexperimental approaches
(e.g. Meuwissen, 1998; Geers et al., 1998).
4. Numerical validation
In this section, a uniaxial tensile test, a simple
shear test, a biaxial test and a cyclic (tension
compression) test will be performed on a perfo-
rated plate, depicted in Fig. 10(a). Results
obtained from direct calculations, where the
complete heterogeneous specimen is discretised,
are confronted with results from calculations in
which the obtained homogenised parameter set is
exploited. The applied meshes for the simulations
are shown in Fig. 10(b) and (c). The shear test, the
biaxial tensile test and the cyclic test are interesting
since the loading history is dierent from the uni-
axial tensile loading on the RVE from which the
eective properties have been extracted. We have
chosen B
0
= 7a and L
0
= 12a. The thickness of the
plate is taken as 0.001 (mm), justifying the plane
stress case.
4.1. Uniaxial tensile test on a perforated plate
The displacements are prescribed in the x
2
-di-
rection. In addition, a geometric imperfection has
been used to give a heterogeneous deformation
pattern, and is given by the expression (Timmer-
mans, 1997)
B
1
(x
2
) = B
0
1
_

1
2
(1 n) cos
px
2
fB
0
_ _
1
___
;
0 6x
2
6fB
0
; (50)
with B
1
the width over the imperfection, B
0
the
initial width of the plate, n = 0:98 a measure of the
imperfection (n = B
1
=B
0
) and f = 0:5 dening the
imperfection length fB
0
.
Fig. 9. Graphical representation of Eq. (49).
Fig. 10. (a) The perforated specimen, (b) the mesh used for the homogenised simulations, and (c) the mesh used for the direct sim-
ulations.
O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759 755
A comparison of the two simulations only is
allowed concerning deformation quantities that
are dened at the edge of the specimens. Thus, the
resulting forcedisplacement curves can be used
for this purpose, and are depicted in Fig. 11. It can
be observed that a good agreement is obtained.
4.2. Simple shear test on a perforated plate
For this test, the displacements on the upper
edge are prescribed in the x
1
-direction. In addition,
the vertical displacements on this edge are sup-
pressed. The resulting forcedisplacement curves
of the direct and homogenised simulations are
depicted in Fig. 12. Due to the deviatoric nature of
the shear test, the resemblance appears quite ex-
cellent, which clearly resides from the fact that our
homogenised model does not incorporate plastic
volume variance. However, at larger deforma-
tions, the inuence of the void growth becomes
more pronounced, thus resulting in a larger dif-
ference between the simulations. Also from this
result, one can conclude that the isotropy as-
sumption for the equivalent continuum can be
justied.
4.3. Biaxial tensile test on a perforated plate
Here, displacements in the x
1
-direction are
prescribed on the right edge, and on the upper
edge, displacement in the x
2
-direction are specied.
The resulting forcedisplacement curves of the
direct and homogenised simulations are depicted
in Fig. 13. Again, the dierence between the two
simulations can be explained by the volume in-
variance assumption of the macroscopic model
and the constant material parameters.
4.4. Cyclic tensioncompression test on a perforated
plate
In this case, alternating positive and negative
displacements in x
2
-direction are prescribed. The
Fig. 11. The resulting loaddisplacement curves for the direct
and the homogenised simulations for the uniaxial tensile test.
Fig. 12. Loaddisplacement curves for the direct and the
homogenised simulations of the simple shear test.
Fig. 13. Loaddisplacement curves for the direct and the
homogenised simulations of the biaxial tensile test.
756 O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759
global forcedisplacement curves are shown in
Fig. 14. Also for this loading scenario, a good
agreement is obtained.
5. Discussion
A homogenisation method is proposed that
provides a way to devise constitutive equations for
heterogeneous strain softening materials. At the
microstructural level, a viscoplastic model is used
to capture the mechanical behaviour of the repre-
sentative volume element. For simplicity, we have
assumed a regular cubic hole stacking, which de-
nes the geometry of the RVE. This assumption is
by no means a limitation of the proposed method.
Exactly the same reasoning could be followed
when using more complex microstructures, such as
viscoplastic inclusions with arbitrary geometry.
The only factor that diers when considering
complex microstructures, will be the computa-
tional time for the microscopic simulations.
For the homogeneous equivalent medium at the
macrolevel, also a viscoplastic constitutive model
is exploited, with yet unknown model parameters.
The values for these eective parameters have been
obtained by separating the three characteristics of
the model, namely the elastic, plastic and viscous
parts. The parameters were tted onto nite ele-
ment calculations on the RVE by applying the
appropriate boundary conditions, which followed
from the micromacro relations. The tting pro-
cess of the viscous parameters revealed that the
chosen macroscopic model is not adequate enough
for describing the mechanical behaviour of the
considered material. In fact, the viscoplastic pa-
rameters may be functions of a deformation
characteristic (for instance, the equivalent visco-
plastic strain rate). Moreover, the assumption of
pressure independent yield appears to be a re-
striction.
Nevertheless, despite these simplications, a
good agreement was obtained between the re-
sponses of the homogenised plate and the perfo-
rated plate. The simple shear test revealed the
inuence of the absence of any pressure dependent
yield criterion in the homogenised model. In ad-
dition, the obtained agreement between the results
of the direct and homogenised simulations of the
considered loading cases, suggests that the isot-
ropy assumption at the macrolevel is justiable.
The occuring stress gradients caused by the holes
indicate that the average only of the state variables
may not be sucient to describe the statistical in-
formation of the RVE accurately. Therefore, ad-
ditional quantities in terms of the variations of the
state quantities could be considered (Van der Sluis
et al., 1999). Clearly, this will result in a higher
accuracy of the statistical information of the state
of the RVE, yielding a better approximation of the
eective parameter set.
References
Auriault, J.L., 1991. Heterogeneous medium: Is an equivalent
macroscopic description possible. Int. J. Engrg. Sci. 29 (7),
785795.
Bathe, K.J., 1982. Finite Element Procedures in Engineering
Analysis. Prentice Hall, Englewood Clis.
Bazant, Z., 1991. Why continuum damage is non-local:
micromechanics arguments. J. Engrg. Mech. 117 (5),
10701087.
Becker, R., Needleman, A., 1986. Eect of yield surface
curvature on necking and failure in porous plastic solids.
J. Appl. Mech. 53, 491499.
Boutin, C., 1996. Microstructural eects in elastic solids. Int. J.
Solids Struct. 33 (7), 10231051.
Brekelmans, W., 1993. Nonlocal formulation of the evolution
of damage in a one-dimensional conguration. Int. J. Solids
Struct. 30 (11), 15031512.
Fig. 14. Loaddisplacement curves for the direct and the
homogenised simulations of the tensile-compression test.
O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759 757
Criseld, M., 1991. Non-linear Finite Element Analysis of
Solids and Structures, Vol. 1: Essentials. Wiley, West
Sussex, England.
Criseld, M., 1997. Non-linear Finite Element Analysis of
Solids and Structures, Vol. 2: Advanced Topics. Wiley,
West Sussex, England.
De Borst, R., 1990. Simulation of localisation using Cosserat
theory. In: Bicanic, N., Mang, H. (Eds.), Computer Aided
Analysis and Design of Concrete Structures, Pineridge
Press, Swansea United Kingdom, Proceedings of SCI-C
1990 Second International Conference held in Zell am See
Austria, pp. 931943.
De Borst, R., M uhlhaus, H.-B., 1991. Continuum models for
discontinuous media. In: Van Mier, J., Rots, J., Bakker, A.
(Eds.), Fracture Processes in Concrete, Rock and Ceramics,
E & FN Spon, London United Kingdom, Proceedings of the
International RILEM/ESIS conference Noordwijk, The
Netherlands, pp. 601618.
Borst, R., Sluys, L., M uhlhaus, H.B., Pamin, J., 1993. Funda-
mental issues in nite element analyses of localization of
deformation. Engrg. Comp. 10, 99121.
Garajeu, M., Suquet, P., 1997. Eective properties of porous
ideally plastic or viscoplastic materials containing rigid
particles. J. Mech. Phys. Solids 45 (6), 873902.
Geers, M., 1997. Experimental Analysis and Computational
Modelling of Damage and Fracture. Eindhoven University
of Technology, Eindhoven.
Geers, M., De Borst, R., Peijs, T., Meuwissen, M., 1998.
Internal length scales for higher-order damage models in
random bre reinforced composites. In: Idelsohn, S., O~ nate,
E., Dvorkin, E. (Eds.), Computational Mechanics New
Trends and Applications. CIMNE, Barcelona, Spain.
Ghosh, S., Moorthy, S., 1995. Elasticplastic analysis of
arbitrary heterogeneous materials with the Voronoi cell
nite element method. Comp. Meth. Appl. Mech. Eng. 121,
373409.
Ghosh, S., Lee, K., Moorthy, S., 1995. Multiple scale analysis
of heterogeneous elastic structures using homogenization
theory and Voronoi cell nite element method. Int. J. Solids
Struct. 32 (1), 2762.
Ghosh, S., Lee, K., Moorthy, S., 1996. Two scale analysis of
heterogeneous elasticplastic materials with asymptotic
homogenization and Voronoi cell nite element model.
Comp. Meth. Appl. Mech. Engrg. 132, 63116.
Gurson, A., 1977. Continuum theory of ductile rupture by void
nucleation and growth: Part I: Yield criteria and ow rules
for porous ductile media. J. Engrg. Mat. Technol. 99, 215.
Haghi, M., Anand, L., 1992. A constitutive model for isotropic
porous elastic-viscoplastic metals. Mech. Mat. 13, 3753.
Hashin, Z., 1983. Analysis of composite materials a survey.
J. Appl. Mech. 50, 481505.
Hill, R., 1965. Continuum micro-mechanics of elastoplastic
polycrystals. J. Mech. Phys. Solids 13, 89101.
Idesman, A., Levitas, V., Stein, E., 1995. Finite-element
simulation of elastoplastic properties of two-phase compos-
ites reinforced by particles. In: Pineau, A., Zaoui, A. (Eds.),
Micromechanics of Plasticity and Damage of Multiphase
Materials, Solid Mechanics and Its Applications, 8390.
Kluwer Academic Publishers, Dordrecht, The Netherlands.
Ju, J., Tseng, K., 1996. Eective elastoplastic behavior of two-
phase ductile matrix composites: a micromechanical frame-
work. Int. J. Solids Struct. 33 (29), 42674291.
Kruch, S., Chaboche, J., Feyel, F., Roux, F., 1998. Two-scale
damage analysis of components reinforced with composite
materials. In: Idelsohn, S., O~ nate, E., Dvorkin, E. (Eds.),
Computational Mechanics New Trends and Applications.
CIMNE, Barcelona, Spain.
Li, J., Weng, G., 1997a. A homogenization theory for the
overall creep of isotropic viscoplastic composites. Acta
Mechanica 125, 141153.
Li, J., Weng, G., 1997b. A secant-viscosity approach to the time
dependent creep of an elasticviscoplastic composite. J.
Mech. Phys. Solids 45 (7), 10691083.
Maugin, G., 1992. The Thermomechanics of Plasticity and
Fracture. Cambridge University Press, Cambridge.
Meuwissen, M., 1998. An Inverse Method for the Mechanical
Characterisation of Metals. Eindhoven University of Tech-
nology, Eindhoven.
Mura, T., 1987. Micromechanics of Defects in Solids. Martinus
Nijho, Dordrecht.
Needleman, A., 1988. Material rate dependence and mesh
sensitivity in localization problems. Comp. Meth. Appl.
Mech. Engrg. 67, 6985.
Nemat-Nasser, S., Hori, M., 1993. Micromechanics: Overall
Properties of Heterogeneous Materials, Elsevier, Amster-
dam.
Owen, D., Hinton, E., 1980. Finite Elements in Plasticity:
Theory and Practice. Pineridge Press Limited, Swansea, UK.
Peerlings, R., Borst, R., Brekelmans, W., Vree, J., 1996.
Gradient enhanced damage for quasi-brittle materials. Int.
J. Num. Meth. Engrg. 39, 33913403.
Peric, D., 1993. On a class of constitutive equations in
viscoplasticity: formulation and computational issues. Int.
J. Num. Meth. Engrg. 36, 13651393.
Perzyna, P., 1996. Fundamental problems in viscoplasticity,
Advances in Applied Mechanics, Vol. 9, 243377. Academic
Press, New York.
Perzyna, P., 1971. Thermodynamic Theory of Viscoplasticity,
Advances in Applied Mechanics, Vol. 11, 313354. Aca-
demic Press, New York.
Ponte Casta~ neda, P., 1996. Exact second-order estimates for the
eective mechanical properties of non-linear composite
materials. J. Mech. Phys. Solids 44, 827862.
Sluys, L., Borst, R., 1992. Wave propagation and localization in
a rate-dependent cracked medium model formulation and
one-dimensional examples. Int. J. Solids Struct. 29 (23),
29452958.
Smit, R., Brekelmans, W., Meijer, H., 1998. Prediction of the
mechanical behavior of non-linear heterogeneous systems
by multi-level nite element modeling. Comp. Meth. Appl.
Mech. Engrg. 155, 181192.
Timmermans, P., 1997. Evaluation of a Constitutive Model for
Solid Polymeric Materials. Eindhoven University of Tech-
nology, Eindhoven.
758 O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759
Toupin, R., 1962. Elastic materials with couple-stresses.
Archives of Rational Mechanics and Analysis 11, 385414.
Triantafyllidis, N., Aifantis, E., 1986. A gradient approach to
localization of deformation I. Hyperelastic materials.
J. Elasticity 16, 225237.
Tvergaard, V., Needleman, A., 1995. Eects of non-local
damage in porous plastic solids. Int. J. Solids Struct. 32
(8/9), 10631077.
Van der Sluis, O., Vosbeek, P., Schreurs, P., Meijer, H., 1999.
Homogenization of heterogeneous polymers. Int. J. Solids
Struct. 36 (21), 31933214.
Vosbeek, P., 1994. A Micromechanical Approach to Deforma-
tion and Failure of Discrete Media. Eindhoven University
of Technology, Eindhoven.
Wang, W., Sluys, L., De Borst, R., 1996. Interaction between
material length scale and imperfection size for localisation
phenomena in viscoplastic media. Euro. J. Mech. A/Solids
15 (3), 447464.
Wu, F., Freund, L., 1984. Deformation trapping due to
thermoplastic instability in one-dimensional wave propaga-
tion. J. Mech. Phys. Solids 32 (2), 119132.
Zienkiewicz, O., Cormeau, I., 1974. Viscoplasticity plasticity
and creep in elastic solids a unied numerical solution
approach. Int. J. Num. Meth. Engrg. 8, 821845.
Zienkiewicz, O., Taylor, R., 1989. The Finite Element
Method Basic Formulation and Linear Problems. Hill,
London UK.
O. van der Sluis et al. / Mechanics of Materials 31 (1999) 743759 759