International Journal of Plasticity 51 (2013) 80–102

Contents lists available at SciVerse ScienceDirect

International Journal of Plasticity

journal homepage: www.elsevier.com/locate/ijplas

A combined incremental-secant mean-field homogenization

scheme with per-phase residual strains for elasto-plastic
composites
L. Wu a,⇑, L. Noels a, L. Adam b, I. Doghri b,c
a
University of Liege, Department of Aeronautics and Mechanical Engineering – Computational & Multiscale Mechanics of Materials, Chemin des Chevreuils 1,
B-4000 Liège, Belgium
b
e-Xstream Engineering, Axis Park-Building H, Rue Emile Francqui 9, B-1435 Mont-Saint-Guibert, Belgium
c
Université Catholique de Louvain, Bâtiment Euler, 1348 Louvain-la-Neuve, Belgium

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents an incremental secant mean-field homogenization (MFH) procedure
Received 12 February 2013 for composites made of elasto-plastic constituents. In this formulation, the residual stress
Received in final revised form 19 June 2013 and strain states reached in the elasto-plastic phases upon a fictitious elastic unloading are
Available online 2 July 2013
considered as starting point to apply the secant method. The mean stress fields in the
phases are then computed using secant tensors, which are naturally isotropic and enable
Keywords: to define the Linear–Comparison-Composite. The method, which remains simple in its for-
Mean-field homogenization
mulation, is valid for general non-monotonic and non-proportional loading. It is applied on
Composites
Elasto-plasticity
various problems involving elastic, elasto-plastic and perfectly-plastic phases, to demon-
Incremental-secant strate its accuracy compared to other existing MFH methods.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction

The direct numerical simulation of composite structures at fine scales being too expensive, emphasize was put during the
last decades on the development of simplified homogenization methods. The latter are either (semi-) analytical or numerical
and predict the macro or meso-scopic response of heterogeneous materials from their micro-structure and constituents
properties at reduced computational cost while maintaining an acceptable degree of accuracy. Kanouté et al. (2009) and
Geers et al. (2010) presented an overview of the different homogenization methods. Among those methods, the mean-field
homogenization (MFH) approach is an efficient semi-analytical framework for the modeling of multi-phase composites. MFH
methods were first developed for linear elastic composite materials by extending Eshelby (1957) single inclusion solution to
multiple inclusions interacting in an average way in the composite. Most common extensions of Eshelby (1957) solution are
the Mori–Tanaka (M–T) scheme developed by Mori and Tanaka (1973) and by Benveniste (1987), and the self-consistent
scheme pioneered by Kröner (1958) and Hill (1965b).
MFH schemes were also extended to the non-linear range to account for non-linear behaviors, such as (visco-) plasticity
or non-linear visco-elasticity, exhibited by the composite’s constituents. Most of these extensions revolved around the def-
inition of a so-called linear comparison composite (LCC) (Talbot and Willis, 1985, 1987, 1992; Ponte Castañeda, 1991, 1992;
Molinari et al., 2004), which is a virtual composite whose constituents linear behaviors match the linearized behavior of the
real constituents for a given strain state. Such a LCC is used in the incremental formulation proposed by Hill (1965a), which
considers linearized relations between the stress and strain increments of the different constituents around their current

⇑ Corresponding author. Tel.: +32 4 366 94 53; fax: +32 4 366 95 05.
E-mail address: [email protected] (L. Wu).

0749-6419/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijplas.2013.06.006
 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102 81

strain states. Thus the former homogenization techniques for linear responses can still be used on the strain increments to
predict the behaviors of highly non-linear composites. Such an approach was applied to predict the meso-scale response of
elasto-plastic composites by Pettermann et al., 1999; Doghri and Ouaar, 2003; Doghri and Tinel, 2005; Pierard and Doghri,
2006b, in which case the behavior of the composite is written as Dr ¼C  tg : De, where De; Dr  tg are respectively macro-
 and C
strain and stress increments, and a tangent operator. This incremental-tangent method can lead to too-stiff results unless
some isotropic projections of the tangent operator are considered during the M–T process, as shown by Doghri and Ouaar
(2003) and Pierard and Doghri (2006b). Another MFH approach is the so-called affine method, which was first proposed
by Molinari et al. (1987, 2004) for visco-plastic materials, and which considers the total strain field instead of strain incre-
ments during the homogenization process. This approach was extended to elasto-plastic materials by Zaoui and Masson
(2002) and Masson et al. (2000). For the affine method, the behavior of the composite is expressed as r  ¼C : e þ s, where
s is the polarization stress and where C can be different from the tangent moduli. Chaboche et al. (2005) showed that this
method can lead to too stiff results when an anisotropic tangent operator is considered in the homogenization process. Dif-
ferent and often accurate affine methods for visco-plastic composites were proposed by Pierard and Doghri (2006a), Mercier
and Molinari (2009) and Doghri et al. (2010). The LCC can also be defined from a secant operator, as initially proposed by
Berveiller and Zaoui (1978) for elasto-plastic materials. In this secant method the operator is the secant joining the origin
to the current strain/stress state and the response of the composite reads r  ¼C  sec : e, which limits the method to monotonic
and proportional loading paths. Recently, Wu et al. (2012) – the authors – have proposed a non-local incremental-tangent
MFH scheme accounting for damage. In that formulation, the incremental MFH approach is extended to account for the dam-
age behavior happening in the matrix-material at the micro-scale. In order to avoid the strain/damage localization caused by
the matrix material softening, a gradient-enhanced formulation (Peerlings et al., 2001; Engelen et al., 2003) was adopted
during the homogenization process. In this formulation, the non-local accumulated plastic strain of the matrix is defined
and depends on the local accumulated plastic strain and on its derivatives through the resolution of a new boundary value
problem following the developments of Peerlings et al. (1996), Geers (1997) and Peerlings et al. (1998).
Most MFH methods only consider first-moment-statistical values of the micro-strain and stress fields during the homog-
enization process. This can lead to poor predictions in the elasto-plastic case, as shown by Moulinec and Suquet (2003). This
motivated to consider second-moment-statistical values (Ponte Castañeda, 1996) during the homogenization process. Such
methods have been proposed for the secant formulations by Suquet (1995), Ponte Castañeda (2002a,b) and for the incremen-
tal-tangent formulation by Doghri et al. (2011). Suquet (1995) actually showed that the variational forms pioneered by Ponte
Castañeda (1992) correspond to a second-order secant formulation, which was called modified secant. Recently, incremental
variational formulations, which also correspond to a second-moment estimation, were proposed for visco-elastic composites
by Lahellec and Suquet (2007a,b), for thermoelastic composites by Lahellec et al. (2011), for elasto-(visco-) plastic compos-
ites by Brassart et al. (2011, 2012), and for elasto-visco-plastic composites with isotropic and kinematic hardening by Lahel-
lec and Suquet (2013). The method proposed by Brassart et al. (2011, 2012) can also be seen as a secant method based on the
elastic trial strain, instead of the total strain as in the original secant formulation, which allows the case of non-monotonic
and non-radial loading conditions to be simulated.
Note that there exist other homogenization methods, which consider non-linear effects, such as the transformation field
analysis proposed by Dvorak (1992, 1994), which is a MFH scheme for elasto-plastic composites where a relaxation stress is
defined due to the irreversible behavior in the matrix, leading to treat the interaction between the phases in a purely elastic
way: r  ¼C el : ðe  eres Þ. Homogenization methods not based on mean-field were also developed such as the method of cells
proposed by Lissenden and Arnold (1997) and Aboudi et al. (2003), the unit cell finite element (FE)-based computations as
performed by Wieckowski (2000), Segurado et al. (2002), Ji and Wang (2003), and Carrere et al. (2004), or again the multi-
scale FE2 method pioneered by Kouznetsova et al. (2002, 2004) as a non-exhaustive list.
Although multiscale homogenization methods in general, and MFH schemes in particular, often exhibit an acceptable,
sometimes even high, level of accuracy to capture the non-linear behavior of composites (Pierard et al., 2007b), some lim-
itations, assumptions or complexity in the formulation remain in the existing methods. For example, as previously said, to
remain accurate in the non-linear range, the incremental-tangent method requires the tangent-operator to be made isotropic
during the M-T process, as discussed by Doghri and Ouaar (2003) and Pierard and Doghri (2006b). Another limit of the incre-
mental-tangent method appears when damage is considered. As explained by Wu et al. (2012), during the softening stage of
the matrix, the fibers should see an elastic unloading due to the damaging process in the matrix, which cannot be modeled
using the incremental approach. As a result, when compared to the direct numerical simulations of a representative volume
element (RVE), the method remains accurate for low fiber ratios or low damage values only. In the variational approach pro-
posed by Brassart et al. (2012), the plastic strain field at the beginning of each time interval is supposed to be uniform within
the RVE, threatening the accuracy of the method in some cases, as when UD composites made of elastic fibers are loaded in
the longitudinal direction.
It is intended in this paper to propose what we believe is a new MFH approach for elasto-plastic materials. In this ap-
proach, at a given strain/stress state of the composite material, an unloading step is applied on the composite material
and the residual stresses are evaluated in both phases. The mean stress fields in the phases are computed using isotropic
secant tensors, which can in turn be used to define the LCC. Contrarily to the incremental-tangent approach, there is no need
to made isotropic the tensor used in the Eshelby term, as it is intrinsically so. In this formulation, the residual strains in the
different phases, and thus the plastic residual strains, are piece-wise continuous, removing a major assumption in the
variational approach proposed by Brassart et al. (2012). Another advantage of the method is expected when damage induced
82 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102

strain softening will be considered in a future work. Indeed, upon strain softening of the matrix, the fibers will be unloaded
by the formulation, which should in turn improve the accuracy of the method for high volume fractions of fibers.
The paper is organized as follows. In Section 2, the key principles of MFH for non-linear behaviors are briefly recalled.
Section 3 presents the proposed incremental-secant MFH for non-linear composites. In this formulation, the residual strain
in a phase appears in the form of a residual stress, obtained after the unloading of the composite. It is shown in Section 4 that
the method predicts accurate results compared to direct numerical simulations for a broad range of elastic and elasto-plastic
composite materials.

2. Generalities on MFH

In this section, the prerequisites to the development of the new incremental-secant MFH are summarized. In particular,
the principle of the MFH method for two-phase composites is recalled.
In the multiscale approach illustrated in Fig. 1, at each macro-point X, the macro-strain e is known, and the macro-stress
r is sought from a micro-scale boundary value problem (BVP), or vice versa. At the micro-level, the macro-point is viewed as
the center of a RVE of domain x 2 x and boundary @ x. Considering adequate boundary conditions (BCs) on the RVE, the Hill–
Mandell condition, expressing the equality between energies at both scales, transforms the relation between macro-strains e
R
and stresses r  into the relation between average strains hei and stresses hri over the RVE, with hf ðxÞi ¼ V1 x f ðxÞdV.
x
Considering a two-phase composite material with the phases volume ratios v 0 þ v I ¼ 1 (subscript 0 refers to the matrix
and I to the inclusions), the macro-strains e and stresses r can be written in terms of the average values in the matrix sub-
domain x0 and in the inclusions subdomain xI as

e ¼ v 0 heix þ v I heix ; and ð1Þ

0 I

r ¼ v 0 hrix0 þ v I hrixI ð2Þ

for both linear and non-linear frameworks.

For simplicity, in the following developments, the notations hixi will be replaced by i . Considering the so-called linear
comparison composite, which is defined in the case of non-linear composites, the relation between the average incremental
strains in the two phases depends on the chosen expressions of the virtual elastic operators C  LCC of the matrix phase and C
 LCC
0 I
of the inclusions phase I, leading to

 
DeI ¼ B I;C LCC  LCC : De0 :
0 ; CI ð3Þ

This equation describes the relation between the strain increment averages per phase through the strain concentration ten-
sor B , which expression depends on the assumptions made on the micro-mechanics. In this paper, we consider Mori and
Tanaka (1973) (M–T) model because it provides good predictions for two-phase composite materials for which the matrix
can be clearly identified, as discussed by Segurado and Llorca (2002). In this case, the strain concentration tensor reads

n h 1 io1
B ¼ I þ S : C LCC
0 : C LCC
I I ; ð4Þ

  LCC 
where Eshelby (1957) tensor S I; C 0 depends on the geometry of the inclusions phase and on the virtual elastic operator
 LCC
C 0 . The expressions of the tensors C LCC and C
 LCC depend on the chosen MFH process. In linear elasticity, these operators cor-
0 I
respond to the elastic material moduli C 0 and C el
el
I . In the classical incremental MFH method for non-linear materials, they
correspond to the so-called comparison tangent operators C  alg and C
 alg , which are uniform per phase, by design.
0 I   iso 
Note that when considering the incremental-tangent formulation, the Eshelby tensor S I;C 0 required to compute the
Mori–Tanaka strain concentration tensor (4) is evaluated from an isotropic part C  iso of C
 LCC , to improve prediction results.
0 0
More details can be found in the work of Pierard and Doghri (2006b).

Fig. 1. Multiscale method.

 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102 83

3. New proposal: incremental-secant MFH

In this section, the new incremental-secant mean-field homogenization (MFH) scheme is developed for elasto-plastic
composites. First, the secant formulation that will be considered to define the LCC is explained. Then, the MFH process is
described in details.

3.1. Incremental-secant moduli for rate-independent models

The secant formulation that will be used in the MFH framework developed herein is introduced in this section. The equa-
tions derived herein can be related to each phase of the composite material. Considering a time interval ½tn ; tnþ1 , with the
total strain tensor en at time tn and the strain increment Denþ1 resulting from the applied loading (for example from a finite
element resolution), yields the strain tensor
enþ1 ¼ en þ Denþ1 ; ð5Þ
at time t nþ1 , see Fig. 2(a). One can assume at time t n a residual strain tensor eres
n that corresponds to an elastic unloading from
the stress state rn to a stress state rres
n . Considering a heterogeneous material, this definition of the residual stress is related
to each material phase but also to the homogenized material. During the homogenization process, the residual stress for the
homogenized material will be null, but this will not be necessarily the case in the different phases. The same definitions also
apply at time tnþ1 .
The secant linearization of the elasto-plastic material is thus carried out in the time interval ½tn ; t nþ1  with the strain
increment Dernþ1 , such that
enþ1 ¼ eres r
n þ Denþ1 : ð6Þ
r
The main idea developed in this paper is to define a LCC, subjected to a strain increment De nþ1 ,
from which the stress ten-
sor is computed. As illustrated in Fig. 2, two methods will be considered in this work: the residual-incremental-secant meth-
od, which evaluates the stress tensor from the residual stress arising upon virtual unloading, and the zero-incremental-
secant method, which evaluates the stress tensor from a zero-stress state. Both methods are now presented.

3.1.1. Residual-incremental-secant approach

Following Fig. 2(a), the new stress tensor could be defined from the stress increment Drrnþ1 , such that
rnþ1 ¼ rres r
n þ Drnþ1 ; ð7Þ
where

Drrnþ1 ¼ C Sr : Dernþ1 : ð8Þ

Sr
In this last equation C is the residual-incremental-secant operator of the elasto-plastic material.
During the elastic regime, the elastic tensor C el can be used in the homogenization procedure. During elasto-plastic flow,
the stress tensor rnþ1 is computed from the unloaded state in the following way:

 Evaluate the trial stress tensor from an elastic response:

 
rtrnþ1 ¼ rn þ C el : Denþ1 ¼ rn þ C el : Dernþ1 þ eres res el r
n  en ¼ rn þ C : Denþ1 : ð9Þ

Fig. 2. Definition of the incremental-secant formulation. (a) Definition of the residual strain and stress and of the residual secant operator. (b) Definition of
the zero-secant operator.
84 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102

 If the trial stress tensor respects the von Mises criterion

   tr eq
f tr rtr
nþ1 ; pn ¼ rnþ1  rY  Rðpn Þ 6 0; ð10Þ
Sr el
then the trial stress (9) is the solution and C ¼ C . In Eq. (10), rY is the initial yield stress, RðpÞ is the isotropic hardening law
in terms of the accumulated plastic strain p, and the superscript ‘‘eq’’ refers to the equivalent von Mises effective stress.
 If the trial stress tensor does not respect the von Mises criterion, i.e. f tr > 0, then apply the plastic correction

rnþ1 ¼ rtrnþ1  C el : Dep ; with Dep ¼ DpN nþ1 ; ð11Þ

where N is the plastic flow direction. In this paper, we assume

 dev
Sr r
3 C : Denþ1
N nþ1 ¼  eq ; ð12Þ
2 C Sr : Der
nþ1

which satisfies N : N ¼ 32. Because of Eqs. (7) and (8), this last equation can be rewritten
 dev
3 rnþ1  rres
n
N nþ1 ¼  eq : ð13Þ
2 rnþ1  rres
n
 dev  
When rres n ¼ 0, this last equation corresponds to N nþ1 ¼ @f ð@rr; pÞ and we have the classical relation. When
 res dev nþ1
rn – 0; N nþ1 is a first order approximation of the normal to the yield surface in the stress space, see Fig. 3. Indeed
for infinitesimal strain increments De ! 0, the reloading process C el : Dernþ1 tends to the unloading increment C el : Deunload ,
see Fig. 2(a), and thus N nþ1 tends to the normal to the yield surface.
At the trial state, using Eq. (9) allows Eq. (12) to be rewritten as
 dev
el r  
3 C : Denþ1
dev
3 rtrnþ1  rres
N tr
nþ1 ¼  eq ¼  n
 eq : ð14Þ
2 C el : Der 2 rtrnþ1  rres
n
nþ1

Note that since C el is isotropic we have C el : ðDpN nþ1 Þ ¼ 2lel N nþ1 Dp and Eq. (11) becomes
 dev
ðrnþ1 Þdev ¼ rtr
nþ1  2lel N nþ1 Dp; ð15Þ

which can be rewritten as

 dev  
res dev
rnþ1  rres
n ¼ rtr
nþ1  rn  2lel N nþ1 Dp; ð16Þ

or, using the definitions (13) and (14), as

h eq i  
res eq
rnþ1  rres
n þ 3lel Dp N nþ1 ¼ rtr nþ1  rn N tr : ð17Þ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflffl
ffl ffl} nþ1
{zfflfflfflfflfflfflfflfflfflfflffl
P0 P0

Since Eqs. (12) and (14) imply N nþ1 : N nþ1 ¼ 32 ¼ N tr tr

nþ1 : N nþ1 , then Eq. (17) results in

Fig. 3. Plastic corrections in the stress space (a) radial return mapping (b) approximation (14).
 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102 85

 eq  
res eq
rnþ1  rres
n þ 3lel Dp ¼ rtr
nþ1  rn and ð18Þ
N nþ1 ¼ N tr
nþ1 : ð19Þ
 
The elasto-plastic scheme consists of solving Eq. (18) and f rnþ1 ; pnþ1 ¼ 0, with pnþ1 ¼ pn þ Dp, in terms of Dp and req
nþ1 .
 Knowing rnþ1 , compute the residual-incremental-secant operator of the linear comparison material as follows. From Eqs.
(7) and (11) one has

Drrnþ1 ¼ C Sr : Dernþ1 ¼ C el : Dernþ1  2lel DpN nþ1 ; ð20Þ

which becomes after using (14) and (19),

2 3
6 I dev : C el 7
Drrnþ1 ¼ 4C el  3lel Dp  Sr
eq 5 : Dernþ1 ¼ C : Dernþ1 : ð21Þ
C el : Dernþ1

For J2-elasto-plastic materials, since C el is isotropic, the residual-incremental-secant operator of the linear comparison mate-
rial C Sr is also isotropic. Moreover, as C el ¼ 3jel I vol þ 2lel I dev , one can directly deduce

C Sr ¼ 3jr I vol þ 2lrs I dev ; ð22Þ

with

jr ¼ jel ; and ð23Þ

el 2 el 2
3l Dp 3 l Dp
lrs ¼ lel   eq ¼ lel   eq : ð24Þ
el
C : De r
nþ1
rnþ1  rres
n

 eq  
 eq
Note that on the one hand, since C el : Dernþ1 ¼ 3lel Dernþ1 , this last result can be rewritten lrs ¼ lel 1  Dp
eq . On
ðDernþ1 Þ
the other hand, in the variational incremental MFH proposed by Brassart et al. (2011) the shear modulus in one phase used
0 1
to define the LCC was found to be lrs ¼ lel @1  q
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Dp
eq 2
 A. Besides this second-moment form, it appears that both
ðenþ1 epn Þ xr

formulations have similarities. Indeed in case of unloading such that rres

n ¼ 0 in the incremental-secant approach, one has

Dernþ1 ¼ enþ1  epn .

 Practically the shear moduli of the virtual elastic material can be obtained by decomposing the increments of the stress
and strain tensors into the hydrostatic and deviatoric parts:
Drr ¼ Drm 1 þ Ds and Der ¼ Dem 1 þ De; ð25Þ
where Dr ¼ m
r Þ; Ds ¼ Dr  Dr 1; De ¼
1
3
trðD r r m m 1
trðD r Þ, r
e and where De ¼ De  De 1, see Appendix A for the notations. In-
3
m

deed, the increments of the von Mises equivalent stress and strain are respectively given by
 1=2  1=2
3 2
Dreq ¼ Ds : Ds and Deeq ¼ De : De ; ð26Þ
2 3
and one has directly1
Dreq
3lrs ¼ : ð27Þ
Deeq
 Evaluate the derivation of the operator (22) following Appendix B.1.

3.1.2. Zero-incremental-secant approach

Anticipating on the development of the new incremental-secant MFH, and as it will be shown in Section 4.1, when defin-
ing the LCC it can be advantageous to modify the residual-incremental-secant approach, following the suggestion in Fig. 2(b).
The approach follows closely Section 3.1.1, but with the omission of the residual stress, in which case the plastic flow direc-
tion (13) is rigorously normal to the yield surface.

3.1.3. Summary of the incremental-secant formulations

The stress tensor of an elasto-plastic phase at time t nþ1 is formulated using the incremental secant operator of the isotro-
pic-linear comparison material as

1
If Deeq ¼ 0, the indefiniteness is solved by considering lrs ¼ lel .
86 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102

( Sr
rres
n þC : Dernþ1 for the residual  incremental  secant method;
rnþ1 ¼ S0
ð28Þ
C : Dernþ1 for the zero  incremental  secant method:
In the forthcoming Section 4.1, a more detailed discussion illustrated by numerical examples will be given to justify the
choice of the incremental-secant operators to be considered in the matrix and in the inclusions phases.

3.2. MFH scheme

In this section, the new incremental-secant MFH is developed. Toward this end, the secant forms (28) will be considered
to define the LCC. Unless the expressions need to be particularized to the residual-incremental-secant or to the zero-incre-
mental-secant forms, the isotropic linear comparison stiffness

C S ¼ 3jI vol þ 2ls I dev ; ð29Þ

Sr S0
will substitute to either C or C . Similarly, ls holds for either l or for l while j ¼ j ¼ j ¼ j , where superscript 0
r
s
0
s,
0 r el

refers to the values of the zero-incremental-secant approach. During the MFH the comparison operators are constructed
as uniform by design on each phase r and are thus denoted C S.
r
Considering a time interval ½tn ; tnþ1 , the known data are the macro-total strain tensor at time tn ; en , the strain increment
Denþ1 , and the internal variables at time t n . The latter include the internal variables, gI n , of the inclusions and, g0 n , of the
matrix constitutive models, but also the residual variables computed from the elastic unloading step (i.e. r  res
n ¼ 0): the resid-
res res res
ual strains in the composite, en , in the inclusions phase, eI n , in the matrix phase, e0 n , and the residual stresses in the inclu-

sions phase, rres res
I n , and in the matrix phase, r0 n .
It needs to be clarified that, in the developed incremental-secant method, the strain increment Denþ1 applied to the RVE is
obtained from the FE resolution while the strain increment Dernþ1 is used to define the LCC in the MFH procedure. This is an
assumption as one cannot prove the existence of a displacement field increment compatible with Dernþ1 .
Combining Eqs. (5) and (6) for the homogenized material, one has

Dernþ1 ¼ en þ Denþ1  eres

n : ð30Þ
res
The explicit evaluation of e is described in the details of the MFH process reported here below.
n
The MFH process is stated by Eqs. (1)–(4). In this formalism, each phase r will see a strain increment Derr nþ1 to be applied
from the elastically-unloaded state, and from which a stress state
 
rr nþ1 ¼ F r Derr nþ1 ; gr n ; eres res
r n ; rr n ð31Þ

and an incremental-secant tensor

 
C Sr ¼ Gr Derr nþ1 ; gr n ; eres res
r n ; rr n ; ð32Þ

which follows Eq. (28), can be obtained from the constitutive model provided in Section 3.1. The system of equations can
thus be rewritten as

Dernþ1 ¼ v 0 Der0 nþ1 þ v I DerI nþ1 ; ð33Þ

r nþ1 ¼ v 0 r0 nþ1 þ v I rI nþ1 ; ð34Þ

with the relation between the strain increments reading

DerI nþ1 ¼ B ðI;C S0 ; C SI Þ : Der0 nþ1 : ð35Þ

To complete these equations, the unloaded state is defined by

r res ¼ v 0 rres
0 n þ v I rI n ¼ 0:
res
ð36Þ

To reach this virtual unloaded state, the composite material and both phases are assumed to obey an elastic behavior. As the
unloaded state is a virtual state used to define the LCC, this assumption can always be made. This elastic unloading is thus
performed from the state at time tn by solving the following system of equations:
   
n ¼ v 0 e0 n  e0 n þv I eI n  eI n ;
en  eres res res
ð37Þ
|fflfflfflfflffl{zfflfflfflfflffl} |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
Deunload
n Deunload Deunload
0n In
   
el
0¼r n ¼ v 0 r0 n  C 0 : De0 n
 res unload
þ v I rI n  C el unload
I : DeI n ; ð38Þ

with the relation between the unloading strain increments reading

Deunload
In ¼ B ðI;C el el unload
0 ; C I Þ : De0 n : ð39Þ
In this paper, a ‘‘First-order moment’’ method is considered, and the MFH process is described as follows.
 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102 87

 Initialize the strain increment in inclusions from the composite strain increment Dernþ1 (30): Dernþ1 ! DerI nþ1 .
 Follow the iterations process (upper indices (i) for values at iteration i of time t nþ1 are omitted for simplicity):
1. Call the constitutive material functions (31) and (32) of the real inclusions material with, as input, the strain tensor
increment in the inclusions phase DerI nþ1 and the internal variables gI n ; eres res
I n ; rI n at time t n . The output is the
updated stress rI nþ1 , the internal variables at time tnþ1 , and the incremental-secant operator C S
I nþ1 for the inclusions
 S el
phase. In case there is no plastic flow, we use C I nþ1 ¼ C I , which is always the case for elastic materials.
2. Compute the average strain in the matrix phase:
Der0 nþ1 ¼ ðDernþ1  v I DerI nþ1 Þ=v 0 : ð40Þ
3. Call the constitutive material functions (31) and (32) of the real matrix material with, as input, the strain tensor
increment in the matrix phase Der0 nþ1 and the internal variables g0 n; eres res
0 n ; r0 n at time t n . The output is the updated
stress r0 nþ1 , the internal variables at time t nþ1 , and the incremental-secant operator C S
0 nþ1 for the matrix phase. In
case there is no plastic flow, we use C S ¼ C el
.
0 nþ1 0
4. Predict the Eshelby tensor SðI;C S
0 nþ1 Þ using the secant operator of the matrix phase.
5. Applying a similar technique to the one detailed by Wu et al. (2012), Eq. (35) corresponds to satisfying F ¼ 0, where
F is the stress residual vector. For a time interval ½t n ; tnþ1 , during which Dernþ1 is constant, compute the stress resid-
ual vector in inclusions as, see Appendix C for details,

S 1 1
F¼C r
0 nþ1 : DeI nþ1  S : ðDerI nþ1  Dernþ1 Þ  C SInþ1 : DerI nþ1 : ð41Þ
v0
6. Check if the residual jFj 6 Tol. If so exit the loop.
7. Else, compute the Jacobian J matrix at constant Dernþ1 , such that dF ¼ J : derI ,2 following the details given in Appendix
C.
8. Correct the strain increment in inclusions

DerI nþ1 DerI nþ1 þ ceI with c eI ¼ J 1 : F; ð42Þ

then start a new iteration (go to step 1).

 After convergence, compute3

1. The homogenized stress
r nþ1 ¼ v 0 r0 nþ1 þ v I rI nþ1 ð43Þ
and the internal variables.
2. The ‘‘consistent’’4 linearization of the homogenized stress from (6) and (28)
 S   S 
 : Der
@ C  : Der
@ C
I I 0 0
dr ¼ v I drI þ v 0 dr0 ¼ v I
 : deI þ v 0 : de0
@ eI @ e0
    
@ C S @ eI @ C S @ e0
¼ v I C SI þ I : DerI : þ v 0 C S0 þ 0 : Der0 : : de: ð44Þ
@ eI @ e @ e0 @ e
This equation implies
 S   S 
C alg  S @ C I : Der : @ eI þ v 0 C S þ @ C 0 : Der : @ e0 ;
nþ1 ¼ v I C I þ I 0 0 ð45Þ
@ eI @ e @ e0 @ e
with @@eeI and @@ee0 reported in Appendix C.
 An unloading step is thus applied here to fit the incremental-secant process, and the obtained results will be kept as inter-
nal variables at time interval t nþ1 . The system of Eqs. (37)–(39), expressed at time t nþ1 instead of time t n , has thus to be
solved. The needed residual variables from unloading are the residual strains in the composite material eres nþ1 , in the inclu-
sions phase eres res res
I nþ1 and in the matrix phase e0nþ1 , as well as the residual stresses in the inclusions phase rI nþ1 and in the
matrix phase rres 0 nþ1 , respectively.
1. The residual strain eres  res
nþ1 (the strain at rnþ1 ¼ 0) of the composite material is calculated from an unloading step,
which is assumed to be a purely elastic process. The unloading operator of the composite is

2
Note that the derivative with respect to Derr has the same expression as the derivative with respect to er .
3
In this formalism, we do not need to evaluate explicitly the strain concentration tensor, nor the macro-moduli, but they follow directly from
n 1  S
o1  1
B ¼ I þ S : ½ðC
S
0 nþ1 Þ : C I nþ1  I  nþ1 ¼ v I C
, and C S  S 
I nþ1 : B þ v 0 C 0 nþ1  : ½v 1 B þ v 0 I , respectively.
4  alg ¼ @ r nþ1 .
In this paper we will use the term ‘‘consistent’’ operator for the derivative of the stress state with respect to the deformation increment C nþ1 @ Denþ1
88 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102

  1
C el el el
nþ1 ¼ ½v I C I nþ1 : B þ v 0 C 0 nþ1  : ½v I B þ v 0 I ; ð46Þ
n 1
o 1
with B ¼ I þ S : ½ðC el 0 nþ1 Þ : C el
I nþ1  I : ð47Þ

Note that C el
0 nþ1 is used in the Eshelby tensor of the elastic unloading step. The residual strain of the composite can be
calculated by

1
eres  unload ¼ enþ1  ðC el
nþ1 ¼ enþ1  Denþ1 nþ1 Þ :r
 nþ1 : ð48Þ
2. The residual strains in the fibers and in the matrix phases are computed following the M–T method, which
yields:

  1
I nþ1 ¼ eI nþ1  DeI nþ1 ¼ eI nþ1  B : ½v I B þ v 0 I
eres unload
: Deunload
nþ1 ; ð49Þ
 1
eres
0nþ1 ¼ e0 nþ1  De unload
0 nþ1 ¼ e0 nþ1  ½v I B þ v 0 I : Dunload
enþ1 : ð50Þ
3. The residual stresses in the fibers and matrix phases can be obtained, respectively, from

el
rres unload
I nþ1 ¼ rI nþ1  C I nþ1 : DeI nþ1 ; ð51Þ
el
r0 nþ1 ¼ r0 nþ1  C 0 nþ1 : Deunload
res
0 nþ1 : ð52Þ

From this detailed incremental-secant MFH process it is clear that the method can be implemented at the material law
level of a FE code. Moreover this homogenized material law calls successively the constitutive models of both the inclusions
and the matrix phases. The constitutive laws of these two materials are the usual elasto-plastic models at two exceptions:
the modification of the return mapping direction and the computation, as an output of the material model, of the secant ten-
 S . For an elastic phase, no modification is required. The incremental-secant MFH is thus simple to implement and is
sor C
computationally efficient as the phases constitutive material models remain classical and independent from each-other.
Note that considering a second-moment method involves deeper modifications of the material models of the different com-
posite material phases.

4. Comparison with direct finite element simulations, fast-Fourier transforms (FFT) methods or experimental results

In this section, the accuracy and the reliability of the proposed incremental-secant method are verified through the com-
parison of the effective response of several two-phase elasto-plastic composites. Reference results are provided either by di-
rect FE or FFT simulations on representative cells of the micro-structure, or by experimental data found in the literature. All
the MFH results presented have converged with the size of the strain increments.
First we study the difference of results obtained by considering the residual-incremental-secant method or the zero-
incremental-secant method for the elasto-plastic matrix phase of composites reinforced by elastic inclusions. Note that
for elastic phases the incremental-secant operator is the elastic operator, as the elastic phase has no need for a virtual elastic
property to be defined. It will be shown that considering the zero-incremental method in the elasto-plastic phase improves
the results accuracy. Then the method is shown to predict results as accurate as other MFH methods, including the incre-
mental-tangent method and the variational method, on a wide variety of composite materials made of an elasto-plastic ma-
trix reinforced by elastic inclusions. Finally the effect of the incremental-secant operator choice for composites made of two
elasto-plastic phases is studied.

4.1. Predictions based on the two different definitions of the incremental-secant moduli – C Sr and C S0 – for composite materials
made of an elasto-plastic matrix reinforced by elastic inclusions

In Section 3.1, we have proposed two different definitions of the incremental-secant moduli: C Sr and C S0 that can be used
to define the LCC. The effects of the different definitions on the predictions are studied herein through a few numerical exam-
ples, which are carried out on some composites with elastic inclusions.
In this case, since the inclusions are made of elastic materials, this phase has no need for a virtual elastic property to be
defined. We will thus study the effect of the operator choice for the matrix phase on several examples to confirm that the
zero-incremental-secant operator C S0 0 should be used for the latter.
Metal Matrix Composites (MMCs) with elastic inclusions and elastic-perfectly-plastic matrix. An elastic-perfectly-plastic
behavior for the matrix phase is considered (R0 ðpÞ ¼ 0) while the inclusions remain elastic. The other material properties are

 Inclusions: EI ¼ 400 GPa; and mI ¼ 0:2;

 Matrix: E0 ¼ 75 GPa; m0 ¼ 0:3; and rY0 ¼ 75 MPa.
 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102 89

Inclusions are spherical with a volume fraction of v I ¼ 15%. The predictions with the two incremental-secant moduli for the
matrix phase, together with FE reference results provided by Brassart et al. (2011), are presented in Fig. 4, which shows that
the zero-incremental-secant method should be used for the matrix to agree with the FE results.
MMCs with elastic inclusions and elasto-plastic matrix. The example consists of an aluminum alloy matrix reinforced with
continuous stiff alumina fibers, whose material properties (Doghri and Friebel, 2005) are

 Inclusions: EI ¼ 344:5 GPa; and mI ¼ 0:26;

 Matrix: E0 ¼ 68:9 GPa; m0 ¼ 0:32; rY0 ¼ 94 MPa; k0 ¼ 578:25 MPa, and m0 ¼ 0:529.

The matrix material follows the hardening law

m
RðpÞ ¼ kp ; ð53Þ
where k and m are the hardening parameters. The volume fraction of the continuous fibers is v I ¼ 55%. The FE predictions
were obtained by Jansson (1992) on a unit cell assuming a periodic microstructure with a hexagonal arrangement. The MFH
results for the two incremental-secant moduli for the matrix phase are presented in Fig. 5(a) for a transerve loading, and in
Fig. 5(b) for a longitudinal loading.
For the transverse loading, see Fig. 5(a), stiffer results are obtained when using C Sr
0 , while good predictions are obtained
when using C S0 0 . Good predictions are observed with both operators for the longitudinal loading, see Fig. 5(b).
The reason why using the zero-incremental-secant formulation for the elasto-plastic phase improves the results accuracy
is the following. As only the matrix phase exhibits a plastic flow, upon the virtual unloading the residual stress in the matrix
is negative (assuming the composite material was under tension and vice versa), while the residual stress in the inclusions
phase is positive. Using the zero-incremental-secant operator, meaning neglecting the residual stress, in the matrix induces
thus more plasticity as the predictor is higher, which counterbalances the over-stiff results predicted by a first-moment
method as the one proposed herein. We think that considering a second-moment approach would solve this issue.
We will show in the next section that for problems with elastic inclusions, using the zero-incremental-secant approach
C S0
0 for the elasto-plastic matrix phase leads to a good accuracy for different materials, inclusion geometries and loading path.

4.2. Predictions for elastic inclusions and with the zero-incremental-secant operator C S0
0 for the elasto-plastic matrix

In this section, the considered composites consist in an elasto-plastic matrix reinforced by ellipsoidal or continuous UD
inclusions with linear elastic behaviors. The inclusions remain elastic and the matrix phase obeys to the zero-incremental-
secant formulation. The predictions of the proposed model are compared to references provided either by direct FE or FFT
simulations on representative cells of the micro-structure or by experimental data gathered from literature.

4.2.1. Responses for short and UD fibers reinforced matrix

The method is here applied to composites with short and UD fibers. In particular, the response in the phases is compared
to reference results.
GFRP: Short glass fibers reinforced polyamide. In this case, the polyamide matrix follows an exponential-linear hardening
law
RðpÞ ¼ k1 p þ k2 ð1  emp Þ ð54Þ
and the material properties are

 Inclusions: EI ¼ 72 GPa; and mI ¼ 0:22;

 Matrix: E0 ¼ 2:1 GPa; m0 ¼ 0:3; rY0 ¼ 29 MPa; k1 0 ¼ 139:0 MPa; k20 ¼ 32:7 MPa, and m0 ¼ 319:4.

Fig. 4. Results for composites with elastic inclusions and elastic-perfectly-plastic matrix.
90 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102

Fig. 5. Comparisons of the secant-incremental methods for Metal Matrix Composites with elastic fibers embedded in an elasto-plastic matrix.

The glass fibers volume fraction is v I ¼ 15:7% and their aspect ratio is a ¼ 15. Aligned fibers are considered and the FE sim-
ulations were performed by Doghri et al. (2011). These FE results are used as reference. Fig. 6(a) shows the macroscopic pre-
dictions of the two MFH formulations under longitudinal (along the fibers direction) uni-axial tension, and Fig. 6(b) shows
the ones under transverse uni-axial tension. Results reported by Doghri et al. (2011) using first-moment and second-moment
incremental-tangent methods are also reported for comparison purposes. On the one hand, for a transverse loading all the
MFH methods predict accurate results, see Fig. 6(b). On the other hand, for a longitudinal loading, although the second-
moment incremental-tangent method predicts a response less stiff than the first-moment schemes, the new scheme exhibits
a better accuracy when compared to the first-moment incremental-tangent method, see Fig. 6(a). When analyzing the aver-
age von Mises stress predicted in the matrix phase,5 the predictions are really accurate for a transverse loading, see Fig. 6(d).
For a loading aligned in the fibers direction, see Fig. 6(c), the newly developed method is as accurate as the results obtained with
the second-moment incremental-tangent method. This behavior is confirmed by the evolution of the average accumulated plas-
tic strain in the matrix phase as shown by Fig. 6(e) and (f).
Continuous UD-fibers reinforced elasto-plastic matrix. The accuracy of the proposed incremental-secant method is assessed
through the comparison with direct FE simulations on periodical cells of a continuous UD fibers reinforced elasto-plastic ma-
trix. This consists of continuous elastic isotropic fibers embedded in a matrix material, which follows a classical J2-elasto-
plastic behavior model. The material parameters, taken from (Wu et al., 2012), are

 Inclusions: EI ¼ 238 GPa; and mI ¼ 0:26;

 Matrix: E0 ¼ 2:89 GPa; m0 ¼ 0:3; rY0 ¼ 35 MPa; k0 ¼ 73 MPa; and m0 ¼ 60.

The matrix material behavior follows the hardening law

RðpÞ ¼ kð1  emp Þ: ð55Þ

The volume fraction of the continuous fibers is v I ¼ 50%. The test consists of a transverse loading of the composite material,
followed by a complete unloading until reaching a zero-strain state. In the direction of the fibers plane-strain conditions are
assumed, while plane-stress conditions are applied in the other transverse direction.
Fig. 7(a) compares the macro-stress evolution obtained with the new incremental-secant MFH and the incremental-tan-
gent MFH to the FE results. During the loading part, the incremental-secant method avoids the over-stiff prediction exhibited
by the incremental-tangent method. This is confirmed when analyzing the evolution of the average stress in the inclusions
phase, see Fig. 7(b), and in the matrix phase, see Fig. 7(c). However, during the unloading part, the stress is slightly lower
than the one predicted by the FE simulation and by the incremental-tangent method. This last method exhibits less error
as it starts from a higher stress at the unloading point, so this compensates for the error due to an over-stiff response.
Fig. 7(d) compares the average value of the effective equivalent von Mises stress in the matrix and Fig. 7(e) illustrates the
accumulated plastic strain evolution in the matrix. Both incremental methods are found to give similar results.

4.2.2. Metal Matrix Composites (MMCs)

Now, the proposed incremental-secant procedure is applied on various Metal Matrix Composites (MMCs), which are
made of an elasto-plastic matrix reinforced by elastic inclusions. The matrix behavior follows the power-law hardening (53).
MMC #1: an aluminum alloy reinforced with stiff ceramic particles I. The properties of the considered material are

 Inclusions: EI ¼ 400 GPa; and mI ¼ 0:2;

 Matrix: E0 ¼ 75 GPa; m0 ¼ 0:3; rY0 ¼ 75 MPa; k0 ¼ 416 MPa, and m0 ¼ 0:3895.

5
For the second-moment method, both the average of the equivalent von Mises stress, and the von Mises stress computed from the average stress tensor in
the matrix are reported.
 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102 91

Fig. 6. Results for the short glass fibers reinforced polyamide test for (a, c, e) a longitudinal traction and for (b, d, f) a transverse traction. (a) and (b)
Composite material response, (c) and (d) evolution of the average of the von Mises stress in the matrix phase, (e) and (f) evolution of the accumulated plastic
strain in the matrix phase.

Inclusions are spherical with a volume fraction of v I ¼ 30%. The macroscopic prediction obtained with the incremental-
secant and the first order incremental-tangent formulations are presented in Fig. 8(a) together with FE predictions, obtained
on a random arrangement of 30 inclusions, reported by Segurado et al. (2002).
MMC #2: an aluminum alloy reinforced with stiff ceramic particles II. The same material system as in case MMC#1 is con-
sidered here again, except for the hardening law. The initial yield stress rY0 is set equal to zero for the computation and the
hardening exponents m0 ¼ 0:05 and m0 ¼ 0:4 of the matrix material are considered successively. The inclusions are aligned
ellipsoids with an aspect ratio a ¼ 3 and the volume fraction of the inclusions is v I ¼ 25%. These material properties corre-
spond to the composites studied by Pierard et al. (2007a), who provided the FE results on unit cells containing 30 inclusions.
A uni-axial tension is carried out in the longitudinal direction of the ellipsoidal inclusions. The results of the two MFH for-
mulations together with the FE results provided by Pierard et al. (2007a) are presented in Fig. 8(b) for m0 ¼ 0:05, and in
Fig. 8(c) for m0 ¼ 0:4, respectively.
MMC #3: an aluminum alloy reinforced with SiC whiskers. The properties of the considered material are

 Inclusions: EI ¼ 450 GPa; and mI ¼ 0:17;

 Matrix: E0 ¼ 68:89 GPa; m0 ¼ 0:33; rY0 ¼ 277:3 MPa; k0 ¼ 592:2 MPa, and m0 ¼ 0:52.

The elastic properties of the whiskers were reported by Christman et al. (1989a) and the properties of the matrix were fitted
by Doghri and Friebel (2005) on the experimental data reported by Christman et al. (1989b). The volume fraction of the
inclusions is v I ¼ 13:2%. The whiskers are assumed to be cylindrical with the length-diameter ratio l=d of 5. For the MFH
computations, the aspect ratio a of the equivalent ellipsoid is obtained by multiplying l=d with a factor of 1.25 as proposed
by Li and Ponte Castañeda (1994) and by Doghri and Friebel (2005).
The experimental data for a uni-axial tension applied along the longitudinal direction of the whiskers are reported by
Christman et al. (1989a,b) and the stress-strain curves are presented in Fig. 8(d) together with the MFH results. This figure
also represents the curve-fitting of the matrix material considered.
92 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102

Fig. 7. Results for continuous-elastic fibers embedded in an elasto-plastic matrix under transverse tension-compression.

MMC #4: a two-phase steel made of ellipsoidal martensitic inclusions embedded in a ferritic matrix. The properties of the two
phases are

 Inclusions: EI ¼ 200 GPa; and mI ¼ 0:3;

 Matrix: E0 ¼ 220 GPa; m0 ¼ 0:3; rY0 ¼ 300 MPa; k0 ¼ 1130 MPa, and m0 ¼ 0:31.

The volume fraction of the inclusions is v I ¼ 25%. The inclusions are successively considered as spherical, a ¼ 1, and as ellip-
soidal with a ¼ 3 and the results are reported in Fig. 8(e) and in Fig. 8(f) respectively. These figures show the comparison of
the two MFH predictions with the FE results, which reference results were obtained by Brassart et al. (2010) on multi-par-
ticle cells.
Fig. 8 shows that the proposed MFH method predicts responses slightly more compliant than with the incremental-tan-
gent method. However the accuracy can be said to be of the same order.

4.2.3. Effect of the triaxiality

In this section, more general loading conditions are applied to test the reliability of the proposed method. We applied a
pure shear loading, a biaxial loading and a plane strain tension/compression successively on a MMC to investigate the effect
of triaxiality on the accuracy of the predictions. The considered composite is a SiC-particles reinforced aluminum matrix
(MMC #5). The elasto-plastic metal matrix follows the power-law hardening (53) and the material properties are

 Inclusions: EI ¼ 400 GPa; and mI ¼ 0:2;

 Matrix: E0 ¼ 75 GPa; m0 ¼ 0:3; rY0 ¼ 75 MPa; k0 ¼ 400 MPa, and m0 ¼ 0:4 or m0 ¼ 0:05.
 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102 93

Fig. 8. Results for the various metal matrix composites tests.

The volume fraction of the spherical inclusions is v I ¼ 15%. The predictions of the incremental secant formulation are pre-
sented in Fig. 9. The reference macro-responses are direct finite element results presented by Brassart et al. (2012), which are
obtained on periodic cells containing 35 randomly dispersed inclusions. The mean field homogenization predictions
obtained by the first-moment incremental-tangent method and by the variational method are also presented in Fig. 9 for
comparison.
For this problem, three different loading conditions are successively applied. First shear loading, with r  12 ¼ r
 21 ¼ r and
the other components of the stress tensor r  being zero, is considered. Then, a biaxial loading is obtained by setting
r 11 ¼ r 22 ¼ r and r 33 ¼ 0. Finally, in the case of plane strain tension/compression, the only non-zero components of e are
e11 and e22 , with e22 computed to satisfy r  22 ¼ 0. Moreover e33 is set to zero, and r
 33 can be measured.
These different loading conditions correspond to different triaxiality states. Indeed, as the macroscopic triaxiality ratio is
defined by T ¼ trðr  Þ=3r eq , one can directly find that shear and biaxial loading conditions correspond to triaxiality ratios of 0
and 2/3 respectively. The uniaxial tension, which has been applied a lot in the previous examples, has a triaxiality ratio of 1/
3. For plane strain tension/compression, the triaxiality ratio is approximately 1 in the plastic regime.
In Fig, 9, the predictions of the proposed method shows an accuracy comparable to the other MFH approaches, for the
different triaxiality states. Moreover, the effect of the strain increment size on the results predicted by the new incremen-
tal-secant method has been studied in Fig. 9(e) and (f) for the last test with the hardening exponent m0 ¼ 0:4. It can be seen
that the results have already converged for a macro-strain increment of De ¼ 0:0001.

4.2.4. Non-monotonic, non-proportional loading path

Eventually the behavior of the incremental-secant MFH is tested with a non-monotonic, non-proportional loading path.
This example was proposed by Lahellec and Suquet (2013) and consists of elastic spherical inclusions embedded in an elas-
tic-perfectly-plastic matrix, with the following material properties:
94 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102

Fig. 9. Results for MMC #5 under different loading conditions.

I ¼ 20 GPa; and lI ¼ 6 GPa;

 Inclusions: jel el

 Matrix: j0 ¼ 10 GPa; l0 ¼ 3 GPa; and rY0 ¼ 100 MPa.

el el

The volume fraction of the spherical inclusions is v I ¼ 17%. The external boundary conditions correspond to constraining
simultaneously all the strain components following

1
eðt Þ ¼ e33 ðtÞ e3  e3  ðe1  e1 þ e2  e2 Þ þ e13 ðtÞðe1  e3 þ e3  e1 þ e2  e3 þ e3  e2 Þ; ð56Þ
2

where the non-monotonic, non-proportional evolution of the two constraints with respect to a fictitious time is illustrated in
Fig. 10a.
This problem was solved using a Fast Fourier Transforms (FFT)-based numerical method and the variational MFH by
Lahellec and Suquet (2013), and the resulting tensile and shear response components are reported in Fig. 10. The new incre-
mental-secant and the incremental-tangent MFH schemes are then applied. The results are illustrated in Fig. 10(b)–(d). It can
be seen that in this test the incremental-secant formulation has an accuracy comparable to the variational method, while the
incremental-tangent method is unable to capture the changes in the of loading direction.
 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102 95

4.3. Elasto-plastic inclusions reinforced elasto-plastic matrix

We have shown that, for composites made of elasto-plastic matrix reinforced by elastic inclusions, considering the zero-
incremental-secant approach in the matrix phase instead of the residual-incremental-secant approach improves the accu-
racy of the results as it counterbalances the over-stiff prediction inherent to a first-moment method.
In this section the case of elasto-plastic inclusions reinforced elasto-plastic matrix composites is considered. For a com-
posite material made of two elasto-plastic phases, four cases according to the combinations of the different incremental-se-
cant moduli for the inclusions and matrix phases are successively considered:

1. C Sr Sr
I and C 0 ,
2. C I and C S0
S0
0 ,
3. C Sr S0
I and C 0 , and
4. C S0 Sr
I and C 0 .

Metal Matrix Composites (MMCs) with power-law hardening in both phases. In this example studied by Brassart et al. (2011),
both matrix and inclusions phases obey an elasto-plastic behavior with a power-law hardening (53). The material properties
for the inclusions and matrix are

 Inclusions: EI ¼ 400 GPa; mI ¼ 0:2; rYI ¼ 75 MPa; kI ¼ 1:0 GPa; and mI ¼ 0:4 or mI ¼ 0:05;
 Matrix: E0 ¼ 75 GPa; m0 ¼ 0:3; rY0 ¼ 75 MPa; k0 ¼ 400 MPa, and m0 ¼ 0:4 or m0 ¼ 0:05.

Inclusions are spherical with a volume fraction of v I ¼ 15%. Uni-axial tension tests are performed, and the macroscopic
responses are predicted for the four different incremental-secant moduli cases with the newly proposed MFH formulation.
The reference results are obtained from the FE simulations reported in Section 6.2 of Brassart et al. (2011) for periodic
composites.
First, in the case of comparable hardening for the matrix and inclusions materials, m0 ¼ mI ¼ 0:05 in Fig. 11(a) and
m0 ¼ mI ¼ 0:4 in Fig. 11(b), the residual stresses remain small in both phases and satisfactory results are obtained for the
four combinations although for m0 ¼ mI ¼ 0:05, see Fig. 11(a), using the residual-incremental-secant operator in both matrix
and inclusions phases overestimates the results slightly more than for the other combinations.

Fig. 10. Results for a non-monotonic, non-proportional loading path. (a) Applied strain components history. (b) Comparisons of the predicted stress
components. (c) Comparison of the predicted stress components history.
96 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102

m I = 0.05

6
m 0 = 0.05

4
σ/σY0 m 0 = 0.4

2 FE (Brassart et al., 2011)

C ISr, C 0Sr
C IS0, C 0S0
C ISr, C 0S0
C IS0, C 0Sr
0
0 0.02 0.04
ε

Fig. 11. Results for the elasto-plastic inclusions – elasto-plastic matrix tests.

Second, in the case of a more severe hardening in the inclusions phase, m0 ¼ 0:4 and mI ¼ 0:05 in Fig. 11(a), the overes-
timation of the predictions when considering the residual-incremental-secant operator in both matrix and inclusions phases
becomes unacceptable. The three other combinations give good predictions.
Third, in the case of a more severe hardening in the matrix phase, m0 ¼ 0:05 and mI ¼ 0:4 in Fig. 11(b), only the solution
obtained when using the residual-incremental-secant method in both phases captures the FE solution with a high accuracy.
In fact, the choice of the incremental-secant moduli to be considered corresponds to an assumption behind the proposed
MFH formulation when defining the LCC. The residual-incremental-secant method should always be considered for the
inclusions phase. Concerning the elasto-plastic matrix phase, the choice of keeping or not the residual stresses is governed
by the relative positions of the stress and residual stress tensors with respect to the stress space origin (zero-stress state):

 The origin lies between the stress and residual stress tensors. When considering a 1D problem in tension, this means that
the stress is positive and the residual stress is negative. In that case, the zero-incremental-secant operator C S0 should be
used in the matrix phase, for the same reason as for elastic inclusions. This is typically the case for composite materials
made of two elasto-plastic phases for which the inclusions phase is stiffer than the matrix phase during the plastic flow,
which is a common case for composite materials.
 Both stress and residual stress tensors lie on the same side with respect to the origin. When consider a 1D problem in
tension, this means that both the stress and the residual stress are positive. In that case the residual-incremental-secant
method should be considered for the matrix phase as using the zero-incremental-secant method would lead to softer pre-
diction. Indeed as illustrated in Fig. 12 for the limiting case of a pure elastic response, the reloading path does not follow
the unloading one if the residual stress is omitted, resulting in a lower stress value. This is typically a configuration of
compliant elasto-plastic inclusions combined with a stiffer matrix response during the plastic flow.

We will now assert this analysis on more examples.

Fig. 12. Configuration in an elastic matrix with the zero-incremental-secant formulation.

 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102 97

Fig. 13. Results for composites with elastic-perfectly-plastic inclusions with a low hardening embedded in an elasto-plastic matrix. (a) Effect of the
incremental-secant operators, De ¼ 0:00001. (b) Residual-incremental-secant operators, effect of the strain increment size D
e.

MMC with elastic-perfectly-plastic inclusions phase. In the first additional example, elastic-perfectly-plastic behaviors for
the inclusions and matrix phases are considered (RI ðpÞ ¼ R0 ðpÞ ¼ 0). The other material properties are the same as for the
previous example, i.e.

 Inclusions: EI ¼ 400GPa;mI ¼ 0:2; and rYI ¼ 75MPa;

 Matrix: E0 ¼ 75GPa;m0 ¼ 0:3; and rY0 ¼ 75MPa.

Inclusions are spherical with a volume fraction of v I ¼ 15%. The predictions with the four combinations of the incremental-
secant moduli together with FE reference results, which are found in Brassart et al. (2011), are presented in Fig. 13(a), which
shows that only the prediction obtained by using C Sr in both phases agrees with the FE results. Moreover, when considering
the residual-incremental-secant operators, the results are found to be quasi-independent of the macro-strain increment size,
see Fig. 13(b).
MMC with low inclusions phase hardening. Another example of composite material made of two elastoplastic phases with a low
hardening inclusion phase was studied by Doghri and Friebel (2005). In this case, the spherical austenite inclusions are more com-
pliant than the ferrite matrix, see Fig. 14. The material properties of both phases, which follow the power-law hardening (53), are

 Inclusions: EI ¼ 179:35 GPa; mI ¼ 0:3; rYI ¼ 202 MPa; kI ¼ 688 MPa, and mI ¼ 0:55;
 Matrix: E0 ¼ 196:85 GPa; m0 ¼ 0:3; rY0 ¼ 600 MPa; k0 ¼ 650 MPa, and m0 ¼ 0:06.

The inclusions volume fraction is v I ¼ 35%. One more time, the solution obtained by using C Sr in both phases agrees with the
FE results provided by Doghri and Friebel (2005), see Fig. 14.

Fig. 14. Results for composites with elasto-plastic inclusions with a low hardening embedded in an elasto-plastic matrix.
98 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102

From the examples above, we can say that good estimations can be obtained by using C Sr Sr
I and C 0 in the proposed MFH
formulation when both phases are elasto-plastic materials and if, during the plastic flow, the inclusions response is not stiffer
than the matrix phase. In other cases, such a combination is too stiff and the zero-incremental-secant operator C S0
0 should be
considered in the matrix phase. However as previously said, it is believed that formulating the framework by considering
second-moment-statistical values will allow considering the residual stress in both phases for any cases.

5. Conclusions

In this work, a new incremental-secant MFH process for composites made of elasto-plastic constituents was proposed. In
this approach, an unloading of the composite material is virtually performed to estimate the residual strains in each phase
before applying a secant approach on the strain increments, which differ in each phase. In order to define the LCC two secant
operators were defined. The first one, the residual-incremental-secant operator, is defined from the phase residual stress.
The second operator, the zero-incremental-secant operator, is defined from a stress-free state in the phase.
The method was then applied on several problems and was compared to other existing MFH methods and to direct FE or
FFT simulations. These examples showed that for composites with inclusions hardening law exhibiting a stiffness lower than
or of the same order as the one of the matrix material, using the residual-incremental-secant operator for both phases leads
to accurate predictions. However, for composites whose elasto-plastic inclusions phase is much stiffer during the plastic flow
than the elasto-plastic matrix material response and for elastic inclusions embedded in an elasto-plastic matrix, the zero-
incremental-secant operator should be used in the matrix phase to avoid over-stiff predictions. As discussed in the paper,
the rigorous criterion is actually based on the relative positions of the stress and residual stress tensors of the matrix phase
with respect to the stress space origin. In case they lie on different sides of the stress space origin the zero-incremental-se-
cant operator should be used in the matrix phase.
With this restriction on the choice of the matrix operator, the method has been shown to predict the macro-stress with an
accuracy level similar to, or better than, the one of the first order incremental-tangent MFH method. In particular, for short
glass fibers reinforced polyamide, the new incremental-secant method has a degree of accuracy higher than the one reached
with the first order incremental-tangent method. The incremental-secant method can also capture the solution under non-
monotonic non-proportional loading, contrarily to the incremental-tangent approach.
The advantages of the method lie in the simple formulation and implementation. Moreover, as the LCC is defined from
secant operators which are naturally isotropic, the method does not require ad-hoc isotropisation of these operators when
computing the Eshelby or concentration tensors needed in the MFH.
In the near future, the method will be extended to the damage case. When considering a composite material whose ma-
trix phase exhibits a damaging process, the inclusions phase can be unloaded during the softening stage of the matrix. The
incremental-secant approach will allow this complex behavior to be captured, ensuring a more accurate prediction of the
scheme as compared to the incremental-tangent method.
Finally, based on the good accuracy obtained by this first-moment method, the method opens perspective for its exten-
sion to a second-moment form.

Acknowledgments

The research has been funded by the Walloon Region under the agreement SIMUCOMP No 1017232 (CT-EUC 2010-10-12)
in the context of the ERA-NET +, Matera + framework.

Appendix A. Tensorial operations and notations

 Dots and colons are used to indicate tensor products contracted over one and two indices respectively:
u  v ¼ ui v i ; ða  uÞi ¼ aij uj ;
ða  bÞij ¼ aik bkj ; a : b ¼ aij bji ;
ðC : aÞij ¼ C ijkl alk ; ðC : DÞijkl ¼ C ijmn Dnmkl : ðA:1Þ

 Dyadic products are designated by :

ðu  v Þij ¼ ui v j ; ða  bÞijkl ¼ aij bkl : ðA:2Þ

 Symbols 1 and I designate the second- and fourth-order symmetric identity tensors respectively:
1
1ij ¼ dij ; I ijkl ¼ ðdik djl þ dil djk Þ; ðA:3Þ
2
where dij ¼ 1 if i ¼ j; dij ¼ 0 if i – j.
 The spherical and deviatoric operators I vol and I dev are respectively
 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102 99

1
I vol  1  1; I dev ¼ I  I vol ; ðA:4Þ
3
so that for symmetric tensors aij ¼ aji we have:
1 1
I vol : a ¼ amm 1; I dev : a ¼ a  amm 1 ¼ devðaÞ: ðA:5Þ
3 3

Appendix B. Derivation of the closed-form expressions for the incremental-secant method

B.1. Residual-incremental-secant operator C Sr

@C Sr
The evaluation of @e
, which is used in the MFH scheme, follows from

@C Sr @ @ Der @ lrs
¼ ð3jr I vol þ 2lrs I dev Þ : ¼ 2I dev  : ðB:1Þ
@e @ Der @e @ Der
@ Dreq eq
Using @ Drr
¼ 32 Ds
Dreq
; @ Ds
@ Drr
¼ I dev ; @@DDeer ¼ 23 De
Deeq
and Eq. (27), this last relation becomes
" #
Sr
@C dev 1 alg 2 r De
¼ 2I  Ds : C  l s ; ðB:2Þ
@e 6lrs ðDeeq Þ2 3 ðDeeq Þ2

where C alg is the derivative of the stress increment with respect to the strain increment, which is obtained from the consti-
tutive law of the material. Due to the modification of the return mapping algorithm, this expression is slightly changed com-
pared to the usual one and reads
 2  2  
2lel 2lel Dp 3 dev
C alg ¼ C el  NN  eq I  N  N ; ðB:3Þ
h rtrnþ1  rres
n
2
 dev
1
ðrnþ1 Þ
with h ¼ 13 N : 3
2 eq
@R
þ 3lel .
ðrnþ1 Þ @p

B.2. Zero-incremental-secant operator C S0

@C S0 @ req eq
The evaluation of @e
is obtained using @r
¼ 32 s @s
req ; @ r ¼ I dev ; @@DDeer ¼ 23 De
Deeq
, the relations

req
3l0s ¼ ; ðB:4Þ
Deeq
and j0 ¼ jel . This operator is readily obtained by
" #
@C S0 dev 1 alg 2 0 De
¼ 2I  s : C  ls : ðB:5Þ
@e 6l0s ðDeeq Þ2 3 ðDeeq Þ2

For this zero-incremental-secant approach, the direction of the normal corresponds strictly to the radial return mapping
assumption, and the classical expression of C alg is recovered:
2 2  
ð2lel Þ ð2lel Þ Dp 3 dev
C alg ¼ C el  N  N   tr eq I NN ; ðB:6Þ
h0 rnþ1 2

with h0 ¼ 3lel þ dR
dp
> 0.

Appendix C. Residual stress vector

The equation to be satisfied at the end of the MFH procedure is Eq. (35). Multiplying Eq. (33) by B ðI; C
S ; C
0
 S Þ and using Eq.
I
(35) lead to

v 0 DerI n þ 1 þ v I B ðI; C S0 ; C SI Þ : DerI n þ 1 ¼ B ðI; C S0 ; C SI Þ : Dernþ1 : ðC:1Þ

With the M–T assumption the strain concentration tensor follows from Eq. (4), and Eq. (C.1) reads
1
DerI n þ 1 þ v 0 S : ½ðC S0 Þ : C SI  I : DerI n þ 1 ¼ Dernþ1 ; ðC:2Þ
or again F ¼ 0 with
100 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102

1
F ¼ C S0 : ½DerI n þ 1  S 1 : ðDerI n þ 1  Dernþ1 Þ  C SI : DerI n þ 1: ðC:3Þ
v0
In order to satisfy F ¼ 0, Eq. (C.3) is linearized as
@F @F @F
dF ¼ : dDerI þ : dDer0 þ : dDer : ðC:4Þ
@ eI @ e0 @ e
When solving F ¼ 0 at constant Der , as v 0 Der0 n þ 1 þ v I DerI n þ 1 is also constant, the iteration process relies on dF ¼ J : deI
with

@F @F @ e0  S h i @ C S v I @ C S0 nþ1
J¼ þ : ¼ C 0 nþ1 : I  S 1  C SInþ1  I nþ1 : DerI nþ1 
@ eI @ e0 @ eI @ eI v 0 @ e0
  r 
DeI n þ 1  Dernþ1 vI @S vI
: DerI nþ1  S 1 :  2 C S0 nþ1  ðDerI n þ 1  Dernþ1 Þ :: ðS 1  S 1 Þ ::  C S : S 1 ; ðC:5Þ
v0 v0 @ e0 v 0 0 nþ1
S
where @@Cerr results from either (B.2) or (B.5). The derivative of the Eshelby tensor is reported in Appendix D.
Once F ¼ 0 is satisfied, the effect on the strain increment in each phase of a variation dDer can directly be obtained by
constraining dF ¼ 0, and Eq. (C.4) leads to6
@F @F @F
0¼ : dDerI þ : dDer0 þ : dDer ; ðC:6Þ
@ eI @ e0 @ e
or again
@ eI @F
¼ J 1 : : ðC:7Þ
@ e @ e
As under these circumstances der ¼ v 0 der0 þ v I derI , this last equation is completed by
@ e0 1 @ eI
¼ ðI  v 1 Þ: ðC:8Þ
@ e v0 @ e

Appendix D. Eshelby tensor and it is derivative

The derivative of Eshelby tensor can be written as

 
@S @S @m @j @ m @ ls
¼  þ : ðD:1Þ
@ Der @ m @ j @ Der @ ls @ Der

One directly has

@j
¼ 0; ðD:2Þ
@ Der
and therefore,
@S @S @ m @ ls
¼  : ðD:3Þ
@ Der @ m @ ls @ Der

References

Aboudi, J., Pindera, M.J., Arnold, S., 2003. Higher-order theory for periodic multiphase materials with inelastic phases. International Journal of Plasticity 19,
805–847.
Benveniste, Y., 1987. A new approach to the application of Mori–Tanaka’s theory in composite materials. Mechanics of Materials 6, 147–157.
Berveiller, M., Zaoui, A., 1978. An extension of the self-consistent scheme to plastically-flowing polycrystals. Journal of the Mechanics and Physics of Solids
26, 325–344.
Brassart, L., Doghri, I., Delannay, L., 2010. Homogenization of elasto-plastic composites coupled with a nonlinear finite element analysis of the equivalent
inclusion problem. International Journal of Solids and Structures 47, 716–729.
Brassart, L., Stainier, L., Doghri, I., Delannay, L., 2011. A variational formulation for the incremental homogenization of elasto-plastic composites. Journal of
the Mechanics and Physics of Solids 59, 2455–2475.
Brassart, L., Stainier, L., Doghri, I., Delannay, L., 2012. Homogenization of elasto-(visco) plastic composites based on an incremental variational principle.
International Journal of Plasticity 36, 86–112.
Carrere, N., Valle, R., Bretheau, T., Chaboche, J.L., 2004. Multiscale analysis of the transverse properties of ti-based matrix composites reinforced by sic fibres:
from the grain scale to the macroscopic scale. International Journal of Plasticity 20, 783–810.
Chaboche, J., Kanouté, P., Roos, A., 2005. On the capabilities of mean-field approaches for the description of plasticity in metal matrix composites.
International Journal of Plasticity 21, 1409–1434.

6
Note that the derivative with respect to Derr has the same expression as the derivative with respect to er .
 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102 101

Christman, T., Needleman, A., Nutt, S., Suresh, S., 1989a. On microstructural evolution and micromechanical modelling of deformation of a whisker-
reinforced metal–matrix composite. Materials Science and Engineering: A 107, 49–61 (Proceedings of the Symposium on Interfacial Phenomena in
Composites: Processing Characterization and Mechanical Properties).
Christman, T., Needleman, A., Suresh, S., 1989b. An experimental and numerical study of deformation in metal–ceramic composites. Acta Metallurgica 37,
3029–3050.
Doghri, I., Adam, L., Bilger, N., 2010. Mean-field homogenization of elasto-viscoplastic composites based on a general incrementally affine linearization
method. International Journal of Plasticity 26, 219–238.
Doghri, I., Brassart, L., Adam, L., Gérard, J.S., 2011. A second-moment incremental formulation for the mean-field homogenization of elasto-plastic
composites. International Journal of Plasticity 27, 352–371.
Doghri, I., Friebel, C., 2005. Effective elasto-plastic properties of inclusion-reinforced composites. Study of shape, orientation and cyclic response. Mechanics
of Materials 37, 45–68.
Doghri, I., Ouaar, A., 2003. Homogenization of two-phase elasto-plastic composite materials and structures: study of tangent operators, cyclic plasticity and
numerical algorithms. International Journal of Solids and Structures 40, 1681–1712.
Doghri, I., Tinel, L., 2005. Micromechanical modeling and computation of elasto-plastic materials reinforced with distributed-orientation fibers.
International Journal of Plasticity 21, 1919–1940.
Dvorak, G.J., 1992. Transformation field analysis of inelastic composite materials. Proceedings: Mathematical and Physical Sciences 437, 311–327.
Dvorak, G.J., Bahei-El-Din, Y.A., Wafa, A.M., 1994. Implementation of the transformation field analysis for inelastic composite materials. Computational
Mechanics 14, 201–228.
Engelen, R.A.B., Geers, M., Baaijens, F., 2003. Nonlocal implicit gradient-enhanced elasto-plasticity for the modelling of softening behaviour. International
Journal of Plasticity 19, 403–433.
Eshelby, J.D., 1957. The determination of the elastic field of an ellipsoidal inclusion, and related problems. Proceedings of the Royal Society of London Series
A, Mathematical and Physical Sciences 241, 376–396.
Geers, M., 1997. Experimental Analysis and Computational Modelling of Damage and Fracture. Ph.D. Thesis. University of Technology, Eindhoven (The
Netherlands).
Geers, M., Kouznetsova, V., Brekelmans, A., 2010. Multi-scale computational homogenization: trends and challenges. Journal of Computational and Applied
Mathematics 234, 2175–2182.
Hill, R., 1965a. Continuum micro-mechanics of elastoplastic polycrystals. Journal of the Mechanics and Physics of Solids 13, 89–101.
Hill, R., 1965b. A self-consistent mechanics of composite materials. Journal of the Mechanics and Physics of Solids 13, 213–222.
Jansson, S., 1992. Homogenized nonlinear constitutive properties and local stress concentrations for composites with periodic internal structure.
International Journal of Solids and Structures 29, 2181–2200.
Ji, B., Wang, T., 2003. Plastic constitutive behavior of short-fiber/particle reinforced composites. International Journal of Plasticity 19, 565–581.
Kanouté, P., Boso, D., Chaboche, J., Schrefler, B., 2009. Multiscale methods for composites: a review. Archives of Computational Methods in Engineering 16,
31–75. http://dx.doi.org/10.1007/s11831-008-9028-8.
Kouznetsova, V., Geers, M., Brekelmans, W., 2004. Multi-scale second-order computational homogenization of multi-phase materials: a nested finite
element solution strategy. Computer Methods in Applied Mechanics and Engineering 193, 5525–5550 (Advances in Computational Plasticity).
Kouznetsova, V., Geers, M.G.D., Brekelmans, W.A.M., 2002. Multi-scale constitutive modelling of heterogeneous materials with a gradient-enhanced
computational homogenization scheme. International Journal for Numerical Methods in Engineering 54, 1235–1260.
Kröner, E., 1958. Berechnung der elastischen konstanten des vielkristalls aus den konstanten des einkristalls. Zeitschrift für Physik A Hadrons and Nuclei
151, 504–518. http://dx.doi.org/10.1007/BF01337948.
Lahellec, N., Ponte Castañeda, P., Suquet, P., 2011. Variational estimates for the effective response and field statistics in thermoelastic composites with intra-
phase property fluctuations. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Science 467, 2224–2246.
Lahellec, N., Suquet, P., 2007a. On the effective behavior of nonlinear inelastic composites: I. Incremental variational principles. Journal of the Mechanics
and Physics of Solids 55, 1932–1963.
Lahellec, N., Suquet, P., 2007b. On the effective behavior of nonlinear inelastic composites: II. A second-order procedure. Journal of the Mechanics and
Physics of Solids 55, 1964–1992.
Lahellec, N., Suquet, P., 2013. Effective response and field statistics in elasto-plastic and elasto-viscoplastic composites under radial and non-radial loadings.
International Journal of Plasticity.
Li, G., Ponte Castañeda, P., 1994. Variational estimates for the elastoplastic response of particle-reinforced metal–matrix composites. Applied Mechanics
Reviews 47, 77.
Lissenden, C., Arnold, S., 1997. Theoretical and experimental considerations in representing macroscale flow/damage surfaces for metal matrix composites.
International Journal of Plasticity 13, 327–358.
Masson, R., Bornert, M., Suquet, P., Zaoui, A., 2000. An affine formulation for the prediction of the effective properties of nonlinear composites and
polycrystals. Journal of the Mechanics and Physics of Solids 48, 1203–1227.
Mercier, S., Molinari, A., 2009. Homogenization of elasticviscoplastic heterogeneous materials: self-consistent and Mori–Tanaka schemes. International
Journal of Plasticity 25, 1024–1048.
Molinari, A., Canova, G., Ahzi, S., 1987. A self consistent approach of the large deformation polycrystal viscoplasticity. Acta Metallurgica 35, 2983–2994.
Molinari, A., El Houdaigui, F., Tóth, L., 2004. Validation of the tangent formulation for the solution of the non-linear Eshelby inclusion problem. International
Journal of Plasticity 20, 291–307.
Mori, T., Tanaka, K., 1973. Average stress in matrix and average elastic energy of materials with misfitting inclusions. Acta Metallurgica 21, 571–574 (Cited
by (since 1996) 1814).
Moulinec, H., Suquet, P., 2003. Intraphase strain heterogeneity in nonlinear composites: a computational approach. European Journal of Mechanics –A/
Solids 22, 751–770.
Peerlings, R., de Borst, R., Brekelmans, W., Ayyapureddi, S., 1996. Gradient-enhanced damage for quasi-brittle materials. International Journal for Numerical
Methods in Engineering 39, 3391–3403.
Peerlings, R., de Borst, R., Brekelmans, W., Geers, M., 1998. Gradient-enhanced damage modelling of concrete fracture. Mechanics of Cohesive-Frictional
Materials 3, 323–342.
Peerlings, R., Geers, M., de Borst, R., Brekelmans, W., 2001. A critical comparison of nonlocal and gradient-enhanced softening continua. International
Journal of Solids and Structures 38, 7723–7746.
Pettermann, H.E., Plankensteiner, A.F., Böhm, H.J., Rammerstorfer, F.G., 1999. A thermo-elasto-plastic constitutive law for inhomogeneous materials based
on an incremental Mori–anaka approach. Computers & Structures 71, 197–214.
Pierard, O., Doghri, I., 2006a. An enhanced affine formulation and the corresponding numerical algorithms for the mean-field homogenization of elasto-
viscoplastic composites. International Journal of Plasticity 22, 131–157.
Pierard, O., Doghri, I., 2006b. Study of various estimates of the macroscopic tangent operator in the incremental homogenization of elastoplastic composites.
International Journal for Multiscale Computational Engineering 4, 521–543.
Pierard, O., Gonzlez, C., Segurado, J., LLorca, J., Doghri, I., 2007a. Micromechanics of elasto-plastic materials reinforced with ellipsoidal inclusions.
International Journal of Solids and Structures 44, 6945–6962.
Pierard, O., LLorca, J., Segurado, J., Doghri, I., 2007b. Micromechanics of particle-reinforced elasto-viscoplastic composites: finite element simulations versus
affine homogenization. International Journal of Plasticity 23, 1041–1060.
Ponte Castañeda, P., 1991. The effective mechanical properties of nonlinear isotropic composites. Journal of the Mechanics and Physics of Solids 39, 45–71.
102 L. Wu et al. / International Journal of Plasticity 51 (2013) 80–102

Ponte Castañeda, P., 1992. A new variational principle and its application to nonlinear heterogeneous systems. SIAM Journal on Applied Mathematics 52,
1321–1341.
Ponte Castañeda, P., 1996. Exact second-order estimates for the effective mechanical properties of nonlinear composite materials. Journal of the Mechanics
and Physics of Solids 44, 827–862.
Ponte Castañeda, P., 2002a. Second-order homogenization estimates for nonlinear composites incorporating field fluctuations: I – Theory. Journal of the
Mechanics and Physics of Solids 50, 737–757.
Ponte Castañeda, P., 2002b. Second-order homogenization estimates for nonlinear composites incorporating field fluctuations: II – Applications. Journal of
the Mechanics and Physics of Solids 50, 759–782.
Segurado, J., Llorca, J., 2002. A numerical approximation to the elastic properties of sphere-reinforced composites. Journal of the Mechanics and Physics of
Solids 50, 2107–2121.
Segurado, J., Llorca, J., González, C., 2002. On the accuracy of mean-field approaches to simulate the plastic deformation of composites. Scripta Materialia 46,
525–529.
Suquet, P., 1995. Overall properties of nonlinear composites: a modified secant moduli theory and its link with Ponte Castañeda’s nonlinear variational
procedure. Comptes Rendus de l Académie des Sciences 320, 563–571.
Talbot, D., Willis, J., 1992. Some simple explicit bounds for the overall behaviour of nonlinear composites. International Journal of Solids and Structures 29,
1981–1987.
Talbot, D.R.S., Willis, J.R., 1985. Variational principles for inhomogeneous non-linear media. IMA Journal of Applied Mathematics 35, 39–54.
Talbot, D.R.S., Willis, J.R., 1987. Bounds and self-consistent estimates for the overall properties of nonlinear composites. IMA Journal of Applied Mathematics
39, 215–240.
Wieckowski, Z., 2000. Dual finite element methods in homogenization for elasticplastic fibrous composite material. International Journal of Plasticity 16,
199–221.
Wu, L., Noels, L., Adam, L., Doghri, I., 2012. Multiscale mean-field homogenization method for fiber-reinforced composites with gradient-enhanced damage
model. Computer Methods in Applied Mechanics and Engineering 233–236, 164–179.
Zaoui, A., Masson, R., 2002. Modelling stress-dependent transformation strains of heterogeneous materials. In: Bahei-El-Din, Y.A., Dvorak, G.J., Gladwell,
G.M.L. (Eds.), IUTAM Symposium on Transformation Problems in Composite and Active Materials, Solid Mechanics and Its Applications, vol. 60, pp. 3–
15.