Vous êtes sur la page 1sur 12

Acta Materialia 55 (2007) 21372148

www.actamat-journals.com

Modeling texture, twinning and hardening evolution


during deformation of hexagonal materials
G. Proust, C.N. Tome *, G.C. Kaschner
MST-8, MS G755, Los Alamos National Laboratory, Los Alamos, NM 87545, USA
Received 20 June 2006; received in revised form 7 November 2006; accepted 8 November 2006
Available online 22 January 2007

Abstract
Hexagonal materials deform plastically by activating diverse slip and twinning modes. The activation of such modes depends on their
relative critical stresses, and the orientation of the crystals with respect to the loading direction. For a constitutive description of these
materials to be reliable, it has to account for texture evolution associated with twin reorientation, and for the eect of the twin barriers on
dislocation propagation and on the stressstrain response. In this work, we introduce a model for twinning, which accounts explicitly for
the composite character of the grain, formed by a matrix with embedded twin lamellae which evolve with deformation. Texture evolution
takes place through reorientation due to slip and twinning. The role of the twins as barriers to dislocations is explicitly incorporated into
the hardening description via geometrically necessary dislocations and a directional HallPetch mechanism. We apply this model to the
interpretation of compression experiments, both monotonic and changing the loading direction, done in rolled Zr at 76 K.
Published by Elsevier Ltd on behalf of Acta Materialia Inc.
Keywords: Twinning; Polycrystal model; Hardening; Hexagonal materials

1. Introduction
Low symmetry aggregates exhibit several crystallographic slip and twinning modes with widely dierent activation stresses and, as a consequence, very anisotropic
plastic properties. Twinning activity, in particular, plays
two important roles: it has a marked eect on texture
evolution and it strongly aects the hardening response
of low-symmetry aggregates. The former eect is due to
the crystallographic reorientation associated with the
twinned portion of the grains, while the latter is associated
with the barriers that the twin lamellae pose to the propagation of dislocations. As a consequence, for a constitutive
plastic description of low-symmetry aggregates to be general and reliable, it needs to be based on crystallography.
Earlier polycrystal models incorporating twinning were
mostly concerned with capturing the texture evolution

Corresponding author. Tel.: +1 505 665 0892; fax: +1 505 667 8021.
E-mail address: tome@lanl.gov (C.N. Tome).

1359-6454/$30.00 Published by Elsevier Ltd on behalf of Acta Materialia Inc.


doi:10.1016/j.actamat.2006.11.017

associated with twinning. In 1978, Van Houtte [1] proposed


a Monte Carlo approach consisting of randomly sampling and reorienting full grains by twinning. In 1991,
Tome et al. [2] proposed a variation of such a model, called
the predominant twin reorientation (PTR) scheme, consisting of reorienting the grain using the most active twin system in it. In 1993, Lebensohn and Tome [3] proposed a
volume fraction transfer (VFT) scheme and implemented
it in a visco-plastic self-consistent (VPSC) code. Within
the VFT scheme, the Euler space is partitioned in cells
and, instead of keeping the grain volume fraction constant
while changing its orientation, as done in PTR, the volume
fraction assigned to each cell evolves with deformation.
The VFT approach permits one to account for the contribution to texture of every twin system in every grain by
transferring volume fractions from one cell to another in
Euler space. A disadvantage of the VFT scheme is that
the grain identity is lost and does not allow one to implement a realistic hardening scheme. An alternative framework was proposed by Kalidindi [4,5] and implemented
within a Taylor code: this approach keeps track of the

2138

G. Proust et al. / Acta Materialia 55 (2007) 21372148

evolving volume fraction of twins in each grain and of their


orientations with respect to the original grain. The matrix
twin crystallographic relation is assumed to be preserved,
and the Cauchy stress in the crystal is obtained as the volume average over matrix and twinned regions. Two drawbacks of Kalidindis model are: (i) grain reorientation
follows only from shear taking place in the matrix, and
(ii) no secondary twinning activity is allowed inside the primary twins.
Models for martensitic transformation in TRIP steels
share many commonalities with the crystallography-based
twin models described above [68] and have been adapted
[9] for modeling twinning. The model by Cherkaoui [9]
applies to face-centered cubic (fcc) systems, treats twins
as an increasing number of small at inclusions which
can be treated via a MoriTanaka approach for composite
systems and does not allow slip or further transformation
inside the twins. While these assumptions are relevant to
low-stacking fault energy fcc systems, they are not applicable to hexagonal close-packed (hcp) systems, where twins
grow to sizes comparable to the grains, and where slip
and secondary twinning activity inside twins plays a nonnegligible role in deformation, as we show here for Zr.
Correctly describing texture evolution is relevant to
understanding the anisotropy associated with plastic forming but accounts only for part of the constitutive description. The other part, describing twin-induced hardening,
is an even more challenging task because it requires us to
understand the interaction between the various slip and
twinning modes. Such an eect has been introduced in
polycrystal models, in a rather empirical way, by assuming
an evolution with strain of the threshold stress of the slip
and twinning modes, and by coupling the two modes using
latent hardening coecients. In Kalidindis approach [4,5],
latent hardening is made a function of the coplanarity
between twin and slip planes, and predictions of mechanical response are made for brass deforming in compression
and shear. A revised version of this model is used by Salem
et al. [10,11] to study texture and stress evolution in Ti
deformed by compression and shear. Tome et al. [12] also
implement a latent hardening scheme in a VPSC code
which used the PTR scheme for handling twin reorientation. They developed constitutive laws for Zr at room
and liquid nitrogen temperature, which reproduce tensile
and compressive response of rolled Zr along dierent directions and reproduce the measured texture as well. The constitutive law was used by Kaschner et al. [13] to study the
four-point bending of Zr bars. The approach utilized in
Ref. [12] for Zr was adapted by Brown et al. to describe
plastic forming of Mg [14] and Be [15].
The eorts described above were reasonably successful
in reproducing the measured stressstrain response and
texture evolution associated with monotonic loading in
tension, compression or torsion. However, the ultimate
challenge of a constitutive description is to predict the plastic response associated with complex loading histories, such
as changes in loading path. In such a situation, latent hard-

ening criteria overlooks the directionality of the grain


microstructure created during the previous deformation
path, and a model with a realistic description of the twin
structure and the twindislocation interaction inside grains
is required.
This paper addresses the challenge described above. We
use the well-established VPSC polycrystal approach as a
platform in which we implement a meso-scale composite
grain (CG) model. Such a procedure, which incorporates
microstructural information into the polycrystal model, is
of general applicability to low-symmetry materials. The
CG enters in the VPSC formulation as an ellipsoidal inclusion formed by alternating layers of twin and matrix
domains. The alternating twinmatrix interfaces provide
a barrier to and dene a directional mean free path for
propagation of slip or twinning. Since strain path changes
alter the prevailing sense of shear in the grain associated
with the previous loading condition, they best reveal the
role that microstructure plays on hardening. We address
here the simulation and interpretation of strain path
changes in rolled Zr tested in compression at liquid nitrogen temperature (76 K). Under this condition one can
induce either compressive or tensile twins, depending on
the testing direction.
This work, which focuses on modeling issues, is part of a
comprehensive microstructural characterization of the
mechanical response of Zr: McCabe et al. [16] report a
transmission electron microscopy (TEM) and electron
backscattering diraction (EBSD) study of the deformation modes as a function of temperature and test direction;
Kaschner et al. [17,18] performed and simulated compression reloads, keeping the test direction constant while
changing the temperature, to study the role of tensile and
compressive twins as barriers to the propagation of dislocations or other twins; and a preliminary version of the CG
model applied to strain path changes was recently reported
by Tome and Kaschner [19].
In this paper. we discuss the issues that need to be
addressed by the model and we highlight the necessity of
carrying out a simultaneous experimental and theoretical
study at several length scales. In addition to mechanical
testing and texture measurements, the necessity to characterize microstructural mechanisms both qualitatively and
quantitatively becomes obvious.
2. Experimental results
Our starting material is a high-purity clock-rolled Zr
crystal bar suciently low in interstitial impurities that
twinning is prevalent under the appropriate conditions.
Cubic samples of initial dimensions 7 8 9 mm3 were
cut from the rolled plate by electron discharge machining
and annealed at 550 C for 1 h. The resultant microstructure is fully recrystallized with equiaxed grains of
1520 lm. The initial texture of the Zr exhibits a strong
basal component perpendicular to the rolling plane
(Fig. 1). The cuboidal shape of the samples permits the

G. Proust et al. / Acta Materialia 55 (2007) 21372148

2139

Fig. 1. Initial (0002) poles of rolled Zr. Stress response associated with monotonic TTC and IPC (open circles). Stress associated with 10% TTC followed
by 20% IPC and 10% IPC followed by 20% TTC (solid lines). All tests performed at 76 K and 103 s1.

reloading of the same samples along a dierent strain direction without having to remachine them. It also allows an
easy preparation of at specimens for performing EBSD
in order to quantify the texture. Neutron diraction is used
to measure intermediate bulk textures on the same sample
between reloads.
The Zr was tested in compression at 76 K at a constant strain rate of 103 s1. This series of tests focuses
on the deformation mechanisms prevalent in Zr at
76 K. At that temperature, both f1 0 
1 2gh1 0 
1 1i tensile
and f1 1 
2 2gh1 1 
2
3i compressive twins are active, which
allows the investigation of their role in plastic deformation. Four dierent tests were performed: monotonic
compression along the through thickness direction
(TTC) up to 30% strain, monotonic compression along
the in plane direction (IPC) up to 30% strain, TTC up
to 10% followed by 20% IPC, and 10% IPC followed
by 20% TTC. The stress strain response for the four tests
is shown in Fig. 1. The basal pole gures, as measured
by neutron diraction at the HIPPO facility (LANSCELANL) at the end of each deformation history, are
shown in Fig. 3.
In our previous studies of Zr [12,16,17], three hardening
regimes were identied in the monotonic loading curves of
Fig. 1: an early slip dominated process (e < 2%) where,
most likely, twins nucleate; next there is a twin-dominated
stage ending at 10% or 20% strain, depending on the type
of twinning, and characterized by an increasing hardening
rate and a rapid increase in the twinned volume fraction.
Finally, twinning tends to saturate and deformation takes
place via slip. This latter stage is characterized by a
decrease in hardening rate, which can be qualitative linked
to slip taking place (either inside the twins or in the matrix)
across much reduced grain dimensions. In what concerns
the texture, it is apparent how twin reorientation substantially modies the initial texture: in the IPC case the c-axis
tends to align with the compressive axes as a consequence
of tensile twin activation, in the TTC case compressive
twins are activated which tends to deplete the center and
form an equatorial ring as viewed in a 0002 projection of
the texture [20].

3. The CG twin model


In previous simulations of Zr, we used a PTR scheme
[12,17] to describe the contribution of twinning to texture
evolution. The PTR scheme was incorporated in our VPSC
polycrystal code and consists of twin-reorienting full grains
once a certain threshold twin fraction (typically 50%) is
reached within the grain. A limitation of such an approach,
as far as hardening is concerned, is that, by considering
either the initial orientation or the twin reoriented grain,
it becomes dicult to account for the directional barrier
eect that twins have upon the propagation of dislocations.
Such an eect was accounted for in an indirect way by
enforcing a strong latent hardening associated with twin
shears.
The CG model that we present here instead accounts for
the original (parent) and twin reoriented (child) fractions,
their interaction and their evolution with deformation.
Here we adopt the philosophy of the PTR scheme, in that
we keep track of the accumulated shear of each twin system
in each grain, and identify as the predominant twin system
(PTS) the one that involves the maximum grain volume
fraction, given by
f PTS DcPTS =S

Here Dc is the shear contributed by the twin system and S


is the characteristic twin shear. We regard the PTS as the
system which is going to form the parallel twin lamellae
in the grain. The experimental evidence indicates that the
thickness of the twin lamellae and their separation are
(for a given strain rate and temperature) approximately
uniform [21,22]. This suggests the introduction of two
parameters in the CG model: the separation d c of the center planes (typically d c = 0.2d g) and the maximum volume
PTS
fraction that may twin-transform inside the grain fmax
(typically 0.5). We will assume that, following twin nucleation,
equally spaced twins of thickness d twin form in a grain of
size d g and grow under deformation to a possible maximum thickness d twin
max . The latter follows from the previous
g
PTS c
two parameters as: d twin
max fmax d (typically 0.1d ). The
schematic of such a CG is shown in Fig. 2.

2140

G. Proust et al. / Acta Materialia 55 (2007) 21372148

Fig. 2. EBSD image of Zr deformed 5% in TTC at 76 K. Schematic of CG and schematic of uncoupled twinmatrix approach showing the characteristic
lengths used in the model in association with twinning. In each case the at domains (or the ellipsoids) representing the twin lamellae and the matrix
lamellae are oriented such that the short direction (or the short axis of the ellipsoid) is along the normal to the K1 plane of the twin (nk1). The long
direction (or the long axis of the ellipsoid) is parallel to the twinning direction g1.

Simple geometric considerations show that the thickness


of the child and the parent domains evolve with the twin
fraction as:
d

twin

PTS c

matr

1  f PTS d c

We dene the PTS by keeping track of the volume fraction


associated with each twin system and assigning the PTS label to the one that rst exceeds a threshold fraction (typically 5% of the grain volume). The idea behind this
procedure is to allow for the nucleation and initial growth
process before committing to any particular twin system.
For the situation depicted in Fig. 2, it is reasonable to
assume that slip occurring in planes parallel to the twin
matrix interface will not be aected by the twin boundary
while slip will become more dicult on planes that intersect
the interface. In a way, twinning introduces a directional
grain size reduction. Once the PTS has been identied in
the grain, its twin plane will have a specic orientation with
respect to the slip and twin planes of the other active systems. As a consequence, a mean free path, d smfp , is introduced (see Fig. 2) for each system s.
d smfp d twin = sin a in the twin and
d smfp

matr

= sin a in the matrix

Here a is the angle made by the intersection of the PTS


plane and the slip or twin plane of the system s. In addition, d smfp < d g is enforced. As a consequence of Eq. (3)
the mean free path for a given system will gradually decrease in the matrix and gradually increase in the twin as
the twin lamellae grow. Such evolution is relevant to the
directional HallPetch hardening of each slip or twin system. However, it is unlikely that the twins will successfully
impede the transmission of dislocations until they involve a
substantial volume fraction of the grain. As a consequence,
when implementing the idea above in the CG model we
introduce the eect of mean free path upon hardening
gradually and do not allow for a full barrier eect until
the PTS has accumulated the maximum allowed fraction
PTS
in the grain fmax
.

4. The hardening law


Within the CG model, three mechanisms contribute to
the hardening of slip and twinning systems inside the
matrix and the twin: the evolution of statistical dislocations
with strain, the evolution of geometrically necessary dislocations (GND) and a directional HallPetch eect associated with the presence of twin interfaces
ss ssSTAT ssGND ssHP

The two mechanisms other than the HallPetch eect


are updated incrementally with deformation. The rst term
in Eq. (4) is a classical saturation Voce law associated with
statistical dislocations, plus a latent hardening eect cou0
pling shear increments Dcs in system s 0 with the increase
in strength in system s:
d^ss X ss0 s0
DssSTAT
h Dc
5a
dC s0
where




Ch0
^ssC s0 s1 1  exp 
:
s1

5b

Here C is the accumulated shear in the grain. An eect


which is central to the CG model comes from considering
a mean free path for each slip and twin system in relation
to the interface posed by the PTS lamella (Eq. (3)). The
dependence of the GND term with the mean free path
has been derived by Karaman et al. [23] and Kok et al.
[24] as:
DssGND

hsm
s
d mfp ssSTAT

ssGND

Dcs

The HallPetch term adopts the familiar form


hsHP

ssHP q
d smfp

What is dierent with respect to Refs. [23,24] is that in writing Eqs. (6) and (7) we utilize a directional mean free path
and make it specic to each system (Eq. (3)). The parameters hsGND and hsHP are empirical and reect the strength of

G. Proust et al. / Acta Materialia 55 (2007) 21372148

the corresponding mechanism. To keep the number of


parameters to a minimum, in this work, we assume that
the hardening parameters that appear in Eqs. (5)(7) are
dierent for each deformation mode but the same for every
system in the mode.
The characteristics of the CG model require us to make
several considerations concerning hardening:
(a) In the past, within the PTR scheme we had to assign a
high latent hardening coecient to the twin systems
to represent the twin barrier eect [12,17]. Now, such
an eect is specically accounted for by the directional dependence of hardening but it only accounts
for the PTS in the grain. Since the rest of the twin systems are allowed to contribute shear, we still allow
0
for hardening coecients hss > 1 for twinslip and
twintwin interactions, except the PTS, where we
use hs,PTS = 1.
(b) The strength of the PTS is set to a high value when
PTS
the maximum fraction fmax
has been reached, in
order to prevent further growth of the twin in the
grain. The rest of the twin systems are allowed to
remain active both, in the parent and the child, subjected to the hardening constitutive laws described
by Eqs. (4)(7).
(c) Basinski et al. [25] have suggested that a crystallographic transformation of glissile dislocations into
sessile should harden the twinned domains.
Micro-indentation measurements of the twins, done
in Hadeld steel [23] and titanium [11], suggest an
increase of about 20% in the yield stress inside the
twins. This eect, however, is dicult to quantify at
the slip system level and, since its inclusion is not critical for the model, it is obviated in this work. All the
same, twins tend to exhibit a larger strength due to
the HallPetch eect.

5. The self-consistent polycrystal model and CG deformation


We implement the CG model in our VPSC code, and the
reader is referred to Ref. [3] and references therein for a
detailed description of the VPSC model. During the simulation we need to calculate the stress and strain rate in the
matrix and twin domains of the CG. The strain rate e_ is
related to the shear rates c_ s contributed by slip and twinning systems through a rate sensitive law
X ms : rn
X
msij c_ s c_ 0
msij
M sec
8
e_ ij
ijkl rkl
s
s
s
s
Here r is the stress tensor and ss is the strength associated
with system s. The strength evolves with deformation and
twinning in the manner discussed above. Within the VPSC
approach we regard each grain as a visco-plastic inclusion
embedded in a visco-plastic eective medium characterized
by a constitutive response

2141

e_ ij M sec r
ijkl  kl

If one solves the problem of a visco-plastic inclusion


embedded in a visco-plastic medium, one obtains a relation
between macroscopic stress and strain rate, and the stress
and strain rate in the grain [3]:
e : r  r
 _e  e_  M

M sec : r  M sec : r

10a

where
e neff I  S1 SM sec
M

10b

is the interaction tensor, S is the visco-plastic Eshelby tensor, and ne tunes the inclusion-matrix interaction to be in
the range secant (ne = 1) to tangent (ne = n). In the present application, we use n = 20 and ne = 10 [26]. Combining Eqs. (8)(10) leads to a localization equation for the
grain stress of the form
e 1 M sec M
e r
rij M sec M
mnkl  kl
ijmn

11

The condition that the average stress over all grains has to
 provides a
be equal to the macroscopic stress hri r
relation for calculating the macroscopic secant compliance
M sec iteratively (self-consistently), namely,
e 1 M sec M
e i Iijkl
hM sec M
ijmn
mnkl

12

The modeling approach described so far is independent


of the CG model for the grain. We use in this work two different methods to calculate the deformation of matrix and
twin. They are briey described below.
5.1. Coupled deformation of twin and matrix: the CG model
In this case, the twin and the matrix are assumed to
interact across the twin interface that separates them.
One may introduce such interaction via a two-site VPSC
model [27] or via a model for lamellar structures proposed
by Lebensohn [28]. Here we implement the latter because it
assumes a planar interface separating twin and matrix,
which is a closer representation of such conguration in
hcp crystals. We dene an orthogonal reference system in
which axes 1 and 2 are contained in the planar interface
and axis 3 is perpendicular to it, and we enforce continuity
of some stress and strain rate components across the interface. Since we further assume that stress and strain rate are
homogeneous in the domain of the twin and the matrix, we
enforce the following conditions over twin and matrix
magnitudes:
T
e_ M
11 e_ 11 ;
T
rM
13 r13 ;

_ T22 ;
e_ M
22 e
T
rM
23 r23 :

_ T12
e_ M
12 e

13

Note that because ours is an incompressible formulation,


only ve of the independent deviatoric components of
stress or strain rate enter in Eq. (13) (the condition
_ T33 follows from incompressibility). For the purpose
e_ M
33 e
of the polycrystal model, we consider a composite grain
with fractions wM and wT assigned to matrix (M) and twin

2142

G. Proust et al. / Acta Materialia 55 (2007) 21372148

(T), respectively. The stress, the strain rate and the secant
compliance of the composite grain are given by the
weighted averages [28]:

the parent to the twin orientation when a threshold is


reached. From a numerical point of view, it only requires
us to duplicate the number of initial orientations.

T T
_ ij
e_ ij wM e_ M
ij w e

5.3. Reorientation of child and parent

rij

wM rM
ij

wT rTij

Note that the strain rate in matrix and twin is given by the
sum of shear rates contributed by slip and twin systems
(Eq. (8)). The twinning contribution, of the form
M tw
_ , may also be written using Eq. (1) as mtw
_ M S tw .
mtw
ij w c
ij w
The latter form is usually treated as a separate transformation term in martensitic transformation models [68] and
some twinning models [9]. The matrix A in Eq. (14) is a
function of the individual secant moduli for twin and matrix (Eq. (8)), and the continuity conditions (Eq. (13)).
The interaction equation (Eq. (10)) contains now to the
CG magnitudes dened by Eq. (14) and, from the point
of view of the self-consistent procedure, the CG is treated
as an eective grain. Observe that, in the limits
wM ! 1 or wT ! 1, the matrix or the twins, respectively,
dominate the eective magnitudes and, as a consequence, the mediumgrain interaction. When the twin fraction is much smaller or much bigger than that of the
matrix, this approach is likely to give a good representation
of the twinmatrix interaction. However, when twin and
matrix fractions become comparable it is likely that accommodation will be limited to the interface, without aecting
the bulk of the twin or the matrix. In such a case, the coupling assumption could be too restrictive, and the uncoupled scheme discussed below may be better suited.

Concerning crystallographic reorientation associated


with slip and twinning we refer the reader to any of the several papers where the kinematics associated with plastic
spin and the inclusion formalism are described (see, for
example, Refs. [3,8]). The issue that we address in this section is whether the twinmatrix crystallographic relation is
preserved during deformation. Crystallographic rotation of
twin and matrix follow from the slip and twinning shears
taking place inside them and, in general, their individual
rotation rates will not preserve the crystallographic twin
matrix relation across the interface. Here one may speculate that if the twin lamella growths in the grain, then such
a relation has to be preserved until growth is completed. As
a consequence, parent and child will co-rotate crystallographically, and we have in the past proposed a simple
approach for enforcing such co-rotation [29]. Past the twin
growth stage, one may assume, based on the fact that twin
and matrix are of comparable sizes, that they reorient independently of each other. While we do not have detailed
experimental evidence about this process, the EBSD characterization [30] indicates that the twinmatrix orientation
relation is preserved (within 15) for deformations of 30%
or less. However, since the main contribution to reorientation comes from twinning and slip does not induce large
changes in the misorientation between twin and matrix,
results are rather insensitive to the assumption adopted.
In our calculations, we reorient the matrix and the twin
independently.

5.2. Uncoupled deformation of twin and matrix

6. Application of the CG model to Zr

In this case, the twin and matrix domains are treated as


two non-interacting orientations (grains) embedded in the
homogeneous eective medium. We still identify the PTS
and account for its orientation, its evolving volume fraction and the evolving mean free path that the PTS denes
for each system. As soon as the twin is created we give it a
at ellipsoid shape. The shortest axis of the ellipsoid is parallel to the normal to the K1 plane of the twin and another
of the ellipsoid axes is parallel to the twinning direction g1
(see Fig. 2). A similarly oriented ellipsoid is created to represent the untwinned region of the crystal as soon as the
PTS
twin volume fraction reaches fmax
. The aspect ratios of
both ellipsoids evolve with deformation. Now twin and
matrix are characterized by independent secant compliances Msec and no explicit twinmatrix interaction is considered to solve the self-consistent equations. The relative
fraction of each phase is updated incrementally with deformation as the grain twins. This approach already represents an improvement upon the PTR scheme, in that it
keeps track of both orientations rather than switching from

The model described above in abstracto is applied here


to describe the constitutive response of pure Zr at 76 K.
We adjust the parameters of the model to a variety of
experimental data: stressstrain curves, textures, twin fractions, observed active modes. The reader should be aware
that the requirement of consistency with such an abundance of experimental data not only reduces the arbitrariness in the optimization of the model parameters but also
supports the validity of the physical mechanism assumed
by the model. TEM and EBSD microscopy [16] indicate
that our Zr deforms at 76 K by f1 0 1 0gh1 1 2 0i prismatic
slip, f1 0 1 2gh1 0 1 1i tensile twins and f1 1 2 2gh1 1 
2
3i compressive twins. Prismatic slip and tensile twinning are
easy to activate but the hard compressive twinning
is required to accommodate compression along the c-axis
of the grains.
Table 1 shows the hardening parameters adjusted using
the CG coupled approach described in Section 5.1. The
various parameters are dened in Section 4. For both slip
and twinning, we use a saturation-type Voce law to

1

M sec wM M sec ;M : A wT M sec ;T : wM : A wT I :


14

G. Proust et al. / Acta Materialia 55 (2007) 21372148

2143

Table 1
Single crystal hardening parameters for zirconium at 76 K
Modes
Prismatic
Tensile twin
Compressive twin

s0 (MPa)
45
105
330

s1 (MPa)
275
20
75

h0 (MPa)
300
60
20

HHP (MPa lm1/2)


65
65
200

HGND (MPa2 lm)


5

1e
3e5
4e5

hs,prism

dc

fmax

1
1
5

N/A
0.2
0.2

N/A
0.5
0.5

The parameter hs,prism represents the latent hardening of deformation mode s due to prismatic slip activity. dc represents the twin separation and fmax the
maximum twin fraction in a grain. See Eqs. (5)(7) for the denition of the other parameters.

describe the hardening until stage III. For slip, we use the
conventional response that follows from dislocation multiplication and recombination: the initial critical resolved
shear stress starts at s0 and increases until reaching the saturation value s0 + s1 (Eq. 5b). For the twinning modes,
however, we use an unorthodox softening response,
obtained by assigning to s1 and h0 negative values (Table 1).
The initial value of the critical resolved shear stress for
twinning is s0, but then it decreases rapidly and saturates
at s0 + s1. Such a response is meant to empirically account
for the twin nucleation stage, typically taking place within
the initial 2% strain. Microscopic observation indicates
that twins always initiate at grain boundaries, which suggests that nucleation is mediated by stress concentrations
at grain boundaries in the initial stages of loading. Further
straining leads to propagation of nucleated twins, which is
the mechanism that we address in our model. Making the
twinning modes harder at the beginning of the simulation
makes it possible to reduce the twinning activity at the start
of the deformation and empirically account for a twin
nucleation phase that reduces the presence of twins at
low strains.
The latent hardening parameters coupling prism slip
with twinning are negative (see Table 1) because the derivative of the Voce law is also negative for both twinning
modes and we want to enforce a positive hardening
through Eq. (5a). The other latent hardening parameters
are equal to 1 to enforce an asymptotic decrease in the twin
strength, as discussed above. This result diers from the
latent hardening coecients reported by Tome et al. [12]
in connection with the PTR scheme, where the coupling
between twinning and the other slip or twinning systems
was mediated by hardening parameters ranging from 2 to
20. In the CG model, such coupling is introduced by explicitly accounting for the presence of the twin interface barrier
rather than by means of ad hoc latent hardening
coecients.
Also, in the previous PTR-based simulations [12,17], the
stage IV of hardening for the deformation systems was
accounted for by a non-saturation Voce law. In the CG
model, it is represented by the GNDs, required to accommodate the lattice curvature caused by non-uniform plastic
deformation [24]. These GNDs act as obstacles to the
motion of other dislocations, hardening the various slip
modes [31,32] and twinning modes [23].
Concerning the HallPetch eect, twin boundaries are
considered in this model as having the same properties as

grain boundaries as far as providing obstacles to the propagation of dislocations is concerned. Therefore, the various
deformation modes become harder as the value of the
mean free path in the matrix decreases due to the presence
of these twin boundaries. The situation reverses inside the
twin domains, where deformation modes become easier
as twins grow. There is little experimental information
available for Zr in what concerns grain size eects upon
the yield stress to justify using separate HallPetch coecients for twins and grain boundaries. The experiments at
76 K done by Song and Gray [33] on Zr with 25 and
75 lm grain sizes, and oriented to activate compressive
twinning, set an upper bound for hHP of about
1000 MPa lm1/2 for the compressive twins. Work realized
by Armstrong et al. [34] on various cubic and hexagonal
metals seems to indicate that the HallPetch coecient is
always greater for twinning than for slip and that the values are independent of temperature and strain rate for all
the deformation modes. Obviously, substantial experimental characterization will be required for elucidating the
HallPetch parameters associated with the other deformation modes and to verify their independence on temperature and strain rate for Zr. The parameters listed in
Table 1 provide a reasonable t to the monotonic loading
tests. We will see that the HallPetch eect plays an important role when strain path changes take place.
For our calculations, we represent the initial texture
(Fig. 1) using 1944 orientations with properly assigned
weights and an initial grain size of 30 lm. We will use as
a benchmark the case in which there is coupling between
the twin and matrix phases (parameters listed in Table 1),
and will discuss the role and inuence of the various model
mechanisms by comparison to this one case.
The predicted stressstrain curves for the monotonic
and strain path change deformations are compared with
the experimental measurements in Fig. 4. The calculated
mode activities in the matrix and twin phases are shown
in Fig. 4 with the evolution of their relative fraction.
Fig. 3 shows the measured and predicted textures at 30%
strain for the four deformation paths studied. Table 2 gives
a comparison of the volume fractions of primary and secondary twins obtained experimentally [30] and from our
benchmark simulation (the twin volume fractions were calculated from the model using only the PTS in the matrix
and in the primary twins).
Two dierent techniques have been used to evaluate
experimentally the twin volume fractions: neutron

G. Proust et al. / Acta Materialia 55 (2007) 21372148

Measured

2144

Predicted

0.7
1.0
1.4
2.0
2.8
4.0

Fig. 3. Comparison of measured and predicted basal pole distributions after 30% accumulated deformation for: (a) monotonic TTC; (b) monotonic IPC;
(c) 10% TTC followed by 20% IPC; (d) 10% IPC followed by 20% TTC. The measured pole gures were obtained by neutron diraction.

Table 2
Measured and predicted primary and secondary twin volume fractions for TTC and IPC
Strains

Volume fraction primary twins


Measured

Predicted

Measured

Predicted

14%
30%
17%
30%

0.40
0.35
0.41
0.48

0.32
0.40
0.30
0.41

0.14
N/A
0.02
N/A

0.02
0.10
0.001
0.04

TTC
TTC
IPC
IPC

Volume fraction secondary twins

The measured values were obtained using neutron diraction at strain 0.3 and EBSD at strain 0.14 for TTC and 0.17 for IPC.

diraction after 30% deformation and EBSD at lower


strains. Each method has advantages and drawbacks. For
example, during TTC, secondary tensile twins form inside
the primary compressive twins. As a consequence, the caxis ips back from the equator to the center of the pole
gure (Fig. 3a), and the twin fraction will be underevaluated when using the pole gures obtained by neutron diffraction. On the other hand, EBSD allows the
identication and quantication of twins within twins,
but EBSD scans are less precise at large deformation
because dislocation structure deteriorates diraction pattern quality. In addition, since EBSD scans are done on a
small window of a cross-section of the deformed sample,
they are not necessarily representative of the bulk texture.
This is especially true for our TTC samples where heterogeneity of the microstructure has been observed. The EBSD
results presented in Table 2 were obtained from the middle
cross-section of the deformed sample, where the presence
of twins is greater. Therefore, the EBSD reported twin fractions are probably an upper bound. Work is in progress to
derive a more precise quantication of twin fraction [30].
The comparison of the measured and calculated results
shows that the CG model predicts accurately the hardening
rate associated with the twinning deformation for both
monotonic deformation processes and predicts satisfactory
the hardening rates for the reloading cases (see Fig. 4). The
predicted yield strengths at reloading are in agreement with
the experimental values, meaning that the eect of microstructure upon shear of new systems is correctly accounted

for by the CG model. As for the evolution of the yield


stress upon reloading, the increase in hardening rate for
both TTC ! IPC and IPC ! TTC is evidence of twinning
being activated. The CG model does predict compressive
twinning for the IPC ! TTC case, but not enough tensile
twinning for the TTC ! IPC case. In the latter case, a
decreasing hardening rate is predicted.
The stressstrain response is only one piece of experimental information. In addition, the predictions of the
model need to be consistent with measured textures and
observed system activity. The deformation modes active
at 76 K in the matrix and in the twins have been experimentally identied and discussed in Ref. [16]. For the TTC
case, the model predicts that deformation is accommodated
by equal amounts of prismatic slip and compressive twinning in the matrix plus an important contribution of prismatic slip and secondary tensile twinning inside the
primary compressive twins (Fig. 4a). Such activity is consistent with the equator that forms in the (0 0 0 2) pole gure, induced by the 64 reorientation by compressive
twinning of the initial orientations (Fig. 3a). Subsequently,
secondary f1 0 1 2gtensile twinning reorients the c-axis by
85 and realigns it again with the compression direction
(Fig. 3b). We would like to highlight the non-negligible role
played by secondary twinning in strain accommodation: we
predict 4% of secondary twinning after 14% TTC (experimentally we observed 14% of secondary twinning in the
middle cross-section of the sample [30] which, we know,
is an upper bound) and 10% after 30% TTC. For

G. Proust et al. / Acta Materialia 55 (2007) 21372148

2145

Fig. 4. Experimental (s) and predicted () stressstrain response for clock-rolled Zr deformed in compression at 76 K and 103 s1. Predicted mode
activity: relative shear contribution by each mode in matrix and twins at each step. (a) Monotonic TTC; (b) monotonic IPC; (c) 10% TTC followed by 20%
IPC; (d) 10% IPC followed by 20% TTC. Also shown as a solid line is the volume fraction of the matrix phase in the aggregate (the twin phase is
complementary).

monotonic IPC, deformation is accommodated by prismatic slip and tensile twinning in the matrix and, past a
strain of 15%, by compressive twinning inside the primary
tensile twins (Fig. 4b). These predicted mode activities are
in agreement with the experimental ndings obtained by
EBSD and TEM [16]. The texture (Fig. 3b) is consistent
with an alignment of the c-axis with the compression direction. The measured and predicted primary twin fractions
are consistent, while the secondary twin fractions are low
in both cases at 17% deformation (Table 2). Note that secondary twinning does not play an important role in IPC by
comparison with TTC for the same percentage of deformation. As discussed above, quantitative measurements of
twin volume fractions are still preliminary due to specic
limitations of the neutron diraction and EBSD techniques, which, respectively, tend to under- and overpredict
the volume fraction of twins. This experimental bias
explains the inconsistency of the primary twin volume fractions observed for TTC at 14% and 30% deformation. For
IPC, the model seems to predict lower volume fractions of
tensile twinning than the measurements by either
technique.
For the reloads, we predict that the microstructure
induced during preload aects substantially the mode
activities by comparison with the monotonic deformation.
For the TTC ! IPC reload, deformation is carried by the
matrix, where prism and tensile twin activities are similar
with the ones observed and predicted during monotonic

IPC. The initial yield upon reload is aected by the Hall


Petch mechanism but the hardening rate is indicative of
twin dominated deformation in the matrix (Fig. 4c),
although the model downplays the latter eect. No further
build-up or shear activity is predicted inside the (compressive) twin phase. The pole gure (Fig. 3c) is consistent with
such an analysis: while the pole gure is similar to the one
obtained after 30% IPC (compare Fig. 3b and c), it shows
that there is less tensile twinning activity than in the monotonic case. It is likely that the microstructure introduced
during TTC pre-loading acts as a barrier for the tensile
twins. The predicted pole gure shows an extra ring at
the equator due to an overprediction of compressive twins
during pre-loading.
For the IPC ! TTC reload, prism slip has hardened the
matrix during IPC preload and compressive twinning
makes a larger contribution to deformation than in the case
of monotonic TTC. Considerable prism activity and secondary tensile twin activity take place in the twin phase
after preload. It is likely that the secondary twinning is
mainly de-twinning of the tensile twins formed during the
IPC preload. The fact that hardening rate is low (Fig. 4d)
and that the nal pole gure (Fig. 3d) shows a reconstituted center component seem to support the idea of detwinning. Experimental and modeling work is underway
to clarify this issue. The remainder of this section focuses
on the eect of the hardening parameters on the simulated
stressstrain curves.

2146

G. Proust et al. / Acta Materialia 55 (2007) 21372148

6.1. Sensitivity of the model to the parameters


In order to characterize the sensitivity of the model to
the various parameters we modify one parameter at a time
while keeping the rest unchanged. The resulting stress
strain curves are presented in Fig. 5 and compared with
the benchmark case in Fig. 4 for the two monotonic and
the two strain path change deformations. The solid lines
in Fig. 5 represent the simulation done using the uncoupled
scheme of Section 5.2, that is, without imposing stress or
strain continuity (Eq. (13)) at the interface between matrix
and twin and by giving an ellipsoidal shape to both
twinned and untwinned regions of the crystal. The
obtained stressstrain responses are the results of the competition between the three following eects: (i) removing
the continuity condition decreases the stress ow; (ii) giving
a shape to the twin and matrix grains increases the stress
ow; and (iii) the deformation mode activity inside the
matrix and twins can vary from the benchmark case. The
role of these eects on the material response varies from
one case to another. The removal of the continuity condition inuences more the monotonic IPC simulation than
the others; the reason for this is the dierence in the number of grains being treated as CGs: 1112 grains out of 1944
for IPC, compared with 764 for TTC. Also, we observe the
greatest variation in the deformation activities for the

800
700
600

Benchmark
no coupling
c
d =0.5
Hgnd=0
Hhp=0

500
400
300
0.0

0.1

0.2

Stress (MPa)

Stress (MPa)

monotonic IPC case: the tensile twinning activity decreases


as the prismatic slip activity increases, which lowers the
stress ow as the critical resolved shear stress for prismatic
slip is smaller than for tensile twinning. Giving an ellipsoidal shape to the twin and matrix grains cannot compensate
the two other eects in the case of monotonic IPC; therefore, we observe a lower stress ow during the uncoupled
simulation. For the three other cases, the deformation
mode activities do not vary much and fewer grains are considered as CGs. Therefore, the shape given to twins and
matrix grains has a predominant eect on the material
response, and we observe an increase of the stress ow.
The uncoupled approach reproduces the rapid increase in
the ow stress experimentally observed during the TTC
reload following IPC. This phenomenon can be attributed
to a rapid increase in the compressive twinning activity in
the matrix and a rapid decrease of the tensile twin activity
in the tensile twins created during the preloading phase.
The ellipsoid shape given to the primary twins facilitates,
at the beginning of the reloading, secondary twinning activity (this can be due to an increase of stress ow in the primary twins); however, shortly after this it becomes easier
for matrix regions to deform by compressive twinning.
The separation of the center planes of the lamellae (d c in
Fig. 2) and the maximum twin volume fraction per grain,
PTS
fmax
, determine the value of the mean free path for all

800

600

400

200

0
0.0

0.3

0.1

Strain

800
700

Stress (MPa)

Stress (MPa)

600
500

0.3

0.2

0.3

800

600

400

200

400
300
0.0

0.2

Strain

0.1

0.2

Strain

0.3

0
0.0

0.1

Strain

Fig. 5. Eect of the various hardening parameters on the simulated stressstrain response of Zr at 76 K. (a) Monotonic TTC; (b) monotonic IPC;
(c) preload in TTC, reload in IPC; (d) preload in IPC, reload in TTC. (- - -) Benchmark simulations using the full CG model; () eliminating coupling at the
matrixtwin interface; (*) increasing the separation d c between the mid-plane of lamellae; (m) removing the geometrically necessary dislocation hardening;
and (h) removing the HallPetch hardening.

G. Proust et al. / Acta Materialia 55 (2007) 21372148

the deformation systems in the matrix and in the twins. The


mean free path controls the magnitude of the GNDs and
HallPetch terms (Eqs. (6) and (7)). The smaller the mean
free path for a specic deformation system, the harder this
system becomes. During the optimization of the hardening
parameters, we observed that the nal twin volume fraction
PTS
is mostly controlled by fmax
and much less by the d c
PTS
parameter. Thus, we kept fmax 0.5 in order to retain a
reasonable prediction of twin volume fraction at 30% strain
and varied d c. In Fig. 5, we compare the benchmark case
(dotted line), for which d c = 0.2, and simulations (represented by the star symbols) run with d c = 0.5. For all the
deformation processes the ow stress decreases and the relative mode activities vary in the matrix and in the twins as
d c changes. For example, in the IPC case the prismatic
activity increases while the tensile twin activity decreases
as d c increases. As a consequence, fewer primary twins
are created by comparison with the benchmark case. The
increase in prismatic activity over twinning is the reason
for the reduction of the ow stress during IPC.
The simulations excluding the contribution of the GND
term (Eq. (6)) are represented by black triangles in Fig. 5.
Excluding this term lowers the ow stress for all the deformation processes, as was to be expected. The monotonic
IPC case is strongly aected, with the hardening rate shifting from increasing to decreasing at high strains (Fig. 5b).
Omitting the GND contribution has the same eect as
increasing the mean free path: prismatic slip activity
increases while tensile twinning activity decreases.
Although the suppression of the GND term does not
greatly aect the stressstrain response for TTC, it does
inuence the nal texture at 30% strain because the tensile
twinning activity inside the primary compressive twins
increases. And, although the model does not account for
secondary twin reorientation, the shear associated with secondary twins reorients the primary twins. Since this phenomenon is not observed experimentally, this example
proves the necessity of preventing excessive secondary
twinning in the model, which can be accomplished through
the GND hardening.
The square symbols in Fig. 5 represent the stressstrain
responses obtained omitting the HallPetch term (Eq. (7))
in the hardening law. To start from the same initial yield
stress, we modify the value of s0. Therefore, only the
HallPetch eect due to the change of the mean free path
caused by twins has been suppressed in the simulations.
The elimination of the HallPetch term from the hardening
law lowers the eective ow stress in all four stressstrain
curves. This procedure aects mostly compressive twinning
in the TTC cases, as it is the deformation mode with the
highest HallPetch coecient. For the IPC cases, the
reduction of the ow stress has two origins. First, prismatic
slip takes the advantage over tensile twinning in the matrix,
inducing a further decrease in ow stress (same phenomenon described for the increase of d c). In addition, compressive twinning is easier than for the benchmark simulation
inside the primary tensile twins, which also causes a

2147

decrease of the ow stress. Moreover, the omission of the


HallPetch eect lowers the values of the reload yield
strength. In the benchmark case, the twins created during
preloading make it harder for the deformation modes to
be activated at reload due to a decrease of the mean free
path. This eect is greater for the TTC ! IPC than for
the IPC ! TTC because the TTC preload induces more
twins than the IPC preload.
7. Discussion
The plastic response of hcp aggregates is complex and
anisotropic due to the variety of crystallographic deformation modes and, in particular, the presence of twinning.
Twinning reorients portions of the grain, poses barriers
to the propagation of slip and other twin systems, and provides domains which may be either hard or soft
depending on the stress state acting on the grain. As a consequence, in hcp aggregates, twinning and hardening evolution are intimately coupled. For a constitutive description
to be reliable such coupling needs to be accounted for
explicitly. In this work, we propose a structural model of
the grain which incorporates the above-mentioned mechanisms and implement such a model in a self-consistent
polycrystal scheme. While the model is general, we dene
specic parameters for pure Zr deforming at 76 K, a temperature at which tensile and compressive twins are present
in addition to prism slip. We choose to perform and simulate strain path changes as a paradigm for revealing the
role of twinning in deformation and as a stricter test of
our constitutive law. Crystallography and relative directionality of slip and twinning are explicitly incorporated
into this CG model.
The advantage and the challenge of our approach is
that the predictions of the model can be compared with a
variety of experimental information, such as system activity, texture, stressstrain response and twin fractions. This
creates, in turn, a synergy between experiments and model:
modeling considerations suggest the experiments, and the
experimental results provide evidence to conrm or revise
modeling mechanisms. We have tried to minimize the number of parameters and the complexity of the mechanisms
associated with the CG model, especially when the experimental information available is not conclusive.
Implicit in our model is the assumption that twins nucleate at the beginning of deformation, most likely due to
stress build-up at grain boundaries. We only claim to
describe the twin propagation and growth that takes place
after nucleation. In addition, we assume uniform twin separation and uniform twin thickness in the grains. These are
reasonable assumptions if one envisages twinning as a
stress relaxation mechanism which, once activated, lowers
the shear stress in the vicinity of the twinmatrix interface
and favors new twinning activity away from the recently
formed twin region. We also consider at most one predominant twin system per grain: while we know from EBSD
that this is not always the case; it is, however, a reasonable

2148

G. Proust et al. / Acta Materialia 55 (2007) 21372148

assumption. In addition, we allow for secondary twins to


contribute shear but we do not consider a layered structure
inside the primary twins because this would complicate the
CG model. Further, it is to be expected that twinning will
not only depend on the relative orientation of the grain
with respect to the external load, but also on the orientation of the neighbors, which will introduce local uctuations. Some of the considerations above may be
incorporated in the model but will require a detailed microscopic and statistical characterization of spatial correlations and observed twin structures.
Concerning the role of twin boundaries as barriers, there
are several related issues. Theoretical calculations justify
using directional hardening for GNDs [24,35]. Concerning
the HallPetch eect for twins, although it has been documented for grain boundaries [33,34], using a directional
HallPetch for describing the eect of the twin interface
upon dislocations is to be regarded as speculative. The simulation results presented here indirectly support the validity
of such a concept, but careful experiments will be required
to characterize the mechanism. A related issue has to do
with transmissibility and reaction of dislocations at twin
interfaces, which will modify the twin-barrier hardening
mechanisms. Transmissibility mechanisms have been characterized or proposed for cubic [22,36] and hexagonal
[37,38] materials. Until we have better experimental characterization, we assume that twin interfaces pose perfect barriers to dislocations. However, since twins do not form a
perfectly layered structure across the whole grain from
the beginning of deformation, we introduce a smooth transition to the full barrier eect, at which point we start treating the grain according to the CG scheme.
The CG model presented in this paper introduces a more
realistic representation of twinning deformation, and
allows us to describe the constitutive response of pure Zr
at 76 K. We not only predict the stressstrain curves for
monotonic deformations, but also the response of the
material associated with changes in the loading direction.
Moreover, this model also predicts the deformation textures, changes in hardening rate due to twinning, and secondary twinning activity inside the rst-generation twins.

Acknowledgements
This work was supported by the Oce of Basic Energy
Sciences, Project FWP 06SCPE401. The authors are grateful to Rodney McCabe for sharing unpublished results of
twin area fraction characterized using EBSD.

References
[1] Van Houtte P. Acta Metall 1978;26:591604.
[2] Tome CN, Lebensohn RA, Kocks UF. Acta Metall Mater
1991;39:266780.
[3] Lebensohn RA, Tome CN. Acta Metall Mater 1993;41:261124.
[4] Kalidindi SR. J Mech Phys Sol 1998;46:26790.
[5] Kalidindi SR. Int J Plast 2001;17:83760.
[6] Cherkaoui M, Berveiller M, Sabar H. Int J Plast 1998;14:597626.
[7] Cherkaoui M. J Eng Mater Technol 2002;124:5561.
[8] Kubler R, Berveiller M, Cherkaoui M, Inal K. J Eng Mater Technol
2003;125:127.
[9] Cherkaoui M. Philos Mag 2003;83:394558.
[10] Salem AA, Kalidindi SR, Doherty RD. Scripta Mater
2002;46:41923.
[11] Salem AA, Kalidindi SR, Seminatin SL. Acta Mater
2005;53:3495502.
[12] Tome CN, Maudlin PJ, Lebensohn RA, Kaschner GC. Acta Mater
2001;49:308596.
[13] Kaschner GC, Liu C, Lovato M, Bingert JF, Stout MG, Maudlin PJ,
et al. Acta Mater 2001;49:3097108.
[14] Brown DW, Agnew SR, Bourke MAM, Holden TM, Vogel SC, Tome
CN. Mater Sci Eng, A 2005;399:112.
[15] Brown DW, Abeln SP, Blumenthal WR, Bourke MAM, Field RD,
Tome CN. Metall Mater Trans A 2005;36A:92939.
[16] McCabe RJ, Cerreta EK, Misra A, Kaschner GC, Tome CN. Philos
Mag A 2006;86:3595611.
[17] Kaschner GC, Tome CN, Beyerlein IJ, Vogel SC, Brown DW,
McCabe RJ. Acta Mater 2006;54:288796.
[18] Kaschner GC, Tome CN, Cerreta EK, McCabe RJ, Misra A, Vogel
SC. Mater Sci Eng, A; in press.
[19] Tome CN, Kaschner GC. Proc. of ICOTOM-14, Mater. Sco. Forum
2005;495497:10016.
[20] Rangaswamy P, Bourke MAM, Brown DW, Kaschner GC, Rogge
RB, Stout MG, et al. Metall Mater Trans A 2002;33:75763.
[21] Remy L. Acta Metall 1978;26:44351.
[22] Remy L. Metall Trans 1981;12A:387408.
[23] Karaman I, Sehitoglu H, Beaudoin AJ, Chumlyakov YI, Maier HJ,
Tome CN. Acta Mater 2000;48:203147.
[24] Kok S, Beaudoin AJ, Tortorelli DA. Acta Mater 2002;50:165367.
[25] Basinski ZS, Szczerba MS, Niewczas M, Embury JD, Basinski SJ.
Rev Metall/Cah Inf Tech 1997;94:103744.
[26] Tome CN. Modell Simul Mater Sci Eng 1999;7:72338.
[27] Lebensohn RA, Canova GR. Acta Mater 1997;45:368794.
[28] Lebensohn RA. Modell Simul Mater Sci Eng 1999;7:73946.
[29] Tome CN, Lebensohn RA, Necker CT. Metall Mater Trans
2002;33A:263548.
[30] McCabe RJ, Proust G, Cerreta EK, Misra A; in preparation.
[31] Arsenlis A, Parks DM. Acta Mater 1999;47:5971611.
[32] Shi MX, Huang Y, Gao H. Int J Plast 2004;20:173962.
[33] Song SG, Gray III GT. Metall Mater Trans A 1995;26A:266575.
[34] Armstrong RW, Zerilli FJ. In: Ankem S, Pande CS, editors.
Advances in twinning. Warrendale (PA): TMS; 1999. p. 6781.
[35] Acharya A, Beaudoin AJ. J Mech Phys Sol 2000;48:221330.
[36] Lee TC, Robertson IM, Birnbaum HK. Metall Trans
1990;21A:243747.
[37] Serra A, Bacon DJ, Pond RC. Metall Mater Trans 2002;33A:80912.
[38] Yoo MH. Trans Metall Soc AIME 1969;245:205160.

Vous aimerez peut-être aussi