fcc
lic
si
F. R
nd U
ed fo
line 1
models such as [2–5] the flow rule is based on the
Orowan equation and the hardening rule is based on
the evolution of dislocation densities. More and more
studies have shown, that it is important to optimize
rate hardening behavior for metal forming applica-
structures or subgrains. As the deformation increases
especially for polyslip the microstructure of the single
crystal can be characterized by cell blocks (CBs) sep-
arated by dense dislocation walls (DDWs). Further-
more these cell blocks consist of ordinary equiaxed
dislocation cells [6].
At somewhat higher strain sections of the DDWs
3–361
* Corresponding author. Tel.: +49-211-6792-393; fax: +49-211-6792-
The constitutive laws for single crystals are very
important as they form the basis for the understand-
ing of poly crystal deformation. For the flow rule of
the single crystal the continuum slip theory has been
well constructed and accepted commonly [1]. Although
in the majority of FEM studies the flow rule is still
described by a simple power law and the hardening
behavior of commercial alloys is represented in terms
of another power law of strain and strain rate em-
pirically, there are more and more authors beginning
to formulate constitutive relations based on disloca-
tion density theory. For most of these more physical
tions to forecast the occurrence of plastic flow local-
ization and fracture phenomena [5]. Such optimized
hardening laws should depend on microstructural state
variables rather than macroscopic quantities such as
accumulated plastic strain and strain rate.
For materials with high stacking fault energy such
as aluminium alloys, the plastic deformation is mainly
caused by dislocation glide along well defined slip
systems. When the deformation is very small the mi-
crostructure of the single crystal consists of tangled
dislocations. After a relatively small amount of plastic
deformation the microstructure will change to cell
A new dislocation density based constitutive model for fcc single crystals at elevated temperatures is developed. In addition to
former composite models [Acta Mater 48 (2000) 4181; J Mater Proc Technol 123 (2002) 155; Acta Metall Mater 41 (1993) 589; Acta
Metal 31 (1983) 1367] it distinguishes the individual slip systems and thus directly accounts for latent hardening. This distinction is
an essential prerequisite for a later implementation into crystal plasticity finite element models. The model is applied to hot
compression tests of aluminium single crystals with their h110i axis parallel to the compression axis. The predicted stress strain
curves fit the experimental observations very well for a strain rate range from 1� 10�5 to 1� 10�1 s�1 and a temperature range from
623 to 723 K using a single set of parameters.
� 2004 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
Keywords: Dislocation density; Internal variables; Constitutive equation; Aluminium; Single crystal; Slip system; Crystal plasticity FEM
1. Introduction the prediction of material strain hardening and strain
A constitutive model for
dislocation densities and its app
of aluminium
A. Ma,
Max Planck Institut f€ur Eisenforschung, Mikrostrukturphysik u
Received 25 March 2004; received in revis
Available on
Abstract
Acta Materialia 52 (2004) 360
333.
E-mail address: roters@mpie.de (F. Roters).
1359-6454/$30.00 � 2004 Acta Materialia Inc. Published by Elsevier Ltd. A
doi:10.1016/j.actamat.2004.04.012
single crystals based on
ation to uniaxial compression
ngle crystals
oters *
mformtechnik, Max-Planck-Str. 1, 40237 D€usseldorf, Germany
rm 14 April 2004; accepted 16 April 2004
0 May 2004
2
www.actamat-journals.com
tend to degenerate to become first-generation micro-
ll rights reserved.
ateri
bands (MB1s). The MB1s are three-dimensional walls
consisting of small pancake-shaped subgrains. In Lef-
fers’s opinion [7] it appears, that the basic wall character
of a DDW is unaffected by its partial transformation to
pieces of MB1s. Therefore, no distinction can be made
between DDWs and MB1s, when one considers the
microstructure for the single crystal. In many experi-
ments the DDWs are found approximately parallel to
the transverse direction (TD) of the rolled sheet, forming
a certain angle with the rolling direction (RD). In Bay’s
experiments [8] at higher strains a second generation of
DDWs was observed intersecting the first ones in some
grains. For Bay et al. [8] two intersecting families of
DDWs/MB1s are the normal situation at higher strains
and the small ordinary equiaxed dislocation cell groups
are divided into big cell blocks. Meanwhile Leffers [7,9]
only considers one predominant family of walls for the
reason of simplicity and the cell blocks are replaced by
band structures.
From the experimental results a single crystal should
be subdivided step by step if one wants to simulate its
physical reactions and texture evolution. In fact for the
FEM calculation this is not a wise choice because it will
render the calculation more and more time consuming
with increasing deformation. Starting from the contin-
uum mechanics view proper internal variables have to be
defined, which can reflect the microstructure evolution
(the ordinary equiaxed dislocation cells, CBs, DDWs)
collectively.
The ordinary equiaxed dislocation cell is composed of
cell walls with high dislocation density and enclosed cell
interiors with low dislocation density. A group of these
ordinary equiaxed dislocation cells, which have almost
the same crystallographic orientation, constructs one
cell block surrounded by those dense dislocation walls.
In their experiments Zehetbauer and Seumer [10] find
that the transformation from ordinary equiaxed dislo-
cation walls to subgrain boundaries or DDWs is the
result of the onset of dislocation climb. So for the reason
of simplicity we assume the physical characteristics of
cell walls and DDWs are the same. In the present work,
we treat one cell block as a perfect single crystal and the
microstructure has only two components: CBs and
DDWs which therefore have a different physical mean-
ing as compared with real structures observed in the
experiment [8], the CBs do not contain any walls and the
DDWs replace the experimental DDWs and cell walls
altogether.
According to experiments for copper the size of the
subgrains will decrease as the strain increases at the
beginning of deformation and saturate toward a value of
about one micrometer after a not very large strain
[11,12]. Analogous studies of the sizes of the CBs and
DDWs for FCC metals show, that both shrink with
increasing deformation, and furthermore the ratio of the
3604 A. Ma, F. Roters / Acta M
two values can be regarded as almost constant during
the deformation process [13]. Based on these results in
this study we assume, that the volume fractions of CBs
and DDWs are fixed.
In this study, we will adopt Roters et al. [3] and
Karhausen and Roters [4] idea, namely, that the im-
mobile dislocations in the CBs are those which have
been locked by other ones with nonparallel Burgers
vectors, and the immobile ones in the DDWs contain
additionally the dislocation dipoles with anti parallel
Burgers vectors.
A proper model should be able to predict the de-
formation work hardening behavior in terms of in-
ternal variables based on the special mechanisms of
microstructure evolution. Similar as the concept of
Roters et al. [3] the interaction of dislocations with
the microstructure can be expressed as: dislocation
sources inside the single crystal generate mobile dis-
locations, at the same time certain parts of the mobile
dislocations change to immobile ones and certain
parts annihilate; those immobile dislocations are
stored in the CBs and DDWs according to different
manners: that is, the dislocation locks are stored in
CBs and DDWs, while the dislocation dipoles are
stored in the DDWs only; the applied stresses cause
the mobile dislocations to ran across the CBs and
DDWs to accommodate the plastic deformation. In
the framework of this paper we only account for the
externally applied stress and the dislocations as stress
sources, i.e., we neglect all other possible stress sour-
ces such as particles, precipitates, etc. Moreover we
only treat the movement of single mobile dislocations
and dislocation dipoles, where the latter does not
contribute to the plastic deformation. We do not treat
the cooperative motion of groups of dislocations, e.g.,
local tilt segments, as this only gives a minor contri-
bution to the overall plastic strain. It is, however,
important for the evolution of the dislocation cell
structure, but, as stated above, we do not treat this
evolution explicitly in the model.
In this paper, we will use the following notation for
the dislocation structure:
• subscripts denote dislocation densities or slip systems:
� M , the mobile dislocation density,
� I , the immobile dislocation density,
� F , the forest dislocation density,
� P , the parallel dislocation density,
� a; b, different slip systems,
• superscripts denote areas of the cell structure:
� B, the dislocation cell block,
� W , the dense dislocation wall.
Of the two regions of the microstructure of disloca-
tions, the CBs contain the lower density of immobile
dislocations for every slip system, and the DDWs con-
tain the higher one. For the initial state without any
deformation it is proper to assume all these immobile
alia 52 (2004) 3603–3612
dislocation densities are identical for all slip systems.
be expressed as
density qPa and the forest dislocation density qFa for slip
system a. qPa is the piercing density perpendicular to ena
which means all the dislocation lines parallel to the slip
plane, and qFa are the dislocation lines parallel to ena.
A. Ma, F. Roters / Acta Materialia 52 (2004) 3603–3612 3605
L ¼ _eFeF�1 þ eFL0eF�1 ð5Þ
with
L0 ¼ �P�1 _P: ð6Þ
In the case of crystal plasticity the following constitutive
assumption is often used to connect the phenomeno-
logical variable and the physical phenomena:
L0 ¼
XN
a¼1
_cafMa; ð7Þ
here fMa ¼ eda � ena is the Schmid tensor for the slip
system a where eda expresses the slip direction and ena the
slip plane normal with respect to the undistorted con-
figuration. N is the total number of slip systems, e.g., for
fcc crystals there are 12 octahedral slip systems,
f111gh110i, i.e., N ¼ 12.
From the Orowan equation the plastic shear rate _ca of
one slip system a is a function of the mobile dislocation
density qMa on that slip system and the average slip
velocity of the dislocation lines
_ca ¼ qMabva: ð8Þ
In order to determine the average dislocation slip ve-
2. The constitutive equations
2.1. The elastic law
Based on the isomorphy assumption, the current
elastic law of the single crystal can be represented by the
referential elastic law through a plastic transformation P
[14]. The elastic law for the current configuration is
T2PK ¼ P~K 1
2
PTCP
�� � I��PT; ð1Þ
where T2PK is the second Piola–Kirchhoff stress tensor,
C ¼ FTF is the right Cauchy–Green strain tensor, F the
deformation gradient, I is the second order identity
tensor, and the fourth order tensor eK is the elasticity
tensor with respect to the undistorted configuration.
From Eq. (1), the Cauchy stress T can be calculated as
T ¼ 1eJ eF eK 12 eFTeF
�h
� I
�ieF ð2Þ
witheF ¼ FP; ð3Þ
andeJ ¼ detðeFÞ: ð4Þ
2.2. The flow rule
The velocity gradient tensor L ¼ _FF�1 is used to ex-
press the loading process. Using Eq. (3) this tensor can
locity va, we have to define the parallel dislocation
Ignoring the mobile dislocations 1 and assuming the
interactions of dislocations of different slip systems have
the same interaction strength, we find
qFa ¼
XN
b¼1
qIb cosðena; enb��� � edbÞ���; ð9Þ
qPa ¼
XN
b¼1
qIb sinðena; enb��� � edbÞ���: ð10Þ
Based on the framework of thermally activated dislo-
cation motion the average velocity va then reads
va ¼ kam0 exp
�
� Qslip
KBh
sinh
jsaj � spass;a
KBh
Va
�
ð11Þ
with
sa ffi eK 12 eFTeF�h � I�i �fMa; ð12Þ
and
spass;a ¼ c1Gb
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qPa þ qMa
p
; ð13Þ
where c1 is a constant, G the shear modulus, b the
magnitude of the Burges vector, h the temperature, KB
the Boltzman constant, m0 the attack frequency and Qslip
the effective activation energy for dislocation slip. sa and
sapass are the external driving force and the athermal re-
sistance or the passing stress for the mobile dislocations.
The jump width ka and the activation volume Va can be
calculated as functions of qFa as
ka ¼ c2ffiffiffiffiffiffiffiqFap ; ð14Þ
and
Va ¼ c3b2ka; ð15Þ
where c2 and c3 are constants.
In Eq. (11) the forward and backward jumps of dis-
locations over the obstacle have been considered. For
the reason of simplicity we will neglect the backward
jumps in the following, so that the dislocation velocity
changes to
va ¼ 1
2
kam0 exp
�
� Qslip
KBh
exp
jsaj � spass;a
KBh
Va
�
: ð16Þ
1 The mobile dislocation density is usually believed to be lower than
the immobile one by at least one order of magnitude.
3. A scaling relation for the mobile dislocation density
When dislocation densities are chosen as internal
variables, their evolution laws are generally formulated
as rate equations containing production terms and an-
c1Gb 1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffip � KBh
2
ffiffiffiffiffiffiffi
qFa
p
¼ 0: ð20Þ
3606 A. Ma, F. Roters / Acta Materialia 52 (2004) 3603–3612
nihilation terms. The evolution laws for the immobile
dislocation densities have been formulated based on
several dislocation reactions such as immobilization of
the mobile dislocations by the formation of dislocation
locks and dipoles and dynamic recovery by dislocation
climb [3]. However, for the mobile dislocation density
there is no such simple model to describe its evolution.
This difficulty arises from the fact, that both the number
of the mobile dislocations and their average slip velocity
effect the plastic deformation rate. One can accommo-
date the external plastic deformation and its changes by
fixing one and changing the other, but in general both
quantities will change at the same time. This makes it
very difficult to get a proper production term for the
evolution law without any additional assumptions.
An alternative idea has been presented in [15] in order
to circumvent the above mentioned difficulties. In this
concept it is assumed, that, although there are many
ways for the dislocation structure to evolve, the proper
one should satisfy, that the given external stress gener-
ates the maximum plastic deformation or the given
plastic deformation rate causes minimum external
stress. 2 This is fulfilled if
ðdsaÞ _ca ¼
osa
oqMa
dqMa þ
osa
oqPa
dqPa þ
osa
oqFa
dqFa ¼ 0: ð17Þ
As the parallel dislocation density qPa and the forest
dislocation density qFa have their independent evolution
laws, 3 which are only a function of the loading history,
and we should not give other constraints to them, we
can only satisfy the criteria partially for one fixed time
step by requesting
osa
oqMa
�
_ca;qPa;qFa
¼ 0: ð18Þ
Eq. (18) can be used as constraint condition for the
mobile dislocation density: First from (8), (13) and (16)
we find the stress strain rate relation
sa ¼ c1Gb
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qPa þ qMa
p þ Qslip
c2c3b2
1
"
þ KBh
Qslip
ln
2_ca
bm0c2
ffiffiffiffiffiffiffi
qFa
p
qMa
!# ffiffiffiffiffiffiffi
qFa
p
;
ð19Þ
and then combining (18) and (19) we get the constraint
equation for qMa
2 This assumption is equivalent to the minimization of global plastic
work.
3 In fact these are functions of the immobile dislocation densities, see
Eqs. (9) and (10).
2 qPa þ qMa c2c3b qMa
Introducing
A ¼ KBh
c1c2c3Gb3
;
(20) can be reformulated as
q2Ma � 4A2qFaqMa � 4A2qFaqPa ¼ 0: ð21Þ
Solving (21) and knowing that qMa > 0 we get
qMa ¼ 2A2qFa þ 2A
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
A2q2Fa þ qPaqFa
q
: ð22Þ
If we take the parameters KB ¼ 1:38� 10�23 J K�1,
0:16 c16 0:5, 16 c26 10, 16 c36 10 and material data
(b;G) for aluminium, the parameter A is in the range of
½10�4; 10�1� for temperatures between room temperature
and 723 K. So it is convenient to simplify Eq. (22) as
qMa � 2A
ffiffiffiffiffiffiffiffiffiffiffiffiffi
qPaqFa
p
; ð23Þ
or
qMa � Bh
ffiffiffiffiffiffiffiffiffiffiffiffiffi
qPaqFa
p ð24Þ
with another constant
B ¼ 2KB
c1c2c3Gb3
:
This means the mobile dislocation density is propor-
tional to the geometric mean of the parallel and the
forest dislocation densities, and has a linear relationship
with the temperature. Eq. (24) is an intrinsic constraint
equation for the dislocation structure. One can easily see
that the dislocation structure with two sets of indepen-
dent internal variables ðqMa; qIa; a ¼ 1;NÞ has been re-
duced to one with only one set of independent internal
variable ðqIa; a ¼ 1;NÞ. 4
From (24) it can be seen that qMa qIa; qPa; qFa. It is
interesting to realize at this point, that if one uses this
assumption in (20), Eq. (24) can be directly derived.
The same principle can be used for a composite
structure. In this case we get
sa ¼ fW c1Gb
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qWPa þ qMa
q(
þ Qslip
c2c3b2
1
"
þ KBh
Qslip
ln
2 _ca
bm0c2
ffiffiffiffiffiffiffi
qWFa
p
qMa
!# ffiffiffiffiffiffiffi
qWFa
q )
þ f B c1Gb
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qBPa þ qMa
q(
þ Qslip
c2c3b2
1
"
þ KBh
Qslip
ln
2 _ca
bm0c2
ffiffiffiffiffiffiffi
qBFa
p
qMa
!# ffiffiffiffiffiffiffi
qBFa
q )
;
ð25Þ
4 Remember that qF and qP are functions of qI (Eqs. (9) and (10)).
of the immobile dislocation density inside the CBs. The
A. Ma, F. Roters / Acta Materialia 52 (2004) 3603–3612 3607
first one is the immobilization of mobile dislocations,
which for pure materials is mainly caused by the inter-
action of the mobile dislocations with forest disloca-
tions. We therefore assume, that the average distance LA
traveled by each mobile dislocation is inversely pro-
portional to the forest dislocation spacing:
LA / 1ffiffiffiffiffiffiffi
qBFa
p : ð29Þ
The frequency of immobilization Fimmob can then be
calculated as
Fimmob ¼ vaLA :
Using the Orowan Eq. (8) the increasing rate for the
immobile dislocations becomes
_qBþIa / qMa � Fimmob;
which can be changed to the final form
_qBþIa ¼ c4
ffiffiffiffiffiffiffi
qBFa
q
_ca ð30Þ
with a constant c4.
The second process is the interaction of mobile dis-
locations of one slip system with the immobile disloca-
and the derivative reads
c1Gb
2
1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qWPa þ qMa
p þ 1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qBPa þ qMa
p !
� KBh
c2c3b2
ffiffiffiffiffiffiffi
qWFa
p
qMa
þ
ffiffiffiffiffiffiffi
qBFa
p
qMa
!
¼ 0: ð26Þ
Eq. (26) cannot be solved analytically, but if one uses
again the approximation qMa qIa; qPa; qFa we get
c1Gb
2
1ffiffiffiffiffiffiffi
qWPa
p þ 1ffiffiffiffiffiffiffi
qBPa
p ! � KBh
c2c3b2
ffiffiffiffiffiffiffi
qWFa
p
qMa
þ
ffiffiffiffiffiffiffi
qBFa
p
qMa
!
;
ð27Þ
and the solution for qM for the composite structure
reads
qMa � Bh
fW
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qWPaq
W
Faq
B
Pa
p
þ f B
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qBPaq
B
Faq
W
Pa
p
fW
ffiffiffiffiffiffiffi
qBPa
p
þ f B
ffiffiffiffiffiffiffi
qWPa
p : ð28Þ
Therefore, it is again possible to remove the mobile
dislocation density as independent internal variable. An
important property of Eq. (28) is that for either fW ¼ 0
or f B ¼ 0 Eq. (24) is recovered.
4. The evolution of the dislocation densities
4.1. The immobile dislocation density inside CBs
There are three processes contributing to the change
tions on the same system. We assume this interaction
mechanism will cause a decrease of the immobile dislo-
cation density. This mechanism is especially important
at low temperatures, where the probability of thermally
activated cross slip of screw dislocations and climb of
edge dislocations are very small. However, at the ele-
vated temperatures treated in this paper, also annihila-
tion by cross slip can be regarded as being included in
this term. Calculating the meeting frequency of one
mobile dislocation with another immobile one of the
same slip system one can derive the nonthermal anni-
hilation rate [3]
_qB�Ia1 ¼ �c5qBIa _ca ð31Þ
with a constant c5.
The third o