Modelling microstructural evolution during

annealing of inhomogeneously deformed

material
M. Rettenmayr and X. Song

A recently developed deterministic/Monte Carlo model is used for studying the interaction between recovery and
recrystallisation under various annealing conditions after inhomogeneous deformation. Locally, the changes of the
stored energy during annealing as well as the temperature and time dependence of the recrystallisation kinetics are
investigated in a parameter study. A higher annealing temperature has a twofold effect on the recrystallisation
process: on the one hand there is a higher nucleation rate, on the other hand the concomitant process of recovery
reduces the driving force for both the formation and growth of the recrystallising nuclei. In the present work, the
evolution of microstructure is simulated as a function of the annealing temperature over a wide range. Graded
recrystallised microstructures, as found in former experiments with cold drawn titanium rods, are reproduced by
simulations. Evaluations of grain size gradient and recrystallised fraction (recrystallisation depth inwards from the
edge of the rods) in experimental and simulated microstructures are in good agreement. MST/5309

At the time the work was carried out both authors are in the Materials Science Department, Darmstadt University of
Technology, D-64287 Darmstadt, Germany (dg9v@hrzpub.tu-darmstadt.de). Dr Song is now in the School of Materials Science
and Engineering, Hebei University of Technology, Tianjin 300130, China. Manuscript received 20 November 2001; accepted
2 May 2002.
# 2003 IoM Communications Ltd. Published by Maney for the Institute of Materials, Minerals and Mining.

dependence on nucleation and growth of recrystallising

Introduction
Recrystallisation is a process that signi® cantly affects the
microstructure and mechanical properties of metallic mate-
rials. Although microstructural evolution after deformation
and subsequent heat treatment leading to recovery and
recrystallisation has been the subject of research for the past
several decades,1 ± 6 a quantitative understanding of the
effects of predeformation and annealing conditions on the
recrystallisation process and the resulting microstructure
has not yet been attained, owing to the intrinsic complexity
of the different governing mechanisms and their interaction.
It is assumed in numerous investigations on recrystal-
lisation that the driving force for recrystallisation, i.e. the
stored energy, results from homogeneous deformation. 1
However, actual processing conditions together with the
geometry of a sample or workpiece generally lead to an
inhomogeneous distribution of the stored energy and hence,
to preferred (non-randomly distributed) nucleation sites
and anisotropic growth of the recrystallising nuclei. As
opposed to recent models accounting for crystal plasticity
where the local deformation in a grain due to its geometry
and orientation is considered,6 in this work macroscopic
strain gradients extending over a large number of grains are
treated. Furthermore, in spite of its obvious technical
importance, the effect of recovery has so far not been
suf® ciently characterised and analysed quantitatively. On
the one hand, dislocation movement as the responsible
mechanism for recovery is also involved in the formation of
nuclei in the deformed microstructure. Recovery should
thus accelerate nucleation. On the other hand, recovery
slows down the nucleation and growth kinetics during
recrystallisation by reducing the available driving force.
While the effect of recovery on nucleation and growth of
recrystallising grains has been described qualitatively and
semi-quantitatively in experimental work, to the knowledge
of the authors there is no extensive study with the aid of
numerical models.
In the present article, equations to describe the effects of
inhomogeneous pre-deformation, recovery and its temperature
grains, are presented and introduced into a Monte Carlo
model. Simulations of microstructure evolution are used to
analyse the kinetics of recrystallisation. The results of
previous experimental investigations7 ,8 on the recrystallisa-
tion behaviour of cold drawn Titanium rods are used as an
example to compare simulated and experimental ® ndings.
In cold drawn rods, the recrystallisation process proceeds
from regions of higher deformation at the surface of the rod
towards regions with lower deformation in the centre. The
characteristics of the experimental recrystallised micro-
structures can be described as: (a) after annealing at low
temperatures, the area fraction of the recrystallised zone
(i.e. recrystallisation depth from the surface of the rod) is
small; (b) at an intermediate temperature, a higher recrys-
tallisation depth, a grain size gradient and a grain aspect
ratio gradient develop from the surface towards the centre
of the rod sample; close to the central region, there is a
sharp change from large elongated grains to unrecrystallised
® ne grains; (c) in the samples annealed at higher tem-
peratures the recrystallisation depth decreases, the grain size
gradient and the grain aspect ratio gradient become less
pronounced; at high temperatures close to the b transition
temperature, large elongated grains are not observed at the
transition of the recrystallised and unrecrystallised zones.
Analytical equations
DEPENDENCE OF STORED ENERGY ON
LOCAL PRE-DEFORMATION
In an inhomogeneously predeformed material, the driving
force for recrystallisation, i.e. the stored energy, varies
locally. The energy introduced into the material during the
deformation process is stored in defects such as vacancies,
stacking faults and, most prominently, dislocations. It is
assumed here that during the heating period to the
annealing temperature, the dislocations undergo early
stages of the recovery process and rearrange themselves in

DOI 10.1179/026708303225009391 Materials Science and Technology February 2003 Vol. 19 173
174 Rettenmayr and Song Microstructural evolution during annealing of deformed material
subgrain (cell) walls. The stored energy will thus be a
function of the size of the subgrains. Applying the empirical
relationship between the subgrain size and strain obtained
by Gil Sevillano et al.,9 the stored energy ED has been
derived1 0 as a function of the true strain e as
e mally activated (as indicated by the exponential function
ED~ c : : : : : : : : : : : : : (1)
ae + b s
where a, b are constants with values of 2. 2 610 ­ 7 and
1 .1 610 ­ 7 respectively (for 2D geometry), and c s is the grain
boundary energy; typically c s ~0.5 J m ­ 1 .
Recovery is assumed to be controlled by thermally
activated glide and/or cross-slip of dislocations. Two pro-
cesses of dislocation movement, particularly annihilation
and rearrangement of dislocations, can contribute to a
reduction of the stored energy. Assuming dislocation move-
ment to be the governing mechanism for recovery, the
decrease of stored energy can be described by the decrease
of the internal stress, as proposed by Cottrell and Aytekin.1 1
Combining the underlying equation with the relationship
between the internal stress after deformation and the
dislocation density on the one hand and the relationship
between the stored energy and the dislocation density on the
other hand, yields an equation describing the evolution of
the stored energy with annealing temperature and time as1 2
" #2
1 =2 1 =2
e C3 kT
ED ~ c s { ln
ae + b C1 C2 G 1=2
where k is the Boltzmann constant, T the absolute
temperature, t the annealing time, G the shear modulus of
the material; C1 is a ® t parameter in the order of 10 ­ 2 7 m 3
that is connected to the temperature effect on the decrease
of the ¯ ow stress during recovery,1 2 C2 and C3 are con-
stants in the order of 0 .5 m ­ 1 /2 and 0 .5, respectively; t0
corresponds to the moment when recovery starts. In the
case of inhomogeneous deformation with a distribution
function as e ~e s (1 ­ x) 4 (derived in Ref. 10, e s is the largest
strain at the surface of the rod, x is the relative distance
from the surface), applying equation (2), the stored energy
at a given position can be calculated at any temperature
and time.
NUCLEATION RATE AS A FUNCTION OF
LOCAL STRAIN, TEMPERATURE, AND TIME
A common expression for thermally activated processes,
consisting of three terms, rate constant, driving force and an
exponential term containing the activation energy, is used to
.
describe the nucleation rate N as1
Ç ~C0 ( ED {E C ) exp({QN =kT )
N :
D
where C0 is a ® tting parameter (10 ­ 4 J ­ 1 s ­ 1 ), E CD is the
critical stored energy below which nucleation will not occur,
that can be calculated from the critical deformation e C (here
e C is set to 2%) based on equation (1), and QN is the acti-
vation energy for nucleation (here 80 kJ mol ­ 1 is used).
Introducing the variation of the stored energy with tem-
perature and time due to the coexisting recovery effect, one
obtains
8"
< e
1 =2
C kT
1 =2
Ç ~C0
N c s { 3
: ae + b C1 C2 G 1=2
9
= con® gurations before and after the reorienting attempt,
e C
{ c s exp( {QN =kT )
ae C +b ; with equal probability. If the selected unit is already
The qualitatively known characteristics of nucleation are
reproduced by equation (4): with increasing local pre-
deformation, the nucleation rate increases as a consequence
of the higher driving force; the nucleation rate decreases
Materials Science and Technology
with time due to the concomitant recovery process that
reduces the stored energy. In contrast to the monotonic
effects of predeformation and time, temperature has a
complex effect on nucleation: on the one hand, there is an
accelerating effect on nucleation, since the process is ther-
on the right side of equation (4)); on the other hand, the
temperature has also an inhibiting effect owing to the
more pronounced recovery that reduces the stored energy
and hence the driving force before a nucleation event
may occur.
Monte Carlo simulation
On the basis of the derived deterministic equations, a 2D
Monte Carlo model for recrystallisation has been deve-
loped. The cylindrical (rod) geometry of the sample is
represented by a narrow band along the radial direction.
The area of the domain is chosen as 150 6200 micro units.
Correlating the length of the simulated domain to the radius
of the rod sample of 2 mm yields the size of one micro unit
corresponding to approx. 10 mm in real dimensions. The
matrix with a given mean grain size (on average a grain
consists of approximately 20 micro units with identical
orientation) is generated with a Monte Carlo algorithm
t
1+ (2) for normal grain growth that is described elsewhere. 1 3
t0 Inhomogeneous predeformation 1 0 is introduced by map-
ping the local strain on each micro unit. The stored energy
varies from grain to grain as well as within a grain, as
a grain consists of several micro units that have been
attributed a local deformation value. The nucleation rate is
calculated according to equation (4) at the given tempera-
ture T, and the number of nuclei for a local zone is thus
obtained. Prior to distributing the nuclei to a certain zone,
an incubation period Dt accounting for the early stages of
recovery that promote nucleation, is introduced, and is
assumed inversely proportional to the actual driving force,
i.e. Dt/ ( ED ­ E CD ) ­ 1 .
The time in Monte Carlo simulations is scaled in Monte
Carlo steps (MCS). Equations (2) and (4) contain the real
time t. By incorporating these equations into the Monte
Carlo model, a second time scale is introduced that is dif-
ferent from the Monte Carlo time scale. One of the
prominent open questions in Monte Carlo modelling is to
describe the correlation between time unit MCS and real
time. This question was not addressed in this current work.
It is not assumed that there is a linear relationship between
the time units. In the present case, the effect of recovery as
described in equation (2) is dominant at the early stages,
: : : : : (3) but exerts only little in¯ uence after a few Monte Carlo
steps.1 2 Thus, a precise knowledge of the correlation
between the time units is certainly desirable but appears
not to be absolutely necessary. A semi-empirical correlation
between the time units will be developed in future work.
The microstructure evolution at a given temperature
history is then simulated by tracking the reorienting
attempts of each micro unit to the orientation state of
one of its nearest neighbours. If the energy state of a micro
unit in the new (tentative) con® guration is lower than in
#2 the former orientation, the reorientation attempt is
t
successful. If reorienting would yield an increase of the
ln 1+
t0 local energy, the orientation of the micro unit remains
unchanged. If the local energy is the same in both
the micro unit reorients or keeps its old orientation
: : : : : : (4)
recrystallised, all reorientation attempts towards the
unrecrystallised state are suppressed. In the case that a
recrystallised micro unit reorients to the state of a
neighbouring recrystallised unit, the procedure represents
growth of the recrystallised grains.
February 2003 Vol. 19
 Rettenmayr and Song Microstructural evolution during annealing of deformed material 175

2 Comparison of grain sizes evaluated from experimen-

tal and simulated microstructures as a function of dis-
tance along the radial direction; experimental values
for titanium rod annealed for 300 min at 850°C14

grain size in the surface layer. The agreement of the

simulated graded microstructure with the experiments in
both the morphology and the quantitative analysis shows
that under the given conditions the recrystallised micro-
structures are well reproduced by the Monte Carlo model
based on the analytical equations.
The local kinetics of recrystallisation is plotted in Fig. 3.
The recrystallisation rate, generally characterised as the
reciprocal of the time for 50% recrystallisation,1 is used to
compare the kinetics of recrystallisation in given zones. Due
to the predeformation, which is highest at the surface and
lowest in the centre, the driving force for nucleation and
growth of the nuclei decreases with increasing distance from
the surface. Consequently, the recrystallisation rate slows
down from the surface towards the centre, i.e. a longer time
is needed for the zone close to the centre to become fully
recrystallised.
Figure 4 shows partially recrystallised microstructures
a simulated, T~850°C, t~180 MCS; b experimental (Ti), that were generated in simulations for a given time interval
annealed at 850°C for 40 min8
at various annealing temperatures. At both lower (Figs. 4 a
1 Recrystallised microstructures and 4 b) and higher temperatures (Figs. 4 e and 4f), the
recrystallised areas are smaller than those at intermediate
temperatures (Figs. 4 c and 4 d). The progress of the
Results recrystallisation front is dependent on predeformation
distribution and temperature. The graded deformation
Simulations were carried out at given constant temperatures determines the initial local driving force for recrystallisa-
in the range 700 °C to 890 °C for various time intervals, tion. At low temperatures, the nucleation rate is dominated
where 700 °C is approximately the lowest experimental
recrystallisation temperature and 890 °C is the b transition
temperature of titanium Grade 2.7 As an example, in
Fig. 1 a a simulated microstructure is shown that is
predicted after annealing at a temperature T~850 °C. In
Fig. 1 b the corresponding experimental microstructure is
shown. The grain size gradient from the surface towards the
centre of the rod sample is a result of the inhomogeneous
pre-deformation, which in turn leads to a varying nuclea-
tion rate along the radial direction. Large grains form at the
recrystallisation front (de® ned as the zone that contains the
interface between recrystallised and unrecrystallised grains),
because at this location the nucleation density is low: the
low predeformation, close to the critical value for recrystal-
lisation, means that only a few nuclei are available, and keep
growing into the unrecrystallised zone.
Quantitative evaluations of the local grain sizes in the
simulated and experimental microstructures after annealing
at 850 °C are compared in Fig. 2. The grain sizes are cha-
racterised by a linear intercept. The mean grain size at a 3 Local recrystallisation kinetics evaluated during the
given distance from the surface, measured perpendicularly time interval 0 ± 180 MCS, at temperature T~850°C; x
to the deformation gradient, was normalised with the mean is the relative distance from the surface

Materials Science and Technology February 2003 Vol. 19

176 Rettenmayr and Song Microstructural evolution during annealing of deformed material
a 750°C; b 775°C; c 800°C; d 825°C; e 850°C; f 875°C
4 Simulated microstructures after annealing for time t~180 MCS at various temperatures: lower (® ne grained) part of
the microstructures represents the unrecrystallised matrix close to the centre of the rod sample
by the slow kinetics due to the exponential temperature
dependence (equation (4)), and recovery has a weak effect
on reducing the driving force for nucleation. With increasing
temperature, recovery becomes more and more prominent
owing to the stronger temperature dependence of recovery
compared to nucleation. At high temperatures, recovery
reduces the local stored energy signi® cantly before the onset
of recrystallisation. If the local stored energy drops below
the critical energy for nucleation, nucleation of recrystal-
lisation is suppressed. This leads to the observed effect
of a decreasing recrystallisation depth obtained at higher
temperatures.
The large elongated grains at the recrystallisation front
(particularly in Figs. 4c and 4 d) form because of the
combined effect of low nucleus density and anisotropic
growth of the recrystallised grains. Low predeformation
close to the critical value for recrystallisation means that
only few nuclei are available, leading to several large grains
in this region after recrystallisation. Because of the ® ner
matrix grains in contact with the recrystallising front (where
a higher amount of grain boundary energy is present) and
some remaining stored energy, growth of the recrystallised
grains occurs at a higher velocity towards the centre of the
rod than in the other directions, leading to an increasing
grain aspect ratio. At higher temperatures, as a result of the
stronger effect of recovery, both the grain size gradient and
the grain aspect ratio gradient become less pronounced. In
particular, the large elongated grains at the recrystallisation
front observed at 800 °C and 825 °C do not appear at 875 °C.
Materials Science and Technology February 2003 Vol. 19
The microstructure morphologies in Fig. 4 agree well with
experimental structures after annealing at the indicated
temperatures.7 ,8
The in¯ uence of temperature on the recrystallisation
kinetics is demonstrated in Fig. 5. In the zone very close to
the surface, e.g. at x~0 .05 in Fig. 5 a, there is the highest
amount of energy stored from predeformation in the
matrix. Although with increasing temperature the stored
energy decreases due to increasingly faster recovery, the
remaining stored energy is still suf® ciently high to provide
the driving force for nucleation and growth of the nuclei.
Thus, the delaying in¯ uence of temperature on recrystalli-
sation through the recovery effect, i.e. the retardation of
nucleation or the reduction of the recrystallisation rate, are
not re¯ ected in the recrystallisation kinetics of this zone
(Fig. 5 a). Similar features to those in Fig. 5a do not appear
in the zone at x~0 .25 in Fig. 5 b. At a relative distance
x~0 .25 from the surface, according to the deformation
distribution,1 0 the local predeformation drops to 32% of the
largest deformation at the surface. In this case, recovery has
a pronounced effect on both nucleation rate and growth
kinetics of recrystallisation.
In the temperature range where recovery leads to a
signi® cant reduction of the stored energy, nucleation may
be delayed because of a longer incubation period. This is
shown in Fig. 5 b by the different starting points of the
recrystallisation curves at different temperatures. At
T~850 °C and T~875 °C, a longer time is needed for this
zone to fully recrystallise, because of the lower energy that
 Rettenmayr and Song Microstructural evolution during annealing of deformed material 177
5 In¯ uence of temperature on kinetics of recrystallisation
at given locations
remains after recovery to drive nucleation and growth. In
the case where the local stored energy is reduced below the
critical value, nucleation will be suppressed and recrystal-
lisation will not occur in the corresponding zone.
Conclusion s
Equations have been derived to describe the effects of
predeformation and recovery on both the stored energy and
the nucleation rate of recrystallisation. Based on the cha-
racteristics of the nucleation kinetics, a Monte Carlo model
was developed. The microstructure evolution during recovery
and recrystallisation has been simulated under various pro-
cessing conditions, and the simulation results were compared
with experimental investigations. The following conclusions
are drawn:
1. The equation for the nucleation rate as a function of
predeformation, annealing temperature and time, contains
Materials Science and Technology
all of the fundamental parameters that exert an in¯ uence on
recrystallisation and thus allows quantitative description of
the nucleation kinetics for a wide range of processing
parameters.
2. The graded recrystallised microstructures are well
reproduced by the simulations qualitatively (morphology)
and quantitatively (mean grain size and grain aspect ratio).
The grain size gradient along the radial direction of the rod
can be attributed to the varying nucleation kinetics that is a
consequence of the inhomogeneous pre-deformation.
3. The graded predeformation and the resulting graded
microstructures in combination with the simulations calcu-
lations give insight into the interaction of different processes
and mechanisms during recrystallisation.
4. Image modelling by the Monte Carlo technique has
been shown to have the potential for predicting varying
microstructures that form due to complex interactions of
several physical processes.
Acknowledgements
Professor E. Exner is thanked for stimulating discussions.
One of the authors (X. Song) is grateful to the Alexander
von Humboldt Foundation for ® nancial support.
References
1. f. j. humphreys and m. hatherly: `Recrystallisation and
related annealing phenomena’ , 173; 1995, Oxford, Pergamon
Press.
2. b. radhakrishnan, g. b. sarma and t. zacharia: Acta Mater.,
1998, 46, 4415 ± 4433.
3. r. l. goetz and v. seetharaman: Metall. Mater. Trans. A,
1998, 29A, 2307 ± 2321.
4. w. j. zhang, u. lorenz and f. appel: Acta Mater., 2000, 48,
2803 ± 2313.
5. d. j. srolovitz, g. s. grest, m. p. anderson and a. d. rollett:
Acta Metall., 1986, 34, 1833 ± 1842.
6. d. raabe: Adv. Eng. Mater., 2001, 3, 745 ± 752.
7. c. mUÈller, s. pascuzzi and p. neuhAÈ usel: Prakt. Metallogr.,
1997, 34, 606 ± 616.
8. c. mUÈ ller and h. e. exner: Z. Metallkd., 1998, 89, 338 ± 342.
9. j. gil sevillano, p. van houtte and e. aernoudt: Prog. Mater.
Sci., 1980, 25, 69 ± 81.
10. x. song, m. rettenmayr, c. mUÈ ller and h. e. exner: Metall.
Mater. Trans. A, 2001, 32A, 2199 ± 2206.
11. a. h. cottrell and v. aytekin: J. Inst. Met., 1950, 77, 389 ±
401.
12. x. song and m. rettenmayr: Mater. Sci. Eng. A, 2002, 153 ±
160.
13. x. song and g. liu: Scr., Mater., 1998, 38, 1691 ± 1696.
14. s. pascuzzi: `In¯ uence of heat treatment on microstructure and
fatigue behavior of titanium grade 2’, Studienarbeit (in
German), Darmstadt University of Technology, Darmstadt,
Germany, 1997.
February 2003 Vol. 19