Professional Documents
Culture Documents
RDK 2012 Acta PhaseFiledGG
RDK 2012 Acta PhaseFiledGG
com
Received 21 December 2011; received in revised form 20 January 2012; accepted 23 January 2012
Available online 3 March 2012
Abstract
The characteristics of 3-D grain growth are investigated by a topological analysis of phase-field simulation results compared with
theoretical mean-field theories. We found that the size distribution of the grains starting from an arbitrary narrow distribution crosses
the self-similar Hillert distribution, and ends in a distribution with relatively longer tails of large grains in which the central peak shifted
towards smaller grain size. The distribution of topological classes, as characterized by the number of facets per grain, is found to be time-
invariant for the process as a whole. The obtained shape function is in good agreement with the analytical distribution function derived
based on the average N-hedron model [Rios PR, Glicksman ME. Act Mater 2008;56:1165]. The volumetric growth rate per topological
class also correlates well with the analytical approach obtained by Mullins [Mullins WW. Acta Mater 1956;3:900]. The relationship
between grain size and its shape, however, deviates from theoretical predictions.
Ó 2012 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
Keywords: Grain growth; Grain size distribution; Phase-field simulation; 3-D von Neumann; Grain shape distribution
1359-6454/$36.00 Ó 2012 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.actamat.2012.01.037
2720 R. Darvishi Kamachali, I. Steinbach / Acta Materialia 60 (2012) 2719–2728
The existence of such a regime, however, requires strict Here, r is the interface tension, and r1 and r2 are the
conditions, which is discussed by Mullins [14]. local radii of curvature. The sign of the velocity is hidden
In our study, we perform large-scale 3-D multi phase- in the sign of the curvature in local coordinates, where a
field simulations of ideal grain growth, i.e. with constant negative sign corresponds to shrinking grains and a posi-
interface energy and mobility. In the phase-field method, tive sign for the growing ones. Burke and Turnbull simplify
grain boundary energy is defined by the gradient energy the case by assuming (i) the interface tension is constant
term between different domains which is built into the free and independent of the radii; (ii) the averaged radius of a
energy functional of the system. Therefore, the existing gra- grain, R, is proportional to the average of the radii of the
dient between domains directly reveals the grain boundary curvature, r 1, where r = C1R; (iii) the rate of change of
position. The Gibbs–Thomson effect, which is naturally the average radius is proportional to the interface velocity,
incorporated in the governing equations of the phase-field, v; dR/dt = C2v; and (iv) the size distribution is self-similar,
guarantees the representation of the curvature-driven i.e. it scales with a typical or average size over time R R.
motion of the grain boundaries. However, it is not enough Employing these assumptions in Eqs. (1) and (2), the rate
for grain growth simulations, since there are considerable of change of the average radius is:
numbers of higher-order junctions which cooperate in the
dR C 2 Mr
growth process. For our propose, the multi phase-field con- ¼ ; ð3Þ
cept [15] is employed which enables us to reconstruct triple dt C1R
lines, quadruple points and higher orders of junctions in a which leads to parabolic kinetic for the growth:
physically sound picture. The validity of this aspect for the
R2 R20 ¼ K 1 t: ð4Þ
multi phase-field model has been examined in previous
2C 2 Mr
studies [16,17]. The most significant limitation for multi where K 1 ¼ is the growth constant. Eq. (4) is asserted
C1
phase-field simulations is the large amount of storage to be true for both 2-D and 3-D grain growth. It should be
which is needed during the computation. We overcome this noted that the above assumptions approximate the grain
problem in the current series of simulations by using a structure built by planar or distorted facets with the picture
dynamic storing algorithm that only reserves the available of spheroids embedded into a uniform medium. We discuss
field variables rather than all field variables. the intimate connection between these assumptions in
In this work, both statistical and topological perspec- Section 3.2.
tives of ideal grain growth are discussed. In this section,
we briefly review the theories on normal grain growth to 1.1.2. Hillert’s approach
which we compare our results. In Section 2, the multi Inspired by Greenwood’s equation for the coalescence
phase-field concept is briefly described and the parameters of particles, Hillert introduced a critical grain size above
of our simulations are given. The results and analysis are which grains grow and below which the grains shrink
discussed in Section 3. We display the role of topological [18]. He started from Eqs. (1) and (2) and introduced the
characteristics of the grain in the process of the growth. growth rate so as to be negative for small grains and posi-
We interrelate the size of the grains with their shape and tive for large ones:
present the grain shape distribution. The results are dis-
dR 1 1
cussed against the background of analytical mean-field the- ¼ aMr ; ð5Þ
dt Rcr R
ories and topological models.
or
1.1. Classical models for grain growth
dR R
R ¼ aMr 1 : ð6Þ
dt Rcr
1.1.1. Burke and Turnbull analysis
The classical model for normal grain growth proposed Here, a is a constant, and the critical grain size, Rcr, is
by Burke and Turnbull [8] deduces a parabolic relationship related to the instantaneous average grain size R, which
for grain evolution. In this model it is assumed that the varies with time. From Eq. (5) small grains are expected
grain boundary migrates toward the centre of its curvature, to shrink at a high rate, and for large grains the growth rate
which in turn reduces the interfacial area as well as its asso- approaches a positive constant value which is relatively
ciated energy. The interface velocity of a single grain is small. Fig. 1 shows the growth rate in arbitrary units for
assumed to be linearly proportional to the relative pres- different grain sizes where Rcr = 5 and aMr = 1.
sure, P, between the inside and outside of the grain: Utilizing the Lifshitz–Slyozov method [19], Hillert ended
up with the following grain growth rate equation:
v ¼ MP ð1Þ
dRcr aMr
where M is constant. The pressure is proportional to the ¼ ð7Þ
dt cRcr
mean curvature:
1 1
P ¼r þ : ð2Þ
r1 r2 1
The grain is projected to an equivalent sphere: 2r ¼ 1
þ r12 .
r1
R. Darvishi Kamachali, I. Steinbach / Acta Materialia 60 (2012) 2719–2728 2721
Fig. 1. (a) The growth rate of different grain sizes presented by Eq. (5) and (b) the schematic diagram of volumetric growth rate derived based on Eq. (9).
Fig. 2. The simulation box of 5123 grid points after 1000, 5000 and 10,000 time steps. The colour coding presents phase-field indices of the grains. The
visual “layering” is only due to the post-processing and the grains have a fully random texture. There are no grains with the same index. (For
interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
lation of 30,000 initial grains in a cubic box of 512 grid cant in the simulation box. Moreover, the same behaviour
points in each direction. Fig. 2 shows three snapshots of has been observed from simulation boxes of different sizes
the simulation box at different time steps. We have chosen and initial distributions. These results show that, although
the physical dimensions for interfacial energy and mobility the grain size distribution crosses Hillert’s prediction, at
of 1 J m2 and 1014 m4 J1 s1, respectively, within the later stages of the growth the distribution function deviates
accepted range for metallic systems. The time step and grid from the mean-field prediction. A tendency to abnormal
spacing are dt = 101 s and Dx = 107 m. grain growth with a bimodal distribution, however, was
not observed. We assume that the final distribution in
3. Results and discussion our simulations fulfils the requirements for self-similarity.
Fig. 3a shows the evolution of the grain size distribution The time evolution of the squared mean grain size is
during the initial steps of the growth. The system evolves shown in Fig. 4. The curve is plotted over first 20,000 time
very rapidly from its initially narrow distribution and steps but note that almost the whole simulation (50,000 ts)
approaches a broader distribution. The x axis in the figure maintains this form. The initial stages of the growth are
presents the relative grain size with respect to the critical omitted.
grain size defined in Hillert’s analysis, 98 R.2 A parabolic equation is obtained by fitting to the curve
After 1500 time steps, the grain size distribution as:
obtained from the simulation matches nicely with Hillert’s
R2 ½m2 ¼ 5:1 1015 t þ 3:7 1013 : ð24Þ
prediction and tends to stay almost unchanged for a rela-
tively long time. In this period of growth, the number of The growth constant is K = AMr = 5.1 1015 m2 s1
existing grains decreases from about 16,000 to 4000. in which A 12. The constant A is defined as 2C C1
2
in the
Fig. 3b compares the grain size distributions for several Burke and Turnbull analysis. From the stability condition,
time steps with the Hillert function. The values are taken a similar constant, is calculated from the mean-field
from individual time steps without averaging. These results approximation as:
confirm the work by Kim et al. [29], who performed similar 2
8 2a
phase-field simulations. ¼ 0:395 ð25Þ
At a later stage, however, the peak of the grain size dis- 9 c
tribution gradually shifts to the left (Fig. 3c). Simulta- which is smaller than the obtained value A. It is worth not-
neously the frequency of big grains at the right-hand side ing that the constants C2 and a in the Burke–Turnbull and
increases and the distribution becomes more symmetric Hillert formulations correct the interface velocity due to
around 1 (Fig. 3d). The rate of change of the grain size dis- the shape effect.
tributions is very slow and the symmetric grain size distri- Since we are able to evaluate Eq. (6) and calculate the
bution also tends to keep its form for a relatively long time. constant a from our simulation, it is easy to test the
Statistically, the number of remaining grains is still signifi- mean-field assumptions in the case of 3-D grain growth.
Fig. 5 shows the normalized growth rate vs. the relative
2
For consistency and fair comparison between figures within this study, grain size for different time steps. Although the data points
we use the relative grain size with respect to the critical value, Rcr, instead are highly scattered, they can still be fitted along a line. The
of average grain size. Note that the Hillert function is not a homogeneous two important characteristics of this fitting are (i) the slope
function of degree 1, and therefore it is not straightforward to reformulate of the line which gives a and (ii) the relative grain radius,
his function for the relative size respect to the average grain size R. See the R
Appendix in Ref. [18]. Rcr
, for which the growth rate vanishes.
2724 R. Darvishi Kamachali, I. Steinbach / Acta Materialia 60 (2012) 2719–2728
Fig. 3. Simulated grain size distributions. (a) The grains evolve toward a smoother function and finally get close to the Hillert function. (b) For a relatively
long period of time, the grain size distribution coincides with the Hillert function. We name this period of growth the “Hillert regime”. (c) The distribution
gradually deviates from the Hillert regime and shifts to the left. (d) While the number of available grains is still considerable, a new, relatively stable
distribution appears. This deviation is found to be independent of the initial configurations.
Fig. 5. The normalized growth rate for different time steps fitted to a linear curve. The colour coding present the topological classes of the grains. For
relatively small grains there is larger deviation from the mean-field approximation. (For interpretation of the references to colour in this figure legend, the
reader is referred to the web version of this article.)
Fig. 6. The evolution of a (a) and critical relative size (b). There is a departure from the mean-field assumptions made by Hillert.
grain growth. This perhaps explains the deviation of the Remarkably, despite all the discrepancies with the
distribution function from the mean-field prediction at mean-field theory, the parabolic kinetics of the grain
later stages, which is a natural result of the geometrical growth with a single growth constant (K) remains valid
complexity of the grains. This will be discussed in more for the entire growth process. If we restart from Eqs. (1)
detail in Section 3.3, where we focus on the topological and (2) and look again at assumptions (ii) and (iii) made
characteristics of the grains during the evolution. by Burke and Turnbull (Section 1.1.1), the property of
2726 R. Darvishi Kamachali, I. Steinbach / Acta Materialia 60 (2012) 2719–2728
parabolic equations requires that constants C1 and C2 have fastest shrinkage rate was predicted to be 0.5Rcr, which is
a fixed relation in order that the growth constant K not hard to distinguish in our simulations. The shape of the dis-
change with time. We have already mentioned that C2 var- tribution changes slightly along the simulation.
ies similarly to a. This is due to the type of correction which
Burke and Turnbull provide for the kinetic Eqs. (3) and (5). 3.3. 3-D von Neumann and topological analysis
We also know that C1 is a geometrical constant which
relates the average radius (calculated from the total volume As presented in Section 1.2, Mullins [11] and Hildefeldt
of the grain) with the average curvature of the grain. Thus, et al. [12] have given analytical expressions for 3-D grain
C1 is expected to vary inversely with C2. Of course, the rate growth. Fig. 8 shows the normalized volumetric rate of
of the changes for C1 and C2 is slow and interconnected. change from our simulation and both these analytical func-
In Fig. 7, the volumetric growth rate is plotted against tions. At the earlier stage of growth, the obtained values
the normalized radius. We have chosen the same time steps coincide with Mullins’s prediction while later on they devi-
as in Fig. 5. The shape of the distributions compares well ate slightly; the slope of the curve decreases slightly with
with the function obtained from the mean-field assumption time. The critical number of faces (ncr), which is the average
(Fig. 1b). This is also comparable with simulation results number of faces per grain at which the volumetric rate
obtained from the vertex model of Barrales Mora et al. changes its sign, is 13.5 for earlier stages. Later on, it
[30], who discussed the variety of growth rates for the increases towards 14, but it is impossible to extract a pre-
grains of the same size. They also plotted the growth rates cise value from the simulations. This value is predicted to
of individual grains vs. their size, and obtained results that be 13.769 by Lazar et al. [31] and 13.397 from the analyses
look similar to ours. From the mean-field theory, the of Rios-Glicksman [24] and Hilgenfeldt et al. [12]. Based on
Fig. 7. Volumetric rate of changes for different time steps. Compare with Fig. 1b. The shape of the distribution changes slightly with time. The colour
coding relates to the topological classes. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this
article.)
R. Darvishi Kamachali, I. Steinbach / Acta Materialia 60 (2012) 2719–2728 2727
Fig. 9. Grain shape distribution at the initial (a) and stable (b) stages of the grain growth simulation.
2728 R. Darvishi Kamachali, I. Steinbach / Acta Materialia 60 (2012) 2719–2728
Fig. 10. Topological analyses. (a) The topological class number (n) of each grain vs. its relative grain size. The black cubes are averaged values. (b) Grain
pffiffiffi
size distributions for three topological classes. (c) Square root of the averaged topological classes ( n) plotted against the relative grain size.