You are on page 1of 14

Journal of Materials Science & Technology 50 (2020) 257–270

Contents lists available at ScienceDirect

Journal of Materials Science & Technology


journal homepage: www.jmst.org

Research Article

Normal and abnormal grain growth in magnesium:


Experimental observations and simulations
Risheng Pei ∗ , Sandra Korte-Kerzel, Talal Al-Samman
Institut für Metallkunde und Materialphysik, RWTH Aachen University, D-52056, Aachen, Germany

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

Article history: Commercial purity as-cast magnesium was hot rolled and subsequently annealed at different tempera-
Received 23 June 2019 tures in order to investigate its grain growth behavior and link it to the texture evolution. Annealing at an
Received in revised form 19 August 2019 intermediate temperature of 220 ◦ C gave rise to abnormal grain growth with a few grains reaching a grain
Accepted 21 September 2019
diameter 10 times larger than the mean. Increasing the annealing temperatureto 350 ◦ C yielded normal
Available online 5 March 2020
grain growth. Both types of grain growth revealed a strengthening of the (0001) 11−20 texture compo-
 
Keywords: nent. It is hypothesized that a dislocation density gradient after recrystallization grants (0001) 11−20
Pure magnesium grains a size advantage during early stages of growth. The type of growth will be, however, determined
Grain growth by the mobility of the present grain boundaries and triple junction drag, which are strongly dependent
Texture on the annealing temperature. The above hypothesis of the interplay between these parameters was
Dislocation density gradient
explored through curvature- and residual dislocation-density-gradient-driven grain growth simulations
Level-set modeling
using a formerly developed level-set approach. The simulation outcome suggests that application of such
a modeling approach in microstructure studies of magnesium can provide valuable new insights into the
problem of grain growth and associated texture evolution.
© 2020 Published by Elsevier Ltd on behalf of The editorial office of Journal of Materials Science &
Technology.

1. Introduction AGG has been widely reported in many metallic materials, e.g.
magnesium alloys [3,4], stainless steels [5], Fe-Si steel [6,7], tita-
Grain growth (GG) behavior during thermomechanical treat- nium alloys [8] and aluminum alloys [9,10]. However, its driving
ments plays a significant role in the investigation of metallic mechanisms are still not entirely clear, and in some cases even con-
materials owing to the close dependence of properties upon grain troversial. In the literature a variety of mechanisms were proposed
size [1]. There are two types of grain growth: normal grain growth to explain this phenomenon, which are summarized as follows:
(NGG) and abnormal grain growth (AGG). NGG takes place when the
average grain size continuously increases with the annealing time
(i) Anisotropic grain boundary characteristics (i.e. mobility and
and the grain size distribution remains self-similar, i.e. the shape
energy) of high and low angle boundaries, as pointed out by
of the size distribution does not change [2]. By contrast, if AGG
Bhattacharyya et al. who studied the grain growth behavior of
occurs, the grain size distribution will vary significantly in terms of
a strongly textured AZ31B magnesium alloy [4].
height and width [2]. AGG is also termed secondary recrystalliza-
(ii) Boundary pinning induced by solute and/or Zener drag, which
tion because only few grains grow rapidly and consume the rest of
was for example reported by Basu et al. in their work on
the microstructure. As a result, a bimodal distribution of fine and
rolled and annealed Mg-Gd and Mg-Dy alloys [3]. Interest-
coarse grain sizes is typically observed during this process until it
ingly, they found that AGG occurred in Mg-Gd but not in Mg-Dy
completes and gives rise to a new unimodal distribution [1,2].
alloy upon annealing at 450 ◦ C for 1 h. This was attributed to
the stronger grain boundary solute segregation effect shown
by Gd in comparison to Dy. As a consequence, at sufficiently
elevated annealing temperatures, grains with a topological
advantage would possess sufficient boundary velocities to
∗ Corresponding author.
overcome solute drag and grow rapidly relative to other grains,
E-mail addresses: pei@imm.rwth-aachen.de (R. Pei),
Korte-Kerzel@imm.rwth-aachen.de (S. Korte-Kerzel),
which would be eventually consumed due to their growth
tasamman@imm.rwth-aachen.de (T. Al-Samman). restriction [3].

https://doi.org/10.1016/j.jmst.2020.01.014
1005-0302/© 2020 Published by Elsevier Ltd on behalf of The editorial office of Journal of Materials Science & Technology.
258 R. Pei et al. / Journal of Materials Science & Technology 50 (2020) 257–270

(iii) Special, coincidence site lattice (CSL) boundaries, as in 13 rapid water quenching. The rolled material was then subjected
(27.8◦ <0001 >) [11] and 11 (35.1◦ <0001>) [12] were also to recrystallization annealing at 200 ◦ C for 1 h, followed also by
highlighted in the context of selective growth in hexagonal water quenching to freeze the obtained recrystallized microstruc-
materials based on the premise of being low energy and high ture. This step was quite challenging because the aim was to attain
mobility boundaries. full recrystallization without triggering grain growth. Therefore,
(iv) Solid-state wetting. The occurrence of AGG in Fe-Si alloys we chose a relatively low annealing temperature to have a bet-
was initially attributed to a grain size advantage neces- ter control of the onset of grain growth, where we ended up with
sary to overcome the Zener pinning force [13–16]. However, almost a fully recrystallized structure, with the exception of a few
other studies argued that this theory cannot explain the phe- grains that remained unrecrystallized. The average grain size of
nomenon because such size advantage was not always evident this microstructure was ∼15 ␮m, which served as the starting
after recrystallization [17–19]. Recently, solid-state wetting microstructure for the following grain growth annealing experi-
was proposed to explain the preferential growth of grains ments, and as input for the simulations. For the grain growth study,
with {110}<001> GOSS orientation in Fe-3%Si alloy [7,20–23]. smaller samples were machined from the recrystallized state and
The basis of this theory stems from the concept of sub- annealed at 220 ◦ C for 10 min, 30 min, 2 h, 8 h, 16 h, 24 h, 3 days
grain boundaries being low energy boundaries, which drives as well as 7 days. Additional annealing experiments at 350 ◦ C for 7
their expansion at the expense of high angle boundaries with days were performed to study the long duration growth behavior
a much higher energy [7,20–23]. Therefore, GOSS-oriented at a higher temperature.
grains, which are rich in low energy boundaries, exhibit pref-
erential growth over other grain orientations.
2.2. Microstructure and texture characterization
(v) Residual dislocation density gradients (DDG). In addition to
grain boundary curvature, differences in the remaining stored
Specimens for optical microscopy and electron back scatter
dislocation energy among adjacent grains after recrystalliza-
diffraction (EBSD) were prepared by grinding and mechanical pol-
tion provide additional driving force for grain boundary motion
ishing, followed by electro-polishing in an electrolyte of ethanol
during further annealing [24,25]. Boundaries of grains of lower
and H3 PO4 (5:3), for 45 min at 2.00 V. For chemical etching a solu-
dislocation density will migrate faster than those that possess
tion consisting of 10 ml H2 O, 10 ml CH3 COOH and 70 ml picral (4%
a higher dislocation density in order to minimize the total free
picric acid in a solution with ethanol) was used. The characteri-
energy of the system. This could eventually lead to AGG after
zation of grain size and grain size distribution of specimens with
long-time annealing treatments.
different annealing conditions was done by means of a Matlab code
(vi) Grain boundary complexion transition. This mechanism has
based on the linear intercept method. LEO-1530 scanning electron
been widely reported for several different doped alumina sys-
microscope (SEM) equipped with a field emission gun operated at
tems [26,27], where AGG was associated with the presence of
20 keV, and HKL Nordlys II EBSD detector was employed to carry
certain grain boundaries that not only had higher mobility,
out the EBSD measurements. The EBSD data was analyzed and
but also microscopically distinguishable structures and com-
visualized by the Matlab toolbox MTEX [30]. The macrotextures of
positions, referred to as complexions [26,28,29]. Based on this,
the investigated specimens were determined by X-ray pole figure
complexions are found to promote AGG if transitions of a frac-
measurements using a Bruker D8 diffractometer equipped with a
tion of grain boundaries dramatically increase their mobility
high resolution area detector. To improve the grain statistics the
relative to the average surrounding grain population.
specimen was oscillated in the diffraction plane during the mea-
surement. Six incomplete pole figures, namely {10 1 0}, {0002},
In the present study, we investigated the grain growth behavior {10 1 1}, {10 1 2}, {11 2 0} and {10 1 3}, were first measured using
in Mg. For this, we used (commercially) pure Mg in order to sys- filtered iron Fe-K˛ radiation at 30 kV and 25 mA. Then, the orienta-
tematically exclude several of the above-mentioned phenomena tion density function (ODF) was determined and used to recalculate
related to the presence of alloying elements in order to improve the complete pole figures.
our understanding of the underlying mechanisms in Mg itself.
The material was hot rolled, recrystallization annealed and then
2.3. Computational procedure
subjected to grain growth annealing treatments at different tem-
peratures for different times in order to quantitatively investigate
For the simulation work in this study, a recently developed Grain
its grain growth behavior and the corresponding texture evolu-
Growth Level Set simulation tool (GraGLeS) [31–33] was applied to
tion. Interestingly, AGG was observed for annealing treatments at
explore the effect of residual dislocation density gradients, remain-
a relatively low temperature of 220 ◦ C but not at higher tempera-
ing in the microstructure after recrystallization, on the grain growth
tures such as 350 ◦ C even after 7 days of annealing. With the use of
behavior during further annealing treatments. This model, which is
experimental characterization techniques combined with level-set
available as an open source code [34], was chosen due to its excel-
grain growth computer simulations, this work attempts to clarify
lent computational stability and good description of the physics
the mechanisms responsible for the aforementioned growth behav-
of grain growth. This being its capability of considering a variety of
ior invoking several potentially important parameters that interact
parameters influencing grain growth, such as anisotropic boundary
differently at varying temperatures.
properties (if required), and finite triple junction mobilities. It also
allows for real-time-scaled simulations and can incorporate both
2. Experiments interface and volume energies as driving forces. The latest devel-
opment of this model featured a 3D approach tailored for parallel
2.1. Specimen processing computing [33]. In that study, the authors addressed a case of mag-
netically driven grain boundary motion in titanium incorporating
As-cast pure Mg (99.95%) was hot rolled at 400 ◦ C (nomi- magnetic energy densities that are dependent on the grain orien-
nal furnace temperature) from 15 mm to 1.2 mm by 7 passes tation with respect to the field direction. For general information
employing 30% thickness reduction per pass. Between the passes on the development of the level-set (LS) method as a determinis-
the rolled specimen was returned into the furnace for 10 min to tic approach for modeling grain growth, the reader is referred to
regain the rolling temperature. The final pass was followed by [32,35–39].
R. Pei et al. / Journal of Materials Science & Technology 50 (2020) 257–270 259

For convenience, the following is a brief reproduction of some between the more mobile high-angle and the virtually immobile
of the information provided in [31–33] regarding the model setup. low-angle boundaries.
Numerical details of the basic and new algorithm will not be   9 

  
addressed here. The LS method is a mathematical framework to   
m  = mmax 1-0.99exp −5  , for  ≤ 15◦ (4)
describe interfaces and their evolution with time. Grain growth is  15◦ 
treated in this method as a competing shrinkage/growth of adja-
cent grains driven by a minimization of grain boundary energy, In the absence of reliable mobility data for pure magnesium,
where the topological transition and the extinction of entire grains the mobility of general high-angle boundaries mmax was assigned
is automatically solved by means of Predictor-Corrector procedure. an estimated value of 4 · 10−14 m4 /Js for 0.53 Tm (220 ◦ C), and
This poses a true advantage over Vertex models since they require 4 · 10−13 m4 /Js for 0.67 Tm (350 ◦ C) on the basis of a study by
explicit instructions to solve such transformations. The computa- Okrutny et al. [41] dealing with modeling of static recrystallization
tional scheme of the basic LS algorithm consists of two main steps: in AZ31 magnesium alloy. There, the authors extracted mobility
values by fitting experimental grain growth kinetics obtained at
(i) Initialization of the input microstructure, where a signed distance different elevated temperatures to Hillert’s grain growth model.
function (SDF) is assigned to each grain. This function repre- Although we should be careful with the aforementioned mobil-
sents both the entire grain and the grain boundary during the ity values, we note that their magnitude is not likely to affect the
evolution algorithm. overall growth trends (abnormal vs. normal growth) but rather the
(ii) Computation of grain growth, where an iterative procedure of temporal rate of growth in the simulated microstructures.
three sequential steps is utilized to process the SDF of each With respect to Eqs. (3) and (4), it is noted that these general-
grain. (1) Predictor step that computes the motion of individual ized approaches do not capture any special conditions that apply,
grains by mean curvature flow of their interface. This motion for example, to coincident site lattice boundaries or to cases where
can be subsequently altered in order to consider interfacial Zener/solute drag necessitates the implementation of anisotropic
anisotropy effects. (2) Corrector step that restores the topo- high-angle boundary energy or mobility. Another limitation of the
logical integrity of the microstructure by removing voids and above equations is that they do not account for a possible depen-
overlaps between neighboring grains. (3) Reinitialization step, dency of the boundary energy and mobility on the grain boundary
where the distance functions resulting from the corrector step plane. Despite the fact that the model at our disposal is capable of
are computed to new interface positions. dealing with anisotropy effects to quantitatively model the rela-
tionship between grain growth and texture in magnesium alloys
The velocity (growth rate) of the migrating boundary at a specific (e.g. Basu et al. in [3]), it will be shown later that the simplified
time t was calculated as: assumptions regarding the mobility and energy of grain boundaries
dr work relatively well in reaching a satisfying agreement between
= = meff (k + pDDG ) , r = r0 for t = t0 (1) experimental microstructures and textures and their simulated
dt
counterparts.
where meff denotes the effective mobility that takes into account
the mobility of triple junctions in addition to the boundary mobil-
3. Results
ity, which can also influence the evolution of microstructure during
growth [32].  denotes the grain boundary energy, k the local cur-
3.1. Grain growth microstructures at 220 ◦ C and 350 ◦ C
vature of the grain boundary, pDDG the additional driving force
for grain growth resulting from orientation-dependent differences
The microstructures of pure Mg annealed at 220 ◦ C for different
in the residual dislocation density, remaining in grains after the
times are shown in Fig. 1. It is obvious that after 8 h of annealing
recrystallization annealing. Thus, during further annealing, recrys-
there are a few large grains surrounded by a much higher pro-
tallized grains with a lower residual dislocation density in their
portion of fine grains. Such configurations of large and fine grains
environment will tend to consume other grains with a higher dis-
remained in the microstructure also for increased annealing times.
location density in order to minimize the total free energy. r is the
Additional annealing for 7 days at the same temperature (220 ◦ C)
grain radius and r0 is the initial average grain size after recrystal-
was conducted to investigate whether AGG comes to completion or
lization. The dislocation density gradient (DDG) driving force was
whether a bimodal microstructure would still exist. Fig. 2 shows the
introduced as:
respective optical microstructure for a large representative surface
pDDG = 0.5Gb2  (2) area of 10 × 8 mm2 . As evident, there was still a bimodal distribu-
tion relating to two groups of rapidly and slowly growing grains.
where G is the shear modulus of pure Mg (16.7 GPa), b the Burg- The average size of small grains remained at ∼ 40 ␮m and that of
ers vector (0.321 nm for < a>-dislocations), and  the difference the very coarse grains at ∼ 400 ␮m, namely, ten times larger.
in the residual dislocation density of adjacent grains of a certain In comparison to recent reports on AGG in Mg alloys, the 220 ◦ C
grain boundary. The estimation of  was based on experimental temperature used in the present study was markedly lower than
EBSD misorientation maps providing information on the geomet- the temperatures used to trigger AGG in the literature [3,4,42].
rically necessary dislocation (GND) densities, and should therefore The results of Wu et al. [42] and Basu et al. [3] showed that AGG
be treated as an order of magnitude estimate, as discussed in Sec- was evident in a magnesium-rare earth (Mg-RE) alloy containing
tion 4.2. The grain boundary energies were assumed to depend on gadolinium, when deformed specimens were annealed at 450 ◦ C
the misorientation angle  between adjacent grains according to – 500 ◦ C for 60 min – 80 min, respectively. For commercial AZ31
the Read-Shockley model [40]. alloys, annealing temperatures between 300 ◦ C and 450 ◦ C are usu-
  
      ally sufficient to induce AGG, yet only if combined with much longer
  = max ◦ 1-ln  ◦  , for  ≤ 15◦
 (3) durations [4]. Therefore, in order to correlate the examined AGG
15 15
behavior in pure Mg with that reported for Mg alloys, an additional
The energy of high-angle grain boundaries ( ≥ 15◦ ) was con- set of annealing experiments was carried out at 350 ◦ C. Fig. 3 shows
sidered constant;  (15◦ ) = max = 0.5 J/m2 . Similarly, the mobility an optical image after annealing at 350 ◦ C for 7 days. The size of
of grain boundaries was considered to be exclusively a function of the observed area was 10 × 6 mm2 (Fig. 3(a)). It is evident from
the misorientation angle, where there was also a clear distinction the high magnification images (Fig. 3(b–d)), that the microstruc-
260 R. Pei et al. / Journal of Materials Science & Technology 50 (2020) 257–270

Fig. 1. Optical microstructures of rolled and recrystallized specimens annealed at 220 ◦ C for different times: (a) – (d) representative micrographs at low magnification; (e) –
(h) specific areas of interest at high magnification.

Fig. 2. Optical microstructure of the specimen annealed at 220 ◦ C for 7 days: (a) Panoramic image of the whole sample area (10 × 8 mm2 ) (RD-TD plane); (b) – (d) magnified
images of several outlined areas in (a) revealing more details with respect to a few abnormally grown grains with an overwhelming size advantage compared to the rest of
the microstructure.
R. Pei et al. / Journal of Materials Science & Technology 50 (2020) 257–270 261

Fig. 3. Optical microstructure of the specimen annealed at 350 ◦ C for 7 days (analogous to Fig. 2): (a) Panoramic image of the whole sample area (10 × 6 mm2 ) (RD-TD plane);
(b) – (d) magnified images of several outlined areas in (a) revealing a more uniform grain size distribution than that in Fig. 2, which is taken as an indication for the absence
of AGG at 350 ◦ C.

ture with respect to the size distribution was much more uniform misorientation axes in a single unit triangle and calculating
 the
in comparison to the 220 ◦ C case, i.e. there were no grains with an rotation angles between the most common 52−7−3 axis and all
overwhelming size advantage. This indicates that grain growth was other axes. Theresults are
 plotted in the bottom part of Fig. 4(c).
initiated globally, which led to NGG. As shown, the 52−7−3 misorientation axis was mostly related
to the other misorientation axes by rotations of ≤ 20◦ .
Analogous to the misorientation analysis of the abnormal grains
3.2. Grain boundary misorientation analysis A–H, fine grains within a chosen area in Fig. 4(b) were also studied
(685 grain boundaries). Similarly, the distribution of the sampled
To examine the role of boundary misorientation in the occur- misorientation axes was more or less uniform within the unit tri-
rence of AGG, EBSD measurements were carried out on the
specimen annealed at 220 ◦ C for 7 days. The results are shown
angle. Almost
 10% of the misorientation axes were clustered about
52−7−3 . Additionally, certain misorientation axes on the arc
in Fig. 4 in terms of inverse pole figures (IPF) and misori-    
between 11−20 and 10−10 were also common with frequen-
entation axis/angle distributions. As shown in Fig. 4(a), eight
cies between 5% and 8%. The misorientation angle distribution of
abnormally-grown grains were marked with letters A to H, and
the fine grains did not show a single distinct peak at 30◦ but rather a
their misorientation data with respect to their direct fine-grained
broad peak between 20◦ and 30◦ . In terms
 of special
 boundaries, the
neighbors (199 grain boundaries in total) was analyzed, as shown
misorientation relationship of 29.9◦ 11−20 found for an appre-
in Fig. 4(c). For visualizing the distribution of misorientations axes
ciable proportion of grain boundaries in Fig. 4 does correspond to
of the selected boundaries, the full circle of all twelve unit tri-
a 15a CSL boundary for c/a=1.620 [11,43]. As this is a minimum-
angles was used to quantify frequent misorientation axes that
energy, and not necessarily a high mobility boundary, its stability
could potentially identify special orientation relationships. The
during grain growth is reasonable.
corresponding colorbar represents the frequency of the plotted
misorientations. For example, if a specific misorientation axis
appears five times in the sampled data, it will be assigned the color 3.3. Texture analysis
‘yellow’. Misorientation axes appearing one time only will be ‘dark
blue’ according to the color bar. In addition to the misorientation It is reported in the literature that few grains with specific orien-
axis distribution the boundary misorientation angle distribution is tations can grow to abnormal sizes, as for example in a Fe-3%Si alloy.
also shown. From the data, approximately 10% (20out of 199)  of the In such material, grains with a GOSS orientation tend to grow faster
plotted misorientation axes corresponded to the 52−7−3 direc- than others [6,7,20–23]. In this work, to evaluate the orientation of
tion. The distribution of misorientation angles revealed a distinct abnormally large and small grains, the microstructure of the speci-
peak at 30◦ . Extended analysis included replotting all measured men annealed at 220 ◦ C for 7 days was divided into two grain groups
262 R. Pei et al. / Journal of Materials Science & Technology 50 (2020) 257–270

Fig. 4. Grain boundary misorientation analysis results of the microstructure annealed at 220 ◦ C for 7 days: (a) ND-IPF map of the sampled area containing both very large and
small grains (step size: 2 ␮m, area: 1600 × 1200 ␮m2 , mean MAD: 0.61◦ ). Blue and white boundaries denote high and low angle boundaries (<15◦ ); (b) selected area from
(a) with only small grains; (c) and (d) misorientation axis and angle distributions of grain boundaries of very large and small grains, respectively. The color coding denotes
the frequency of plotted misorientation axes in the unit triangles.

using a cutoff with respect to the average grain size in the analyzed between small grains) are separated by 30◦ , which correlates well
EBSD map (33.9 ␮m). When the grain area was more than four times with the findings in Fig. 4 with respect to the misorientation angle
larger than the average one, the respective grain was assumed to distribution. The resultcould also suggest that in this particular
be a large grain. The analysis is shown in Fig. 5(a) and (b) in terms case the (0001) 11−20 orientation bears a certain growth advan-
of IPF-coloring maps with reference to the rolling direction  (RD). tage among other basal-oriented grains. In order to validate such
Most of the very large grains were colored green ( 11−20 RD), growth advantage statistically, another two EBSD maps were per-
whereas the small grains
 were highlighted
 in both green and blue formed and altogether more than 8000 grains were indexed. The
exhibiting both 11−20 RD and 10−10 RD orientations. Cor- corresponding pole figures of large and small grains are shown in
respondingly, the basal and prismatic pole figures were plotted for Fig. 5(c) and (d), and intensities along the outmost circle (˛ = 90◦ )
both datasets to locate the position of maximum basal and pris- on (11 −2 0) plane with ˇ increased are plotted in Fig. 5(e). Both of
matic pole density. A comparison of the basal pole figures showed the pole figures and intensity  distributions confirmed the growth
that the basal poles were very closely aligned with ND but the tex- advantage of (0001) 11−20 orientated grains.
ture strength was different. Obviously, the few large grains dictated The macrotextures of the specimens annealed at 220 ◦ C and 350
the measured sharp basal texture. Interestingly, when compar- ◦ C for 16 h and 7 days are given in Fig. 6. In comparison to the fully

 polefigures (either one) the very large grains


ing the prismatic recrystallized (initial) state, the bulk texture did not change dur-
were of (0001) 11−20 orientation, and the small grains of (0001) ing the annealing treatments at 220 ◦ C regardless of the annealing
 
10−10 orientation. These two orientations (which coexist also duration (Fig. 6(a–c)). The growth advantage seen in the EBSD data
R. Pei et al. / Journal of Materials Science & Technology 50 (2020) 257–270 263

Fig. 5. Grain orientation analysis results (220 ◦ C / 7 days) for very large grains and small grains in terms of RD-IPF maps (a) and (b) along with the corresponding basal and
prismatic pole figures; (c) and (d) pole figures of large and small grains of more than 8000 grains gotten by multi-EBSD measurements; (e) intensity on (11 −2 0) plane along
the outmost circle (˛ = 90◦ ) with angle ˇ increased. Texture intensity is given in terms of multiples of a random distribution.

for the annealing condition of 220 ◦ C / 7 days was not visible in long times up to 7 days. Texture analysis revealedindications
 of
the XRD data. At 350 ◦ C and after 7 days ofannealing, there was an a certain growth advantage for basal grains with 11−20 || RD
obvious strengthening of a (0001) 11−20 type texture, where the orientation. Boundary misorientation analysis of a microstructure
 
11−20 directions became strongly aligned with the RD (Fig. 6(d)). comprising the two grain groups of the very large and the much

 a frequent
smaller ones revealed  misorientation angle of 30 and
misorientation axis 52−7−3 associated with the very large grains
4. Discussion and their direct neighbors. This section discusses the mechanisms
that could explain the rather unexpected occurrence of abnormal
To summarize the results reported above, a very different grain grain growth in deformed pure magnesium at certain annealing
growth behavior was observed upon annealing of hot rolled, com- conditions, where a size advantage would not be a sufficient rea-
mercially pure magnesium at 220 ◦ C (AGG) and 350 ◦ C (NGG) for son for the overwhelming growth of a few grains at the expense
264 R. Pei et al. / Journal of Materials Science & Technology 50 (2020) 257–270

Fig. 6. XRD bulk textures of selected specimens: (a) as-recrystallized; (b) annealed at 220 ◦ C for 16 h; (c) annealed at 220 ◦ C for 7 days; (d) annealed at 350 ◦ C for 7 days.
Texture intensity is given in terms of multiples of a random distribution.

   
of other much finer grains. The discussion presented is supported into either (0001) 11−20 or (0001) 10−10 groups based on
by level set computer simulations of grain growth, implementing the smallest rotation to reach either orientation. The correspond-
anisotropic microstructural conditions, such as different bound- ing size distributions in terms of grain number and area fractions
ary mobility, triple junction drag and different densities of stored
are plotted
 in Fig. 7(c)  and (d),
 respectively. Evidently, the (0001)
dislocations after recrystallization. 11−20 and (0001) 10−10 oriented grains exhibit almost iden-
tical distributions for the whole range of grain sizes measured. This

4.1. Possible mechanisms for the occurrence of AGG in pure is important
 because it proves that the larger grains in the (0001)
11−20 texture during AGG at 220 ◦ C (Fig. 5) do not originate from
magnesium

the recrystallized condition.  also applies to the global strength-
This
ening of the (0001) 11−20 component during NGG at 350 ◦ C,
The driving force for grain growth is the minimization of the
grain boundary energy, which greatly contributes to the reduc- 
which cannot be attributed to an initial size advantage in the (0001)
11−20 distribution present in the recrystallized condition.
tion of the total free energy of the system [1]. During grain growth,
From Eq. (1), the driving force for grain growth is directly pro-
grains merge in order to decrease the total grain boundary area. This
portional to the grain boundary energy and the boundary curvature.
process can give rise to an initial size advantage for some grains,
If the microstructure comprises a certain proportion of special low
so that they grow faster at the expense of smaller neighboring
energy boundaries, the characteristics of grain growth could exhibit
grains [44]. However, this cannot explain why  in some
 particular an anisotropic behavior. Although we have seen some indications of
cases only a few grains, mostly with (0001) 11−20 orientation,
CSL grain boundaries in the microstructures of our annealed spec-
were seen to preferentially grow to overwhelming sizes. The ini-
imens, we cannot claim with confidence that this mechanism is
tial recrystallized microstructures of such cases do not show any
the main driver for the AGG behavior reported here. Due to their
obvious peculiarity with respect to orientation, grain size or grain
lower energy, it is not unreasonable to assume that CSL boundaries
orientation spread (GOS) distributions. An example is shown in
can survive long duration annealing treatments, and thus remain
Fig. 7 of a sample subjected to recrystallization annealing at 200
◦ C for 1 h. The EBSD maps in Fig. 7(a) and (b) show the orientation
in the microstructure. From the perspective of anisotropic bound-
ary properties, different relative mobility values could potentially
of recrystallized grains with respect to RD, and the correspond-
lead to an abnormal growth behavior. When the movement of grain
ing GOS values, respectively. To quantitatively assess the role of
boundaries is restricted by some factors, like solute or Zener drag
grain size and size distribution in selecting texture components, the
resulting from boundary interaction with solute atoms or precipi-
recrystallized grains (GOS < 1) shown in Fig. 7(a) were categorized
R. Pei et al. / Journal of Materials Science & Technology 50 (2020) 257–270 265

 annealing
Fig. 7. Initial microstructure of the recrystallized condition at 200 ◦ C for 1 h prior to grain growth   presented
treatments  in terms of (a) RD-IPF EBSD map (step
size: 1.5 ␮m, area: 1800 × 1500 ␮m2 , mean MAD: 0.68◦ ); (b) GOS map up to 5◦ . (c) & (d) (0001) 11−20 and (0001) 10−10 size distributions of GOS <1◦ grains depicted
in terms of grain number and area fractions.

tates, AGG can occur. This was for example reported by Basu et al. [3] Fig. 8(b), where grain A grows at the expense of grain A* [47]. It was
who investigated the effect of solute segregation to grain bound- found that this mechanism necessitates the presence of some low
aries on the grain growth behavior and resulting weak textures energy boundaries, such as sub-grain boundaries, which enhance
in magnesium-rare earth alloys. They attributed the occurrence of solid-stated wetting conditions around triple junctions. An exam-
AGG in Mg-Gd alloy to the hypothesis that at a sufficient annealing ple of this is found for grains with GOSS orientation in silicon steel
temperature some grain boundaries are able to overcome the pin- [7,21,47].
ning effect and grow to larger sizes compared to other boundaries From the above, it is thus to be expected that as a consequence
that remain pinned. Having used pure magnesium in the current of grain growth driven by solid-state wetting, the fraction of sub-
study, the pinning mechanism does not offer a suitable explana- grain boundaries should increase in the annealed microstructures
tion for the occurrence of AGG. It is noted that the material still has [23,48]. However, as for our data, a comparison of the misorienta-
impurities but we believe their concentration is too low to induce tion angle distributions (␪ ≤ 15◦ ) for as-recrystallized specimens
significant pinning effects. After all, solute or Zener drag should (initial state) and those annealed at 220 ◦ C for 7 days (final state)
not be dependent on the  grain orientation
 since it was mostly the revealed no significant variation (Fig. 9). According to Shim et al.
grains with the (0001) 11−20 orientation that seemed to grow [20], low energy boundaries are frequently found in silicon steel
abnormally. annealed microstructures because the GOSS-oriented grains, which
Another mechanism that has been proposed to understand the typically have a lower stored energy after deformation, rather
occurrence of AGG in Fe-Si alloys is the so-called solid-state wet- recover than recrystallize. In our material, due to the small varia-
ting. The theoretical description of this mechanism can be found tion in the texture, it is not likely that some grain orientations only
in [45,46], whereas Fig. 8 schematically illustrates its basic prin- undergo recovery while other orientations undergo recrystalliza-
ciple. The numbers associated with the grain boundaries denote tion. This would eliminate the possibility of having a considerable
their energies. The sum of energy of grain boundary AA* and grain fraction of remnant sub-grain boundaries after recrystallization
boundary BA* is smaller than the energy of grain boundary AB so that is sufficient to trigger solid-state wetting during further grain
that the boundaries AA* and BA* tend to expand at the expense of growth annealing treatments.
the boundary AB. This means that grain A* will grow and eventu- From another perspective, it might be possible that  the differ-

ally consume grain A, as illustrated in Fig. 8(a) [47]. Based on the ent fiber components of the basal texture, i.e. (0001) 11−20 and
 
same principle of energy minimization, a different case is shown, in (0001) 10−10 , have some variations in their remaining stored
266 R. Pei et al. / Journal of Materials Science & Technology 50 (2020) 257–270

Fig. 8. Schematic illustration of the solid-state wetting mechanism of grain growth in silicon steel reproduced from [47]. The numbers at the grain boundaries denote
examples of their energies.

Fig. 9. Comparison of the fraction of low angle boundaries (up to 15◦ ) present in (a) the initial recrystallized condition, and (b) in the annealed condition at 220 ◦ C for 7 days.

energy upon recrystallization (even though it is much lower) that low grain boundary mobility and high  triple  junction drag at the
could influence their grain growth behavior based on the DDG this temperature. Eventually, such 11−20 grains would evolve
mechanism described earlier in the context of possible AGG drivers. into abnormally grown grains with much larger sizes than the sur-
 Zhu etal. reported
In fact, in a relevant study of a Zr-2Hf alloy [12], rounding grains. In this scenario, AGG will cease when the very
a favorable increase in the intensity of the 11−20 fiber over large grains come to impingement. Note that in the experiments
 
the 10−10 fiber during primary recrystallization and normal reported above, the corresponding XRD bulk texture did not change
grain growth. In the current case, when recrystallized specimens too much in comparison to the recrystallized texture (Fig. 6(b) and
are annealed at low temperature (220 ◦ C), stored dislocations are (c)), which can be related to the small fraction of abnormally grown
grains in the annealed microstructure at 220 ◦ C. It is noted that the
not likely to have enough energy to move and are thus  trapped

inside grains. From the discussion above, grains with 11−20 || average grain size in this case remained below 50 ␮m.
  On the other hand, when the specimens are annealed at a higher
RD and 10−10 || RD orientations are likely to exhibit a gradi-
temperature (350 ◦ C), dislocations trapped inside grains can be
ent in their residual dislocation density,producing
 a driving force expected to be more mobile and thereby be easily absorbed by their
for a selective grain growth to consume 10−10 grains and elim-
nearest grain boundaries decreasing the anisotropic nature of grain
inate their higher dislocation density. Moreover, in addition to
growth arising from DDG. In addition, thermal activation by rais-
the DDGinduced driving force, a resulting grain size advantage of ing the temperature is also expected to drastically decrease the
11−20 grains could also accelerate their further growth, while
triple junction drag [49] and increase the grain boundary mobil-
the other grains remain comparably small because of relatively
R. Pei et al. / Journal of Materials Science & Technology 50 (2020) 257–270 267

Fig. 10. Schematic illustration for abnormal and normal grain growth occurring in annealed pure magnesium at 220 ◦ C and 350 ◦ C, respectively. The length and the thickness
of the arrowsrepresent
 the magnitude of the impact of boundary mobility and DDG at each temperature. The number of G2 grains was restricted to those with almost an
ideal (0001) 11−20 orientation (c.f. Section 4.2).

ity. As aresult, the difference in the growth rate between (0001) higher residual stored energy than recrystallized grains. Among
11−20 grains and their neighbors should decrease, preventing the group of recrystallized
 grains we further distinguished between
the occurrence of AGG. With respect to texture evolution, there was (0001) 11−20 grains (G2) having a competitive advantage dur-
 basal texture and a
a significant strengthening of the recrystallized ing growthand all other
 recrystallized grains (G3). The selection of
clear transition from a basal fiber to a (0001) 11−20 component the (0001) 11−20 grains was based on a criterion of maximum 5◦

(Fig. 6(d)). This is likely to emanate from a remaining DDG effect angle between the c-axis and ND and maximum 3 angle between
that provides grains of that orientation with an advantage during the 11−20 axis and RD.
growth. Fig. 10 provides a schematic illustration of the two cases The next step in the approach was to derive a rough estima-
discussed above with the conditions that lead to AGG and NGG at tion of the dislocation density of the aforementioned three grain
220 ◦ C and 350 ◦ C, respectively. groups. Stored dislocations in the material after annealing consist
of geometrically necessary dislocations (GNDs), which are typically
immobile, and therefore less prone to rearrangement and annihi-
4.2. Level-Set computer simulations of grain growth lation events during recovery [50]. Unlike GNDs, which are formed
to account for lattice curvature during deformation, statistically
The hypothesized roles of local dislocation density, triple junc- stored dislocations (SSDs) originating from statistical dislocation
tion drag and grain boundary mobility for the specific texture and entanglements [51] can undergo recovery resulting in reduction of
grain size distributions found in the experiments, can be inves- their density in the annealed microstructure [1]. In this work, the
tigated further by Level-Set simulations that allow systematic kernel average misorientation (KAM) function was used to esti-
variation of each parameter. mate the GND densities for G1, G2 and G3 grains. Details of this
Initial microstructures for the simulation were obtained by EBSD calculation method can be found in [52,53]. It is noted that the stan-
measurements after recrystallization annealing at 200 ◦ C for 1 h dard EBSD employed for the local orientation measurements has
(Fig. 7). As shown in Fig. 7, the recrystallized microstructure can an angular error of 0.5◦ -1◦ , which is thought to result in artificially
be firstly divided into two grain groups based on their GOS values. high GND densities, particularly in a recrystallized material with a
Grains with GOS>1◦ (group1, hereafter G1) are assumed to be recov- low total dislocation density. Our calculations, which are therefore
ered, whereas grains with GOS<1◦ are considered recrystallized. likely to be an overestimate, showed that the average GND densities
Although we did not employ high-resolution diffraction measure- in the aforementioned three groups of grains were 1 × 1013 /m2 , 6
ments to estimate the stored energy in different grain groups, × 1012 /m2 and 9 × 1012 /m2 , respectively. Since it was not possible
we can reasonably assume that G1 grains are associated with a
268 R. Pei et al. / Journal of Materials Science & Technology 50 (2020) 257–270

Fig. 11. Simulated grain growth microstructures and their corresponding textures at different time increments matching the experimental annealing times employed. (a, f)
input EBSD microstructure; (b - e) anomalous grain growth behavior equivalent to annealing at 220 ◦ C; (g – j) uniform grain growth equivalent to annealing at 350 ◦ C. The
color of grain boundaries denotes their relative mobility (red is maximum, purple is minimum).

to know the relation between the GND density and total disloca- 350 ◦ C. In Fig. 11(b)–(e) some grains start to develop a significant
tion density, the latter was assumed to be 10 times larger than the size advantage and proceed by eclipsing their neighboring grains,
estimated GND density at 220 ◦ C. Hence, for the simulations at 220 resulting in a heterogeneous microstructure with very large and
◦ C, the total dislocation density in G1, G2 and G3 was taken as 1.0 × very small grains. In Fig. 11(g)–(j) the simulations show a uniform
1014 /m2 , 6 × 1013 /m2 and 9 × 1013 /m2 , respectively. To incorpo- growth behavior with increasing annealing time. With respect to
rate the role of temperature in the simulations, the values assigned texture, the simulated microstructures at different time steps were
to the dislocation densities, triple junction drag and grain bound- exported back into the file format of the EBSD data used to pop-
ary mobility were adjusted according to the trends shown in Fig. 10. ulate the simulations, which allowed for further texture
analysis
For the 350 ◦ C case, the density of residual stored dislocations was using the MTEX software. The reproduced basal and 11−20 pris-
one tenth of that used in the simulations of the 220 ◦ C case, which  figures
matic pole  in Fig. 11 show a gradual strengthening of the
should be treated as an order of magnitude estimate. The bound- (0001) 11−20 component with increasing annealing time. This
ary mobility was increased by a factor of 10 and the triple junction qualitative trend is in line with the experimental XRD and EBSD
drag effect decreased by a factor of 1.3. These values showed a good measurements (Figs. 5 and 6). In terms of a quantitative agreement,
qualitative agreement with the target textures and microstructures it was difficult to obtain comparable texture intensities due to the
(grain size distributions) obtained from experiments. limited number of remaining grains in the simulations at advanced
Grain growth simulation results of evolved microstructure and time steps, particularly in the case of NGG that would of course lead
texture at 220 ◦ C and 350 ◦ C are displayed in Fig. 11 for different to overstrengthening of the texture.
time steps matching the experimental annealing times based upon Fig. 12 presents a quantitative analysis of the simulation results
a similar average grain size to the one obtained in the experiments. shown in Fig. 11. Fig. 12(a) shows a comparison of the temporal
The different colors assigned to grain boundaries denote their rel- grain size evolution for the two cases corresponding to 220 ◦ C and
ative mobility, with ‘red’ corresponding to maximum and ‘purple’ 350 ◦ C. For determining the grain growth behavior, i.e. distinguish-
to minimum levels. The input EBSD microstructure with its unique ing normal from abnormal growth, different criteria can be used.
set of measured Euler angles is shown in Fig. 11(a) and (f). In both The most popular ones are based upon appearance of a bimodal
cases of grain growth, the level set method, with the assumptions grain size distribution or variation in the breadth of the normalized
made for each regarding the dislocation density gradient, triple grain size distribution (loss of self-similarity) [2,4]. However, due
junction drag and boundary mobility was able to reproduce a very to the limited number of grains in the present simulations, we used
similar growth behavior observed experimentally at 220 ◦ C and another method instead, which examines the ratio D/<d>, where D
R. Pei et al. / Journal of Materials Science & Technology 50 (2020) 257–270 269

Fig. 12. Quantitative analysis of the simulated growth behavior shown in Fig. 11. (a) temporal grain size evolution for the two cases corresponding to 220 ◦ C and 350 ◦ C; (b)
variation of D/<d> with time; (c) variation of the derivative of D/<d> with time. D is the diameter of the largest grain and <d> is the average grain diameter of all grains.

is the diameter of the largest grain and <d> is the average grain stored in grains after recrystallization. Other potential mecha-
diameter of all grains [5,21,47,54–56]. Fig. 12(b) shows the tempo- nisms, such as special boundary misorientation relationships,
ral evolution of D/<d> for the two cases resembling annealing at 220 initial size advantage emerging from recrystallization, solid-state
◦ C and 350 ◦ C. During the first 50 min, the initial D/<d> ratio does wetting effects were discussed and considered less likely to play
not exhibit a strong variation, which is thought to be a period in an important role in the present case.
which the group of large recovered grains (G1) are eaten up by the  force
(4) The driving  induced by a dislocation density gradient grants
recrystallized grains due to their different stored dislocation den- (0001) 11−20 grains a size advantage during early stages of
sities. This process would slightly decrease the D/<d> ratio. After growth. The type of growth will be, however, dictated by the
50 min of annealing (the beginning of the second period) the two present grain boundary mobility and triple junction drag, which
cases reveal a completely different evolution of D/<d> with increas- are highly dependent on the annealing temperature.
ing annealing time. For the 220 ◦ C case, the D/<d> ratio increases 
(5) For annealing 
conditions where the temperature is low enough,
up to 12, while the D/<d> ratio for the 350 ◦ C case drops to 4. A (0001) 11−20 grains would thus have a higher chance of rapid
large D/<d> ratio indicates presence of island grains with a grain growth and becoming abnormally large, while most of the other
size D many times larger than the average grain size <d>, which is grains grow at a much slower rate and remain comparably small.
in turn indicative of anomalous growth. Thus, analogously, a small This is due to a generally low boundary mobility and high triple
ratio of D/<d> is characteristic for normal grain growth. For vali- junction drag at low temperatures, where abnormal grain growth
dation purposes, the experimental D/<d> data at 220 ◦ C was also occurs because normal grain growth is globally restricted. At a
plotted in Fig. 12(b), which reveals a similar trend to attain high much higher annealing temperature, the anisotropic nature of
values during period 2 annealing that spans between 50 min and 7 grain growth arising from orientation-dependent density gradi-
days. In a similar fashion to Fig. 12(b), (c) now presents a compari- ents will be extenuated because the residual dislocations inside
son of the derivative of D/<d>. This analysis considers the changes in grains are more likely to be mobile and be absorbed by their sur-
D/<d> that take place during successive time increments. A positive rounding grain boundaries. Additionally, growth conditions will
derivative of D/<d> is used as in indicator of AGG, which is observed be globally enhanced by an increase in the average boundary
for the simulated and experimentally obtained data at 220 ◦ C (inset mobility and a reduction of triple junction drag. This will lead
in Fig. 12(c)). The inflection point in the positive regime of AGG can to normal grain growth.
be interpreted as the impingement of the anomalously large grains (6) The above hypothesis of the interplay between dislocation
and resumption of NGG. Eventually, a steady state is reached after density gradient, boundary mobility, triple junction drag and
very long annealing times, where D/<d> becomes constant. annealing temperature was explored through level set model-
ing of grain growth incorporating those aspects. The simulation
5. Conclusions outcome reproduced the same trends for the normal and abnor-

mal grain growth occurring at the two investigated temperatures, 
In this work, commercial purity Mg was hot rolled and subse- including the characteristic strengthening of the (0001) 11−20
quently annealed at different temperatures for different times in component. This suggests that application of this modeling
order to investigate its normal and abnormal grain growth behav- approach in microstructure studies of magnesium alloys can pro-
ior and link it to the corresponding texture evolution. The following vide valuable new insights into the problem of grain growth and
main conclusions are parsed from the findings obtained by exper- associated texture evolution.
imental microstructure characterization and level set computer
simulations of grain growth. Acknowledgement

(1) Annealing at an intermediate temperature of 220 ◦ C gave rise to R.S Pei is grateful for financial support from the Chinese Schol-
abnormal grain growth with a few grains having a grain diam- arship Council (CSC).
eter 10 times larger than the average. Increasing the annealing
temperature to 350 ◦ C yielded normal grain growth.
(2) 
For both types of grain growth, strengthening of the (0001) References
11−20 texture component was coupled with a competitive
[1] F.J. Humphreys, M. Hatherly, Recrystallization and Related Annealing
growth advantage for grains with that orientation.   Phenomena, 2nd ed., Pergamon Press, Oxford, 2004.
(3) It is hypothesized that the fast growth rate of (0001) 11−20 [2] G. Gottstein, Physical Foundations of Materials Science, Springer-Verlag,
grains is a result of gradients in the residual dislocation density Berlin Heidelberg, 2004.
270 R. Pei et al. / Journal of Materials Science & Technology 50 (2020) 257–270

[3] I. Basu, K.G. Pradeep, C. Mießen, L.A. Barrales-Mora, T. Al-Samman, Acta [30] F. Bachmann, R. Hielscher, P.E. Jupp, W. Pantleon, H. Schaeben, E. Wegert, J.
Mater. 116 (2016) 77–94. Appl. Crystallogr. 43 (2010) 1338–1355.
[4] J.J. Bhattacharyya, S.R. Agnew, G. Muralidharan, Acta Mater. 86 (2015) 80–94. [31] M. Kühbach, C. Mießen, L.A. Barrales-Mora, G. Gottstein, Proceedings of the
[5] M. Shirdel, H. Mirzadeh, M.H. Parsa, Mater. Charact. 97 (2014) 11–17. 6th International Conference on Recrystallization and Grain Growth (REX&GG
[6] Z.Q. Liu, P. Yang, W.M. Mao, F.E. Cui, Acta Metall. Sin. 51 (2015) 769–776. 2016), John Wiley & Sons, Inc., 2016, pp. 35–42.
[7] Y. Wang, Y.B. Xu, Y.X. Zhang, S.Q. Xie, Y.M. Yu, G.D. Wang, Mater. Res. Bull. 69 [32] C. Mießen, M. Liesenjohann, L.A. Barrales-Mora, L.S. Shvindlerman, G.
(2015) 138–141. Gottstein, Acta Mater. 99 (2015) 39–48.
[8] N. Bozzolo, N. Dewobroto, T. Grosdidier, E. Wagner, Mater. Sci. Eng. A 397 [33] C. Mießen, N. Velinov, G. Gottstein, L.A. Barrales-Mora,Modell. Simul. Mater.
(2005) 346–355. Sci. Eng. 25 (2017), 084002.
[9] N. Takata, F. Yoshida, K. Ikeda, H. Nakashima, H. Abe, Mater. Trans. 46 (2005) [34] C. Mießen, A Simulation Software for Anisotropic Grain Growth in 2D, on,
2975–2980. 2017 https://github.com/GraGLeS/GraGLeS2D.
[10] Y.L. Wang, G.J. Huang, X. Li, Q. Liu, Optoelectron. Adv. Mater. 9 (2015) [35] M. Elsey, S. Esedoglu, P. Smereka, J. Comput. Phys. 228 (2009) 8015–8033.
696–703. [36] M. Elsey, S. Esedoglu, P. Smereka, Acta Mater. 61 (2013) 2033–2043.
[11] A. Ostapovets, P. Molnar, P. Lejcek, Mater. Lett. 137 (2014) 102–105. [37] B. Scholtes, M. Shakoor, A. Settefrati, P.O. Bouchard, N. Bozzolo, M. Bernacki,
[12] K.Y. Zhu, B. Bacroix, T. Chauveau, D. Chaubet, O. Castelnau, Metall. Mater. Comput. Mater. Sci. 109 (2015) 388–398.
Trans. A 40A (2009) 2423–2434. [38] B. Scholtes, R. Boulais-Sinou, A. Settefrati, D. Pino Muñoz, I. Poitrault, A.
[13] M. Hillert, Acta Metall. 13 (1965) 227-&. Montouchet, N. Bozzolo, M. Bernacki, Comput. Mater. Sci. 122 (2016) 57–71.
[14] P.R. Rios, Acta Mater. 45 (1997) 1785–1789. [39] H. Hallberg,Modell. Simul. Mater. Sci. Eng. 21 (2013), 085012.
[15] P.R. Rios, Acta Mater. 42 (1994) 839–843. [40] W.T. Read, W. Shockley, Phys. Rev. 78 (1950) 275–289.
[16] P.R. Rios, Scripta Mater. 38 (1998) 1359–1364. [41] P. Okrutny, S. Liang, L. Meng, H. Zurob, Magnesium Technology 2014, Springer
[17] N.C. Pease, D.W. Jones, M.H.L. Wise, W.B. Hutchinson, Met. Sci. 15 (1981) International Publishing, Cham, 2016, pp. 149–153.
203–209. [42] W.X. Wu, L. Jin, Z.Y. Zhang, W.J. Ding, J. Dong, J. Alloys Compd. 585 (2014)
[18] P. Lin, G. Palumbo, J. Harase, K.T. Aust, Acta Mater. 44 (1996) 111–119.
4677–4683. [43] G. Ibe, K. Lücke, Archiv für das Eisenhüttenwesen 39 (1968) 693–703.
[19] N. Chen, S. Zaefferer, L. Lahn, K. Gunther, D. Raabe, Acta Mater. 51 (2003) [44] M.A. Steiner, J.J. Bhattacharyya, S.R. Agnew, Acta Mater. 95 (2015) 443–455.
1755–1765. [45] H.K. Park, H.G. Kang, C.S. Park, M.Y. Huh, N.M. Hwang, Metall. Mater. Trans. A
[20] H.S. Shim, N.M. Hwang, Korean J. Met. Mater. 52 (2014) 663–687. 43A (2012) 5218–5223.
[21] K.J. Ko, A.D. Rollett, N.M. Hwang, Acta Mater. 58 (2010) 4414–4423. [46] N.M. Hwang, S.B. Lee, D.Y. Kim, Scripta Mater. 44 (2001) 1153–1160.
[22] H.K. Park, S.D. Kim, S.C. Park, J.T. Park, N.M. Hwang, Scripta Mater. 62 (2010) [47] K.J. Ko, P.R. Cha, D. Srolovitz, N.M. Hwang, Acta Mater. 57 (2009) 838–845.
376–378. [48] C.S. Park, T.W. Na, H.K. Park, D.K. Kim, C.H. Han, N.M. Hwang, Philos. Mag. Lett.
[23] H. Park, D.Y. Kim, N.M. Hwang, Y.C. Joo, C.H. Han, J.K. Kim, J. Appl. Phys. 95 92 (2012) 344–351.
(2004) 5515–5521. [49] D. Mattissen, D.A. Molodov, L.S. Shvindlerman, G. Gottstein, Acta Mater. 53
[24] G. Abrivard, E.P. Busso, S. Forest, B. Appolaire, Philos. Mag. 92 (2012) (2005) 2049–2057.
3618–3642. [50] M.F. Ashby, Phil. Mag. A 21 (1970) 399–424.
[25] G. Abrivard, E.P. Busso, S. Forest, B. Appolaire, Philos. Mag. 92 (2012) [51] U.F. Kocks, Phil. Mag. A 13 (1966) 541–566.
3643–3664. [52] S.I. Wright, M.M. Nowell, D.P. Field, Microsc. Microanal. 17 (2011)
[26] A.K. Lawrence, A. Kundu, M.P. Harmer, C. Compson, J. Atria, M. Spreij, J. Am. 316–329.
Ceram. Soc. 98 (2015) 1347–1355. [53] M. Kamaya, Ultramicroscopy 111 (2011) 1189–1199.
[27] P.R. Cantwell, M. Tang, S.J. Dillon, J. Luo, G.S. Rohrer, M.P. Harmer, Acta Mater. [54] M. Shirdel, H. Mirzadeh, M.H. Parsa, Metall. Mater. Trans. A 45A (2014)
62 (2014) 1–48. 5185–5193.
[28] S.J. Dillon, M.P. Harmer, Acta Mater. 55 (2007) 5247–5254. [55] P.R. Rios, Acta Metall. Mater. 40 (1992) 2765–2768.
[29] S.J. Dillon, M. Tang, W.C. Carter, M.P. Harmer, Acta Mater. 55 (2007) [56] A.D. Rollett, D.J. Srolovitz, M.P. Anderson, Acta Metall. 37 (1989)
6208–6218. 1227–1240.

You might also like