Available online at www.sciencedirect.

com

ScienceDirect
Acta Materialia 69 (2014) 236–245
www.elsevier.com/locate/actamat

In situ electron microscopy investigation of void healing

in an Al–Mg–Er alloy at a low temperature
M. Song a, K. Du a,⇑, S.P. Wen b, Z.R. Nie b, H.Q. Ye a
a
Shenyang National Laboratory for Materials Science, Institute of Metal Research, Chinese Academy of Sciences, Shenyang 110016, People’s Republic
of China
b
School of Materials Science and Engineering, Beijing University of Technology, Beijing 100022, People’s Republic of China

Received 25 September 2013; received in revised form 10 January 2014; accepted 2 February 2014
Available online 28 February 2014

Abstract

Using in situ transmission electron microscopy and electron tomography, we have studied the healing process of submicron-scale
voids embedded in a cold-rolled Al–Mg–Er alloy. The results show that voids are healed successfully within 50 min at a relatively
low temperature of 453 K. Quantitative analysis of the in situ micrographs reveals that the void-healing process involves several stages:
an initial fast-healing stage, then a slow-healing stage, and finally a rapid healing near the end. The different healing rates are likely
related to varying surface curvatures due to the evolution of void geometry during the healing process. Because the voids are embedded
inside Al alloy grains, lattice diffusion is considered to dominate the entire healing process. Mg enrichment was observed at the healed
voids immediately after the healing. This indicates that the faster diffusion of Mg atoms in the Al matrix enhances void healing in the
Al–Mg–Er alloy. Post-mortem experiments verify that voids in bulk cold-rolled Al–Mg–Er alloy can also be healed by 1 h of annealing at
473 K, consistent with the in situ experiments. The fatigue crack growth resistance and plasticity of the cold-rolled Al alloy are improved
significantly after annealing at 473 K.
Ó 2014 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.

Keywords: In situ; Al–Mg–Er alloy; Void healing; Transmission electron microscopy (TEM); Electron tomography

1. Introduction materials properties and help develop materials processing

Extensive studies [1–3] have revealed that void defects
usually cause stress concentration and crack nucleation
around them under external loadings, which would greatly
shorten the incubation period of crack initiation. Voids
also have adverse effects on fatigue crack propagation
and crack closure, whereas most mechanical fractures in
engineering are reportedly associated with fatigue crack
initiation and growth in metals [4]. Therefore, investiga-
tions on the evolution of voids in materials have attracted
significant interest [5–8], and these investigations will facil-
itate an understanding of the recovery mechanism of
⇑ Corresponding author. Tel.: +86 24 8397 0725; fax: +86 24 2389 1320.
E-mail address: kuidu@imr.ac.cn (K. Du).
techniques.
In recent years, studies on void healing have reported
significant progress, particularly in the areas of theoretical
analysis and simulation. Key issues with void healing
involve the provision of material and energy. The main
sources of energy include heating [9], electropulsing [10]
and external loading [11]. The sources of material primarily
involve surface [12], lattice [13] and interface diffusion [14].
Simulations of pure metals have demonstrated that temper-
ature has a significant effect on the process of void healing.
When the temperature is reduced, the void size reduction
rate decreases dramatically. Only for pure metals without
pre-existing dislocations can voids be healed at relatively
low temperatures, such as 400 K for Cu nanoplates [15]
and 673 K for body-centered cubic Fe crystals [16]. Based

http://dx.doi.org/10.1016/j.actamat.2014.02.004
1359-6454/Ó 2014 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
 M. Song et al. / Acta Materialia 69 (2014) 236–245
on the difference in chemical potentials between a spherical
void surface and the grain boundary interface, Sun [13] has
simulated the shrinkage of a spherical void at the center of
a quasi-spherical grain dominated by lattice self-diffusion
in pure Fe at high temperatures ranging from 973 to
1173 K for a-Fe and from 1200 to 1400 K for c-Fe and pre-
dicted that only small, micron-sized spherical voids could
be wholly healed within hours at high temperature. A large
number of experiments on inorganic non-metallic materials
have shown that void healing can occur under annealing at
high temperatures, ranging from 1073 to 1723 K [17–20]. In
contrast, a relatively small number of experiments have
been performed on void healing in metals or alloys, and
these mainly involve post-mortem observations at high
temperatures [21]. Takahashi et al. [22] have analyzed the
shrinkage process of artificially introduced microscale
cylindrical holes in Cu at high temperatures by post-mor-
tem scanning electron microscopy (SEM) observations.
They investigated void junctions by SEM and concluded
that voids with different morphologies could be predicted
by suitable models. Nevertheless, because of the small size
(<10 lm) of voids and the surface imaging nature of SEM
techniques, it is difficult to observe the void-shrinkage pro-
cess directly by SEM. Therefore, few experimental in situ
studies on void healing have been reported. In this study,
in situ heating transmission electron microscopy (TEM)
has been employed to study the void-healing process in a
cold-rolled Al–Mg–Er alloy [23] at low temperatures. Via
in situ heating at 453 K we have observed the healing pro-
cess of voids embedded in an Al–Mg–Er alloy in real time
and analyzed the healing kinetics accordingly.
2. Experimental
2.1. Sample preparation
The experimental Al–Mg–Er alloys were produced by
the semicontinuous chill-casting method. The as-cast alloys
were homogenized at 743 K for 24 h to eliminate interden-
dritic segregation, followed by hot-rolling at 743 K, anneal-
ing at 453 K for 2 h and cold-rolling into sheets of 6 mm
thickness. The thickness reductions brought about by the
hot-rolling and cold-rolling processes were 92% and 67%,
respectively. The chemical composition of the alloy was
analyzed by inductively coupled plasma spectroscopy as
5.80% Mg, 0.67% Mn, 0.31% Er, 0.10% Zr (wt.%), and bal-
ance was Al. To compare with in situ heating TEM sam-
ples, annealed (CA) Al alloy samples were prepared from
the cold-rolled (CR) alloy by annealing at 473 K for differ-
ent durations (10, 20, 30, 40, 50 and 60 min, respectively).
2.2. Performance tests
All mechanical tests of the CR and the CA alloys were
carried out under the same experimental conditions in an
ambient environment. The tensile and fatigue specimens
were machined parallel to the rolling direction. Tensile
237
properties were determined with a loading rate of
2 mm min1. Fatigue crack growth rates were measured
by using standard center crack tensile (CCT) samples.
The test dimensions and notch lengths of the CCT samples
are 225 mm  60 mm  6 mm and 1.5 mm, respectively. A
sinusoidal load–time function was adopted, with the
stress ratio R set to 0.1 and the stress in the interval of
1.6  104–2.0  104 N. The oscillation frequency was
15 Hz.
2.3. Microstructure characterization
In situ heating experiments were conducted on a
JEM2100 transmission electron microscope with a Gatan
model 628 single-tilt heating holder at an accelerating volt-
age of 200 kV. A Tecnai G2 F30 transmission electron
microscope, equipped with a high-angle annular-dark-field
(HAADF) detector and an X-ray energy-dispersive spec-
trometry (EDS) system, was used at 300 kV for bright-field
imaging, HAADF scanning transmission electron micros-
copy (STEM) imaging, and composition analysis. The elec-
tron tomography was performed on the Tecnai G2 F30
equipped with a Fishione 2040 tomography sample holder,
and the data was acquired using Inspect3D software for
STEM tomography. 3-D reconstruction and volume analy-
sis were performed with Amira 5.0.1 software (Visage Imag-
ing, Berlin). During the experiment, the sample was tilted
from –60° to +60° in steps of 2°. TEM thin-foil specimens
parallel to the rolling plane were prepared by cutting, grind-
ing, polishing and ion-milling. A SUPRA35 field emission
scanning electron microscope, operated at 20 kV, was used
to investigate the fracture surface of fatigue cracks.
3. Results
3.1. Microstructure of the cold-rolled Al alloy
Bright-field TEM images show the distribution of gray
lath-like precipitates and “white features” between them
in the present cold-rolled Al–Mg–Er alloy (Fig. 1a). Based
on a large number of EDS analyses (more than 20), the
chemical composition of the lath-like precipitates was
revealed to be Al and Mn with an atomic percentage ratio
of Al:Mn of approximately 6:1. Hence, the lath-like precip-
itates are presumably Al6Mn and will be described as such
hereinafter. Between the Al6Mn precipitates, the “white
features” have a projected dimension of about
400 nm  50 nm. A large number of TEM images revealed
that the “white features” were located only between Al6Mn
precipitates and had projected dimensions of the order of
tens of nanometers to submicrometers. HAADF STEM
imaging shows that the “white features” in bright-field
TEM images turn into areas of dark contrast in HAADF
STEM images (Fig. 1b). Fig. 1c shows the intensity profile
between the two arrows in the HAADF STEM image
(Fig. 1b). The intensity profile shows a “deep well” charac-
teristic, as distinguished from a “shallow dish” feature. In
238 M. Song et al. / Acta Materialia 69 (2014) 236–245

Fig. 1. Microstructure of the CR alloy. (a) A bright-field TEM image of the CR alloy showing the gray lath-like precipitates and “white features” between
them. (b) An HAADF STEM image of the CR alloy showing bright-contrast Al6Mn precipitates and voids between them. (c) Image intensity profile
between the two arrows in (b).

addition, the EDS counts generated from the “white fea-
tures” are far below those of the surrounding Al matrix.
Based on the above observations, the “white features”
should correspond to voids embedded in the CR alloy.
Because Al6Mn precipitates are nondeformable particles
[24], which are much harder than the surrounding Al
matrix [25,26], it is easy for Al6Mn precipitates to be bro-
ken during the cold-rolling deformation. When the broken
fragments move along with the Al matrix, voids would be
formed between them [27].
The 3-D geometry of voids was determined by electron
tomography [28] (Fig. 2). Fig. 2c shows 3-D reconstructed
surface renders of the void (gold) and the Al6Mn precipi-
tates (purple) on two sides of the void; everything is embed-
ded in an electron-transparent Al matrix. Fig. 2d–f shows
surface renders of the void viewed from different directions
by rotating the sample (see also Supplementary Movie S1).
When the tilting angle is 0°, the projected void shows a
dog-bone shape with a length of 180 nm and a projected
width varying from 30 nm at the center to 60 nm at the
ends (Fig. 2d). With an increase in the tilting angle, the pro-
jected width increases up to 110–180 nm (Fig. 2f). It is clear
from the 3-D surface renders that there are a considerable
number of surface sites with a small curvature radius,
which is related to void healing, as discussed below.
3.2. In situ heating experiment
To investigate the evolution of voids embedded in the
CR alloy during an annealing treatment, in situ heating

Fig. 2. HAADF STEM images and surface renders of a void between Al6Mn precipitates embedded in the Al matrix. (a) HAADF STEM image of the
void with 0° tilting. (b) HAADF STEM image of the void with 50° tilting by rotating clockwise around the arrow indicated in (a). (c) Electron
tomography image of the void (gold) and the Al6Mn precipitates (purple) at two sides of the void; everything is embedded in an electron-transparent Al
matrix. (d–f) Surface renders of the void, obtained by electron tomography, shown by rotating counterclockwise around the x-axis indicated in (c). (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
 M. Song et al. / Acta Materialia 69 (2014) 236–245
TEM has been employed to study the void-healing process.
Voids embedded inside Al grains that were 2–3 lm away
from the sample perforation were chosen for the in situ
heating experiment at 453 K so that this experiment could
reflect the void evolution in the bulk material. Fig. 3 shows
snapshots of a void during the in situ heating. During the
initial stage of this experiment, it is obvious that voids heal
quickly at sites with a smaller surface curvature radius
(Fig. 3a and b). During this stage, small isolated voids
are formed at the necking region of the original voids
and they disappear after a short time (Fig. 3a–c), behavior
that is similar to the characteristics described by Gupta for
the observation of cylindrical voids in polycrystalline
alumina [29]. Then, the healing rate slows down. After
the voids reach a certain size, the speed of healing
increases, and the voids are healed instantaneously
(Fig. 3e–h, see also Supplementary Movie S2).
3.3. Variation of the projected area and volume of voids with
the annealing time
Due to necking, the void in Fig. 3a was divided into two
parts, A and B (see also Fig. 4a). For each part, Fig. 4a
shows how the projected areas A of voids A and B vary
with the heating time. The curves reveal that the void-heal-
ing rate varies significantly in different periods. At the
beginning of the healing process, the healing rate is high;
then it slows down; but, towards the end of healing, the
healing rate increases significantly. The evolution of void
projected area with heating duration was also measured
for 20 voids by in situ heating experiments. Table 1 shows
that the initial size (the mean projected area is 2716 nm2
with a standard deviation of 894 nm2) as well as the time
Fig. 3. TEM images showing the healing evolution of voids embedded in the CR alloy during the in situ heating experiment.
239
needed to complete the healing (the mean healing time is
2205 s with a standard deviation of 750 s) varied consider-
ably for these voids. The scatterings of the initial void size
and the healing time are possibly due to the varied size and
orientation of Al6Mn precipitates in Al alloy and the differ-
ences in fracture surfaces of the Al6Mn precipitates after
cold rolling, as these determine the geometry and size of
the voids. In order to clearly illustrate the healing kinetics,
the void projected area and the healing duration have been
normalized based on the initial size (denoted by the initial
projected area here) of the voids and the heating duration
needed for completing the healing, respectively, and then
averaged and plotted as an AN–tN curve in Fig. 4b. Here,
error bars indicate the standard deviation of normalized
projected areas at different heating times. The time deriva-
tive of the normalized projected area (inset of Fig. 4b)
shows that the AN–tN curve can be divided into three dis-
tinct stages: an initial fast-healing stage (from s0 to s1), then
a slow-healing stage (from s1 to s2), and finally a rapid-
healing stage (from s2 to s3). Table 1 summarizes the pro-
jected area and heating duration of the voids at the begin-
ning and the end of the void healing, as well as two nodes
for the three distinct healing stages, respectively.
In order to reveal the change in the volume properties of
voids as heating progresses, the in situ heating experiment
has also been combined with electron tomography charac-
terization. When a void in a TEM specimen reached a cer-
tain healing stage during the in situ heating experiment, the
specimen was transferred to the tomography specimen
holder and characterized there by electron tomography.
In this way, the projected area and volume of the void
were measured from HAADF STEM and reconstructed
3-D images, respectively. After the electron tomography
240 M. Song et al. / Acta Materialia 69 (2014) 236–245
Fig. 4. (a) The void projected area as a function of time during the in situ heating experiment. (b) The normalized projected area (AN) as a function of the
normalized heating duration (tN) measured from 20 voids by in situ heating experiments. (Inset) the derivative of the normalized projected area with
respect to the normalized heating duration. s1 and s2 are the ending times of the first and the second stages, respectively.
Table 1
Mean values and standard deviations of void projected areas and heating
durations measured from 20 voids by in situ heating experiments. The
values correspond to nodes between three distinct healing stages (s1, s2), as
well as the beginning (s0) and the end (s3) of void healing process. t,
heating duration; rt, standard deviation of the heating duration; A, void
projected area; rA, standard deviation of the projected area.
Normalized heating duration Time (s) Projected area (nm2)
tN t ± rt A ± rA
s0 = 0 0 2716 ± 894
s1 = 0.26 565 ± 192 1914 ± 605
s2 = 0.92 2036 ± 693 664 ± 207
s3 = 1 2205 ± 750 0
observation, the TEM specimen was transferred back to
the heating holder for further in situ testing. Hence, the
variations of void projected area and volume were determined
for different heating durations. Fig. 5 shows that the void
was divided into two parts, C and D, after 2 min of
in situ heating at 453 K. Electron tomography images
(Fig. 5b–d, f–h, j–l, n–p, r–t and v–x) present the evolution
of the 3-D geometries of voids C and D with heating time.
According to the variation of the projected area with heat-
ing time (Fig. 6), Fig. 5a would be roughly associated with
the first healing stage, and Fig. 5e, i and m with the second
stage for void C. Similarly, for void D, Fig. 5a and b would
be associated with the first healing stage, and Fig. 5i, m and
q with the second stage. The change in void thickness
(Fig. 5h, l and p) along the STEM viewing direction is
insignificant over 40 min of heating. Fig. 5q–u corresponds
to the third stage, where void D changed from ellipsoidal to
spherical in shape with a rapid decrease in void thickness.
Fig. 6 shows the variation of void volume with the heat-
ing time. In agreement with the projected area, the void
volume clearly shows an initial fast-healing stage and then
a slow-healing stage. During the final rapid-healing stage,
the volume change accelerated, particularly when
approaching the end of void healing, although the change
in healing rate between the second and third stages is less
significant in the volume–heating duration curves com-
pared to the projected area. It should be noted that,
because heating and cooling procedures were needed when
transferring the specimen between the in situ heating pro-
cess and electron tomography characterization, the actual
heating duration the TEM specimen experienced deviated
slightly from the nominal heating duration, especially in
the later healing stages.
3.4. Composition analysis
After 40 min of heating at 453 K, bright-field TEM and
HAADF STEM imaging show that voids are healed suc-
cessfully, but EDS elemental maps indicate an inhomoge-
neous distribution of Al and Mg (Fig. 7a and d).
Compared with the Al alloy matrix, there is enrichment
in Mg and deficiency in Al in the regions where the voids
used to be. With prolonged annealing, the segregation
decreases (Fig. 7b and e). After 2 h of annealing at
453 K, both Al and Mg are distributed homogeneously
(Fig. 7c and f).
3.5. Post-mortem experiment
The annealed bulk Al–Mg–Er alloys, which were
obtained by annealing the CR alloy at 473 K for different
durations, have also been investigated. HAADF STEM
images of the CR alloy (e.g. Fig. 8a) show that dog-
bone-shaped voids were located between Al6Mn precipi-
tates. Annealing for 20 min (Fig. 8b and c) caused the void
regions attached to Al6Mn precipitates and with small radii
of curvature to shrink, and also healed the narrow necking
regions of voids, which also had small radii of surface cur-
vature. The number of voids as well as their size (denoted
by the projected area) decrease as the heating duration
(e.g. Fig. 8d–f) is extended from 30 to 50 min. After
60 min of annealing, the HAADF STEM images (e.g.
 M. Song et al. / Acta Materialia 69 (2014) 236–245 241

Fig. 5. HAADF STEM images and surface renders of voids obtained at different in situ heating durations at 453 K. (a, e, i, m, q, u) HAADF STEM
images of the voids obtained at 2, 5, 13, 40, 67 and 75 min of heating, respectively. (b, f, j, n, r, v), (c, g, k, o, s w) and (d, h, l, p, t, x) Surface renders of
voids C and D (indicated in (a)), at different heating durations, shown by rotating counterclockwise around the x-axis indicated in (b) for 0°, 45° and 90°,
respectively.

Fig. 8g) show no voids between Al6Mn precipitates. The

post-mortem experiment results indicate that voids in the
bulk CR Al–Mg–Er alloy can also be healed after 1 h of
low-temperature annealing. The healing process is consis-
tent with the in situ heating of thin-foil samples (e.g.
Fig. 3a–h), while the void geometries during the annealing
process are qualitatively consistent with those of the in situ
heating experiments.

3.6. Mechanical properties of the CR and CA alloys

The mechanical properties and fatigue crack growth rate

Fig. 6. The projected area and volume of voids C and D indicated in
Fig. 5a determined from HAADF STEM and reconstructed 3-D images,
respectively, as functions of time during the in situ heating experiment.
were characterized for the CR and CA alloys; all tests were
performed under the same experimental conditions. The
results are summarized in Table 2. The tensile strength of
the CR alloy is higher than that of the CA alloy, whereas,
as expected, the fracture elongation of the CR alloy is
much less than that of the CA alloy. Nevertheless, the fati-
gue crack growth rates da/dN of the CA alloy are much
242 M. Song et al. / Acta Materialia 69 (2014) 236–245

Fig. 7. The enrichment of Mg at the healed voids. (a–c) EDS maps of Al (right) and Mg (middle), respectively, obtained from the respective areas of the
HAADF images (left). (d–f) Al and Mg EDS line-scan profiles scanned along the red dashed line of the HAADF images in (a–c), respectively.

Fig. 8. HAADF STEM images of the annealed bulk Al–Mg–Er alloys after different annealing times.

Table 2
Mechanical and fatigue properties of the CR and the CA alloys obtained at room temperature. DK, stress intensity factor range.
Specimen Tensile strength (Mpa) Fracture elongation (%) Fatigue crack growth rate da/dN (mm/cycle)
DK = 6.0 Mpa m1/2 DK = 10.0 Mpa m1/2 DK = 20.0 Mpa m1/2
5 4
CR 515 9.0 4.5  10 2.6  10 1.8  103
CA 445 16.5 2.4  105 1.8  104 0.8  103

lower under the same stress intensity factor range DK. The CR alloy and the CA alloy taken from corresponding loca-
results indicate that the plasticity of the CR alloy has been tions. The results suggest that the fracture surface of the
improved significantly after annealing for 1 h at 473 K. CR alloy is rougher due to the large local stress concentra-
Fig. 9a and b shows the fatigue fracture surfaces of the tion in the CR alloy.
 M. Song et al. / Acta Materialia 69 (2014) 236–245
Fig. 9. SEM micrographs of fatigue fracture surfaces of the CR and the CA alloys taken from the corresponding locations.
4. Discussion
As described above, the void-healing process can be
divided into several apparent stages that are closely related
to the variation of the surface morphology of the voids
during the annealing. The results of the in situ experiment
indicate that voids tend to heal first at the corners with a
smaller surface curvature radius in the initial fast-healing
stage. This can be explained, approximately, by the con-
verted GibbsThompson equation [30]:
lr ¼ lb þ 2rV m =rRT : ð1Þ
Here, lr and lb, represent the chemical potentials of voids
with a radius r and of infinite size, respectively; r is the spe-
cific surface energy; Vm is the molar volume of the material;
r is the radius of a void; R is the gas constant; and T is the
absolute temperature. Therefore, the voided regions with a
smaller curvature radius have a larger driving force for dif-
fusion and, consequently, a greater healing rate. The nar-
row neck of the void, which also has a small surface
curvature radius, shrinks and is divided into several small
voids, and these small voids soon disappear. As a result,
the void evolves into two ellipsoid-like parts, where each
part is attached to an Al6Mn precipitate. Therefore, the
remaining curvature radii r are the larger ones for the void
surface, and the curvature radii of the ellipsoid-like voids
are more uniform than the initially dog-bone-shaped void.
Consequently, the healing process enters the slow-healing
stage. As the ellipsoid-like voids continually shrink during
the second stage, the average curvature radius r decreases
with time t. The reason for the significant increase in the
healing rate is associated with the small size of the voids
entering the rapid-healing phase. Because of the small ra-
dius of curvature in this phase, the transport of materials
is significantly increased per unit area at the same time.
The thickness along the viewing direction decreases
quickly, while the ellipsoidal voids become spherical in
shape. Meanwhile, because the average curvature radius r
decreases with time t in this phase, the driving force for dif-
fusion increases. Fig. 10 shows a schematic of the evolution
of voids during the in situ experiment. Fig. 10a–c and c and
243
Fig. 10. Schematic model illustrating the evolution of void healing during
the in situ experiment.
d correspond to the initial fast-healing stage and the slow-
healing stage, respectively. Fig. 10d, e and e, f show the
final rapid-healing stages of voids A and B.
Electron tomography combined with in situ heating
confirms that the variation in void volume during the heal-
ing process can be divided into different stages, most signif-
icantly the initial fast-healing stage and the slow-healing
stage. This is consistent with the relationship revealed by
the variation in projected area with heating time under con-
tinuous in situ heating observations. Although the distinc-
tion between the second and third stages is not so clear
from the void volume, a tendency to accelerated healing
is obvious near the end of void healing (e.g. Fig. 6). This
is likely due to the small surface curvature radii of the voids
in this phase, which increase the driving force for diffusion.
The difficulty of identifying the third healing stage is also
attributed to the experimental setup. As previously dis-
cussed, cooling and heating processes occur when the spec-
imen is transferred from the in situ heating experiment to
the electron tomography observation. This would influence
the actual heating experienced by the voids being observed,
thus creating error in the estimated heating durations for
void volume–heating time curves. This uncertainty over
the time-scale assessment is particularly critical for the
determination of the third healing stage, due to its short
duration (3 min, as shown in Table 1). In this regard, fur-
ther research with in situ 3-D observation would be helpful
when electron tomography processing has been developed
to be fast enough to capture void evolution in real time.
244 M. Song et al. / Acta Materialia 69 (2014) 236–245
Fig. 11. The diffusivity of diffusants in the Al matrix from data given in
Refs. [33–36].
Despite the fact that there are significant differences in
the healing rates of the different stages, the overall evolu-
tion of void healing involves lattice, interface and surface
diffusion but presumably is controlled by the lattice diffu-
sion. During the process of void healing, grain boundaries
may serve as an inexhaustible source of atoms and a sink
for vacancies decomposed by voids, with the driving force
provided by the difference between the chemical potentials
of the grain boundary interface and the internal surface of
voids [13]. The transportation of materials is controlled by
lattice diffusion. The role of internal surface diffusion is to
keep the spherical void shape unchanged during the healing
process [31]. Based on the above analysis, the voids embed-
ded inside grains can only be eliminated by lattice diffusion
because of the transportation of atoms and vacancies. Fur-
thermore, lattice diffusion is much slower than interface
diffusion [32]. Hence, the overall evolution of voids in the
present research is controlled by lattice diffusion.
After 40 min of in situ heating at 453 K, the elemental
maps indicate that the filled void areas are deficient in Al
and enriched with Mg compared with the surrounding
matrix; This inhomogeneous distribution can be eliminated
by increasing the in situ heating time to 2 h. Fig. 11 shows
the diffusion rate of diffusants in the Al matrix. Previous
experimental data [33–36] indicate that there is a linear
relationship between ln D and 1/T for Mg in an Al matrix
between the temperatures of 573 and 928 K. When this lin-
ear relationship is extrapolated to 453 K, the diffusion rate
of Mg in an Al matrix is revealed to be much higher
(6 times) than the self-diffusion of Al atoms. Therefore,
the enriching of Mg in the healed voids is associated with
the higher diffusion rate of Mg compared with Al in the
Al matrix. This indicates that the faster diffusion of Mg
atoms in the Al matrix enhances the void healing in the
Al alloy, which is particularly important at low
temperatures.
It is noteworthy that compared with the CA alloy, both
the higher fatigue crack growth rates and greater roughness
of the fatigue fracture surface are likely related to a
significant extent to the large number of voids in the CR
alloy. Owing to the local stress concentration around voids
in the CR alloy, cracks are likely to nucleate more easily
under external loadings. Meanwhile, voids have adverse
effects on the fatigue crack propagation path or crack clo-
sure. Therefore, successful healing of voids after annealing
at 473 K for 1 h results in a significant improvement in the
performance of the CA alloy.
5. Conclusions
The alloy studied here has a large number of voids
between the Al6Mn precipitates; these voids are embedded
in the matrix of the CR alloy, with the projected dimen-
sions from tens of nanometers to the submicrometer scale.
The voids can be healed successfully after 50 min of in situ
heating at 453 K. The progress of void healing can be cat-
egorized into several apparent stages: an initial fast-healing
stage, a slow-healing stage and a rapid-healing near the
end. The void-healing process is completed by a combina-
tion of lattice, interface and surface diffusions, and is con-
trolled by the lattice diffusion.
The filled void areas were deficient in Al and enriched in
Mg compared with the surrounding matrix, but after 2 h of
in situ heating at 453 K, the Al and Mg were distributed
homogeneously in the matrix. The enrichment of Mg is
related to the higher diffusion rate of Mg in the Al matrix,
which enhances the void healing in the Al alloy and is par-
ticularly important for void healing at low temperatures.
Post-mortem experimental results indicate that voids in
the CR alloy can be healed by an annealing treatment at
473 K for 1 h, consistent with the in situ experiments.
Voids have adverse effects on the fatigue performance of
the CR alloy, whereas the plasticity of the CR alloy can
be improved significantly after annealing at 473 K for 1 h.
Acknowledgements
The authors acknowledge financial support from the
Special Funds for the Major State Basic Research Projects
of China (Grant No. 2012CB619503) and the Natural Sci-
ences Foundation of China (Grant Nos. 51171188,
51390473, 51221264, and 11332010). The authors are also
grateful to J.K. Qiu, H. Xu, J. Wang, J.P. Cui, Y.T. Zhou,
J.F. Liu, N. Lu and Y.J. Xu for stimulating discussions,
and to X.L. Liu for providing the alloy samples.
Appendix A. Supplementary material
Supplementary data associated with this article can be
found, in the online version, at http://dx.doi.org/10.1016/
j.actamat.2014.02.004.
References
[1] Lados DA, Apelian D. Mater Sci Eng A 2004;385:187.
[2] Li P, Maijer DM, Lindley TC, Lee PD. Mater Sci Eng A 2007;460:20.
 M. Song et al. / Acta Materialia 69 (2014) 236–245
[3] Caton MJ, Jones JW, Boileau JM, Allison JE. Metall Mater Trans A
1999;30:3055.
[4] Suresh S. Fatigue of materials. 2nd ed. Cambridge: Cambridge
University Press; 1998.
[5] Petrov AI, Razuvaeva MV. Phys Solid State 2007;45:907.
[6] Salac D, Lu W. Int J Solids Struct 2008;45:3793.
[7] Nosonovsky M, Amano R, Luccia JM, Rohatgib K. Phys Chem
Chem Phys 2009;11:9530.
[8] Gittins A. Acta Metall 1968;16:517.
[9] Takahashi Y, Inoue K, Nishiguchi K. Acta Metall Mater
1993;41:3077.
[10] Qin RS, Su SX. J Mater Res 2002;17:2048.
[11] Beere WB, Greenwood GB. Met Sci 1971;5:107.
[12] Makin MJ. J Nucl Mater 1978;71:300.
[13] Sun J. J Mater Eng Perform 2002;11:322.
[14] Ma Q. Acta Metall 1998;46:1669.
[15] Wang MF, Du GJ, Xia DY. Key Eng Mater 2013;531:454.
[16] Wei DB, Han JT, Tieu K, Jiang ZY. Scr Mater 2004;51:583.
[17] Choi SR, Tikare V. Mater Sci Eng A 1993;171:77.
[18] Ando K, Houjyou K, Chu MC, Takeshita S, Takahashi K, Sakamoto
S, et al. J Eur Ceram Soc 2002;22:1339.
[19] Nam KW, Kim JS. Mater Sci Eng A 2010;527:3236.
245
[20] Liu SP, Ando K, Kim BS, Takahashi K. Int Commun Heat Mass
2009;36:563.
[21] Brett J, Seigle L. Acta Metall 1963;11:467.
[22] Takahashi Y, Ueno F, Nishiguchi K. Acta Metall 1988;36:3007.
[23] Meng G, Li BL, Li HM, Huang H, Nie ZR. Mater Sci Eng A
2009;516:131.
[24] Bae DH, Ghosh AK. Acta Mater 2002;50:511.
[25] Cao B, Joshib SP, Ramesh KT. Scr Mater 2009;60:619.
[26] Dai LH, Ling Z, Bai YL. Compos Sci Technol 2001;61:1057.
[27] Azuma M. Structural control of void formation in dual phase steels,
PhD thesis, Technical University of Denmark; 2013.
[28] Midgley PA, Ward EPW, Hungrı́a AB, Thomas JM. Chem Soc Rev
2007;36:1477.
[29] Gupta TK. J Am Ceram Soc 1978;61:191.
[30] Peng ZA, Peng XG. J Am Chem Soc 2002;124:3343.
[31] Zhang HL, Sun J. Mater Sci Eng A 2004;382:171.
[32] Yu HH, Suo Z. J Mech Phys Solids 1999;47:1131.
[33] Volin TE, Balluffi RW. Phys Status Solidi 1968;25:163.
[34] Dais S, Messer R, Seeger A. Mater Sci Forum 1987;15:419.
[35] Fujita T, Horita Z. Philos Mag A 2002;82:2249.
[36] Rothman SJ, Peterson NL, Nowicki LJ, Robinson LC. Phys Status
Solidi B 1974;63:29.