Materials Science & Engineering A 777 (2020) 139091

Contents lists available at ScienceDirect

Materials Science & Engineering A

journal homepage: http://www.elsevier.com/locate/msea

On the long-term correlations in the twinning and dislocation slip dynamics

E. Agletdinov a, D. Drozdenko b, c, P. Harcuba b, P. Dobron
� b, D. Merson a, A. Vinogradov d, *
a
Institute of Advanced Technologies, Togliatti State University, Togliatti, 445020, Russia
b
Faculty of Mathematics and Physics, Charles University, 121 16, Prague, Czech Republic
c
Magnesium Research Center, Kumamoto University, Kumamoto, 860-8555, Japan
d
Department of Mechanical and Industrial Engineering, Norwegian University of Science and Technology - NTNU, Trondheim, 7491, Norway

A R T I C L E I N F O A B S T R A C T

Keywords: The present work aims at gaining new insights into the dynamics of the dislocation slip and twinning processes,
Mechanical twinning which govern the majority of possible scenarios of plastic deformation in structural metals and alloys. Using
Dislocation slip model magnesium single crystals and employing the statistical analysis of acoustic emissions (AE) generated in
Acoustic emission
the course of plastic deformation, it was demonstrated that twinning is a process with a memory of the past
Magnesium
Single crystal
whereby the twinning events affect the occurrence of successive events. As opposes to this, the basal dislocation
slip appears as a substantially random intermittent process comprising of independent elementary slip events
without long-term correlations between gliding dislocations or slip bands emerging due to the collective dislo­
cation glide along basal planes. The single crystals were chosen with axial orientations best suitable to facilitate
either basal dislocation slip or tensile twinning under compressive loading, which significantly simplified the
interpretations of AE findings. Besides, the AE results were fully corroborated in situ by microstructure obser­
vations using scanning electron microscopy imaging.

1. Introduction: motivation and methodology in Ref. [4]. To separate these deformation mechanisms and eliminate the
short-term slip-twin interactions, the use of a properly oriented single
Deformation twinning is a displacive transition which occurs by crystal is indispensable [9,10]. Single crystals of magnesium are highly
high-speed collective displacements of a large number of atoms across anisotropic due to the large difference in critical resolved shear stress
many scales in space and time [1,2]. In time, the mechanical twinning (CRSS) for the deformation systems, that makes the activation of
may evolve through several stages involving nucleation of tiny embryos deformation mechanisms predictable and controllable. Therefore, to
and their coalescence before the twin nucleus shoots through a perfect validate the methodological approach proposed in Ref. [4], in the pre­
crystal as a propagating lamella [3]. Using temporal statistical analysis sent work, a comparative analysis was made of the temporal structure of
of the acoustic emission (AE) time series acquired during tensile defor­ the AE time series arising in magnesium single crystals where either
mation of the polycrystalline magnesium alloy ZK60, some of the pre­ dislocation slip or deformation-induced twinning were promoted as
sent authors [4] have demonstrated recently that the mechanical dominant deformation modes during compression loading. Specifically,
twinning belongs to a wide class of non-Poisson random processes. In the compression along the axis 1122 is associated with the prevalence of
other words, twinning manifests itself as a process with a memory of the the basal <a> slip having the highest Schmid factor of 0.5, while the
past. The alloy ZK60 is a typical hexagonal close-packed (hcp) metal compression along the 1120 axis triggers extension twinning as the
exhibiting strain hardening which involves the activation and interplay primary deformation mode. The detailed statistical AE analysis was
between the dislocation slip and twinning – the two primary deforma­ corroborated by the in situ secondary electron (SE) imaging providing
tion mechanisms for many hcp metals and alloys. The concurrent insight into the microstructure evolution during loading.
operation of these mechanisms substantially complicates the analysis
and the interpretation of the results. In particular, the highly possible
dislocation-twin interactions, which are common in deformed poly­
crystalline hcp magnesium [2,5–8], could affect the conclusions drawn

* Corresponding author.
E-mail address: alexei.vinogradov@ntnu.no (A. Vinogradov).

https://doi.org/10.1016/j.msea.2020.139091
Received 5 December 2019; Received in revised form 7 February 2020; Accepted 10 February 2020
Available online 11 February 2020
0921-5093/© 2020 Elsevier B.V. All rights reserved.
E. Agletdinov et al. Materials Science & Engineering A 777 (2020) 139091

2. Materials and methods electron microscope Zeiss Crossbeam Auriga equipped with the defor­
mation stage MTEST Quattro (Materials Testing system ADMET). The in
2.1. Materials and testing situ deformation tests were periodically paused for obtaining the SEM
images of the surface morphology at different strains and magnifica­
The magnesium single crystals (of 99.95% commercial purity) were tions. Prior to observations, the surface was ground and polished using a
grown by a modified vertical Bridgman technique from monocrystalline diamond paste down to the ¼ μm grade, and, finally, electrolytically
seeds. The specimens (of 5 � 5 � 6 mm3) specifically oriented for uni­ polished using the Struers Lectropol in AC2 electrolyte at 18 V, 40 � C
axial compression along the 1122 axis or the 1120 axis to promote either for 30 s.
basal slip or twinning, respectively, were prepared from the as-cast in­
gots by spark erosion. The desired initial orientation of the specimens 2.2. AE analysis of arrival times of event in time series
was confirmed by means of X-ray diffraction (Panalytical X-ray θ 2θ
diffractometer, CuKα radiation). AE resulting from localised sources due to rapid stress relaxation in
The mechanical loading, accompanied by AE measurements, was the solids resembles seismic time series and appears as a set of electric
performed using the Instron 5882 universal testing machine at the pulses at the sensor output as is illustrated in Fig. 1 representing the
constant nominal strain rate of 10 3 s 1 at room temperature. different typical fragments of the AE behaviour observed in the Mg
The AE activity was monitored by a computer-controlled setup based single crystals tested in the present work. Following the data processing
on the PCI-2-based (Physical Acoustic Corporation) board operating in a strategy proposed in Ref. [4] for the assessment of possible correlations
continuous thresholdless mode with 2 MHz sampling frequency. The in the behaviour of deformation-induced defects, the AE time series
sensor - a miniature piezoelectric transducer MST8S (Dakel-ZD Rpety, reduces to a point process which is characterised solely by a set of arrival
Czech Republic) having a diameter of 3 mm and a reasonably flat times of events ft0 ; ​ t1 ​ ::: ti ​ ::: ​ tN g shown schematically by dashed
response in the 100–600 kHz frequency range - was attached to the lines in Fig. 1b, d. In practice, it is more convenient to investigate the
holder in close proximity to the specimen. The signal was pre-amplified distribution of time intervals or inter-arrival times between the succes­
by 40 dB before an acquisition. The full scale of the A/D converter used sive events Δti ¼ tiþ1 ti , c.f. Fig. 1b, d. Assuming that the sources of
in the acquisition chain was �10 V providing the total dynamic range of local stress relaxation resulting in the AE events are independent, a
100 dB. The background input noise during the tests did not exceed 10 Poisson-type time series [11] is anticipated in the form of a train of
μV peak-to-peak. Details of the AE signal processing are given in the Dirac’s δ-impulses δðt tk Þ with amplitudes Uk
following sub-sections.
X
N
To verify the outcomes of the indirect AE analysis and to provide a AðtÞ ¼ Uk δðt tk Þ (1)
deeper insight into the microstructure evolution during compressive k¼1

deformation of magnesium single crystals, the in situ scanning electron The inter-arrival time intervals Δti should, therefore, obey an expo­
microscopy (SEM) imaging, using secondary and backscattered electron nential distribution with the probability density function (PDF)
detection, was performed in the chamber of the filed-emission scanning

Fig. 1. Typical fragments (sectioned from the raw AE streaming data with different time scales) of AE time series exemplifying the AE response dominated by
dislocation slip (green) (a, b) and twinning (red) (c, d) in the magnesium single crystal deformed along the 〈1120〉 axis, c.f. Fig. 3 and refer to the section Results for
details. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

2
E. Agletdinov et al. Materials Science & Engineering A 777 (2020) 139091

1
ρðΔtÞ ¼ expð Δt = ΔtÞ ¼ λ expð λΔtÞ (2)
Δt

where Δt is the average time interval between the pulses in a time-series

(averaging is done over the time interval T) and λ ¼ 1=Δt is the average
intensity (also termed as count rate or activity) of the pulse flow on the
chosen time interval T.
Since the properties of any random process are fully determined by
the PDF of the descriptive random variable, the AE stream was sectioned
into a set of successive realisations and the PDF ρðΔtÞ was obtained for
each j-th realisation. The statistical goodness-of-fit χ [2] test was applied
to probe the agreement between the inter-arrival times Δ t and the
Poisson distribution (2) with the count rate λj ¼ Nj =T for each realisa­
tion j 2 ½1; K�.

2.3. Non-supervised cluster analysis of AE data

In magnesium polycrystals with typical grain sizes in the 10-μm

range and above, the main deformation mechanisms - dislocation slip
and twinning - co-exist in the course of plastic deformation. As opposes
to this and as discussed above, the contributions from both mechanisms
are well-separated in the deformation history of the suitably oriented
single crystals. The dislocation motion along the slip planes and twin­
ning differ significantly in their source dynamics and, therefore, they
generate AE with different waveforms (low-amplitude waveforms fluc­
tuating in a wider range of frequencies, which are produced by dislo­
Fig. 2. AE activity in the Mg single crystal deformed along the 1122 axis,
cation slip, can be compared to high amplitude transients for twinning promoting the basal slip: (a) AE count rate and raw AE signal correlated with
[12,13]). To discriminate quantitatively between the sources of different the deformation curve; (b) non-supervised cluster analysis of AE signal.
origin in random AE time series, the proposed by Pomponi and Vinog­
radov adaptive sequential k-means (ASK) signal clusterisation algorithm
[14] has been proven effective. For the cluster analysis, the acquired AE
stream was sectioned into successive realisations of 4096 samples. The
Fourier power spectral density (PSD) function was calculated for each
realisation by the Welch technique and normalised to unite area for
comparison. The ASK procedure groups together the signals with similar
PSDs while disjoins the signals having statistically different spectra. The
symmetrical version of the Kullback-Leibler distance is used in the
present work as a measure of similarity/dissimilarity between normal­
ised PSDs (see [12,15–17] for mathematical details and implementation
of the ASK algorithm and illustrative case studies).

3. Results and data analysis

3.1. Statistical analysis of arrival times in AE time series

Raw streaming AE data (coloured in grey) superimposed with the

loading curves are shown in Figs. 2 and 3 in parallel with the main re­
sults of the statistical analysis of AE time series. The coloured symbols
depict the count rate λ, i.e. the number of detected events per unit time,
of the AE flow, calculated for statistically representative samples of the
time series. The colour of the symbol corresponds to the result of the
Pearson’s test for each sample: the green coloured triangles correspond
to the fragments of the AE stream where the distribution of the inter-
arrival times agrees with the exponential Poisson distribution, Eq. (2),
while the red colour marks the sections with the non-Poisson behaviour
of the sources. In other words, the green colour denotes the samples with Fig. 3. AE activity λ (triangles) in the Mg single crystal deformed along the
the Poisson process of random and independent events, while the red 1120 axis, promoting twinning: (a) AE count rate and raw AE signal correlated
colour indicates the samples with a statistically significant deviation with deformation curve; (b) non-supervised cluster analysis of AE signal. The
from the Poisson process, which heralds the existence of a temporal dashed line indicates a cross-over between the two primary deformation modes,
correlation between the events in the time series. The AE pulse flow is which coincides with the inflexion point on the loading diagram.
intense in both specimens tested. On the time scale shown in Figs. 2 and
3, the AE stream looks like a continuous noise signal which never van­ corresponding to green- (a,b) and red- (c,d) coloured fragments (Poisson
ishes in the entire interval of observation. However, on the smaller time and non-Poisson, respectively) of the AE activity shown in Fig. 3 for the
scales, AE for both specimens tested appears as pulse trains consisting of magnesium single crystal deformed along the 1120 axis. The arrival
well-separated transients of different amplitudes as has been clearly times for each transient including those with the low signal-to-noise
illustrated in Fig. 1 showing typical fragments of AE time series

3
E. Agletdinov et al. Materials Science & Engineering A 777 (2020) 139091

ratio, c.f. small transients shown in Fig. 1c, were picked using the al­ plateau with little work hardening due to profuse tensile twinning is
gorithm described in [18, 19]. Typical experimental histograms of the observed on the stress-strain curves in Fig. 3. The AE flow on this stage is
waiting time distributions, i.e. the probability density functions of the particularly intense and a broad peak of the AE count rate is observed
inter-arrival times, are shown in Figs. 4 and 5 for both specimens tested. from 0 to 120 s. Within the same time interval, the χ 2 goodness-of-fit test
Each figure consists of eight sub-plots referring to different deformation suggests that the Poisson model does not represent the experimental
stages indicated by time marks on each sub-plot. The solid lines super­ data as indicated by the red symbols on the AE intensity diagram. This
imposed onto the experimental probability density histograms represent behaviour is at sharp contrast with what has been seen above for the
the theoretical exponential functions according to Eq. (2), which are 1122 oriented crystal and is indicative of existing temporal correlations
anticipated for the Poisson process of the same intensity λ as that ob­ in the spontaneous fluctuating behaviour of emitting defects. It has been
tained from the respective sample distribution (notice that these are not proposed recently [4], that the twinning process in magnesium alloys
the fitting curves). Besides the quantitative changes in PDFs, which can be regarded as a non-Poisson self-exciting process with the memory
trivially caused by the changes in AE activity, the qualitative changes of the past. The present results obtained on suitably oriented single
occur as reflected by the Pearson’s goodness-of-fit p-test values for the crystals fully corroborate this conclusion leaving no doubts that the
experimental histogram shown on the corresponding diagrams. The long-term correlations are seen in the collective dynamics of twins.
p-value less than the 0.05 significance level suggests that one cannot These should not be confused with the strong short-term avalanche-like
accept the null hypothesis stating that the data are consistent with a correlations between elementary twinning dislocations forming each
specified Poisson distribution and that there is a significant difference individual twin; these correlations are observed in the nano-and sub-­
between the observed and the expected Poisson frequency. microsecond range riving rise to high amplitude transient AE signals
The crystal, which is loaded in compression along the 1122 direction with sharp fronts and wide power spectra [20]. The distribution of the
deforms mainly by basal slip. The remarkable result is that in this case time intervals between the events, Fig. 5b–f, measured with a
the statistical AE analysis consistently shows that the dislocation slip sub-millisecond resolution suggests further that the observed correla­
generates acoustic emissions following the Poisson’s behaviour featured tions are caused by the influence that twinning events exert on their
by the exponential distribution of the inter-arrival times, Fig. 4. This nearest neighbours, as it is expected in the random self-exciting Hawkes
behaviour is persistent and is observed throughout the experiment, point process [21,22]. The essential property of the Hawkes process is
except the very initial stage of loading, where multiple “false” AE events that the occurrence of any event in a random sequence increased the
appeared due to initial self-adjustments and rubbing in the deformation probability of further events. Due to this feature, this process has found
rig at small loads, Fig. 1. Therefore, this early deformation stage was numerous applications in modelling the arrival of random events with
excluded from the analysis. the clustering/branching effect in economics, biology and many other
For the crystal 1120 deformed in compression, the typical initial disciplines [22]. The characteristic time scale where the remarkable

Fig. 4. The evolution of the distribution function of time intervals between successive events in the AE time series corresponding to Fig. 2. Time marks corresponding
to some fragments are shown in each sub-figure.

4
E. Agletdinov et al. Materials Science & Engineering A 777 (2020) 139091

Fig. 5. The evolution of the distribution function of time intervals between successive events in the AE time series corresponding to Fig. 3. Different time intervals j 2
½1; K� are shown in each sub-figure.

deviation of the distribution of interarrival times is observed from the During compression along the 1122 axis, three clusters correspond­
Poisson exponential behaviour is found to be of the order of few milli­ ing to different AE sources have been revealed by the ASK procedure,
seconds or less. At the mature stage of the uniform deformation Fig. 2b. Here and in what follows, the clusters are labelled on the first-
(approximately 120–200 s), the dynamics of the AE events changes so come-first-numbered basis. Cluster 1 (green line), corresponding to the
that the transients appear to be distributed according to the Poisson first active AE mechanism is characterised by the low amplitude AE and
model as is indicated by green triangles on the tail of the AE activity lower frequency components prevailing in the PSD, which is typically
curve in Fig. 3. At this stage, the exponential (Poisson) distribution of the associated with the dominance of dislocation slip in the deformation
time intervals is consistently observed, Fig. 5g and h. process. One can notice, that this cluster evolves steadily throughout the
Results of the statistical analysis of inter-arrival times in AE time test as it is reasonably expected for the basal slip active in single crystals
series are supplemented by the independent characterisation of under­ with this orientation. Cluster 2 (red line), which comes into play notably
lying mechanisms by means of the cluster analysis of the AE datasets as later in the deformation history, is composed of high-frequency transient
discussed in the following section. signals with high peak amplitudes, which are assumed naturally from
the preponderance of mechanical twinning over slip. From the kinetics
of individual clusters, which is explicit in Fig. 2b, it is obvious that this
3.2. Non-supervised cluster analysis of AE
cluster progresses significantly slower compared to Cluster 1 (basal slip).
Moreover, after around 170 s, this cluster does not grow any more, i.e.
Results of the application of the ASK clusterisation algorithm to the
the associated mechanism ceases to emit elastic waves. Cluster 3 (cyan
AE streaming data are summarised in terms of the number of class
line) appears almost immediately after Cluster 2 with a just slight delay,
members accumulated during loading time as shown in Figs. 2b and 3b
i.e. the mechanism responsible for Cluster 3 triggers shortly after several
for both differently oriented single crystals. This procedure reveals that
first twins show up. It then keeps growing, though one can notice that
the AE signals fall into few primary categories (three and two for the
the rate of new members acquisition in this class reduces at approxi­
crystals with 1122 and 1120 axial orientation, respectively) with sta­
mately the same time as Cluster 2 saturates, thus indicating the intimate
tistically different features (the realisations corresponding to the back­
interconnection between the underlying mechanisms. Considering that
ground electrical noise, which is inevitable in the highly sensitive AE
features of Cluster 3 bear plenty of similarity to those of Cluster 1
setups, were also identified and filtered out). These groups are distinct
assigned to dislocation slip, and Cluster 3 most likely represents the slip
by their PSDs having different relative fractions of low and high-
in the twinned regions.
frequency components. The characteristic statistical features of these
The AE cluster pattern in the single crystal deformed in compression
groups do not differ fundamentally from those discussed early for typical
of crystal along the 1120 axis is different, Fig. 3b. In this case, the ASK
AE mechanisms originating from dislocation slip and twinning (see, e.g.
procedure reveals two distinct clusters. The activated first Cluster 1 (red
[4,12,15]) – low amplitude AE signals with relatively wide and strongly
line) is comprised of AE signals with high peak amplitudes (similarly to
fluctuating spectra are associated with dislocation activity, while the
Cluster 2 in the 1122 oriented crystal), and, therefore, it is most
high-amplitude transients with the compact frequency distribution in
reasonably assigned to mechanical twinning. The number of members in
the power spectra are related to the twinning dynamics.

5
E. Agletdinov et al. Materials Science & Engineering A 777 (2020) 139091

this cluster gradually increases with strain/time before reaching satu­ taken at 25 s). These deformation twins continue growing fast with
ration at approximately 120 s. At about 35 s, Cluster 2 (green line) further loading (c.f. the SEM image taken at 50 s). At 75 s after the
associated with basal slip comes into play as a minor contributor to the beginning of the test, the crystal is nearly completely twinned (this was
overall AE signal. However, this cluster then evolves rapidly so that a confirmed by the electron back-scattered diffraction measurements of
crossover between the two modes occurs at approximately 85–90 s – the crystal orientation). The slip traces resulting from profuse disloca­
exactly at the time when the inflexion point is observed on the loading tion motion become visible much later than mechanical twins (c.f. the
diagram as indicated by the dashed line, Fig. 3a. As deformation pro­ SEM image taken at 75 s). As f1012g1122 extension twinning results in
ceeds, the dislocation slip plays a more and more significant role in the the reorientation of the initial crystal lattice by 86.3� , the observed slip
strain hardening process, and, finally, it becomes a dominant mecha­ lines can be reasonably assigned to the basal slip in the twinned volume.
nism of energy dissipation and strain accommodation in the deforming
crystal. 4. Discussion

Plastic deformation during compression of the magnesium crystal

3.3. Microstructure development: in situ SEM imaging along the 1122 axis is governed by dislocation motion on basal planes.
This deformation mode activates most easily due to the highest Schmid
The in situ microstructural observations corroborate the conclusions factor being of 0.5. The analysis of the AE time series unequivocally
drawn from both independent methods of AE data analysis discussed suggests that the observed long individual slip lines on the surface,
above. During compression loading of the magnesium single crystal Fig. 6, are detectable sources of AE associated with the collective motion
along the 1122 axis, Fig. 6, only individual slip lines appear on the free of dislocations, similarly to what has been observed on even a smaller
surface along the basal plane. As deformation proceeds, those lines scale in locally loaded pure copper [23]. However, it should be noted
become more and more visible. The important difference with the that unlike what is commonly seen in pure fcc metals, where AE appears
deformation process in polycrystals, where the slip lines emerge as a continuous signal, the raw AE signal in magnesium is transient
randomly and independently here and there in differently oriented regardless of whether it was generated by twins or moving dislocation,
grains, all slip lines in the single crystal belong to the same family of Fig. 3. It has long been understood that mechanical twinning, being a
basal planes. Although this axial orientation of the single crystal is very fast process of crystal lattice re-orientation involving cooperative
theoretically “ideal” for basal slip, a few tiny f1012g1122 extension motion of a large number of twinning dislocations, generates very
twins are observed on the mature deformation stage at about 200 s. powerful high-amplitude AE bursts, which, unlike most other AE sour­
However, with further loading, those twins barely grow in width (c.f. the ces, can be audible: “tin cry” or “crackling” sound are well-known ex­
SEM image taken at 300 s), thus confirming that this orientation is not amples of this kind of behaviour [24]. In the present work, a few tiny
favourable for twinning. At a higher magnification, the dislocation slip tensile twins are found in the single crystals specifically oriented for
lines can be resolved in the twinned regions, as is shown in the inset in basal slip. The in situ microstructure observations reveal them un­
Fig. 6. doubtedly albeit with the notable time lag compared to AE data indi­
In case of compression loading of the magnesium single crystal along cating the early appearance of twins as represented in Fig. 1b: c.f. AE –
the 1120 axis, Fig. 7, the wide f1012g1122 extension twins can be 50 s vs SEM imaging – 200 s. This time difference should not be
observed already at the beginning of deformation (c.f. the SEM image confusing and could be explained by the complexity of the twinning

Fig. 6. A sequence of SEM images showing the microstructure evolution observed in-situ during compressive loading of the Mg single crystal along the 1122 axis
favourable for the basal <a> dislocation slip. All images refer to the same surface fragment: visible changes in the grey contrast are due to the SEM settings and the
variation of the working distance during loading. The appearance of the tiny twin is highlighted by the white arrow. The inset to the “300 s” image showing the
magnified fragment of the deformation twin reveals the traces of the secondary slip in the twinned region.

6
E. Agletdinov et al. Materials Science & Engineering A 777 (2020) 139091

Fig. 7. A sequence of SEM images showing the microstructure evolution observed in-situ during compressive loading of the Mg single crystal along the 1120 axis
favourable for f1012g extension twinning.

process on the one hand and by the limitations of direct in situ obser­ vation of f1012g〈1011〉 extension twins easily occurs when the
vations on the other. As it has been mentioned in Section 3.2, the compression load is applied normally to the c-axis [9,28,29]. In this
extension twins grow thickness-wise relatively slowly even at the situation, the Schmid factors for the extension twinning system, pris­
mature loading stage (c.f. images taken at 200–300 s in Fig. 6). On the matic and basal slip systems are of 0.5, 0.43 and 0, respectively [30].
other hand, the AE method is sensitive to twin nucleation, while the twin Therefore, in the present work, extension twinning is the dominant
thickening does not produce detectable AE due to the significant dif­ deformation mode in the single crystals compressed along the 〈1120〉
ference in the velocity of twin nucleation (propagation in length) and axis. Further plastic straining of the re-oriented (twinned) grains is
thickening (propagation in width) [25]. Therefore, there is a reasonably accommodated by the <a> dislocation slip on the basal and prismatic
expected delay between the time of twin nucleation detected by AE and planes. The mechanisms predicted from the Schmid factor analysis
the time when the twin becomes wide enough for reliable recognition at appear in harmony with both the AE results and the in situ SEM obser­
relatively small magnifications used in the present in situ observations. vations. While the high-amplitude signals prevailing at the beginning of
Moreover, one should bear in mind that AE is a technique receiving loading, Fig. 3a, and the ASK analysis collectively signify twinning ac­
information about evolving defects from the entire specimen volume, tivity indirectly, the direct in situ microstructure observation, Fig. 7,
while the SEM imaging technique is limited to the 2D field of view firmly entrenches the AE-based hypothesis and strengthens the conclu­
capturing only a fraction of the deforming surface. Therefore, the freshly sions drawn from the AE method. Furthermore, the gradual involvement
nucleated twins, which are detected by AE already at 50 s, might be of the dislocation slip as a complementary accommodation mechanism
invisible at the surface or they are too small to be resolved under the is clearly observed as the deformation proceeds beyond the inflexion
SEM imaging conditions applied. Besides, we should underline that the point on the strain hardening curve. Once activated, dislocation slip
statistical AE analysis strongly suggested that twinning served only as an followed by storage mediates the strain hardening process giving rise to
occasional contributor to the overall AE in the magnesium crystal the sigmoidal shape of the strain hardening diagram in Fig. 2.
deformed in compression along the 1122 axis. These observations The AE signal clustering procedure reveals several interesting fea­
appear to be in excellent agreement with the works of Chapius and tures in the evolution of underlying deformation mechanisms in both
Driver [26] and Molodov et al. [27], where anomalous extension twins differently oriented specimens. Once the signals are associated with
in plane-strain (channel-die) compressed magnesium single crystals respective sources, the activity of different deformation mechanisms can
preferentially oriented for basal slip have been observed and discussed. be traced separately as it is demonstrated in Figs. 1b and 2b in terms of
Besides basal slip entailing a rotation of the c-axis towards the the cumulative number of class members for both types of single crystals
compression direction, the twins were found parallel to the constraint in studied. Importantly is that the proposed non-supervised clustering
order to compensate the deformation heterogeneity with respect to the procedure, unlike the conventional k- or c-means algorithms, does not
slip-induced orientation change. The twins variants formed in sym­ require the user to define the number of derived clusters a priori, i.e. it is
metrical pairs [27] to mutually compensate their shear components in data-driven, and the number of clusters is determined solely on the basis
the lateral constraint 〈1010〉 direction. The appearance of twins close to of statistical differences between incoming signals and their PSDs.
the loading platens in the present work can be reasonable for the same Therefore, it is interesting to notice that in the single crystal oriented for
reason - compensation of the basal slip induced rotation of the crystal. basal slip, the ASK procedure returns three clusters corresponding to
In sharp contrast with the behaviour of the 〈1122〉 crystal, the acti­ different AE sources. They have been identified above and correspond to

7
E. Agletdinov et al. Materials Science & Engineering A 777 (2020) 139091

the basal slip, twinning, and the dislocation slip in the twinned fraction. usually of the order of few nanoseconds as is commonly estimated by
All these mechanisms are represented in Fig. 6. Overall, the AE flow was discrete dislocation dynamics simulations [36]. The sub-microsecond
nearly uniformly composed of the signals caused by dislocation motion time scale of the individual slip line formation has been demonstrated
with only a small contribution from occasional twins (Fig. 1b shows that by rapid video imaging in the “classic” experiments by Neuha €user and
the slip-related clusters are by far more representative than that for co-authors [37,38], and this time scale agrees reasonably with short
twins). low-amplitude transients recorded during dislocation slip line genera­
Unlike the single crystal oriented for basal slip, the twin-oriented tion during scratch testing of pure copper polycrystals [23].
sample behaves in a different way upon loading. According to the ASK
analysis Fig. 2b, plastic deformation in the single crystal loaded along 5. Conclusions
the 〈1120〉 axis commences by activation of mechanical twinning, and
after of 35 s the dislocation slip comes into play. Despite the coexistence The acoustic emission technique powered by the advanced signal
of both deformation mechanisms on the mature stage of straining, the categorization technique was employed to probe the dynamics of
statistical analysis of AE time-series, Fig. 2a, reveals the exponential dislocation slip and twinning in magnesium single crystals suitably
distribution in the occurrence of inter-arrival times between transients oriented to give preference to one of these two primary deformation
after 120 s. Some “time lag” visible between the crossover point in modes. It was unambiguously demonstrated that twinning behaved as
Fig. 2b and the appearance of green symbols indicating the Poisson-type an intermittent process with the temporal correlation between the
time series in Fig. 2a is a simple consequence of the statistical procedures events having a relatively long (of sub-millisecond and millisecond
involved: the deformation twinning remains very pronounced even after range) memory of the past. As opposes to this, the basal dislocation slip
120 s and the existing memory of the past in the twin-related AE com­ manifested itself as a substantially random process of independent
ponents determines the results of Pierson’s goodness-of-fit test until the elementary events with no temporal correlations between gliding dis­
dislocation slip becomes truly dominating. locations or the slip bands emerging to a free surface due to collective
Before concluding this work, some of its implications for future dislocation slip along the basal planes.
research are to be outlined. The main thrust of the present study was to For the first time, the classification power of the adaptive sequential k-
provide solid experimental evidence in support of the hypothesis of the means (ASK) algorithm was demonstrated in the model test program that
long-term correlations existing between the mechanical twins. The self- enables predicting the deformation mechanisms with very high confi­
exciting nature of these correlations is revealed by the statistical analysis dence and verifying them in direct in situ SEM observations.
of AE time series at the sub-millisecond and millisecond scale. Not only The signatures of twinning have been revealed by AE in the single
this suggests that the twins do not evolve independently, but that crystal “ideally” oriented for basal slip and prompted detailed micro­
initiation of a parent twin gives birth to a series of new twins with a scopic observations confirming the appearance of few tiny extension
locally accelerated rate. An important corollary from this observation is twins. They do not appear to contribute much to the deformation and
that the temporal twin-twin interaction and correlation has to be taken rather are developed to compensate for the crystal rotation induced by
into account when it comes to modelling the mechanical behaviour of basal slip.
materials that exhibit profuse twinning activity. Within a contemporary The non-supervised ASK cluster analysis of AE time series is thus
purely phenomenological approach, twin boundaries, acting as impen­ proven capable of predictive discrimination between underlying defor­
etrable obstacles for dislocation glide, are involved in strain hardening mation mechanisms, the accurate timing of their activation and
through the so-called dynamic Hall-Petch concept [31,32], that bridges evolution.
the time/strain dependent volume fraction of twins F and the time/­
strain dependent evolution of the dislocation density. Up to date, the Data availability
twinning kinetics is most often derived from the Olson-Cohen [33]
assumption according to which the twin volume fraction F as a function The experimental data that support the findings of this study are
of strain ε obeys the first-order differential equation dFð available from the corresponding author, AV, upon reasonable request.
dε ¼ mð1 FÞ.
εÞ

Here ð1 FÞ refers to the fraction of the untwinned material where twins

may form with the rate determined by the microstructurally sensitive Declaration of competing interest
factor m. Although this simple equation captures the ascending trend in
FðεÞ to saturation in a general manner, it does not imply any twin-twin The authors declare that they have no known competing financial
interactions. The significance and implications of these interactions on interests or personal relationships that could have appeared to influence
twinning accumulation will be unfolded a dedicated forthcoming pub­ the work reported in this paper.
lication where appropriate modifications of the existing approach will
be proposed and presented in a more elaborate way. CRediT authorship contribution statement
We should also notice that the present results showing that the
dislocation slip is a memoryless process at a time scale above a milli­ E. Agletdinov: Methodology, Data curation, Software, Formal
second or so (as resolved by the present signal detector) is in fair analysis, Validation, Visualization, Writing - original draft. D. Droz­
agreement with the model and experimental results presented by denko: Data curation, Visualization, Investigation, Writing - original
Vinogradov et al. in Ref. [34] where acoustic emission mediated by draft. P. Harcuba: Visualization, Investigation. P. Dobron � : Investiga­
dislocation slip in pure copper was considered as a continuous tion. D. Merson: Project administration, Supervision. A. Vinogradov:
Ornstein-Uhlenbeck random process. Unlike the present work, where Conceptualization, Methodology, Writing - review & editing,
the burst events can be individually detected, separated from each other Supervision.
and associated with a point process, the low-amplitude AE transients in
pure fcc and bcc metals overlap, forming a continuous random noise-like Acknowledgements
signal. The autocorrelation function of the Ornstein-Uhlenbeck [35]
process is exponentially decaying so that the autocorrelation in the The support from the Ministry of Education and Science of Russia
high-frequency AE signal caused by dislocation sources approaches zero through the State Assignment according to the contract No.
at the scale of approximately 1 ms, thus signifying a lack of long-term 11.5281.2017/8.9 is gratefully appreciated. This work was also partly
correlation between dislocation avalanches. The millisecond time financially supported by the Czech Science Foundation, Czechia, under
scale is by far larger than the dislocation mean free flight time, which is the grant No. 17-21855S and the European Regional Development Fund
(ERDF) under the project CZ.02.1.01/0.0/0.0/15_003/0000485.

8
E. Agletdinov et al. Materials Science & Engineering A 777 (2020) 139091

References [20] C.B. Scruby, H.N.G. Wadley, J.J. Hill, Dynamic elastic displacements at the surface
of an elastic half-space due to defect sources, J. Phys. Appl. Phys. 16 (6) (1983)
1069–1083.
[1] R.W. Cahn, Twinned crystals, Adv. Phys. 3 (12) (1954) 363–445.
[21] A.G. Hawkes, Spectra of some self-exciting and mutually exciting point processes,
[2] J.W. Christian, S. Mahajan, Deformation twinning, Prog. Mater. Sci. 39 (1–2)
Biometrika 58 (1) (1971) 83–90.
(1995) 1–157.
[22] A.G. Hawkes, D. Oakes, A cluster process representation of a self-exciting process,
[3] L. Jiang, M.A. Kumar, I.J. Beyerlein, X. Wang, D. Zhang, C. Wu, C. Cooper, T.
J. Appl. Probab. 11 (3) (1974) 493–503.
J. Rupert, S. Mahajan, E.J. Lavernia, J.M. Schoenung, Twin formation from a twin
[23] A. Vinogradov, A.V. Danyuk, D.L. Merson, I.S. Yasnikov, Probing elementary
boundary in Mg during in-situ nanomechanical testing, Mater. Sci. Eng., A 759
dislocation mechanisms of local plastic deformation by the advanced acoustic
(2019) 142–153.
emission technique, Scripta Mater. 151 (2018) 53–56.
[4] A. Vinogradov, E. Agletdinov, D. Merson, Mechanical twinning is a correlated
[24] C.R. Heiple, S.H. Carpenter, Acoustic emission produced by deformation of metals
dynamic process, Sci. Rep. 9 (1) (2019) 5748.
and alloys - a review: Part I and II, J. Acoust. Emiss. 6 (3) (1987) 177–237.
[5] S. Ni, Y.B. Wang, X.Z. Liao, R.B. Figueiredo, H.Q. Li, S.P. Ringer, T.G. Langdon, Y.
[25] A. Vinogradov, E. Vasilev, M. Seleznev, K. M� athis, D. Orlov, D. Merson, On the
T. Zhu, The effect of dislocation density on the interactions between dislocations
limits of acoustic emission detectability for twinning, Mater. Lett. 183 (2016)
and twin boundaries in nanocrystalline materials, Acta Mater. 60 (6) (2012)
417–419.
3181–3189.
[26] A. Chapuis, J.H. Driver, Temperature dependency of slip and twinning in plane
[6] Y. Chang, D.M. Kochmann, A variational constitutive model for slip-twinning
strain compressed magnesium single crystals, Acta Mater. 59 (5) (2011)
interactions in hcp metals: application to single- and polycrystalline magnesium,
1986–1994.
Int. J. Plast. 73 (2015) 39–61.
[27] K.D. Molodov, T. Al-Samman, D.A. Molodov, G. Gottstein, On the role of
[7] K.D. Molodov, T. Al-Samman, D.A. Molodov, Profuse slip transmission across twin
anomalous twinning in the plasticity of magnesium, Acta Mater. 103 (2016)
boundaries in magnesium, Acta Mater. 124 (2017) 397–409.
711–723.
[8] F. Wang, C.D. Barrett, R.J. McCabe, H. El Kadiri, L. Capolungo, S.R. Agnew,
[28] P.G. Partridge, The crystallography and deformation modes of hexagonal close-
Dislocation induced twin growth and formation of basal stacking faults in {1012}
packed metals, Metall. Rev. 12 (1) (1967) 169–194.
twins in pure Mg, Acta Mater. 165 (2019) 471–485.
[29] M.R. Barnett, Twinning and the ductility of magnesium alloys. Part I: “Tension”
[9] E.W. Kelley, W.F. Hosford, Plane-strain compression of magnesium and magnesium
twins, Mater. Sci. Eng. 464 (1–2) (2007) 1–7.
alloy crystals, Trans. Metall. Soc. AIME 242 (1) (1968) 5.
[30] G.S. Kim, Small Volume Investigation of Slip and Twinning in Magnesium Single
[10] D. Drozdenko, J. Bohlen, F. Chmelík, P. Luk� a�c, P. Dobro�
n, Acoustic emission study
Crystals, Ph.D. Thesis, Universite de Grenoble, Grenoble, France, 2011.
on the activity of slip and twin mechanisms during compression testing of
[31] O. Bouaziz, N. Guelton, Modelling of TWIP effect on work-hardening, Mater. Sci.
magnesium single crystals, Mater. Sci. Eng., A 650 (2016) 20–27.
Eng., A 319–321 (2001) 246–249, 0.
[11] J.F.C. Kingman, Poisson Processes, Clarendon Press, Oxford, 1993.
[32] S. Allain, J.P. Chateau, O. Bouaziz, A physical model of the twinning-induced
[12] A. Vinogradov, D. Orlov, A. Danyuk, Y. Estrin, Effect of grain size on the
plasticity effect in a high manganese austenitic steel, Mater. Sci. Eng., A 387–389
mechanisms of plastic deformation in wrought Mg–Zn–Zr alloy revealed by
(2004) 143–147, 0.
acoustic emission measurements, Acta Mater. 61 (6) (2013) 2044–2056.
[33] G.B. Olson, M. Cohen, Kinetics of strain-induced martensitic nucleation, Metall.
[13] K. M� athis, G. Csisz�
ar, J. Capek,
� J. Gubicza, B. Clausen, P. Luk�
a�s, A. Vinogradov, S.
Trans. A 6 (4) (1975) 791–795.
R. Agnew, Effect of the loading mode on the evolution of the deformation
[34] A. Vinogradov, I.S. Yasnikov, Y. Estrin, Stochastic dislocation kinetics and fractal
mechanisms in randomly textured magnesium polycrystals – comparison of
structures in deforming metals probed by acoustic emission and surface
experimental and modeling results, Int. J. Plast. 72 (2015) 127–150.
topography measurements, J. Appl. Phys. 115 (23) (2014) 233506.
[14] E. Pomponi, A. Vinogradov, A real-time approach to acoustic emission clustering,
[35] G.E. Uhlenbeck, L.S. Ornstein, On the theory of the Brownian motion, Phys. Rev.
Mech. Syst. Signal Process. 40 (2) (2013) 791–804.
36 (5) (1930) 823–841.
[15] A. Vinogradov, D. Orlov, A. Danyuk, Y. Estrin, Deformation mechanisms
[36] M. Hiratani, H.M. Zbib, Stochastic dislocation dynamics for dislocation-defects
underlying tension–compression asymmetry in magnesium alloy ZK60 revealed by
interaction: a Multiscale Modeling Approach, J. Eng. Mater. Technol. 124 (3)
acoustic emission monitoring, Mater. Sci. Eng., A 621 (2015) 243–251.
(2002) 335–341.
[16] A. Vinogradov, E. Vasilev, D. Merson, Y. Estrin, A phenomenological model of
[37] M. Kügler, A. Hampel, H. Neuh€ auser, Slip line kinetics during deformation of Cu-Al
twinning kinetics, Adv. Eng. Mater. 19 (1) (2017) 1600092.
single and polycrystals, Phys. Status Solidi Appl. Res. 175 (2) (1999) 513–526.
[17] A. Vinogradov, E. Vasilev, M. Linderov, D. Merson, In situ observations of the
[38] U. Hoffmann, A. Hampel, H. Neuh€ auser, In situ optical microscopic recording of
kinetics of twinning–detwinning and dislocation slip in magnesium, Mater. Sci.
slip line development during plastic deformation of Cu-15at%Al single crystals
Eng., A 676 (2016) 351–360.
with high resolution - estimation of dislocation velocities, J. Microsc. 203 (1)
[18] E. Agletdinov, E. Pomponi, D. Merson, A. Vinogradov, A novel Bayesian approach
(2001) 108–118.
to acoustic emission data analysis, Ultrasonics 72 (2016) 89–94.
[19] E. Agletdinov, D. Merson, A. Vinogradov, A new method of low amplitude signal
detection and its application in acoustic emission, Appl. Sci. 10 (1) (2019) 73.