Page 130 - Read Online
P. 130
Page 10 of 19 Chen et al. J Mater Inf 2023;3:10 https://dx.doi.org/10.20517/jmi.2023.06
[161]
Figure 5. (A) Three-stage deformation of heterogeneous materials. Reproduced from Ref. . CC BY 4.0; (B) schematic illustration of
[162]
the heterogenous deformation strengthening mechanism. Reproduced from Ref. . CC BY 4.0.
[155]
data cleaning ; (2) data augmentation with binary/ternary eutectics; (3) engineering of data features which
may lead to improved predictability of the ML models even from the small dataset, such as Fuzzy C-means
[156]
clustering function (FCM) and Genetic Programming-based feature extraction using Rough Set Theory
[157] [158]
(GPRST)” . In the literature, Bhowan et al. proposed new parameters to mitigate the issue of
imbalanced data. It includes (1) the average mean square error (AMSE), which uses the average MSE for
each data class instead of the overall MSE for all data, (2) the incremental-reward-assigned accuracy (Incr),
which can differentiate different models with similar accuracy by assigning a higher weighted factor to the
outputs closer to the target value, and (3) the correlation-ratio-based parameter (Corr), which uses the
separability of outputs for different data classes to evaluate the classifier performance. In our opinion, it is
plausible to extend the finding of Bhowan et al. to the data-driven design of EHEAs, which, however,
remains to be an open issue.
Currently, the design of eutectics is still limited in the dual-phase structure while only a few multi-phase
[59]
eutectics were found, e.g. triple-phase eutectics , which makes the database significantly biased towards
dual-phase eutectics and makes it difficult to find multi-phase eutectics using supervised machine learning
models. To solve this problem, one method is to enlarge the database by including more multi-phase
eutectics, which, however, is time-consuming. The other one is to use generative machine learning models,
[159] [160]
such as Variational Autoencoder (VAE) or Generative Adversarial Network (GAN) , to generate
multi-phase eutectics even with the data from binary eutectics. To our best knowledge, this has not been
explored yet for EHEAs.
MECHANICAL PROPERTIES OF EUTECTIC HIGH ENTROPY ALLOYS
Similar to conventional eutectic alloys, EHEAs usually show lamellar or rod-like microstructure with
alternating soft and hard phases. Such a heterogeneous microstructure (HS) can provide a unique strain
hardening capability through the asynchronous plastic deformation of the soft and hard phases during
plastic deformation, which is termed the heterogenous deformation induced (HDI) strengthening
mechanism [161-163] . To rationalize the HDI effect, it was proposed that geometrically necessary dislocations
(GNDs) will be generated during the plastic flow in a heterogeneous microstructure, which pile up along the
interface between the hard/soft phase to maintain the overall deformation compatibility, as illustrated in
Figure 5. As a result, this produces the back stress in the soft phase and forward stress in the hard phase,
leading to the synergy that a more plastic flow is facilitated even at higher flow stress. Consequently, EHEAs
[47,65,164-171]
usually show a balanced combination of strength and ductility, as shown in Figure 6 . To
characterize the effect of HDI strengthening, various experimental techniques have been utilized, such as
electron back scattered diffraction (EBSD) [172-174] , transmission electron microscopy (TEM) [175,176] , digital