Page 91 - Read Online

Page 16 of 45 Mooraj et al. J Mater Inf 2023;3:4 https://dx.doi.org/10.20517/jmi.2022.41

Figure 7. (A) Constructed molecular dynamics model of Fe 80-x Mn Co Cr HEAs with FCC lattice structure and distribution of
10
10
x
elements in the model. This figure is quoted from Pan et al. [121] , copyright 2022, Elsevier; (B) simulation cell used in LAMMPS to
calculate generalized stacking fault energy of the Co-Cr-Fe-Ni system. This figure is quoted with permission from Jarlov et al. [122] ,
copyright 2022, Elsevier; (C) elemental distribution in the Co-Cr-Ni MEA system with different compositions produced by MD
simulation. This figure is quoted with permission from Li et al. [123] , copyright 2021, Elsevier. FCC: Face-centered cubic; HEA: high-
entropy alloy; MD: molecular dynamics.
achieving the highest number of deformation twins. This result illustrates the power of MD simulations to
optimize the alloy composition based on yield strength and tailoring of the simulated deformation
mechanism.

While MD simulations can be powerful tools to predict the material properties of alloys, the high
computational cost associated with these simulations makes it difficult to rapidly produce large datasets for
high-throughput studies . On the other hand, ML techniques are known for their potential to quickly and
[123]
efficiently process and output huge amounts of data and thus offer a means to overcome the low data
output of MD simulations. Li et al. combined high throughput MD simulation with ML to leverage both
techniques’ strengths to explore an extensive data set and provide accurate and detailed information on the
material properties . MD simulations can produce highly accurate predictions of yield strength, but the
[123]
data produced by these simulations have high dimensional input-low dimensional output characteristics.
These properties make it challenging to produce mathematical models to predict the correlation between
input factors and yield strength. On the other hand, ML techniques can produce enormous amounts of data.
Still, their accuracy requires a large and robust set of training data that experimentation cannot do. Thus, Li
et al. utilized high-throughput MD simulation to produce an extensive training data set to train an ANN
that can almost fully explore the composition space of the Co-Cr-Ni medium entropy alloy (MEA)
system . Figure 7C shows examples of different MD simulation models prepared for this study. The
[123]
predictions made by the ANN were shown to be highly accurate. This work highlights the potential for high
throughput MD simulations used in tandem with ML techniques to produce vast amounts of highly