A voting-based ensemble classifier to predict phases and crystal structures of high entropy alloys through thermodynamic, electronic, and configurational parameters

Abstract

This study aims to predict the various phases present in high entropy alloys (HEAs) and consequently classify their crystal structure employing multiple machine learning (ML) algorithms utilizing five thermodynamic, electronic and configurational parameters which are considered to be essential for the formation of HEA phases. The properties of a high entropy alloy can eventually be traced through accurate phase and crystal structure prediction, which is essential for selecting the ideal elements for designs. Twelve distinct ML algorithms were executed to predict the phases of HEAs, adopting an experimental database of 322 different HEAs, involving 33 amorphous (AM), 31 intermetallics (IM), and 258 solid solutions (SS) phases. Among the twelve ML models, Cat Boost Classifier displayed the optimum accuracy of 98.06 % for phase predictions. Further, crystal structure classification of the SS phase (body-centered cubic- BCC, face-centered cubic- FCC, and mixed body-centered and face-centered cubic- BCC+FCC) has endeavoured for better microstructure evolution using a different database containing of 194 additional HEAs data with 61 FCC, 76 BCC, and 57 BCC+FCC crystal structures and in comparison to the other models tested, the Gradient Boosting Classifier evolved with the highest accuracy of 86.90 %. An ensemble classifier was also introduced to improve the performance of the ML models, resulting in an accuracy increase to 98.70 % and 86.95 % for phase and crystal structure predictions, respectively. Additionally, the influence of parameters on model accuracy was determined independently.

Keywords

• High entropy alloy
• Machine learning
• Computational metallurgy
• Phase predictions
• Ensemble classifier