Predicting elastic properties of refractory high-entropy alloys via machine-learning approach

Abstract

Refractory high entropy alloy (RHEA) is an important family of alloy with excellent mechanical properties and broad industrial applications. Different chemical and stoichiometric compositions make RHEA a highly tunable material with a huge space for optimization. However, the virtual screening of RHEA is extremely nontrivial, considering the high cost of conventional first-principles calculations. In this work, we compute the ground state and the elastic properties of 2487 RHEAs using exact muffin-tin orbital method combined with coherent potential approximation (EMTO-CPA). Using these data, we train an accurate and robust random forest model that can predict the elastic properties of RHEA systems much more efficiently. Furthermore, we also examine the correlation between seven independent input features and the elastic properties of the alloy. Through this analysis, we identify the most important features that control the elastic properties of RHEA, paving the road to a rational design of the RHEA materials.