Machine learning-accelerated prediction of mechanical and microstructural properties of BCC Fe–Cr–Ni–Al high-entropy alloys across the full compositional space

Abstract

This study employs machine learning to accelerate the computation of mechanical properties and microstructural characteristics of BCC FeCrNiAl high-entropy alloys across the entire compositional range. Initially, 49 representative compositional points were selected within the compositional space, encompassing unary to quaternary alloys. Based on ab initio calculations using density functional theory, theoretical Young's modulus, bulk modulus, brittleness-toughness indicators, deviation of atomic positions, shape deformation ratio, and density were obtained. The Categorical Boosting algorithm was then used to develop a composition-property model, predicting properties across the full compositional space. Alloy properties were mapped onto a 2-D plane for systematic analysis, and the SHAP method was used to quantify the influence of individual elements. It was found that fine-tuning the composition can achieve simultaneous optimization of strength and toughness. Specifically, adjusting the Cr content at high FeNi levels effectively enhances ductility, while keeping Al content within the maximum permissible range is crucial for engineering applications that require a balance between strength and density. Further exploration was conducted on the characteristics of the FeAl(CrNi)x series alloys. This study highlights the potential of machine learning to accelerate ab initio calculations methods for complex alloys, improving efficiency and providing a basis for designing and optimizing FeCrNiAl HEAs.