Investigating nanostructure-property relationship of WTaVCr high-entropy alloy via machine learning optimized reactive potential

Abstract

In spite of the promising prospects of the WTaVCr refractory high-entropy alloy (RHEA), the nanoscale structure-property relationship remains largely unexplored. This study introduces a supervised machine learning (ML) framework to develop a charge-transfer ionic potential (CTIP) for the W/Ta/V/Cr/O multi-component system. This novel approach demonstrates exceptional advantages in optimizing numerous parameters of a highly flexible yet physically rigorous potential, leveraging a modified distributed breeder genetic algorithm (DBGA) to balance search comprehensiveness and efficiency. The robustness of developed potential is verified through cross-validation against first-principles predictions on various properties of metals, alloys and oxides. Subsequently, dynamic simulations of annealing, mechanical loading, surface oxidation and radiation collision are conducted utilizing CTIP. Results of these simulations align with previous experiments about nanoscale chemical ordering, plastic deformation behavior, oxidation mechanisms and radiation tolerance. These findings not only further corroborate the reliability of CTIP potential, but also uncover atomic-scale insights that are experimentally unattainable, thereby enhancing the understanding of the nanostructure-property relationship.

Keywords

• High-entropy alloys
• Reactive interatomic potential
• Machine learning
• Transformation induced plasticity
• Oxidation
• Radiation resistance