Intelligent Design and Simulation of High-Entropy Alloys via Machine Learning and Multiobjective Optimization Algorithms

Abstract

High-entropy alloys (HEAs) are innovative metallic materials with unique properties and wide potential applications. However, the compositional complexity of HEAs poses a great challenge to investigate the physical mechanisms controlling their performance. Herein, we propose a novel framework composed of high-entropy alloys design and simulations (HEADS) that combines machine learning (ML), molecular dynamics (MD), and multiobjective optimization algorithm (MOOA). When considering the disordered characteristics of high-entropy alloys, this framework initially predicts the phase structure of high-entropy alloys with different compositions by using ML and subsequently performs theoretical modeling. Tensile simulations were conducted via MD to generate the mechanical property data, which served as the foundation for further optimization. Within this framework, deep neural network (DNN) models conduct multitask regression to fit the data obtained from the MD simulations, thereby developing an accurate performance prediction model. This model was employed as the fitness function in the multiobjective optimization algorithm to optimize the elastic modulus (EM) and ultimate tensile strength (UTS) of HEAs. The framework is validated using the FeNiCrCoCuAlMg alloy and supports flexible weight assignments for EM and UTS, allowing tailored optimization based on specific application requirements. HEADS framework can provide a robust strategy to accelerate the development of high-performance HEAs and offer new insights for engineering applications requiring advanced materials with optimized properties.