Accelerating the development of Fe–Co–Ni–Cr system HEAs with high hardness by deep learning based on Bayesian optimization

Abstract

Machine learning (ML) is progressively supplanting conventional trial-and-error approaches for designing alloys with desirable properties. In this study, four ML regression models were utilized to identify high-entropy alloys (HEAs) with high hardness within the Fe–Co–Ni–Cr system. The Bayesian optimized deep learning (BO-DL) method yielded the highest prediction accuracy (R² = 0.93). Notably, the BO-DL method is no longer limited to a single HEA system and now can target different alloy systems composed of more elements with reliable prediction results. Furthermore, a genetic algorithm was utilized to search for HEAs with high hardness. The accuracy and reliability of the predictions were experimentally verified. As-cast Fe₅Co₂₀Ni₁₀Cr₃₀Al₅Ti₃₀ HEA exhibited a remarkable hardness of 890 HV, which is one of the highest for alloys in the Fe–Co–Ni–Cr system. The methodologies and framework proposed in this study can serve as a blueprint for facilitating the design of HEAs.

Graphical Abstract