Application of Machine Learning in Exploration of Uncharted Territory of Multi-Principal Element Alloys

Abstract

Multi-principal element alloys (MPEAs) present an enormous (~ 10⁸ distinct types of compositions), largely uncharted compositional landscape, of which only a small fraction has been discovered to date. The critical task for materials scientists and metallurgists is to identify potential compositions with tailored properties for specific application while minimizing the reliance on laborious and energy-intensive experiments. In this study, we developed a robust machine learning framework utilizing nature-inspired optimization method to navigate potential MPEA candidates. Cuckoo Search Optimization (CSO), a metaheuristic algorithm based on the brood parasitism behavior of cuckoo birds, was utilized to generate novel MPEAs with targeted mechanical properties (yield strength, ultimate tensile strength and elongation). A dataset of 700 instances, sourced from experimental literature on MPEAs, was utilized. CSO explored the search space to generate novel MPEAs with good combinations of mechanical properties (yield strength (YS), ultimate tensile strength (UTS) and elongation (????), by iteratively replacing inferior solutions with better ones, converging towards optimal or near-optimal solutions. Pareto front solutions were then identified using Pareto dominance, ensuring that no individual objective among the multi-objective criteria (YS, UTS and ????) could be improved without compromising another. Thus, this research leverages the potential of machine learning in accelerating the discovery of high-performance MPEAs, paving the way for future innovations in materials design and engineering.