An end-to-end machine learning framework exploring phase formation for high entropy alloys

Abstract

Exploring the rules of high entropy alloys (HEAs) phase formation has clear guiding significance for the design of new alloys. An end-to-end framework was proposed to select the feature subset and machine learning (ML) model from the feature pool and model pool, respectively. In this framework, each model in the pool is to determine its materials feature subset based on the feature importance. The final model was confirmed through the evaluation of the fitting result of every model and its feature subset. This method extracts important factors affecting the phase formation of HEAs. The results show that the chosen model could classify 430 HEAs into five phases, with test accuracy of 87.8%. And the model analysis suggests that the formation of single-phase solid solution is often inhibited when the atomic size difference is greater than 8.295%.