Machine learning and visualization assisted solid solution strengthening phase prediction of high entropy alloys

Abstract

Solid solution strengthening (SSS) influences the mechanical properties and microstructure of multi-principal high entropy alloys (HEAs). Thus, given the vast compositional space of HEAs, accurately predicting the microstructure of HEAs is regarded as a challenge. Data-driven machine learning (ML) is seen as a powerful tool for predicting the subtle relationship between the composition and structure of HEAs, but the difficulty in interpreting ML models hinders improvement in the performance of phase prediction models. In this work, ML is applied to the modeling of phase prediction, and a complete workflow of ML-based alloy structure prediction is proposed. In this paper, we train an artificial neural network (ANN) model from 12 ML models in a feature pool of 30 material descriptors, which achieves an F1 score of 0.94 on independent test sets. Shapley Additive exPlanations theory is applied to explain how models make classification predictions for phase structures. Finally, the effect of empirical parameters on the phase structure of HEAs was verified by visualization from the ML perspective.