Machine Learning Based Prediction of Phases in High-Entropy Alloys

Abstract

Intro: The vastness of high-entropy alloys (HEAs) compositional space and breadth of microstructures makes it possible to tailor properties required by an application. Predicting the phase formation is therefore important in the design of future HEAs. Machine learning based studies on high-entropy alloys are still relatively scarce and non-standardized. Objectives: this paper reports on the implementation of ML-based classifiers techniques compares the performance of Decision Tree (DT) and Random Forest (RF) for the prediction of phase formation in HEAs. Results: A new dataset based on 1460 microstructural observations collected from 418 peer-reviewed studies was curated. It contains 36 metallurgy-specific predictor features and a dependent variable, which referred to the phase formation. Based on recursive feature-elimination algorithm an expansive collection of predictive feature for future studies is proposed in addition to widely employed features such difference in the electro-negativities (Δχ) and atomic size (δ), valence electron concentration (VEC), mixing enthalpy (ΔHm), and configuration entropy (ΔSm). The RF model yields higher discriminative performance compared to the DT classifier as well as other tree-based benchmark models. The RF model should be the first choice for investigators when building classification. All models shows lower classification accuracy when intermetallic compounds. Summary: These results suggest classic machine learning algorithms may be sufficient to develop high quality predictive models of phase formation in HEAs.