Machine Learning-Based Classification, Interpretation, and Prediction of High-Entropy-Alloy Intermetallic Phases

TL;DR

This paper develops machine learning models using physics-informed features to classify and predict high-entropy alloy phases, achieving high accuracy and aiding alloy design.

Contribution

It introduces feature engineering-assisted ML models that effectively classify and interpret high-entropy intermetallic phases with improved accuracy.

Findings

ML models classify phases with 80-94% accuracy

Combining phase-diagram and physics-based features enhances prediction

Synthesized 86 alloys to validate model predictions

Abstract

The design of high-entropy alloys (HEA) with desired properties is challenging due to their large compositional space. While various machine learning (ML) models can predict specific HEA solid-solution phases (SS), predicting high-entropy intermetallic phases (IM) is underdeveloped due to limited datasets and inadequate ML features. This paper introduces feature engineering-assisted ML models that achieve detailed phase classification and high accuracy. By combining phase-diagram-based and physics-based features, it is found that the ML models trained on the Random Forest (RF) and Support Vector Machine (SVM) regressors, are able to classify individual SS and common IM (Sigma, Laves, Heusler, and refractory B2 phases) with accuracies ranging from 80 - 94%. The machine-learned features also enable the interpretation of IM formation. Furthermore, the efficacies of the RF, SVM, and neural network (NN) models are critically evaluated. The phase classification accuracies are found to decrease upon utilizing the NN model to train the datasets. The accuracy of the model prediction is validated by synthesizing 86 new alloys. This approach provides a practical and robust framework for guiding HEA phase design, particularly for technologically significant IM phases.