Machine Learning-Based Prediction of High-Entropy Alloy Hardness: Design and Experimental Validation of Superior Hardness

Abstract

The primary aim of this study is to predict the hardness of high entropy alloys and identify optimal alloy compositions with superior hardness through machine learning techniques. To enhance the accuracy of predictions, a dual-layer algorithmic machine learning model was employed and augmented with Shapley Additive Explanations (SHAP) analysis to increase the model’s interpretability. During model development, multiple machine learning algorithms were evaluated, and innovatively, a combination of the three most optimal model outcomes was incorporated into the prediction process, thus improving the accuracy of hardness predictions. Furthermore, using the Al–Co–Cr–Fe–Ni system as an example, an HEA with a predicted hardness of 776HV was identified from 820,000 datasets. This sample was fabricated using two different preparation techniques and subsequently validated through experimental testing.