Prediction of mechanical properties of high entropy alloys based on machine learning

Abstract In recent years, the ideal- properties (young’s modulus, yield strength, toughness) and advanced application potential of high-entropy alloys (HEAs) have attracted numerous researchers. However, due to their unique structure and multiple structural combinations, it is challenging to explore the impact of various factors on their mechanical performance solely through experiments. This study considers the concentrations of five high-entropy alloy atoms and working temperature as input parameters. Molecular dynamics (MD) simulations and machine learning (ML) algorithms are employed to predict the tensile properties of FeNiCrCoCu HEAs, including Young’s modulus ( E ) and toughness ( uT ). A dataset of 1000 HEAs is generated through MD simulations, and feature selection is conducted using principal component analysis and Spearman correlation analysis. XGBoost, RF, DT, LGBoost, and AdaBoost are utilized to predict the mechanical properties of HEAs, comparing the impact of the two feature selection methods on prediction outcomes. During ML model training, 10-fold cross-validation and grid search are employed to obtain the best models and parameters. Root mean squard error ( RMSE ), coefficient of determination ( R 2 ), mean absolute error ( MAE ) and relative absolute error ( RAE ) are used as evaluation metrics. Results indicate that Spearman correlation analysis for feature selection outperforms principal component analysis, and XGBoost demonstrates superior predictive performance for the mechanical properties of HEAs compared to other models. Predictions for E are more accurate than those for uT , with R 2 exceeding 0.9 for four out of the five ML models. This work may provide a new feature selection method for studying the mechanical properties of HEAs through ML. In the future, this method can be applied to other research areas of HEAs compositions, providing theoretical support for experiments. It can then be further applied to critical fields such as biomedical and aerospace industries.