Machine Learning Assisted Design of High Entropy Alloy Composition and Hardness Prediction

Abstract

High-entropy alloys (HEAs) are multicomponent systems that have attracted significant attention due to their superior mechanical properties as compared to conventional alloys. Among these properties, hardness plays a vital role and is strongly influenced by the selection and concentration of principal alloying elements. However, predicting the hardness of HEAs is challenging due to the complex and nonlinear relationship between composition and mechanical behavior. In this study, an artificial neural network (ANN) model was developed using experimentally reported hardness data for HEAs composed of Fe, Co, Ni, Cr, V, Mn, Al, Nb, and Cu. The model achieved high prediction accuracy, with adjusted R² values of 0.9592 and 0.9023 for the training and testing datasets, respectively. A user-friendly graphical interface was also developed to support the practical application of the model. The model was further employed to evaluate the effect of individual alloying elements on hardness using the Index of Relative Importance (I_RI). Results showed that Al had the highest positive influence on hardness, while Fe exhibited the most negative impact. Elements such as Al, Cr, Nb and V were found to enhance hardness, whereas Co, Cu, Mn, Ni, and Fe tended to reduce it. Finally, the developed model proposed HEA compositions 30Co–10.5Ni–21.1Cr–7Mn–25Al and 16Fe–27.54Co–47.1Cr–6Mn–13.65Nb with a predicted hardness of 733.67HV and 963.8HV, respectively. The predicted hardness was found near to experimental values.

Graphical abstract