Machine learning of hardness and composition in Al–Co–Cr–Fe–Ni–Mo–Ti high-entropy alloys

Abstract

In the present research, machine-learning (ML) models are used to train a seven-component Al–Co–Cr–Fe–Ni–Mo–Ti system after adding Mo and Ti and removing Cu from six-component Al–Co–Cr–Cu–Fe–Ni high-entropy alloys (HEAs). The composition and hardness of the Al–Co–Cr–Fe–Ni–Mo–Ti HEAs are predicted and screened using five ML algorithms: a multilayer perceptron (MLP), random forest (RF), support vector machine (SVM), extreme gradient boosting tree (XGBoost), and decision tree (DT). The XGBoost algorithm produces the best prediction results, with a coefficient of determination (R²) of 0.95 and a root mean square error (RMSE) of only 43.37. It is found that there is no relationship between the predicted hardness and the Al content in the novel HEAs when the predicted hardness is higher than 800 HV. To evaluate the validity of the ML model, two alloys with greater predicted hardness were chosen for fabrication and hardness measurements. Both alloys were made up of two phases with body-centered-cubic (BCC) structures, and the predicted values were 858 HV and 816 HV, which are in good agreement with the experimental test values of 851 HV and 786 HV, respectively. High-hardness seven-component Al–Co–Cr–Fe–Ni–Mo–Ti HEAs with very little Co content and without a brittle σ phase are worth further study for applications.