Evading the strength-ductility trade-off dilemma in high-entropy alloys by multi-objective machine learning

The strength-ductility trade-off has limited the potential of many structural materials, especially in high-entropy alloys (HEAs). Here, we study the yield strength and ductility of HEAs with an experimental dataset consisting of 144 samples using multi-objective machine learning. First, we construct a feature pool including 20 features from the phase and mechanical properties of these HEAs, and utilize feature engineering to screen and rank the features. Then, the multi-objective random forest and MultiTaskLasso algorithms are chosen to train and predict the yield strength and ductility. The differences between multi-task Lasso and Lasso are compared. Moreover, through the interpretable feature analysis method—shapley additive explanation, the influence of the feature value on the mechanical properties of HEAs is analyzed.