Machine learning approach for predicting the fracture toughness of bulk metallic glasses

This study developed a machine learning (ML) model to accurately predict the fracture toughness, K-Q, of bulk metallic glasses (BMGs). To achieve this goal, this work established a dataset of 85 BMGs from the literature comprising the chemical composition and fracture toughness of these BMGs. Various ML algorithms were then employed to develop predictive models for fracture toughness based on the atomic chemical concentration alone. Among the eight ML models that were compared, the extremely randomised trees regression (ETR) model performed the best performance, with the highest determination coefficient (R-2) of 0.779 and the lowest mean absolute error (MAE) of 11.826 MPa root m on the testing set. The generalizability was shown by the pseudo-ternary diagram AlxNiy-4.5Y4.5Co98.5-x-yLa1.5, which demonstrated how the ETR model predicts K-Q based on chemical composition. These findings indicate that the ETR model is very reliable and has great potential for predicting the fracture toughness of BMGs.