Data-driven machine learning prediction of flow curve and Newtonian to non-Newtonian flow regime transition in bulk metallic glasses within supercooled liquid region

Deformation behavior prediction in bulk metallic glasses (BMGs) was explored through the application of data-driven machine-learning techniques in this study. Multiple ensemble models were trained using a dataset containing 26 distinct systems and 10,386 samples subjected to varying test conditions. Among these models, extremely randomized trees emerged as the most accurate, achieving an impressive R2-score of 0.991. Leveraging the well-developed model, precise predictions of flow stress for specific BMG compositions became feasible by considering deformation strain rate, temperature, true strain, and additional features enhancing the model`s adaptability across various BMGs, not confined to specific systems or bases. Furthermore, the model of this study adeptly generates contour plots that illustrate the relationship between strain rate, temperature, and peak stress, revealing flow regime transitions in addition to the correlation between steady flow stress and strain to investigate the deformation behavior of the metallic glasses. This capability stands as a valuable asset for the streamlined design of new BMGs for net-shape-forming applications, obviating the need for additional experimental endeavors.