Exploration of characteristic temperature contributions to metallic glass forming ability

Various combinations of characteristic temperatures, such as the glass transition temperature, liquidus and crystallization have been proposed predictions forming ability metal alloys. We used statistical approaches from machine learning to systematically explore a wide range possible temperature functions for predicting in form critical casting diameter, Dmax. Both linear non-linear models were learn on largest database Dmax values date consisting 747 compositions. find that no combination temperatures features offers better prediction model than themselves, regression suffer poor performance standard metrics like root mean square error (minimum value 3.3±0.1 mm data with deviation 4.8 mm). Examination errors vs. size suggest larger may improve results, although significantly here would likely be required. Shifting focus categorization can weakly distinguish glasses above below our database’s median 4.0 mm, F1 score 0.77±0.02 this categorization. The overall weak results suggests cooling rate might target prediction.