Design of Fe-based bulk metallic glasses for maximum amorphous diameter (Dmax) using machine learning models

Abstract

The development of bulk metallic glasses (BMGs) is a topic of current interest due to the unique set of properties that distinguish them from their crystalline counterpart and make them attractive in industrial applications as both structural and functional materials. Currently, a great effort is being made to model and quantify the glass forming ability of the amorphous in an alloy, as well as in tuning their properties in view of the final application of the material. In this work we have used two machine learning techniques, multiple linear regression and tree boosting, to predict the maximum amorphous diameter of Fe-based BMGs, exclusively from the alloy’s chemical composition. The modeĺ s predictive power is characterised by a predicted-R2 of 0.71 (predicted-R = 0.84) and a training-R2 of 0.90 (training-R = 0.95) over a set of 480 alloys present in the dataset. Learning curves are employed as part of a comparative prediction analysis of the two techniques and to help decide the modelling aspects on which effort should be invested in the future. Selected examples using pseudo-ternary diagrams for the design of new Fe-based BMGs are presented, where the potential of the model becomes clear.