Machine Learning-Guided Exploration of Glass-Forming Ability in Multicomponent Alloys

Abstract

The prediction of glass-forming ability (GFA) in alloy systems is a challenging problem in material science as well as for metallurgical applications. In this study, we build artificial neural network (ANN) models to investigate the GFA of multicomponent alloys, based on the datasets assembled from ternary alloys as well as quinary alloys prepared by magnetron sputtering. Through training the ANN models with different combinations of datasets, we tackle the problem of the influence of the data source on the model performance, especially the generalizability of the models in predicting the GFA in unseen multicomponent alloy systems. The ANN model trained on a combined dataset exhibits the best performance, specifically low root mean square error in leave-one-alloy-system-out validation and high model robustness, for several CoCrFeNi-based multicomponent alloys. To further verify the ANN models, we synthesize CoCrFeNi-Mo metallic thin films by magnetron co-sputtering and characterize the structure and phase information via x-ray diffraction and electron microscopy. The outcomes of our experiments agree reasonably well with the ANN model predictions, indicating that the data-driven machine learning approach can be a useful tool in the future design of multicomponent amorphous alloys.