Robustness of machine learning predictions for Fe-Co-Ni alloys prepared by various synthesis methods.

Developing high-performance alloys is essential for applications in advanced electromagnetic energy conversion devices. In this study, we assess Fe-Co-Ni alloy compositions identified in our previous work through a machine learning (ML) framework, which used both multi-property ML models and multi-objective Bayesian optimization to design compositions with predicted high values of saturation magnetization, Curie temperature, and Vickers hardness. Experimental validation was conducted on two promising compositions synthesized using three different methods: arc melting, ball milling followed by spark plasma sintering (SPS), and chemical synthesis followed by SPS.