Robustness of machine learning predictions for FeCo-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. The results show that the experimental property values of arc melted samples deviated less than 14% from predicted values. This work further explains how structural variations across synthesis methods impact property behavior, validating the robustness of ML-predicted compositions and highlighting a pathway for integrating processing conditions into alloy development.