Accelerated design of Fe-based soft magnetic materials using machine learning and stochastic optimization

To significantly expedite the material discovery and design process, we demonstrated a machine learning study of the Fe-based soft magnetic materials database composed of published experimental results, which can be used to efficiently understand and optimize different properties of soft magnetic materials, thus accelerating the design process of next-generation soft magnetic nanocrystalline materials. Various soft magnetic properties, including magnetic saturation, coercivity, and magnetostriction, were studied by different machine learning approaches. Machine learning regression models were trained to predict soft magnetic properties, where random forest shows the best performance. Stochastic optimization was then used to discover new material chemical compositions and secondary processing conditions in order to optimize corresponding properties based on different applications.