Machine Learning-Assisted Design and Synthesis with Excellent Performance of Fe-Based Amorphous/Nanocrystalline Soft Magnetic Materials

Soft magnetic materials with high saturation magnetic induction strength hold significant potential for various applications, particularly in the electromagnetic field. By harnessing the power of machine learning (ML) as an alternative to traditional trial-and-error experimentation, not only can we minimize experimental costs but also achieve faster and more precise designs of soft magnetic materials that meet specific requirements. In this study, we employ ML models to predict the amorphous formation ability (GFA), saturation magnetization strength (Bs), and coercivity (Hc) of alloys. Through meticulous hyperparameter tuning, the eXtreme gradient boosting (XGBoost) model emerges as the top performer, exhibiting a high determination coefficient (R2) and a low root mean square error (RMSE). Leveraging the characteristic importance insights provided by the XGBoost model, we successfully design and fabricate high-performance Fe-based amorphous alloys with a composition of Fe77B15Co6Si1P0.6Nb0.4. In the as-cast state, the Fe77B15Co6Si1P0.6Nb0.4 alloy demonstrates a saturation magnetization of Bs=1.77T and a coercivity of Hc=28.34A/m. After undergoing a suitable annealing treatment, the alloy's saturation magnetization increases to Bs=2.01T, and its coercivity enhances to Hc=36.23A/m. Remarkably, these observed variations in Bs and Hc for the Fe77B15Co6Si1P0.6Nb0.4 alloy align perfectly with the predictions made by the XGBoost model. Consequently, this study contributes significantly to the efficient development of high-performance soft magnetic materials.