Novel intelligent design method of soft magnetic metallic glasses with excellent magnetic properties by utilizing machine learning technology

In this study, a machine-learning (ML)-assisted compositional screening strategy is developed for Fe-based amorphous/nanocrystalline soft magnetic alloys. Typical ML algorithms including random forest (RF), k-nearest neighbor (kNN), and support vector machine (SVM) are applied to establish the ML-assisted compositional screening model of the Fe–based amorphous/nanocrystalline alloy, and grid search, data augmentation together with normalization are utilized to optimize the model parameters and original dataset to maximize the predictive performance of the ML models. Results indicate that the RF model exhibits the best predictive performance and generalization ability in alloy composition prediction, achieving the highest coefficient of determination (R² = 0.931), the lowest root mean squared error (RMSE = 0.014), and optimal cross-validated performance metrics. Based on the feature engineering analysis derived from the RF model, a Fe 81.09 Si 2.86 B 13.04 P 1.8 C 0.82 Cu 0.39 amorphous alloy ribbon is designed and fabricated by melt spinning, which shows excellent magnetic properties including high Bs of 1.67 T together with low Hc of 3.6 A/m after annealing at 703.15 K for 2min. These results demonstrate that ML-assisted compositional screening can provide a useful preliminary route for developing Fe-based amorphous/nanocrystalline soft magnetic alloys when coupled with experimental kinetic optimization.