Performance prediction models for sintered NdFeB using machine learning methods and interpretable studies.

Various features can benefit the sintered NdFeB material modeling process, as they provide more dimensional information related to the target and make the model more accurate. In this work, by introducing composition and process features as input, we successfully built a sintered NdFeB performance prediction model by comparing different machine learning models with good generalization capability, high accuracy, and sound interpretation compared to previously published work. In addition, using the Shapley additive interpretation (SHAP) method, the unexplainable problem of ML models is solved by evaluating the contribution of the features in the regression model to the results. The intuitive SHAP value plots showed the complex relationship between input variables and magnet performance. Finally, we used the above machine learning model to complete the process framework for evaluating the performance of sintered NdFeB materials. Our work is expected to accelerate performance screening and material development of sintered NdFeB. [Display omitted] • A machine learning model has been developed to predict the remanence and coercivity properties of sintered NdFeB. • The developed models demonstrate high accuracy and generalization on the respective test sets. • Interpretation of machine learning models using SHAP analysis. • The machine learning model is utilized to propose a material design framework for sintered NdFeB. [ABSTRACT FROM AUTHOR]