Optimizing Coercivity in Nd-Fe-B Magnets Through Grain Boundary Diffusion: A Machine Learning Approach for Tb-Based Diffusion Source Design

Abstract

Grain boundary diffusion is a promising solution to reduce heavy rare earth usage in high-performance Nd-Fe-B magnets. However, optimizing diffusion source composition is complex due to the wide range of available options. In this study, machine learning was employed to design the Tb-based diffusion source in sintered Nd-Fe-B magnets. Five regression models were evaluated and the support vector regression achieved the highest predictive accuracy. After model iteration, three diffusion sources (Tb21, Tb49, and Tb83) were systematically investigated to reveal the insights between Tb content, diffusion depth, and coercivity enhancement. Experimental validation confirmed that Tb21 provided the 9.5 kOe coercivity enhancement while optimizing heavy rare earth utilization by forming a uniform, continuous Tb-rich shell. Microstructural analysis using EPMA and TEM further demonstrated the influence of diffusion-induced core-shell structures on magnet performance. This study highlights the potential of machine learning for guiding diffusion source selection, reducing experimental costs, and accelerating the development of high coercivity Nd-Fe-B magnets.

Keywords: Nd-Fe-B magnets, Grain boundary diffusion, Machine Learning, Coercivity enhancement