Machine learning guided design of Ce-rich rare earth permanent magnets with outstanding comprehensive magnetic properties

Abstract

The development of high-performance permanent magnets that utilize abundant rare earth elements is essential for addressing supply chain vulnerabilities in critical applications. This study presents a comprehensive machine learning framework for accelerating the design of sustainable Nd–Fe–B permanent magnets using light rare earth elements. Through systematic processes of model training, evaluation, screening, and optimization, XGBoost coupled with particle swarm optimization emerged as the optimal model, achieving an exceptional remanence prediction accuracy of 98% for experimental samples. The machine learning-designed Ce-rich Nd–Fe–B magnets with high abundance rare earth elements exhibited remarkable magnetic properties of 12.7 kG remanence, 12.7 kOe coercivity, and 40.1 MGOe energy product. Multi-objective optimization using Pareto frontier analysis revealed that the developed magnets approach the theoretical Pareto limits under restrictive compositional constraints (greater than 9.5 wt% Ce and free of heavy rare earth elements). This work demonstrates the significant potential of machine learning in guiding the development of high-performance Ce-rich permanent magnets, establishing a robust framework for sustainable permanent magnet production. This framework not only ensures competitive performance but also optimizes the use of abundant rare earth resources, providing critical insights for addressing resource scarcity in advanced materials applications.