[P2] AI, Simulations & Theory

[P2-6] Machine learning guided design of high-performance RE-Fe-B with abundant rare earth substitution

*Zheng Wang^1,2, Jing Wang^1,2, Fengxia Hu^1,2, Baogen Shen^1,2,3,4 (1. Beijing National Laboratory for Condensed Matter Physics,Institute of Physics, Chinese Academy of Sciences, Beijing (China), 2. School of Physical Sciences, University of Chinese Academy of Sciences, Beijing (China), 3. Ganjiang Innovation Academy, Chinese Academy of Sciences, Ganzhou, Jiangxi (China), 4. Songshan Lake Materials Laboratory, Dongguan, Jiangxi (China))

Keywords:

RE-Fe-B,machine learning,high-abundance rare earth elemenets,electronegativity

The optimization of high-abundance RE-Fe-B (RE = PrNd, La, Ce) permanent magnets with optimal magnetic performance has been a key research focus to reduce reliance on critical rare earth elements. However, traditional trial-and-error methods are constrained by experimental costs and time limitations. Recently, machine-learning methods have been widely applied in scientific research and have become essential tools for accelerating materials design [1-4]. Here, we propose a machine-learning approach to accelerate the design of melt-spun high-abundance (PrNd,La,Ce)-Fe-B ribbons[1]. Based on a database incorporating elemental electronegativity, composition and corresponding magnetic properties, we develop novel machine-learning models that combine heuristic optimization algorithms and ensemble strategies to enhance both model accuracy and generalization ability. Based on the established machine-learning models, we systematically predict and investigate the relationship between rare earth content, electronegativity parameters and overall magnetic performance (coercivity, remanence and maximum energy product) of RE-Fe-B. Our analysis reveals the correlation between electronegativity parameters and magnetic performance, suggesting their potential as new indicators for magnetic performance in RE-Fe-B (Fig.1a). Further predictions for representative compositions of (PrNd_xLa_yCe_1-x-y)₁₂Fe₈₂B₆ reveal a high-abundance rare earth compositional range with optimal overall magnetic performance, where the La content ranges from 25% to 40%, and Ce content reaches up to 20% (Fig.1b). Experimental validation supports the accuracy of our predictions, with seven randomly selected compositions achieving over 90% accuracy in comparison with predicted and experimental values.(Fig.1c) To identify the most cost-effective composition, a cost-performance analysis within this high-abundance compositional range (Fig.1d) determines that (Pr,Nd)_8.1La_3.6Ce_0.3Fe₈₂B₆ retains 86.3% of the overall magnetic performance while reducing material costs by 31.3% compared to compositions without La and Ce. Three other cost-effective compositions are also discovered, achieving cost reductions of 31.4%, 28.9%, and 31.3% while retaining 81.2%, 80.6% and 82.1% of overall magnetic performance, respectively. These findings advance the optimization of RE-Fe-B compositions with high-abundance rare earth elements, demonstrating the enormous potential of machine-learning approach in the design and development of high-performance and cost-effective RE-Fe-B permanent magnets.

Reference
[1] Z. Wang, J. Wang, F.X. Hu, B. G. Shen, et al. Acta Mater. Under revision.
[2] Z. Rao, P.Y Tung, R. Xie, et al. Science, 2022, 378(6615): 78-85.
[3] J. Yang, F. Shi, C. Zhou, et al. Adv. Funct. Mater., 2024, 34(52): 2411170.
[4] D. Xue, P. V. Balachandran, J. Hogden, et al. Nat. Commun., 2016, 7(1): 1-9.

Download high resolution images(7.4 MB)