Machine Learning Model for Nd₂Fe₁₄B-Based Permanent Magnets

Area of Science:

• Materials Science
• Computational Materials Science
• Magnetism

Background:

• Neodymium iron boron (Nd2Fe14B) magnets are critical for modern technologies.
• Increasing demand for magnet recycling necessitates understanding the impact of diverse impurities.
• Impurity elements like Cr, Mn, and Sm complicate magnetic property prediction.

Purpose of the Study:

• To develop an efficient machine learning (ML) model for predicting magnetic property changes in Nd2Fe14B magnets.
• To accurately predict alterations in saturation magnetization (Ms) and magnetocrystalline anisotropy (K1).
• To create a versatile model capable of handling numerous substitutional elements and concentrations.

Main Methods:

• Implementation of an efficient machine learning model.
• Utilizing feature engineering to address challenges posed by specific impurity elements.
• Training a single, comprehensive model for diverse impurity scenarios.

Main Results:

• Demonstrated the sufficiency of relatively simple ML models for complex magnetic property prediction.
• Successfully predicted magnetic property changes across a wide range of impurity elements (>20).
• Developed a unified model effective for various impurity concentrations.

Conclusions:

• The developed ML model efficiently predicts magnetic property changes in Nd2Fe14B magnets with impurities.
• Feature engineering effectively mitigates challenges from complex magnetic elements, enabling a single predictive model.
• This approach supports the recycling of permanent magnets by predicting property variations due to impurities.