Machine learning-guided rare earth recovery from NdFeB magnet waste: Model development, parameter influence analysis
Boyang Xu1, Shanshan E2, Jia Liu3
1Key Laboratory of Farmland Ecological Environment of Hebei Province, College of Resources and Environmental Science, Hebei Agricultural University, Hebei, Baoding, 071000, People's Republic of China.
Journal of Environmental Management
|May 3, 2025
Summary
Recycling rare earth elements (REEs) from electric vehicle magnet waste is crucial. This study uses machine learning to optimize REEs recovery, revealing complex influencing factors for efficient and intelligent recycling.
Area of Science:
- Materials Science
- Chemical Engineering
- Data Science
Background:
- Electric vehicle expansion generates significant NdFeB magnet waste, rich in strategic rare earth elements (REEs) like Nd, Pr, and Dy.
- Traditional REEs recovery methods face challenges in optimization, cost-effectiveness, and environmental risks due to waste variability and process complexity.
- Understanding multi-parameter influences on REEs leaching is complex with conventional approaches.
Purpose of the Study:
- To develop an intelligent system for efficient rare earth element (REE) recovery from NdFeB magnet waste using machine learning.
- To bypass extensive experimental optimization and elucidate the complex mechanisms influencing REEs leaching.
- To create a user-friendly tool for guiding efficient REEs recovery processes.
Main Methods:
- Utilized a dataset of 9650 records with 24 input features (waste properties, technological parameters) and 5 output features (REEs and Fe leaching efficiencies).
- Developed and compared 20 machine learning models using four algorithms, with XGBoost showing superior performance.
- Employed R-squared values (0.80-0.99) across training, testing, validation, and cross-validation sets to assess model accuracy.
Main Results:
- The XGBoost model achieved high prediction accuracy (R² 0.80-0.99), demonstrating its effectiveness for REEs recovery.
- Successfully elucidated the intricate influence of waste properties, calcination, and acid-leaching parameters on REEs leaching rates.
- Developed a graphical user interface (GUI) validated by experiments for practical application in guiding REEs leaching.
Conclusions:
- Machine learning offers an efficient and intelligent approach to rare earth element (REE) recycling from NdFeB magnet waste.
- The developed model and GUI can significantly improve optimization efficiency, reducing costs and environmental impact.
- This study provides a pathway for sustainable development in the electric vehicle industry through effective waste management and resource recovery.
