Intelligent metal recovery from spent Li-ion batteries: machine learning breaks the barriers of traditional optimizations

Abstract

As the retirement of new energy vehicles peaks, efficient recycling of spent lithium-ion batteries (LIBs) is significant for environmental sustainability. Roasting integrated with water leaching is a popular process for metal recovery from spent LIBs. However, traditional optimization experiments face the challenges of lengthy and costly procedures due to the variability of waste materials and the intricate interplay between process variables. This study breaks through the barriers by introducing machine learning (ML) to establish a smart prediction model for efficient metal recovery from LIBs. Based on the 8921 data collected, the model incorporates 18 input features, encompassing waste particle size, components, and roasting–water leaching parameters, and predicts Li, Co, Mn, and Ni recovery efficiencies. Four ML algorithms are compared to determine the best prediction models with R² values of 0.81–0.98 in the training and test datasets. The intricate interaction mechanisms of each feature with metal recovery were revealed, providing a deeper understanding of the recovery process. Furthermore, we developed a user-friendly GUI that instantly suggests optimal parameters for maximizing the metal recovery efficiency, simply by inputting waste particle sizes and components. Finally, the reliability and practicability of the GUI are verified by experiments. This work dispenses with the traditional and extensive optimization experiments and decreases the recovery costs, which achieves efficient and intelligent spent LIB recycling.