The electric vehicle revolution is generating an impending wave of retired lithium-ion batteries, presenting both challenges and opportunities for sustainable resource management. A critical barrier to effective battery reuse lies in their inconsistent performance. The inconsistent performance presents a major obstacle to the reuse of batteries in energy storage systems and other secondary applications. This study addresses this challenge by developing an innovative two-stage machine learning framework to optimize the sorting process for retired batteries, significantly enhancing their potential for secondary applications. In the first stage, density-based spatial clustering of applications with noise is applied to discharge capacity and ohmic resistance to filter out abnormal cells and estimate the number of clusters. In the second stage, kernel principal component analysis is used to reduce the dimensionality of discharge voltage curves, followed by high-precision grouping with whale-optimization algorithm - fuzzy C-means clustering (WOA-FCM). Experiments on 80 retired batteries demonstrate that WOA-FCM achieves a 55.17% higher calinski-harabasz score than K-Means, 21.90% improvement over gaussian mixture models and 31.86% improvement over self-organizing map, indicating superior inter-cluster separation and intra-cluster compactness. Additionally, WOA-FCM converges 40.91% faster than traditional FCM. The regrouped batteries exhibit markedly reduced variance in static and dynamic characteristics such as capacity, resistance, and voltage plateau. This ensures improved consistency, safety, and economic value of reassembled modules. The proposed framework provides a practical and scalable solution for efficient sorting and echelon utilization of retired batteries, supporting the circular economy of the electric vehicle industry. • Two-stage framework involves static features and dynamic feature. • DBSCAN is used for preliminary anomaly detection and cluster estimation. • KPCA reduces the dimension of discharge voltage curves. • WOA-FCM improves accuracy of retired battery sorting. • WOA-FCM converges 40.91% faster than traditional FCM.
Wu et al. (Tue,) studied this question.