Fast sorting of retired lithium-ion batteries: An EIS and VAE-FINCH based soft clustering method

The rapid growth of electric vehicles has led to an increasing number of retired lithium-ion batteries, highlighting the need for efficient sorting methods for second-life applications. Traditional methods depend on time-consuming charge–discharge cycling procedures, limiting their practical application in large-scale battery processing facilities. To address this challenge, we propose a fast-sorting framework that integrates electrochemical impedance spectroscopy (EIS) with an ensemble variational autoencoder (VAE)-clustering approach. First, we employ EIS to rapidly and comprehensively extract battery health-related data. Second, we develop an integrated architecture combining a VAE with a first integer neighbor clustering hierarchy (FINCH) algorithm, where the VAE incorporates reconstruction loss with specialized loss functions for state of health and direct current internal resistance. Third, we implement an ensemble-based soft clustering strategy that calculates probability distributions for robust sorting results. Experimental validation on a self-tested 381 retired batteries shows that the proposed method outperforms traditional sorting techniques in clustering accuracy and is approximately 8 times faster than charge and discharge-based methods. • SOH-DCIR constrained VAE-FINCH enables parameter-free soft clustering. • Method achieves up to 66.76% reduction in SOH standard deviation within clusters. • Ensemble strategy provides probability-based grouping for flexible battery regrouping. • Experimental validation on 381 retired batteries demonstrates superior sorting accuracy.