Fast Clustering of Retired Lithium-Ion Batteries for Secondary Life with a Two-Step Learning Method

Secondary utilization of retired lithium-ion batteries (LIBs) from electric vehicles could provide significant economic benefits. Herein, based on a short pulse test, we propose a two-step machine leaning method, which combines unsupervised K-means clustering and supervised Gaussian process regression for sorting and estimating the remaining capacity of retired LIBs simultaneously. First, the pulse test to reflect battery aging is detailed, and the significance of the screening process in clustering batteries is validated by the poor clustering accuracy of over 500 unscreened batteries and the various thermal performance of six types of batteries. However, unsupervised K-means can sort out the same type of batteries, which is further verified by the Gaussian mixture model. Furthermore, the remaining capacity of various types of LIBs is given by supervised Gaussian process regression with a correlation coefficient of over 98%. Finally, an automatic sorting machine is designed to corporate with the fast-clustering method, improving the sorting efficiency of retired LIBs.