A Flexible Screening Scheme for Retired Lithium-Ion Batteries Based on Novel Capacity Indicator and Random Forest Algorithm

As a large number of lithium-ion batteries are disassembled from electric vehicles, the screening and regrouping of retired batteries are increasingly crucial. Currently, known screening methods such as direct measurement methods for retired batteries exist the problem of low efficiency due to the long test time, and indirect methods lack generalization because of a single dataset. In this article, a flexible screening scheme for two mainstream types of retired batteries based on a random forest (RF) algorithm and new feature is proposed. First, retired battery modules are disassembled into battery cells. Subsequently, the incremental capacity (IC) curves of all batteries are gained by differentiating the capacity–voltage curves, and new curves, namely, incremental IC (IIC) curves, are acquired by calculating the numerical differentiation of IC curves. Second, two RF models are designed with LiNCM and LiFePO4 batteries data, respectively. The peak coordinates of IC and IIC curves are taken as inputs to the model. Finally, a voltage test system is built to evaluate consistency among the retired batteries. The experimental results of 108 samples show that overall screening accuracy reaches 97.2%. In addition, the voltage consistency after screening is improved than before, and the maximum standard deviation drops by up to 14 times.