Retired LiFePO₄ Battery Sorting Using Electrochemical Impedance Spectroscopy and Multi-Modal Feature Fusion

The cascading utilization of retired lithium-ion batteries faces the challenge of high-precision detection cost constraints. This study proposes a sorting method for retired LiFePO₄ batteries based on electrochemical impedance spectroscopy (EIS) using multi-modal feature fusion and deep clustering. Compared to traditional charge-discharge curve, EIS offers rapid detection and in-depth characterization capabilities, providing key insights into the internal electrochemical processes of retired batteries. In this work, EIS is converted into two-dimensional images, with features automatically extracted via the residual network. By integrating the physical features highly correlated with battery aging, which are extracted from the equivalent circuit model and distribution of relaxation times, the limitations of single-feature characterization for the complex aging characteristics of batteries are overcome through feature fusion, thus enhancing the robustness of subsequent battery sorting results. Ultimately, the partitioned confidence maximization clustering algorithm is adopted for battery sorting. It can effectively capture battery intrinsic properties from the fused features and achieve accurate sorting of retired batteries under the tested conditions. The proposed method is validated on a EIS dataset of 374 LiFePO₄ batteries tested at 25 °C and 0% state of charge. Experimental results demonstrate that the proposed method achieves values of 0.667, 103.061 and 56.033 for the adjusted rand index, capacity consistency and voltage consistency, respectively, representing an improvement over traditional clustering methods and single-feature methods.