Abnormal Battery Identification via Graph Signal Processing Method

In recent years, the penetration of electric bicycles (EBs) is increasing rapidly in China because of their low prices and convenience. However, fires caused by battery failure in both charging and discharging procedures are reported frequently, leading to property losses and human injuries even life loss. Therefore, early anomaly detection during the charging and discharging procedure of an EB is of great significance for reducing the fire hazard by identifying its aging status and predicting faults. In this paper, we firstly define a series of features indicating the abnormal events on the data collected by the battery management system during EB discharging periods, then apply graph signal processing concepts to detect abnormal discharging events, contributing to further fault location, retrofit and life assessment on battery cells, etc. In the experiments, the proposed method is validated on thousands of discharging cycles collected in the real world under two evaluation metrics and outperforms two benchmarks based on fuzzy c-means and affinity propagation.