Data-Driven Fault Diagnosis and Early Safety Warning for Lithium-Ion Batteries

With the rapid expansion of the Electric Vehicles (EVs) market, early detection of battery faults is increasingly vital to ensure the safety of both individuals and property. This study proposes an effective method for detecting battery faults using machine learning based voltage analysis technology. Initially, the voltage at each sampling point is normalized to enhance the detection of subtle changes indicative of battery faults. Subsequently, a sliding window approach is utilized to extract autoencoder features and standard deviation features within small time intervals. These features are then input into the LOF algorithm to accurately identify abnormal battery cells. Furthermore, a detailed exploration of the influence of autoencoder feature dimensions and sliding window lengths on fault detection outcomes is presented. Through the analysis of the sudden battery faults, potential battery faults of the EVs, the reliability and effectiveness of the proposed method are validated.