An early fault detection method of series battery packs based on multi-feature clustering and unsupervised scoring

With the rapid adoption of electric vehicles (EVs), ensuring the safety of power batteries is critical for reliable operation. Accurate assessment and early detection of lithium-ion battery faults are essential to prevent significant losses and safety incidents. However, accurately identifying and distinguishing potential faulty cells at an early stage remains challenging, as many existing methods fail to meet the requirements for real-time monitoring and reliability. To address these challenges, a novel early fault detection algorithm based on multi-feature clustering fusion for unsupervised scoring and hierarchical warning is proposed in this paper, aiming to detect, locate and tract battery fault cells in real-world scenarios. Initially, features characterizing early cell information from different perspectives are extracted. Then, unsupervised iterative scoring of battery cells is performed using clustering algorithms and a sliding window technique. Finally, a hierarchical warning mechanism for fault detection is implemented based on real-time scoring results and cumulative values to ensure high detection accuracy. The proposed algorithm is validated using three different datasets of early faults. It extends the warning time for abnormal cells by over ten days relative to battery management systems. This significantly enhances early fault detection capability, contributing to the safe and efficient operation of EVs.