Early Diagnosis of Battery Faults Through an Unsupervised Health Scoring Method for Real-World Applications

Battery fault diagnosis is critical to ensure the safe and reliable operation of electric vehicles or energy storage systems. Early diagnosis of battery faults can enable timely maintenance and reduce potential accidents. However, the lead time for detection is still relatively insufficient, and the identification of target vehicle with unidentified fault type has generally been neglected. To fill the gap, an unsupervised health scoring (UHS) method for early diagnosis of battery faults is proposed in this article. First, considering the properties of field data, new features and four types of feature sets related to battery health and fault status are derived for each cell. Then, a novel strategy is proposed to transform a typical classification problem into a quantitative scoring problem by performing multiple clustering. To produce ample clustering results, three cluster algorithms based on different principles are used and the features are randomly divided into feature subsets. By coupling temperature information, early determination of thermal runaway faults can be achieved. Finally, the real-world cloud data of three typical accidents are employed for verification, the results indicate that the proposed approach can innovatively achieve the detection of the abnormal cells at the level of days in advance, demonstrating excellent performance.