Multi-fault diagnosis of lithium battery packs based on comprehensive analysis of locally weighted Manhattan distance and voltage ratio

Abstract

The diagnosis of faults in lithium-ion battery packs is pivotal to ensuring the operational safety of electric vehicles. A fault diagnosis method is introduced to address the lack of a discharge phase and the existence of a single fault type in traditional diagnostic methods. This method relies on a comprehensive assessment involving locally weighted Manhattan distance and voltage ratio anomalies. The KD-Tree-MLLE dimensionality reduction technique is utilized for fast data processing, which improves the computational efficiency by 1.3 % compared to the original MLLE algorithm. The fault detection method utilizing locally weighted Manhattan distance can meet the same threshold evaluation criteria as the charging phase. This approach resolves the constraint of a single phase and accomplishes comprehensive fault detection. The precision and efficacy of the methodology are confirmed through F1-score evaluation. Accuracy and recall rates are compared under varying thresholds during the charging and discharging phases to determine the optimal threshold value. The ratio between the voltage data of the defective battery and the average voltage data of the normal batteries is calculated, and the comprehensive anomaly analysis method based on the voltage ratio and temperature is proposed. The use of comprehensive data such as voltage ratio and temperature to judge the type of fault can simultaneously achieve the judgment of a single fault and mixed fault types. The applicability of locally weighted Manhattan distance under dynamic circumstances and the efficacy of the voltage ratio analysis method across charging and discharging phases are confirmed through the establishment of an experimental and simulation platform dedicated to lithium-ion batteries. Findings indicate that all faults in both phases can be accurately located when the threshold is set to 0.17. This affirms the superior accuracy of the proposed method in detecting and localizing individual faults and combinations of faults quickly and reliably.