Battery leakage fault diagnosis based on multi-modality multi-classifier fusion decision algorithm

Abstract

With the rapid development of the new energy vehicle industry and the overall number of electric vehicles, the thermal runaway problem of lithium-ion batteries has become a major obstacle to the promotion of electric vehicles. During actual usage, the battery leakage problem leads to the degradation of the system performance, which may cause arcing, external short circuit or even thermal runaway. Therefore, it is essential to analyze the internal mechanism of electrolyte leakage phenomenon and design the corresponding fault diagnosis algorithm. This work tests the disassembled leaking battery module of the practical vehicle. The incremental capacity analysis of the charging process indicated that the battery had capacity loss, and the voltage signal trend analysis of the discharging process found that the leaking battery had higher voltage difference slope. The detection results based on the electrochemical impedance spectrum (EIS) test corroborate the anomaly internal impedance of the battery. Moreover, an incremental capacity bar graph analysis method based on cloud data is designed and the ohm resistance is calculated by a simplified subspace identification algorithm. On this basis, the threshold alarm information is incorporated to form a feature matrix, and a machine learning fault diagnosis algorithm based on multi-modality multi-classifier fusion decision framework is proposed, which is capable of scoring and quantifying the hazard level of fault types such as electrolyte leakage, and achieve up to 26 days advance warning on cloud-based real-vehicle data.