ABSTRACT With the widespread adoption of electric vehicles (EVs), battery failures have become a significant concern. To safeguard the lives and property of drivers, accurate and prompt diagnosis of battery failures is crucial. This paper proposes a novel battery fault diagnosis method based on the Relative-Range-Feature (RRF) and an improved Theil index, utilizing actual operating data from EVs. Firstly, the trend component of the voltage is extracted utilizing Singular Spectrum Analysis (SSA) to eliminate noise from the original voltage data. Secondly, to address the challenge of inconspicuous early fault features, a new RRF feature extraction method is introduced. This method significantly amplifies the feature differences between normal and faulty batteries. An improved Theil index is proposed to achieve prompt fault detection with high robustness. Finally, a faulty battery localization algorithm that combines Multidimensional scaling (MDS) and Mahalanobis distance (MD) is proposed. This algorithm excels in distinguishing faulty batteries from normal ones. Validation using actual vehicle data demonstrates that the proposed algorithm offers significant advantages in reliability, effectiveness, and robustness. Compared to the information entropy and correlation coefficient methods, this method demonstrates superior applicability.
Read full abstract7-days of FREE Audio papers, translation & more with Prime
7-days of FREE Prime access