A Multi-Fault Diagnosis Method for Lithium-Ion Battery Packs Based on Multi-Source Feature Map Fusion

Fault diagnosis is crucial for maintaining the safe and stable operation of lithium-ion batteries. However, the voltage and temperature during minor fault occurrence do not exceed safety thresholds. The traditional threshold method is difficult to detect minor faults. Therefore, a multi-fault diagnosis method based on multi-source feature map fusion is proposed in this paper. Firstly, a 1-Dimensional Convolutional Variational Autoencoder (1-D CVAE) is applied to expand the voltage and temperature sequence data of the fault occurrence. Then, the voltage and temperature deviation sequence are used to generate the Generalized S-Transform (GST) and Gramian Angular Summation Field (GASF) map, respectively. Next, GST map and GASF map are fused to form a new map feature G-G by concatenation technology, which represents different types of fault features. Finally, a residual network is employed to classify the G-G map for diagnosing External Short Circuit (ESC), Internal Short Circuit (ISC), and Poor Contact (PC) faults. Experimental results demonstrate the Fault Detection Rate (FDR) and Detection Accuracy Rate (DAR) reach up to 99.92% and 99.95%, respectively, which are 4.17% and 4.07% higher than the latest deep learning methods.