A Fault Rapid-Accurate-Quantitative Integration Diagnosis Method for Lithium-Ion Battery Packs

Fault diagnosis techniques for lithium-ion batteries are essential for enhancing the safety of electric vehicles (EVs). Existing fault diagnosis methods rely on each cell voltages, which cannot be applied practically. The reason is that EVs only provide battery module total voltage and extreme cell voltages. Moreover, it is difficult for a single diagnostic method to guarantee the high fault detection rate (FDR) and high detection accuracy rate (DAR). Therefore, a rapid-accurate-quantitative integration diagnosis strategy based on extremum voltage sequences is proposed to solve the above problems. First, the difference sample entropy (DSE) rapidly detects suspicious battery faults to ensure high FDR. Then, the correlation coefficient method precisely diagnoses suspicious faults to significantly improve DAR. Finally, the deep neural network is used to quantify the defined state of fault (SOF) for the first time. The SOF can indicate the fault degree. Experimental results show that the proposed method can detect the minor short-circuit and open-circuit faults using only the extremum and mean voltages, achieving 98.15% FDR and 98.18% DAR. It is approximately 20% and 22% higher than that of the conventional methods, respectively. In summary, the presented approach shows practical application prospects, which can significantly improve the safety of EVs.