Abstract:

The timely detection and accurate identification of various fault types in battery packs are critical to the safe operation of electric vehicles and battery energy storage systems. Early fault diagnosis is the key to preventing the thermal runaway of lithium-ion batteries in a battery management system. However, existing fault diagnosis methods cannot identify early faults through weak voltage fluctuations. This study proposes an early multiple-fault diagnosis method based on the charging voltage to detect fault batteries with slight voltage changes. First, the battery data are preprocessed, and the features are extracted. The current, state of charge, temperature, and total voltage are selected as the input variables. Next, a voltage prediction model based on the extreme gradient boost (XGBoost) algorithm is established. Compared with the long short-term memory, support vector machine, and random forest methods, the XGBoost-based voltage prediction method effectively reduces the calculation time while improving the prediction accuracy. The voltage residuals obtained through measurement and prediction are used as the fault features, and the mean normalization (MN) method is employed to amplify them. The early faults of the battery packs are classified according to the MN value characteristics of different faults. Finally, density-based spatial clustering of applications with noise (DBSCAN) is used to automatically classify and locate the multiple faults of the battery packs. In conclusion, the clustering method based on DBSCAN identifies multiple faults within 250 s and realizes the accurate classification and location of potential battery unit faults.

Key words: fault diagnosis, charge voltage, battery pack, voltage prediction, mean normalization