A novel intelligent method for fault diagnosis of electric vehicle battery system based on wavelet neural network

Abstract This paper proposes a method of fault detection of Lithium-ion batteries based on wavelet-neural for guaranteeing the safety and reliability of electric vehicles (EVS). In actual operation of electric vehicle, many factors, such as electromagnetic interference, road condition, and driving habit can make the battery fault system complex, non-liner or multi-parameter coupling, which makes it difficult to identify faults accurately only by a single parameter. The data of voltage fluctuation is obtained through the simulation of the battery charging and discharging experiment under vibration environment. The proposed method eliminates the voltage signal noise by decomposing and reconstructing discrete wavelet transform (DWT). The parameters of voltage, voltage difference (VD), covariance matrix and variance matrix are used as input values of general regression neural networks (GRNN) to classify the fault status. Through in-depth analysis of the correlation degree between the parameters and the fault signal, we find that the VD value has a strong correlation with fault diagnosis. The experimental data shows that the method proposed in this paper can significantly improve the efficiency and precision in the classification of fault degree.