A neural network-based method with data preprocess for fault diagnosis of drive system in battery electric vehicles

The dynamic and system reliability of driving system in battery electric vehicles (BEVs) highly depend on the fault diagnosis technology. In this paper, we provided a new data compression approach and validated it on a method based on neural network (NN) to detect both failures' types and degree in drive system. In time-/frequency domain several statistical features were extracted from signals acquired during the simulation with injection of faults. A brief method was introduced to preprocess training data with a comparison to the standard deviation-based method, via analyzing the linear relationship between features and patterns to be classified. In addition, the diagnostic NN's configuration was optimized by the design of experiment. Results indicate the proposed method for data preprocess can significantly improve the efficiency and precision in categorizing all the faults sample especially for fault degree considered in this study.