Abstract: Under dynamic working conditions, the transient signal of the electric vacuum pump assisted braking system of new energy vehicles has multi-modal nonlinear coupling characteristics, which can easily lead to misjudgement of fault modes. Therefore, this study proposes a new fault diagnosis method. Firstly, by real-time monitoring of core parameters such as the maximum vacuum degree and pumping rate during the operation of the vacuum pump, a dynamic characteristic model of the electric vacuum pump assisted braking system is constructed. Secondly, starting from the aspects of current and pressure, the transient characteristics of the electric vacuum pump assisted braking system are obtained, and continuous transient signals of the system operation are generated; Then, to address the multi-modal nonlinear coupling characteristics of the signal, the transient signal is decomposed using wavelet analysis theory, and the energy entropy features of each frequency band signal are extracted. By finely characterizing the inherent nature of the fault features, the identification and discrimination of the fault features are improved, and a vacuum failure feature library is established; Finally, an intelligent fault diagnosis model is constructed by combining ResCNN network and XGBoost. Energy entropy features are input into the model, and deep level representative features are further explored through convolutional pooling operation. Then, relying on the automatic classification ability of XGBoost, the deep level features are efficiently and accurately classified to output the fault diagnosis results of the braking system. The results show that the ACC value of this method reaches 0.97, which realizes the accurate description of the running state of the braking system, effectively shortens the vacuum establishment time, and reduces the pressure fluctuation and energy consumption of the system.

Key words: electric vacuum pump braking system, vacuum dynamics, deep learning, vacuum failure, fault diagnosis