Research On Fault Diagnosis Methods Of High-speed Train Bogie Based On DDF-denseNet

With the extensive implementation of high speed railway acceleration,higher requirements for safety and stability are put forward in the operation.Therefore,how to monitor the safety and health of high-speed train(HST)more efficiently has become an important problem to study.As the only connection between the track and the train body,the health of bogie is closely related to safety of the train.Due to the complexity of bogie fault mechanism and the difficulty of visualization of fault features,it is hard to extract key fault information by traditional methods.Therefore,it is of great significance to propose an efficient fault diagnosis method for HST bogies.In this paper,an intelligent fault diagnosis model based on Dense Net is proposed to key components of bogie in high-speed train.The main contents are as follows:Firstly,the bogie and its related research are described in detail.The key components of the bogie are introduced to different dimensions,and the simulation scheme and fault data of the bogie based on SIMPACK are introduced.Furthermore,the simulation data is analyzed by time-frequency transformation and EEMD decomposition.The foundation for the following theories is established in this chapter.Secondly,the fault diagnosis model of HST bogies based on the gate current unit(GRU)is proposed.This model has superior processing ability for time series data,and it can deeply dig for the correlation information on the fault signal.For the original time-domain signal,the GRU can achieve multi-channel feature fusion,and it has a good classification effect on single fault with different speeds.GRU model can extract time-domain data features effectively,but it ignores the key information in frequency domain.In order to fully explore the fault features of data,a dual domain fusion dense convolutional network(DDF-Dense Net)model is proposed.DDFDense Net can combine time and frequency domains to extract complex and abstract highdimensional features,so that the nature of fault signals can be understood more deeply.The experimental results show that the DDF-Dense Net has high performance in both completely damaged recognition and fault location discriminate tasks of failure components.Then,for the performance degradation problems that occurred in the HST bogie,a double domain fusion network based on a classification algorithm is built.The performance degradation of the two key components is evaluated and predicted at different speeds,and the robustness of the model is tested.The performance of the model is evaluated by multiple assessment indicators.Finally,the fault diagnosis model and experimental results are summarized.The shortcomings and problems with the experiment are further analyzed and the future research direction is prospected.