A Feature Extraction Method and Its Application on Fault Diagnosis for High-Speed Train Bogie

In the process of fault diagnosis for high-speed trains, problems such as high data dimension, large data volume and complex actual working conditions are often encountered. After considering the similarity of the running states of the adjacent points of the adjacent coach and combining with the characteristics of time-varying evolution of the trajectory data in the actual running data samples, a feature extraction method based on DTW(Dynamic time warping) is proposed for the fault diagnosis for high-speed train bogie. This method takes the dynamic evolution characteristics and the traditional static characteristics together as the input of the fault diagnosis model, and comprehensively considers the running state of high-speed trains. Firstly, the sampled data of the sampling points over a period of time are connected as a dynamic trajectory, and the DTW algorithm is used to draw the similarity matrix of the data trajectory of some adjacent sampling points. Secondly, a new characteristic, the deviation degree is defined. The deviation degree of this component is calculated on the basis of similarity matrix by referring to the calculation method of K-L divergence(Kullback-Leibler divergence). Considering the characteristics of time - varying evolution of data samples, this method also reduces the dimension and computational complexity of data to a certain extent. Finally, a simulation experiment is carried out based on the actual operation data of a certain type of high-speed train. The deviation degree and static characteristics are input into the fault diagnosis model to verify the effectiveness of the algorithm in the fault diagnosis for high-speed train bogie.