Data-driven Based Incipient Fault Real-time Diagnosis Of High-Speed Railway Traction System

Reliability improvement of electric traction system is extremely important for the safe operation of high-speed trains.The electric traction system of high-speed train is mainly composed of transformer,rectifier,inverter,traction motor and et al.Due to the complex working environments such as high temperature and electromagnetic interference,various types of incipient faults will appear in the electric traction system with time goes by.If such kinds of incipient faults can not be timely detected and diagnosed,it will bring enormous danger to the safe operation of high-speed trains.Therefore,the incipient fault detection and diagnosis of high-speed train electric traction system have important theoretical significance and engineering application value.In this thesis,the structure of high-speed train electric traction system and common faults are introduced firstly.Then several data-driven based incipient fault diagnosis schemes are proposed for incipient fault detection and isolation for high-speed train electric traction devices.Finally,the effectiveness of proposed methods is verified on TDCS-FIB(Traction Drive Control System-Fault Injection Benchmark)simulation platform co-developed by our research team and Central South University.The main works are presented as follows:(1)In this thesis,a real-time incipient fault diagnosis of high-speed train electric traction system based on Deep-PCA and KLD(Kullback-Leibler Divergence)is proposed.Firstly,the six modes corresponding to the six working conditions of inverter are constructed.Then,incipient fault detection and diagnosis are realized based on the improved Deep-PCA theory in each mode.Finally,the KLD in information field is applied on fault isolation.(2)In view of the small amplitude characteristics of incipient faults,a method focusing on signal to noise ratio enhancement is proposed to improve the incipient fault detection and diagnosis performance in this thesis.Firstly,an intelligent decomposition levels selection scheme is proposed by quantifying the similarity of detail components,and the selection of noise threshold is determined and optimized to further improve the accuracy of noise reduction.Then,a continuous wavelet transform-based fault information enhancement approach is proposed.Finally,the detectability and isolatability of Deep-PCA are analyzed at the same time.(3)In this thesis,the data filling of large number continuous lost signals in traction system of high-speed train is realized on the basis of the combination of kernel function and Makima(Modified Akima)filling method via weighting coefficients.Based on which,the fault detection ability is improved by combining Deep-PCA and MSPCA(Multi-Scale Principal Component Analysis).Finally,accurate fault isolation is realized by increasing the fault information in data set based on wavelet transform.