Application Of Blind Source Separation Algorithm In High Speed Train In The Monitoring Data

Under the objective environment of the rapid development of economy, the high-speed trains are gradually demanded by most of people. Safety is the prerequisite condition for developing high-speed train, so it is important and necessary to research the safety performance assessment based on monitoring data of high-speed train. There are more researches on signal fault condition of high-speed train than on multiple fault conditions of high-speed train which can bring about more serious consequences than by signal fault condition of high-speed train. The research on multiple fault conditions based on monitoring data will face with more complex question, such as the critical problems how to ensure the sensor channels used to analyze high-speed train and their quantities, how to select limited diagnostic methods for multiple fault conditions, and how to improve the selected diagnostic methods for multiple fault conditions to make them more practical. Aim at the feature of distribution and signal of high-speed train sensor, the thesis will combine the outstanding hybrid separation characteristics of blind source separation algorithm with the perspective of single channel and multiple channel blind source separation algorithm to analyze the multiple fault conditions of high-speed train.This paper introduces the type fault of high speed train, such as the leakage of air spring, lateral absorber fault, anti-yaw absorber, the leakage of air spring fault mixed with lateral absorber fault, the leakage of air spring fault mixed with anti-yaw absorber. The thesis is aimed at selecting suitable sensor data, which is going to be preprocessed such as smoothing, data Prewhitening and normalization at first, and then separates multiple fault signal with corresponding blind source separation algorithm, and evaluates the effects of blind source separation based on the way of feature extraction. The thesis contains comprehensive data, including complete demolished data, incomplete demolished data and parameter gradient data of multiple faults, of which complete demolished data means demolishing all corresponding components of one operating condition completely, and incomplete demolished data means demolishing one component which will happen probably while the analyze of parameter gradient data refers to a kind of process that the component changes gradually from regular to totally fault.Basing on monitoring data of high-speed train and distribution feature of sensor, the author selects blind source separation algorithm to research, of which KICA algorithm based on information theory and joint approximative diagonalization of eigenmatrix based on algebraic structure will be mainly analyzed. KICA algorithm is applied to blind source separation algorithm of single channel to analyze independent element in the gathered signal of single sensor. Here are the steps:first of all, EEMD will be considered as a kind of method to rise dimensions and PCA as estimation method of the quantities of blind source. And then, combining with KICA algorithm, the author will find suitable objective function and optimization algorithm. Finally, the author will extract and select suitable feature and identify it with SVM. Joint approximative diagonalization of eigenmatrix is applied to blind source separation algorithm of multiple channel. Selecting suitable multiple sensor data for the algorithm is very crucial, with which the author will analyze the signal after preprocessing to find target matrix and the algorithm with algebraic structure of it, and finally extract the blind source signal and its estimation respectively and select suitable feature to evaluate the effect of blind source separation algorithm. With above research, the author will use the algorithm to analyze the simulation data of complete demolished data, incomplete demolished data and parameter gradient data.