The Status Recognition Method Of High Speed Train Based On Fractal And Singular Spectrum Analysis

The security state of high-speed train in the process of long-term service will be changed. A large number of train vibration data collected by the sensors monitoring which contain a mass of train status information can reflect the running state of the train. The approach of extracting the characteristic parameters of high-speed train running state possesses great significance for the inversion and identification of high-speed train running state. Therefore, the status recognition methods based on the fractal and singular spectrum analysis of high speed train state have been proposed, and the characteristic parameters reflecting the running state information for the status recognition of the high-speed train have been extracted. The main work of the thesis are as follows.1. Four common fractal dimensions including capacity dimension, information dimension, correlation dimension and the Hausdorff dimension are analyzed and extracted as features. The sensitivity and stability of the characteristics of fractal dimension for the various fault state of high-speed train are compared. The results show that the fractal dimensions can characterize high-speed train fault state, and own certain stability.2. The high-speed train vibration data have multifractal properties based on multifractal theory. High dimensional feature vectors have been formed with the characteristics of the generalized dimension spectrum parameters, the mass exponent spectrum parameters, the multifractal singularity spectrum parameters. Finally the Support Vector Machine is adopted to achieve the status recognition of the seven kinds of states which contain normal state of train, air spring loss of gas, lateral damper demolition, anti-hunting demolition, air spring loss of gas and lateral damper demolition, air spring loss of gas and anti-hunting demolition, lateral damper demolition and anti-hunting demolition.3. In order to highlight the fractal characteristics of high-speed train vibration signals, the multifractal of the high-speed train has been analyzed based on the theory of multifractal detrended fluctuation analysis. High dimensional feature vectors have been formed with the generalized Hurst dimension spectrum parameters, the mass exponent spectrum parameters and the multifractal singularity spectrum parameters. The Relief algorithm, Mahalanobis distance and Fisher ratio are used for features ordering. The result of feature selection comes from three orderings’weighted average. Choosing the optimal feature subset assess the state of the high-speed train with Support Vector Machine. The experimental results show that the method based on MF-DFA is more effective than the multifractal method for status recognition of high-speed train.4. In order to accurately locate the point of abnormal vibration of high-speed train, The amended criterion of singular spectrum analysis has been proposed to select the number of useful signals which reconstitute the space based on the theory of singular spectrum analysis. The actual monitoring data from the body and frame, and simulation data of suspension parameters failure of the high-speed train have been calculated. The results showed that singular spectrum analysis can accurately detect the abnormal vibration point of the high-speed train.This work was supported by National Natural Science key Foundation of China. (No.61134002)...