Research On High-Speed Train State Analysis And Fault Diagnosis Based On Fractal Theory

With the rapid development of high-speed train equipment technology and the continuous increase of the train speed in recent years, a greater challenge is presented for the train safety warning and health maintenance. The deterioration of the security performance is brought out such as the parts wear accelerated and the body vibration intensified with the long-term and high-speed service of train. A large number of train vibration data collected by the sensors contains a wealth of train state information. How to use the monitoring data to identify the high-speed train running state possesses great research value and engineering significance for ensuring the safety of the high-speed train. Considering the nonlinear and nonstationary characteristics of the high-speed train monitoring data, the feature analysis of the data and the decision schemes of state identification based on the fractal theory have been proposed in this thesis. The main works are as follows:(1) The research status of the train state assessment drived by the data and the fault diagnosis technology based on fractal are summarized. The bogie structure and vibration characteristics are analyzed briefly. The monitoring data proved to be self-similar fractal characteristics by the Hurst index. Finally, the train state analysis schemes based on the single fractal and multifractal are proposed.(2) Single fractal characteristics of four typical train states (normal, air spring loss of gas, yaw damper failure, lateral damper failure) monitoring data is analyzed based on the box dimension. By comparing the box dimension the train state is identified. Finally, the train state recognition scheme based on EMD and box dimension is presented for overcoming the limitations that the box dimension only represents the overall fractal and improving the state recognition capability of the box dimension. Experiments show that the scheme can effectively improve the train state recognition rate.(3) It has been showed that the multifractal characteristics of different running states have significant differences by the comprehensive analysis of the quality index spectrum, multifractal spectrum and generalized dimension spectrum of the four typical train states monitoring data. The faults more serious, the multifractal is stronger. SVM is used to recognize train state based on the combined features vector of the multifractal spectrum parameters and generalized dimension spectrum parameters. PCA is introduced to eliminate the related interference among the features and the multifractal state recognition scheme is proposed based on PCA and SVM. Experiments show that the scheme can effectively improve the train state recognition rate.(4) By analyzing the monitoring data multifractal characteristics of the train complex fault (parameter gradient fault, mixed fault and single part demolition fault) condition, feature distribution under the parameter gradient fault and the impact on train of mixed fault and single fault and the distribution of train safety features under single part demolition fault are studied. The results show that the train vibration still shows certain regularity under the complex fault condition.