Processing Of The Monitoring Data Of High-Speed Train Based On Wavelet Transform

With the speed of high-speed train continues to improve, the security state of train in the process of long-term service will change and make the train safety warning and health maintenance poses special challenges. During the high-speed train running the sensors monitor a large number of vibration data, how to identify the train running status based on tracking and monitoring data and make security state assessment are critical. Vibration monitoring data is nonlinear and non-stationary signals, wavelet analysis has good time-frequency localization advantages. It is widely used in the signal characteristics analysis. Therefore, in this thesis, analyzing characteristic of security state assessment data of high-speed train based on wavelet analysis, and studying the four states about normal state of train, air spring loss of gas, anti-hunting demolition and lateral damper demolition.This paper first introduces the source of monitoring data and the relationships between them, and then starting from the time domain and frequency domain data for monitoring high-speed trains under different conditions conducted a preliminary analysis, they are low-frequency vibration signals the vibration frequency is mainly concentrated in the0-20Hz, and finally through the wavelet packet envelope spectrum analysis of the data obtained characteristic frequency of vibration of three fault condition, and the statistics of the characteristic frequency of the channel number in the channel for the follow-up study provide a basis for selection.On the basis of wavelet transform, using wavelet packet transform to decomposite the band of signal which is more than wavelet transform, and according to the definition of wavelet packet entropy to extract wavelet packet features characteristic of monitoring data. This method decomposite train vibration signal in different frequency bands using wavelet packet, and then obtain the wavelet packet entropy feature vector of each band.and then enter them into the SVM to verify the validity of feature extraction to verify the effective of the feature extraction, and gives a variety of different channel conditions of the recognition rate.Using the method of combining wavelet packet decomposition with autoregressive spectral to extract wavelet packet autoregressive spectral band energy features of the conditions of high-speed train vibration signal. The method of vibration signal by wavelet packet decomposition into different frequency bands, and then build from the regression model for each spectral band signal to strike their autoregressive spectral parameters, the last band to strike their energy as SVM feature vectors state recognition, giving a recognition rate of a single type of channel conditions and mixed failure.