Research On High-speed Turnout Vibration Signal Endpoint Detection And Distortion Of Identification

High-speed turnout is a fragile part in railway line, its enormous deformation and damage severely endanger rail safety, so the research on turnout flaw identification is a hot issue. In the research, acceleration sensors collect a lot of data, if all the data is used in high-speed turnout damage detection, it will bring problems of slow speed and poor real-time, so signal endpoint detection and effective vibration signal extraction is of great significance. In addition, because of the sensor malfunctions in signal acquisition part, turnout signal will be mixed with the distorted phenomenon, it is difficult to provide an objective and truthful information for the turnout damage detection, which ultimately affects the rail detection, so the distortion identification of high-speed turnout vibration signal is of great significant.The main contributions in this thesis are as follows:(1) In the part of high-speed turnout vibration signal endpoint detection, because of the non-stationary characteristic of turnout vibration signal, characteristic energy based on the Hilbert Huang transformation is adopted as the different characteristic between vibration signal and noise, and order statistic filter is applied to the characteristic to enhance the identification degree between vibration signal and noise, then the fuzzy c-means clustering is carried out on the energy characteristic to estimate threshold adaptively. Finally, the method of this thesis and the method based on short-term energy and the method based on sub-band spectral entropy are compared, the simulation results prove that the thesis method is feasible, and the effective vibration signal extracted by endpoint detection can improve the turnout injury identification rate to a certain extent and reduce the turnout injury identification time.(2) In the part of turnout vibration signal distortion identification, firstly, on the basis of the measured normal turnout vibration signal, the distortion signals of sensors drift, stuck fault in sensors, and shock interference are simulated. According to the non-stationary of four signals, three characteristics are extracted based on EMD:the intrinsic mode energy entropy, characteristic energy and residual energy. Finally, the turnout vibration signal distortion identification is carried out respectively by nearest neighbor algorithm based on FCM-subtraction clustering, by nearest neighbor algorithm based on FCM-FCM clustering and by BP neural network method, the comparing results verify that the result by nearest neighbor algorithm based on FCM-subtraction clustering is closer to the actual result.Finally, this thesis’s work is summarized and further work in the future is discussed.