Flaw Detection In High Speed Railway Based On Vibration Signals

Railway is the main artery of transportation, which is very important for the development of the economy. With the improvement of high-speed railway, the existing flaw detection methods utilizing the rail flaw inspection car and rail flaw detector are low efficiency on the rail. New requirements are proposed on the existing rail flaw detection methods. To solve this problem, this paper proposes a new flaw detection method based on the feature extraction and classification of vibration signals.In order to analyze the characteristics of the rail flaw, we establish the rail model under the ANSYS platform by finite element method and select the monitoring point and flaw structure. By using incentives on the rail, we can get the vibration signals of the rail which include the different flaw positions and the different monitoring points.We study the HHT in order to analyze the vibration signals of the rail. EMD generating undesirable IMFs at the low-frequency region is a shortcoming, we select the IMFs which have strong correlation with the original signal and include the flaw information. Based on frequency response function, the vibration signals are analyzed and the relationship among frequency, amplitude and time are obtained.Then we selcet the appropriate monitoring points and frequency analysis method. We study the NTF algorithm and related theory, then we give the detailed steps of the NTF algorithm. The amplitude frequency signals, monitoring points and flaw types are used to build the tensor signal of the rail for the actual situation. Meanwhile, we use the NTF to decompose the tensor signal and get the corresponding tensor base and the characteristic coefficient matrix of the rail flaw.Support Vector Machine classification method and Relevance Vector Machine classification method are discussed; Meanwhile, in order to expand the two classifier to multiple classifier, we select the one to one method to build the multiple classifier after the analysis of advantages and disadvantages; Then, the signal-to-noise ratio and the number of classification vector are compared for SVM and RVM. The feature vector of rail flaw are classified through the RVM and We obtain the satisfactory results; Finally, we can decide the decision-making criteria of rail flaw for the existence and position of the flaw.In summary, We study on how to build rail system model based on the theory and simulation, and use the corresponding signal processing method and classification method to realize real-time online testing for the high speed railway. Low efficiency and the bad real-time problems of the current rail detection are solved. A new flaw detection method for high speed railway based on the feature extraction and classification of vibration signals is proposed.