Research On High-Speed Turnout Vibration Signal Recovery And Flaw Identification

Turnout is the key component of the railway line and the weak part of the railway, relating to train speed and safety performance closely. Therefore, the study on flaw detection of turnout has great significance for efficient and safe operation of train.The vibration signals generated by train have the advantages of convenient collection, flexible measurement and wide range of applications, containing the abundant damage information of rail. To analyze it, the information of damage characteristics can be extracted effectively. In view of this, a new rail damage detection method based on vibration signals is studied. In addition, in signal acquisition process, data loss may appear due to sensor failure, it affects the accuracy of injury identification. Therefore, a vibration data recovery method based on compressed sensing theory is also studied.The main research contents in this thesis can be outlined as follows:(1) Considering missing data recovery of rail vibration signals, a data recovery method based on K-SVD and ROMP is proposed. The method references the thought of data reconstruction in compressed sensing, observation matrix is constructed accord with the data loss model, and the effective dictionary which can represent signal sparsely is trained by K-SVD. At last, ROMP algorithm with high reconstruction speed and precision is selected for signal reconstrction. According to characteristics of rail vibration signals, features of time domain and frequency domain are chosen as the evaluation index. The repair results of rail vibration signals show that the method can effectively improve the statistical indicators, and superior to the traditional method based on DCT and OMP.(2) Considering flaw feature extraction and condition monitoring of high-speed turnout, a turnout flaw detection based on CEEMD singular entropy and LSSVM is proposed. It makes full use of advantages of CEEMD to analysis non-stationary signal and the function of singular entropy to measure complex sequence, the IMF singular entropy is extracted which can better reflect the characteristics of the rail flaw. And the data fusion of different points can reduce uncertainty and fuzziness of flaw information. Finally, the flaw eigenvectors are inputted LSSVM to train and test. The analysis of vibration signals on simulation experiment platform show that it has better recognition result than the method based on EMD and EEMD. In addition, the proposed method is immune to noises and has good stability when the signal-to-noise ratio is higher than 20 db.Finally, this thesis’s research content is summarized and and further work in the future is discussed.