Research On Vibration Signal Based High-Speed Turnout Damage Identification

The high-speed railway turnout is the basic equipment and weak link in high-speed railway lines and limits the train speed. Compared with the basic rail, the curved blade withstand greater lateral forces, and under the action of factors as the interaction between wheel-rail, the external loads, environment and natural disasters, the turnout is prone to bend to level, arch back, peeling, cracks and even fracture. The damages above make the turnout’s structure performance decline, and seriously affect the security of the train operation. In view of this, the high-speed turnout damage identification based on vibration signals studied in this thesis.Turnout vibration signals are important information in the high-speed turnout damage monitoring. As the signals are interfered with strong noise during the process of field acquisition and transmission, the accuracy of damage identification based on noisy vibration signals is declined seriously. To overcome this problem, an effective denoising method based on wavelet threshold for turnout vibration signals is proposed in this thesis.The main works studied in this thesis include the following aspects:(1) The research on denoising of high-speed turnout vibration signals based on wavelet threshold is discussed. The wavelet threshold denoising method in this thesis is based on analysis of the relevant denoising method at home and abroad and the classical theory of wavelet threshold denoising put forward by D. L. Donoho. The wavelet basis, decomposition scale, threshold criteria and threshold function are empirically discussed for wavelet threshold denoising. Finally, the optimal denoising method is determined by contrast analysis.(2) The research on turnout damage identification based on the frequency response function (FRF) and principal component analysis (PCA) is discussed. As used for the damage identification, FRF contain plentiful information from turnout vibration signals. Compared with the modal parameters, FRF can reflect the turnout vibration characteristics more directly and accurately. Around the FRF measured from turnout, the method of measurement, assessment and estimation are studied. And the approach of constructing the FRF matrix and damage identification matrix are also studied in turnout damage identification. By introducing the theory of PCA, the principal components extraction, the ellipse control and clustering analysis are studied. The results on real data exhibit that the proposed method can identify the damage preliminary.(3) The research on turnout damage identification based on the basis of the theory of wavelet reconstruction by single branch and Mahalanobis distance (MD) is discussed. Firstly, the wavelet reconstruction algorithm by single branch on the basis of the wavelet transform is studied. Then, the improvement from original single-signal single-FRF to single-signal multi-FRF is studied combined with the analysis of FRF. That is, the multi-FRF is got by dividing the single-signal to multi-signal on the different frequency band, which makes the feature information of the turnout damage more clear on the different frequency band. Thirdly, under the condition of not doing subtraction operation on FRF before and after the turnout damage directly, a novel approach is proposed by means of calculating MD between principal matrix before and after the situation of the turnout damage. Experimental results have proved that the method not only can identify the position of turnout damage, but also can estimate the degree of turnout damage preliminary.Finally, the research work of this thesis is summarized, and the future research direction is indicated.