Research On The High-speed Rail Surface Defects Detection Method Based On Photoacoustic Signals

Railway plays an important role in our life as the artery of our national economy.In recent years, with the improvement of high-speed train running speed, people pay more and more attention to the safety operation of the high-speed train. Among the major railway traffic accidents, a lot of damages are due to the rail fracture which has caused lots of train derailments, so the NDT methods research of rail damage has become very important. At present, the main detection method is based on ultrasonic signal to test the rail damage,it has many advantages such as high accuracy and simple operation characteristics of surface condition, but the surface or near surface of the micro crack detection effect is not very good. A high-speed rail surface defects based on photoacoustic signal of nondestructive detection method is proposed in this paper in order to solve this problem.This technology detects the ultrasonic signal generated by the laser which reflects the organization or the laser energy absorption differences of the specimens. This technology uses the characteristics of the high resolution of the optical signal, so it has a higher sensitivity to the detection of the tiny crack. Photoacoustic signals can be reconstructed as a picture which reflects the initial distribution of light absorption.Through the picture, the defect of rail surface can be seen clearly which enhances the accuracy of the rail damage detection. The contents of this dissertation mainly include the following aspects.At first,the software COMSOL is used to build the rail surface model and obtain the original sound field under laser irradiation;Then using the K-wave photoacoustic toolbox simulates transmission of the photoacoustic signal in the rail and receive photoacoustic signal which in different surface defect cases and in different locations.Then, the photoacoustic images are reconstructed by using the method of time reversal according to the simulation of photoacoustic signal. The accuracy of the results and the influence of sensor parameters for the final reconstruction results were analyzed.Using empirical mode decomposition method to decompose the original photoacoustic signal,and then the corresponding feature parameters were acquired from time domain and time-frequency domain, with which the rail defect feature parameter library was constituted.At last, support vector machine(SVM) classification method is expounded. Then the rail surface defects are classified and recognized preliminary by the SVM classification method, and the grid search method is used to optimize the characteristic parameters of support vector machine to improve the classification accuracy.