Research On On-rail Sensing Algorithm For Wheel Defects Of High-speed Train

Railway has been the most significant method of transportation in China. As the destiny and speed of railway transportation increase, the safety of railway transport becomes more and more important. In the vehicle and track system, the speed of relative motion between the wheel and rail is very high, and the dynamic load is quite big. So the defects are more likely to appear on wheels. Once there are defects on wheels such as flat, out-of round and eccentricity, impact between the wheel and rail may be produced, which is higher than normal one in a few times, even one hundred times. The vehicle and rail will be serious damaged. So the find of wheel defects in time is of great importance.In this article, vehicle and track vibration characteristics are analyzed based on vehicle-track coupling system mode and the dynamic equation of vehicle and track vibration is built. Newmark predictor-corrector integration is chosen to compute dynamic equation. IFFT method is used to simulate the track geometry irregularity. The paper realizes mathematic model of wheel defects and get the simulation data under wheel defects. The signal processing algorithm is used to extract the features of vibration signal. The paper comes up with a new method to extract information. First, using empirical mode decomposition (EMD) decomposes the rail vibration signal, and gets different modal components. In the intrinsic mode function, Fractal dimension and bispectrum is used to representative the information of wheel defects. Bispectrum is analysed by Gray Level-gradient Co-occurrence Matrix. Fractal dimension and the information of Gray Level-gradient Co-occurrence Matrix make up the track vibration characteristics.To recognize the wheel defects, the article proposes an algorithm to recognize and estimate the wheel defects, used support vector machine which uses vibration characteristics to train SVM. And SVM is promoted in this article, using PSO (particle swarm optimization) to optimize the parameter of SVM and GPU to accelerate the train speed of SVM.The simulation result shows that the algorithm the paper came up with is effectivity to recognize the wheel defects such as flat, out-of round and eccentricity, recognition rate can be achieved100%. the correct rate is91.46%.