Research On Wheel-Rail Surface Defect Detection Based On EMD Algorithm

With the development of rail transit modernization,the traditional transportation system will face many new problems.The higher the speed is,the more prominent the safety problem is.During the operation of trains,it is necessary not only to keep the vehicle from derailment and make it run safely,but also to ensure the running stability and comfort.The faster the vehicle runs,the stronger the vibrations will be caused by minor irregularities and surface defects between vehicle and track.Since the vehicle-track coupling system is a non-linear dynamic system,all kinds of small vibrations will lead to stronger vibrations which will ultimately affect the comfort of passengers and the integrity of cargo.In this sense,it has a significant importance to detect surface defects between vehicle and track.There are many kinds of detection methods for wheel-rail surface defects,such as non-destructive testing,machine vision,vibration technology.The vibration technology is the most popular one which can guarantee the detection speed and the convenience of equipment installation.The vibration signal analysis is an on-board condition monitoring method in service.This method has the characteristics of minimal effects on vehicle system,convenience for equipment installation and less influences on the environment.The vibration of the vehicle system is non-linear and non-stationary because of the non-linearity of the vehicle-track coupling system and the discontinuous impact of wheel-rail surfacet defects.However,since most of time-frequency analysis methods have their own limitations,the Hilbert-Huang Transform(HHT)method is widely used to analyse non-linear and non-stationary signals.The HHT method includes the Empirical Mode Decomposition(EMD)and Instantaneous Frequency(IF)calculations.In this thesis,the EMD method is used to decompose the vibration signals of the vehicle system.The goal is to apply the EMD method to detect wheel-rail surface defects in the vehicle-track coupling system.Based on the EMD method,this thesis studies the adaptive demodulation method for detecting wheel-rail surface defect in vehicle-track coupling system.The contents are as following:(1)The simulation model of wheel-rail surface damage is developed and solved by the new explicit integral method.A passenger vehicle and ballasted track are selected.The simulation model is employed to analyze the vibration response of the vehicle-track coupling system under various damage excitations.(2)In view of the discontinuity of vibration and shock in practical working conditions,a generalized intrinsic mode function is proposed.The EMD method based on generalized intrinsic mode function(EMD-GIMF)is used to analyze the vertical acceleration signal of bogie frame of vehicle with wheel flats(single flat and dual flats).And then the generalized intrinsic mode functions(IMF)are demodulated to obtain the final envelope spectrum by Hilbert Transform.It is proved that the discontinuity of vibration signal can be solved by EMD-GIMF in the signal analysis of vehicle-track coupling system.The feature frequency of wheel flat fault can be extracted and the wheel flat can be identified effectively.(3)The possibility of applying Ensemble Empirical Mode Decomposition(EEMD)method to detect multi wheel-rail surface defects is studied.Then EMD based on binary time scale(EMD-BTS)is proposed.This method can effectively solve the modal aliasing phenomenon in multi wheel-rail surface defects detection.It is also used to decompose the vibration data with wheel flat and track harmonic irregularity,wheel flat and rail corrugation.The cross-covariance method is used to remove the pseudo-IMF components.Finally,the sensitive components are selected according to the energy entropy of IMF.The energy operator spectrum of the sensitive components is calculated.The results can identify the feature frequencies of two kinds of wheel-rail surface faults simultaneously.(4)In order to identify the single wheel-rail surface defect and multi wheel-rail surface defects automatically,the EMD-BTS method is used to decompose vibration signals.After removing the pseudo-IMF components caused by over-decomposition,the sensitive components are picked out.The AR model of sensitive components is built.The auto-regressive coefficients and residual variances of AR model are selected to construct the feature vectors.Extreme Learning Machine(ELM)is used for fault pattern recognition.The results show that it can identify wheel-rail surface defects automatically,and has high accuracy and practical values.