A Preliminary Study On Rail Corrugation Detection Method Of Metro Lines Based On Data–model Dual Drive

Rail corrugation is a type of damage form that widely exits in in metro lines.Its initiation and evolution mechanism is complex and it is difficult to control.It is one of the main reasons that causes abnormal vibration of wheel–rail system,increases wheel–rail noise and shortens the service life of vehicle and track components.Severe rail corrugation may even lead to train derailment accidents.Timely grinding of rail can restrict the development of rail corrugation to the maximum extent and effectively reduce the series of adverse effects caused by rail corrugation.However,how to make a reasonable grinding strategy(e.g.grinding period and grinding amount)is still worth in-depth discussion.Therefore,the efficient and accurate detection of rail corrugation is an important prerequisite to make the grinding work more scientific.It is of great scientific and engineering significance to carry out the research on the detection method of rail corrugation.In this thesis,a series of field tests were carried out on a metro line in China.Through combing field investigation and numerical simulation,the intelligent classification,wavelength recognition and depth prediction model of rail corrugation was constructed,and a preliminary study on rail corrugation detection method of metro lines based on dual drive of data and model was conducted.The main work and conclusions of this thesis are as follows.(1)The measurements of rail irregularity,wheel out-of-roundness and vehicle vibration were carried out on a metro line in China.The dynamic response characteristics of the vehicle caused by rail corrugation excitation were analyzed systematically when the wheel surface is in good condition.Through data analysis and mining,it was found that there is an obvious correlation between vehicle speed,the wavelength and depth of rail corrugation,and the vertical vibration of the axle box.Grasping the interaction law of the above variables can provide an important support for the research of rail corrugation detection.In addition,the wheel–rail wear data measured on site can provide necessary parameter input for dynamic simulation modeling,and the vehicle vibration data can also be used to verify the accuracy of the established model.(2)A one dimensional convolutional neural network(1DCNN)model for intelligent classification and wavelength recognition of rail corrugation was developed based on the data.The intelligent classification and wavelength recognition process of rail corrugation based on 1DCNN was introduced in detail,including data mining and “spatial domain cutting”,nonlinear mapping between vibration and corrugation signal,establishment of sample set,structural design and training of 1DCNN,state classification and wavelength recognition of rail corrugation.The results show that the 1DCNN with appropriate frame structure and configuration parameters can effectively,quickly and stably carry out intelligent classification,wavelength recognition and accurate positioning of rail corrugation.High detection accuracy can be maintained under the complicated operation environment and time–varying speed conditions,which is stable at 99.2%(standard deviation of 0.1).The detection time for each sample is less than 0.2 ms,which meets the timeliness requirement of online monitoring of rail corrugation.Meanwhile,the positioning resolution can be adjusted arbitrarily by setting the length of “spatial window” in the intelligent detection of rail corrugation.(3)Due to the limitation of conditions,the field test cases are often relatively single and cannot cover the diversity of rail corrugation(the combination of different wavelengths and depths).Therefore,a vehicle–track coupled dynamic model considering the flexibility of wheelsets and track structure was established to simulate the axle box dynamic response at different speed under the rail corrugation excitation with different wavelengths and depths.The Kriging Surrogate Model(KSM)was used to construct the response surface of the root mean square(RMS)value of axle box vertical vibration acceleration varying with the vehicle speed and the depth of rail corrugation.The measurable axle box acceleration and velocity signal during the vehicle operation were put into the KSM response surface,and then the particle swarm optimization algorithm(PSO)was introduced to solve the depth of corrugation.The results show that the prediction model of rail corrugation depth based on KSM–PSO can well reflect the characteristics of actual rail corrugation depth.Its relative error is less than 15%,and the average relative error is only 6.75%.