Research On Identification Method Of Metro Wheel Polygonisation Based On Vibration Acceleration Of Axle Box

During the long-term operation of metro,the wheels are prone to polygonal wear.Wheel polygonal wear is one of the many forms of wheel irregularity in metros.which can make the interaction between the wheel and rail more intense,and the vibration of the wheelset,axle box,rail and other vehicle track components will intensify.Long-term periodic wheel–rail vibration will cause fatigue failures of vehicle track components(such as loose bolts,damaged bearings,broken rail fasteners,etc.)and lead to loud noises,threatening the safe operation of trains and affecting the comfort of passengers.At present,the main measure to solve the wheel polygonal wear is wheel repair,and the premise of the wheel repair strategy is to master the characteristics and status of the wheel polygonal wear.Therefore,how to accurately and timely identify the polygonal wear of wheels is of great significance for reducing the operating cost of metro vehicles and improving operating safety.Aiming at the polygonal wear of metro,this thesis establishes a rigid–flexible coupling dynamics model of metro vehicle–tracks,analyses the effect of polygonal wear on wheels on the dynamic response of wheel-rail system,and builds a polygonal wear classification and wave depth prediction model.The main work and conclusions are as follows:(1)Based on the finite element software ANSYS and dynamic analysis software SIMPACK,a metro vehicle–track rigid–flexible coupling dynamic model considering the flexibility of the wheelset and track was established,and the simulation model was verified using the vertical acceleration of the axle box and the vertical displacement of the rail fieldtested,the results showed that the established metro vehicle–track rigid–flexible coupling dynamics model is accurate and reliable.(2)Using the rigid–flexible coupling dynamics model of metro vehicle–track,the effect of wheel polygonal wear on the vertical force of wheel–rail and the vertical acceleration of axle box,and the phase correspondence between it and the vertical force of wheel–rail or vertical acceleration of axle box was analyzed.The results showed that the wheel–rail vertical force reaches the maximum value when the passing frequency of the wheel polygonization is close to the P2 resonance frequency(63 Hz),while the vertical acceleration of the axle box appears local peaks when the polygon passing frequency is close to the P2 resonance frequency,and the first-order transverse bending mode(93 Hz)and the first-order vertical bending mode(88 Hz)of the wheelset in the restrained state.In a certain wheel polygon wave depth range,the maximum wheel-rail vertical force and the peak axle box vertical acceleration increase almost linearly with the increase of wheel polygon wear wave depth.But when the wave depth is greater than a certain value,the wheel–rail separation phenomenon will occur.The peak acceleration of the box will increase sharply.In the time domain,when the polygon passing frequency is near the resonant frequency of P2,the linear correlation between the changing rate of polygon wear amplitude(time-varying rate of wheel out-of-roundness,the first derivative of wear amplitude to time)and the wheel–rail vertical force is the most obvious,and in the range of 60 ～ 95 Hz,the linear correlation between the time-varying rate of wheel out-of-roundness and the vertical acceleration of the axle box is the most obvious.(3)Based on the rigid–flexible coupling dynamics model,a dataset for the polygonal wear fault classification was built,and a support vector machine(SVM)classifier with output of fully connected layer of the convolutional neural network(CNN)as input was built for the polygonal wear fault classification and identification.The results showed that the model that inputs the features of the fully connected layer of CNN into the SVM for training and classification achieves a recognition rate of 99.82% for normal wheels,low-order polygon wheels,high-order polygon wheels and random non-circular wheels.The recognition rate is also above 80% for samples with noise in the test set.Adding a small amount of noise with different levels in the training set,it still has a recognition rate of more than 90% in the test set with low signal-to-noise ratio.Compared with CNN,DBN,and SVM,its generalization performance and reinforcement learning ability have obvious advantages,and the model also achieved good results in the field-tested data.(4)The vertical acceleration response of the axle box under the polygonal wear excitation of different speeds,different orders and different wave depths was obtained by using the rigid–flexible coupling dynamics model of metro vehicle–track.A kriging surrogate model was used to construct the response surface between the root mean square value of the vertical acceleration of the axle box and the wave depth or vehicle speed.Based on the particle swarm optimization algorithm,the root mean square value of the vertical acceleration of the axle box and the vehicle speed are used as input coordinates to substitute into the response surface constructed by the kriging model,and then the polygon wear wave depth is reversely solved.The results showed that,through simulation verification,the maximum relative error of the model’s wave depth prediction is less than 2%,and the average relative error is only 0.50%.