Identification On Wheelset Hunting State And Its Application On Small-amplitude Hunting Forewarning Before Turnout Passing

A certain high-speed train of China is in abnormal small-amplitude hunting vibration of the wheelset on the ordinary line due to the failure of the secondary lateral damper,and evolves into hunting instability after passing the turnout.The effect of turnout on the evolution of the small-amplitude hunting limit cycle is studied in this thesis.To address the problem that the existing monitoring standard uses the framing acceleration signal as the monitoring quantity,and its characteristics may not be equivalent to the hunting state of the wheelset.Therefore,a method to identify the hunting state of the wheelset based on the simulation data of the lateral displacement of the wheelset is proposed.Firstly,the simulation model is used to generate the wheelset displacement data for hunting state classification,which needs to contain the data of three states: normal hunting,abnormal small hunting and hunting instability.For the data form the vehicle system following the subcritical bifurcation type,an unsupervised learning approach is used in this thesis to classify the hunting classification labels of the data samples based on the limit cycle theory.The classification results are based on the wheelset displacement,which cannot be economically and accurately measured in online monitoring yet.Therefore,a One Dimension Convolutional Neural Networks(1D CNN)classifier is built in this thesis,and the hunting state identification method is applied to the online monitoring work.Finally,based on the measured data and joint simulation,it is proved that the method can identify the abnormal small-amplitude hunting and provide warn before the turnout passing.The research contents and conclusions of this thesis are as follows:1.According to the abnormal hunting case of a high-speed train on a certain line in China,the vehicle-turnout dynamic simulation model is established,and the influence of turnout excitation on abnormal small-amplitude hunting of the vehicle is studied.The research results show that the increase of excitation when the vehicle directly passes through the movable track turnout will reduce the critical speed of the limit cycle of the vehicle system,and it makes the unstable limit cycle evolves faster.As the initial lateral displacement of the wheelset hunting increases,the evolution of the unstable limit cycle is further accelerated.2.The vehicle system which obey to subcritical bifurcation cannot provide hunting sort label directly when it provides simulation data.According to the limit cycle theory,nine indexes of time domain class and frequency domain class are extracted from simulation signal samples of the lateral displacement of the wheelset.After that,the spectral clustering analysis is used to divide the unknown lateral displacement signal samples of the wheelset of the simulation model into three hunting states,which are normal hunting class and abnormal small-amplitude hunting class and hunting instability class.Finally,the accuracy of classification results is verified based on the limit cycle theory,and its reliability is compared based on the existing standards.After expanding the sample,its promotional value is also proved.3.Since the current online monitoring method cannot accurately measure the wheelset lateral displacement,in this paper,a one-dimensional convolutional neural network is built.The input of the 1D CNN model is the vibration acceleration signal of the chosen measure point on the axle box,and the output is the classification result of the hunting state of the wheelset.The classifier can find the implicit relationship between the indirect signal of the axle box component and the hunting state by extracting the characteristics from the sensor signal samples.The test accuracy of neural network is 96 % after 50 iterations of training data.This paper uses the network to design the wheelset hunting state identification method,then the function and pseudo-codes of each part above are also introduced.4.The identification method of wheelset hunting state is verified by the actual data of lateral vibration acceleration from the axle box on a certain high-speed train in China,which proves that the identification method can provide warning information of hunting,and have no missing report about hunting instability.After that,the SIMAT co-simulation model is established.It is verified that after adopting the identification method in this paper and the braking control of the abnormal small-amplitude hunting,indexes of lateral stability are improved,which includes the derailment coefficient,the wheel unloading rate,the overturning coefficient,the wheelset lateral force and the wheel-rail vertical force.