Research On The Detection Method For The Global Wheel Out-roundness Of Urban Rail Train

With the continuous increase of the speed of urban rail vehicles,the dynamic interaction between wheel and rail is intensified,and the phenomenon of wheel global non-roundness occurs frequently,which has a great impact on the safe operation of urban rail trains.Therefore,it is of great significance to study the detection method of wheel global non-roundness.On the basis of summarizing the research results at home and abroad,this paper presents a method for detecting the global non-roundness of urban rail train wheels based on vibration signals,The main work is as follows:Firstly,based on SIMPACK,the vehicle-track coupling dynamic model is established,and the wheel global roundness is used as the input signal to simulate,and the wheel-rail vibration signal is obtained.Secondly,the global non-roundness fault detection system of urban rail train wheels is designed.including requirement analysis,system overall design,key hardware selection and field system installation,which provides field data sources for subsequent algorithms.Thirdly,vibration signal denoising based on improved wavelet threshold is studied.An improved wavelet threshold de-noising method is proposed to solve the problem of noise in the original vibration signal.Finally,experiments show that the method has better de-noising effect.Fourthly,the feature extraction of vibration signal based on IMPE is studied.Multi-scale permutation entropy is introduced to feature extraction of wheel global non-roundness faults.However,the stability is not good in the process of calculating the entropy value,so an improved multi-scale permutation entropy is proposed.Finally,the feasibility and validity of the method are verified by simulation signals and measured data.Finally,the fault identification method based on extreme learning machine is studied.The feature vectors of normal and faulty wheels are input into the extreme learning machine for training,and then the classification results are tested.Compared with the BP neural network,the accuracy and superiority of the extreme learning machine in identifying the global non-roundness faults of wheels are proved.