Research On Fault Diagnosis Of High-speed Railway Turnout Based On Multi-channel LSTM-VAE

The rapid development of high-speed railways in the world has made it easier for people to travel.With the continuous speeding up of high-speed railways,its safety issues have become the focus of attention.As a conversion component that determines the running direction of high-speed trains,the safety and reliability of speed-increasing turnouts is the top priority to ensure the normal operation of high-speed trains.Therefore,how to diagnose the failure mode of the speed-increasing turnout during the service period in advance and eliminate the hidden danger of accident in time has important practical significance.This research takes the S700 K speed-increasing turnout in a certain section of Urumqi Railway Bureau as the research object,and proposes a turnout fault diagnosis model based on the multi-channel neural network model,in order to provide technical support for the troubleshooting and management of the speed-increasing turnout.First,through the comprehensive analysis of the composition structure of the speed-increasing switch,the principle of the switch machine,and the power curve of the switch;the operation current image and power image waveform of the speed-increasing switch are connected with the health status of the switch,and the corresponding fault type of the switch is matched to construct A sample set of 400 faults composed of 7 types of speed-increasing switch faults.Then,on the premise of data preprocessing of the fault sample set,the data feature value is extracted through wavelet transform,and then the high-dimensional data is reduced by the CCA method to obtain feature factors containing 10 types of features as the subsequent neural network model The input vector of multivariate features.Secondly,relying on the advantages of long and short-term memory neural network(LSTM)in time series prediction,a multi-channel LSTM-VAE turnout fault diagnosis model is established,and multi-variable feature parameters containing current characteristics are selected as the data input for multi-channel LSTM model training.Carry out multivariate prediction of switch current and power data;perform cosine similarity analysis on the predicted switch current and power data and healthy data,and bring the fault state data into the variational automatic encoder(VAE)for fault classification.Finally,this article takes 400 sets of operating current and power curves of a switch machine in a certain industrial and electrical section into the model for example verification.The research results show that the switch operating current values obtained based on the multi-channel LSTM-VAE model are all within a reasonable error range,and the fault classification effect reaches more than 95%.