Research On Fault Diagnosis Of ZDJ9 Switch Machine Based On Deep Learning

With the rapid development of railway in China,a huge railway transportation network has been formed.As one of the three basic equipment of railway signal,switch machine plays an extremely important role in the safe transportation of railway.Because the switch machine works in the outdoor,the use scene is complex,has the problem of high action frequency and high failure rate.At present,there are microcomputer monitoring systems that can monitor the status of the switch machine equipment in real time,but the maintenance of the switch machine is still manual,which has the problems of low efficiency,high cost,and high probability of missed diagnosis and misdiagnosis.A lot of research has been carried out to improve the diagnosis efficiency of the switch machine.However,in the actual production and transportation,there are many types of failures of the switch machine,and the failure data is not easy to obtain.Due to the limitation of data,many diagnostic methods are constrained and ineffective.Aiming at the above problems,this thesis carried out the research on the switch machine fault data enhancement and the intelligent diagnosis of switch machine based on deep learning.The specific research contents are as follows:(1)By optimizing the network structure and objective function of the generative adversarial networks,a ZDJ9 type switch machine fault data enhancement model based on AC＿WGAN is proposed;multiple evaluation indicators are used to optimize the network parameters;By comparing with other generation adversarial networks and traditional generation methods,the result shows that the samples generated by AC＿WGAN generation model are of higher quality and better performance,which solves the problem of insufficient field monitoring fault data of the switch machine.(2)Compare the diagnosis effect of convolutional network and recurrent network from the aspects of accuracy rate,loss value,ROC curve and confusion matrix through experiments,combine a single model,and apply the residual idea to the recurrent network,this thesis proposes a method based on RESNET-Res BIGRU mixed residual model of switch machine fault diagnosis method,the experimental results show that the model has a higher diagnostic accuracy,which can provide a certain guiding significance for the fault diagnosis of switch machine.