Research On Fault Diagnosis Based On Switch Operating Current

The switch is a guarantee for the safety and efficient operation of the railway signal system,which has a great influence on the safety and efficiency of railway transportation.At present,China’s railway system mainly adopts the way of periodic maintenance and window maintenance for turnouts to ensure the safety of turnout system.For this fault analysis and fault location methods,There are many problems,such as low efficiency,heavy workload,no guarantee of timeliness and accuracy and so on.First,because the change trend of the turnout’s operating current can reflect the current operating state of the turnout,this thesis proposes a switch fault feature extraction method based on CEEMD,in which the rough set attribute reduction,information entropy,and adaptive method to determine the parameters and other knowledge theories are applied.Then the multi-dimensional analysis method is used to display the fault feature extraction results in two dimensions,and the effectiveness of fault feature extraction is analyzed.Finally,a convolutional neural network model is built to diagnose faults.The experimental results prove the feasibility of this fault diagnosis method.Second,Point at the different between the switch action current data and the traditional depth residual network about the input data form,we adjust the internal structure of residuals learning module,and design a turnout fault method based on deep residual network structure.In order to improve the performance of the system,the network parameters are optimized,the fault diagnosis results are analyzed with different activation functions,and the most suitable activation function is selected.Through the experiment simulation,the feasibility of the turnout fault diagnosis is proved.The diagnosis efficiency of the depth residual network is better compared with the shallow residual net and the fully connected neural network.