Research On Switch Machine Fault Diagnosis Method Based On Neural Network

In recent years,domestic rail transit has developed rapidly.In the past five years,the investment in rail transit has been over 800 billion yuan a year.As of 2019,the operating mileage of domestic railways and other railways has reached 139,000 kilometers.5,737 kilometers,and it is expected that more than 28,000 kilometers of rail transit lines will be planned in 70 cities in the future.As a "steering wheel" for rail transit,turnout switch machines have a growing demand for rail equipment such as switch machines.In the face of such a huge railway system,in order to ensure the safety and efficiency of rail transit transportation,it is even more necessary to be relatively complete,Automated and intelligent monitoring and diagnosis system.In the past,maintenance,repair,and supervision of turnout switches rely on the analysis of observation data by engineering and technical personnel to determine the working status of the equipment.However,the judgment and reliance on manual knowledge and experience not only has a long diagnosis and evaluation cycle,but also Very easy to miss or misjudge.Based on the analysis of the domestic and foreign related switch machine fault diagnosis methods and other signal recognition studies,this paper combines the power signal feature extraction and neural network technologies in the case of non-intelligent means of current switch machine fault diagnosis methods.The research on fault diagnosis methods was carried out.Firstly,by analyzing the working process of the switch machine,a neural network fault diagnosis method based on the shape characteristics of the turnout signal is proposed.By studying the internal relationship between the working state,the working process and the power curve,the physical process and Power signal shape feature mapping theory.Through feature extraction,the power curve shape feature set is established and the feature vector is calculated.The feature vector is modeled by a neural network,and finally the signal to be inspected is identified and diagnosed.The data samples collected in a certain experimental platform were trained and tested,and then real-world tests were performed on the sample data in the field.The accuracy of the test results was above 98% and 93%,both of which achieved the effect of engineering applications.This method is effective and practical.Secondly,in view of the instability of the "manual" feature extraction diagnosis method,using the characteristics of the convolutional neural network,from the perspective of "automatic" feature extraction and diagnosis,the fault diagnosis of the convolutional neural network based on the feature map of turnouts is studied method.Through multi-scale analysis,one-dimensional power signals are transformed into two-dimensional feature maps.Through the information feature modeling and classification recognition capabilities of multi-channel convolutional neural networks,fault diagnosis models of convolutional neural networks for feature maps at different time scales are constructed.Through experiments and proofs,the correctness rate of the fault diagnosis test set has been increased to 99.50% and 97.50%,to a certain extent,the problems of instability of the "manual" feature diagnosis method have been overcome,and the feasibility of the method has been verified.Improved diagnostic performance over “manual” feature fault diagnosis.