Research On Fault Diagnosis Of Turnout Based On RBFNN

Turnout is a common electromechanical signaling device that plays an important role in the development of railways.As a key component of the turnout,the reliability of the turning machine has a non-negligible impact on railway driving efficiency and driving safety.At present,the turnout still relies on experienced signal technicians to monitor the analysis of the motion curve of the turnout machine of the centralized monitoring system,which not only requires a lot of human resources but also has the disadvantage of high alarm leakage rate.Due to the complexity of its own structure and the harsh working environment,the difficulty of manual diagnosis is exacerbated by factors such as the complexity of the turning machine.Therefore,intelligent turnout fault diagnosis methods have become more and more hot topics in research.In this thesis,the power curve of the S700 K turning machine and the power curve of various faults are analyzed,and the data pre-processing of the extracted power is proposed,and a fault diagnosis method based on radial base neural network is proposed.The main research contents of the thesis are as follows:First of all,the body structure,working principle and control circuit of the S700 K rotary machine are expounded,and then the power curve of the rotary machine under normal operation is analyzed,the data is collected and listed the common 8 kinds of rotary machine fault power curves,and the power curves of different fault types are analyzed according to the working principle and action steps of the rutting machine.Secondly,the sample imbalance problem of the normal curve and the fault curve of the turner is handled by using the synthetic minority oversampling technology.According to the steps of the turning machine action,the power curve is divided into multiple regions from the time domain and value domain angles,and the fault characteristic parameters of the torque of the turning machine in each region are calculated separately.The Fisher criterion function is used to select feature parameters,reduce high-dimensional features to low dimensions,and finally extract 8-dimensional feature data collection by principal component analysis.Finally,due to the advantages of radial base neural network such as fast training speed and strong classification ability,the radial base neural network is selected as the classifier of the power curve of the turner to construct a PCA-RBF turnout fault diagnosis model.The PCA-RBF turnout fault diagnosis model is trained by the principal component analysis method using the resulting torque characteristic vector of the turnout as input,and finally the trained model is tested using the test set.After several experiments,the results show that the proposed method has fast training speed,high diagnostic accuracy and certain feasibility in turnout fault diagnosis.