High Speed Railway Switch Fault Diagnosis Based On Kernel Methods

Switch, as one of the key equipment for railway, its technical level intensively reflects the development level of railway track of a nation. Railway switch in China has entered the stage of the wide application of high speed switch, and the technical maturity and security have been greatly improved. But the current switch monitoring method depends on manual browsing switch operating current data of Centralized Signal Monitoring system, which has the disadvantage of low intelligence and high missing alarm rate. It causes that the monitoring and fault diagnosis is one of the most important and heavy work in the railway field.Fault diagnosis for high speed railway switch mainly involves two problems. One is an efficient fault diagnosis algorithm, which makes the fault diagnosis model give a diagnosis result with high accuracy timely and quickly. The difficulty lies in the fault diagnosis model which needs high adaptability and fault-tolerance, so that it can adapt to the different working conditions of different switches and make intelligent judgments. The other one is the accurate description of the features corresponding to switch different operating modes, it makes the input of the fault diagnosis model is the most concise and efficient features combination which can represent switch working states.To solve the above two problems, this paper puts forward two kinds of machine learning algorithms based on kernel function method:SVM and kernel Fisher, and respectively uses them for the switch fault diagnosis of the mechanical structure and control circuit through the switch action current curve and the location voltage. This paper puts forward the fusion feature representation based on the intelligent partitioning, Fisher feature selection method and principal component analysis (PCA) to represent, select and extract the characteristics of the switch current curve. In this paper, the main work is as follows:(1)Specify the actual demand of railway switch failure diagnosis combining with the working principle and the monitoring principle of high speed railway switch. Put forward the classic fault modes to study in this paper, and analyze their fault reasons and fault characteristics. Put forward intelligent partitioning of the fusion feature representation, according to the fault characteristics of different failure modes.(2)Use Fisher feature selection method to filter all the characteristics to leave the ones which can help the fault diagnosis; Use PCA and KPCA method to extract the characteristics from the selected to make least and most effective features to represent each switch movement curve.(3)Use SVM method as fault diagnosis classifier to diagnose the characteristics of the switch current curve data to complete the implementation and verification of fault diagnosis. Validation shows that when the algorithm uses 3d characteristics as input, the accuracy can reach 96.667%, which training time and accuracy can meet the demand of the scene. As the same time, the meaning of heuristic algorithm for the optimization for kernel parameter and penalty parameter is studied, and an optimization algorithm combined with genetic algorithm and particle swarm optimization is proposed.(4)Use kernel Fisher discriminant method as fault diagnosis classifier to diagnose according to the switch position voltage data, and complete the implementation and verification of the fault diagnosis. Validation shows that the algorithm accuracy can reach 87.5%, significantly higher than the researches by other methods in the past.In this article, the kernel method is studied, optimized and applied through the method of KPCA, SVM and kernel Fisher. Verification results show that the kernel method has advancement in the pattern recognition problems and feasibility in the high speed railway switch fault diagnosis.