Research On Intelligent Fault Diagnosis Method In Point Based On Neural Network

Along with the speed of train rising year by year, the point as railway key equipment is required stricter request. This paper is a project based on point monitoring system which proved large amounts state data. Based on these state data from point monitoring system, the neural network technology can be used on the point intelligent fault diagnosis. Due to the research of this field is scarce, few experience can be used to reference. This paper is only a preliminary attempt of point intelligent fault diagnosis. Providing a feasible method for intelligent fault diagnosis in point is the purpose of this paper.This paper firstly introduces the definition and principle of neural network, then, uses two typical neural networks which are the BP neural network and RBF neural network to expound the network structure and learn algorithms and its application. In order to build the structure of the neural network system, through the introduction of the point shifting system structure, this paper analysis the principle of various typical faults and classify the various desultorily faults to be unified. Then point monitoring system is introduced which provide various point monitoring data. Finally, this paper structures the BP neural network model and RBF neural network mode respectively using the MATLAB. After a lot of experiments and network performance comparison, the anticipated diagnosis requirements are basic achieved and the theory research on intelligent fault diagnosis in point is completed.The research is carried out in these several aspects:According to the fault mechanism of point and neural network structure characteristics, all point faults are divided into three categories. Every kind structures a junior neural network. The system framework is built by these junior neural networks.Research on Various BP algorithms.Every junior neural network is trained and tested by four common BP algorithms respectively. The advantage and disadvantage have been found after testing.Research on the method to diagnose fault by the BP algorithm. For each junior network, the paper calculates the least range of nerve cell's number which consist of hidden layer. Later, the paper analyses the influence of network when some different number of nerve cell consisting of hidden layer, then creates the best neural network by the L-M algorithm.Research on the method to diagnose fault by the RBF algorithm. For each junior network, the best number of nerve cell consisting of hidden layer and spread by some tests. Soon afterwards the best RBF neural network is built.The theoretical research and extensive experiments proved it is practical and feasible that the neural network technology applies to the fault intelligent diagnosis in point This method can rapidly and effectively diagnose the reason of fault and provide technical support for maintenance personnel.