Study On Railway Switches Diagnosis System Based On Deep Belief Networks

As the development of China’s railway industry especially the high-speed rail and motor train unit picks up momentum,China has ranked world row in this area.As the high-speed rail is constantly picked up speed,people increasingly lay emphasis on the railway equipment security.Besides,how to find and solve problems in the traditional way gradually fail to meet the current need.Therefore,the hot spot researched by experts is the intellectual method of early fault discovery and prompt solutions.“The early warning system of railway fault diagnosis” program is made by the Academy of Sciences of Hebei Province in cooperation with Shijiazhuang Railway Electric Business.Against the background of the program,this thesis selects railway switch,the most important part in railway equipment,as the research target and aims to form a high-accuracy railway switch fault diagnosis program by starting from the deep learning of new areas.The main research points of this thesis are as follows:(1)On the basis of understanding the railway switch fault mechanism,this thesis sums up and sorts out the switch fault types with the help of references and the Electric Business staff.Railway switch faults usually reflect in switch action electric current curves,but the electric current data of different switches represents various high dimensions,which makes the following researches inconvenient.Under this circumstance,this thesis comes up with the spline interpolation algorithm and aims to obtain concordant electric current data with low dimensions after data-processing based on this algorithm.(2)Based on the deep understanding of neural networks,this thesis advances a new type of Deep Belief Network(DBN)model according to the characteristics of switch electric currents.This thesis also alters the DBN model top classifier by experiments and selects radial basis function(RBF)neural networks as the final choice in order to achieve higher identifying accuracy and a more optimized structure of the neural network model mentioned above.(3)Combined with the mechanism of fault diagnosis and expert system and DBN model,this thesis designs and develops a new type of railway switch fault diagnosis system which could constantly realize the ability of automatic training and learning through switch electric current data.This new system could become an intellectual system that keeps learning and developing when its error recognition and correction mechanism constantly improves the fault diagnosis accuracy.In the end,it concludes the contribution and limitation of the research and look forward to the further studies.