Fault Diagnosis Expert System For Automatic Block With Jointless Frequency-Shift System Based On Neural Networks

Railway signal devices play an increasingly important role in railway transportation with the railway electrification process of China moving forward, the train speed on the rising and the traffic density on the increasing. To guarantee the traffic safety of the railway, railway signal devices are necessary to be int the normal operation. However, there are still many problems in the fault diagnosis and maintenance of signal devices to be solved. Therefore develop a set of fault diagnosis expert system is of great significance.This paper focuses on the research of fault diagnosis expert system for ZPW-2000A signal device. Firstly I designed a hybrid algorithm BP-LM-PSO-GA algorithm which meet the requirements of effectiveness and quickness in neural network training by reading a lot literature and books about BP algorithm, particle swarm optimization algorithm, LM algorithm and genetic algorithm. I verified the professionalism of the hybrid algorithm by a group of typical samples and a typical neural network and tested the interoperability of the hybrid algorithm through six test functions which are commonly used to verify the performance of the algorithm. Secondly, the expert system's basic concepts, principles and composition were introduced. The difference and association between neural networks and expert systems were analysised. The structure of neural networks-expert system and modules of the knowledge base of expert systems, human interface, inference engine, dynamic database, knowledge acquisition were designed. The fault feature signals were summarized after the failure mechanism of ZPW-2000A system was studied.According to actual needs, parallel neural network and serial neural network which would be trained respectively designed and named creatively. Taking the scalability of dealing with the problem and the convenience of expansion into account, unified modules of parallel neural network and serial neural network were designed and implemented respectively.In order that new neural network which could handle new issues and would be put into the overall network conveniently can easily be designed.The expert system should not only have a good diagnostic capabilities, but also hierarchy. Two interfaces which are expert interface and operator interface were designed. The main role of the expert interface is neural network training, weights and thresholds storage and database management. The main role of the operator interface is fault diagnosis, history fault datas inquiry and so on.Finally, the fault diagnosis expert system for automatic block with jointless frequency-shift system based on neural networks which had a good human-machine interface based on relational database was given. And then the various functional modules of the system were given presentations. Finally, that the neural network expert system could make an intelligent, real-time, and accurate diagnosis of the railway signal device failures was verified by example. The feasibility of design ideas was proved by the running results.