Research On Fault Diagnosis Based On RBF Neural Network For Condensation And Feed-Water System

The main function of condensation and feed-water system of secondary loop in nuclear power plant is to cool the exhausted steam that come from a steam turbine group and to send feed water to a steam generator incessantly. So it is important for the security of the whole plant that condensation and feed-water system run normally. Because there are many components which make up the system, the possibility breaking down is relatively big; In order to reduce the fault rate, one of the keys is to know accurately the state of the system in time. In the paper, RBF artificial neural network was used to construct a fault diagnosis system which could realize the function of off-line and online diagnosis, so the state of the system could be grasped in time.Before establishing the fault diagnosis system, the dynamic characteristic of the condensation and feed-water system should be studied well. So, first, mathematics models of the system were constructed with module modeling method, and were simulated on computer. And then based on the models and combined the field experts' knowledge 10 common systematic faults of condensation and feed-water system were simulated; the simulation result showed these fault models were correct, which gave a solid foundation for taking fault signs.At present, the main mode of fault diagnosis applying artificial intelligent methods has two methods: expert system and neural network (NN). In the paper, a fault diagnose model was set up with RBF neural network; compared with BP neural network, RBF network have some advantages over BP network which not only have the nature of best local approximation, but also have the nature of universal approximation which is obtained by the summation of local approximation, which establish the foundation for the application of RBF network. Furthermore, because of the use of dynamic designing method of RBF network, not only the scale of it was smaller, but also the generalization ability of it was higher.With Visual C++, The fault diagnosis system of condensation and feed-water system in a nuclear power installation was constructed based on RBF neural network. The fault diagnosis system of condense and feed water adopt mutualinterface between man and computer, interface is friendly and simple.