Detection of malfunctions in the secondary system of a nuclear power plant by neural networks

In nuclear power systems the early identification of transients caused by malfunctions is of great importance for preventing the development of serious accidents and performing adequate operation and maintenance practice. For this reason, various plant system parameters are monitored to provide information about the systems state. This information can be used to detect and classify the transients that occur inadvertently in a nuclear system. In this paper, a neural network methodology is developed which exploits the information provided by the online monitoring of different variables for classifying transients that can occur in the secondary system of a boiling water reactor (BWR). The initiating events of the transients in the scope of this work consist of leakages, of different sizes, performed in different locations of the secondary system. Each initiating event considered defines a class of transients which can be identified by analyzing the evolution in time of a group of variables monitored. In this work a three layered feed-forward neural network has been trained to assign an integer value, corresponding to the class of transient number, as output when the neural network is fed with the evolution of different variables of the system monitored online. This methodology has been applied to the identification of transient classes using data provided by the simulator HAMBO. This simulator has been developed to reproduce the behaviour of Forsmark nuclear power plant. Five classes of transients were considered, corresponding to leakages in different locations of the secondary system, and the evolution of ten variables, corresponding to temperatures and level positions of different control valves has been monitored. Using this information, the network was trained, and the trained system has been successful in classifying new transients belonging to any of the classes considered and also in identifying as ''un-known'' transients belonging to other classes not considered in the training phase. (author)