Development of a hybrid intelligent system for on-line monitoring of nuclear power plant operations

Nuclear power plant operation is a complex task which requests high reliability, highly skilled operators and advanced analytical tools. Because of system complexity, operators and system managers have difficulty to analyze the great quantity of collected data to formalize correct and timely decisions, which is very important, either in normal operation or at emergency situations. Automated equipment health monitoring becomes necessary for the given complex conditions of the systems, structures, and components, SSC, in the nuclear power plants, NPP. The purpose of equipment health monitoring is to relate the relevant features to a set of variables, which define the current health state of the SSC. The real-time fault diagnosis, and decision-making tasks become cumbersome due to complexity and high load of monitored data. This paper introduces a hybrid intelligent system as a solution to address the difficulty of these tasks within the complexity of equipment health monitoring for SSC in the NPP. The developed methodology combines several artificial intelligence techniques, as Bayesian networks, influence diagrams, neural networks and fuzzy logic. Two modules, expert system module and the neural network module, are incorporated into the complete system. The proposed hybrid architecture has desirable properties inherited from both fields of numeric neural networks and symbolic expert systems. The expert system module of the hybrid system has been completed and validated on the bearing system of horizontal charging pumps in nuclear power plants. The neural network module is in development stage.