Nuclear power plant diagnostics study at the Midland training simulator

Training simulators provide a real world environment for testing advanced diagnostic and control systems as an aid to nuclear power plant operators. The simulators not only duplicate the hardware din the actual control room, allowing for analysis of man-machine interface, but also represent the dynamic behavior of the reference plant in real-time, in a realistic manner. Training simulators provide the means to representing the reference plant operations in a wide range of operation conditions including off-normal and emergency conditions. Transient events with very low probability of occurrence can then be represented and used to test the capabilities of advanced diagnostic and control systems. For these reasons, full-scope operator training simulators have been used as a test bed for a number of advanced diagnostic concepts. The University of Michigan and Consumers Power Company have been collaborating in a program devoted to the development and study of advanced concepts for automatic diagnostics and control of nuclear power plants. The program has been focused on the use of the full-scope operator training Midland Nuclear Power Plant Unit 2 (MNP-2) Simulator for development, testing, and verification of advanced diagnostics concepts. In their current efforts, the authors have developed two artificial intelligent (AI) diagnostic concepts that have been applied to the MNP-2 Simulator: the systematic generation and updating of a rule-based knowledge system for nuclear power plant diagnostics and a nonlinear parameter estimation algorithm called the simulation filter. The simulation filter algorithm is used with the MNP-2 Simulator to improve the simulation of the Three Mile Island Unit 2 (TMI-2) accident. 11 refs., 4 figs