Simulation-Based Expert System for Nuclear Reactor Diagnostics: Final Report, September 15, 1985-April 30, 1988.

摘要

An expert system methodology combining model- and rule-based algorithms has been developed for diagnosis of off-normal events in nuclear power plants. A rule-based reasoning system is used in a fuzzy logic structure to analyze plant data heuristically to form a set of hypotheses about a particular transient. Hypothesis testing and fault magnitude estimation are performed with simulation programs to model plant behavior. A new parameter estimation algorithm, called the simulation filter, has been developed for accurate simulation of plant dynamics with low-order physical models of plant components. A critical safety function approach is used to delineate the complexity of a power plant. The simulation-based expert system methodology has been incorporated as the Rx code. Validity of the Rx methodology has been tested through simulation of the Three Mile Island accident. Methods of systematically generating diagnostic rules through pattern recognition and fault tree encoding have been developed and demonstrated on a preliminary basis.