Systematic generation of rules for nuclear power plant diagnostics

The knowledge base of an expert system is generally represented by a set of heuristic rules derived from the expert's own experience and judgmental knowledge. These heuristic or production rules are cast as if (condition), then (consequence) statements, and represent, for nuclear power plant diagnostic systems, information connecting symptoms to failures. In this paper, the authors apply an entropy minimax pattern recognition algorithm to automate the process of extracting and encoding knowledge into a set of rules. Knowledge is extracted by recognizing patterns in plant parameters or symptoms associated with failures or transient events, and is encoded by casting the discovered patterns as production rules. The paper discusses how the proposed method can systematically generate rules that characterize failure of pressurizer components based on transient events analyzed with a pressurizer components based on transient events analyzed with a pressurizer water reactor simulator program.