A NEURAL NETWORK APPROACH TO NUCLEAR PLANT MONITORING

THE MAJOR OBJECTIVE OF NUCLEAR PLANT MONITORING SYSTEMS IS TO ENABLE THE OPERATOR TO MAKE CORRECT DECISIONS THAT ENSURESAFE PLANT OPERATIONS WHILE PROVIDING A RELIABLE ENERGY SOURCE. THE USUAL DIFFICULTIES ENCOUNTERED BY NUCLEAR PLANT OPERATORS ARE INCORRECT ASSUMPTIONS OF PLANT STATUS, INSUFFICIENT INFORMATION FOR FAULT DIAGNOSIS, AND INNACCURATE PREDICTION OF SCENARIOS THAT WOULD RESULT FROM VARIOUS COURSES OF ACTION. TO ALLEVIATE THESE DIFFICULTIES,THE HYBRID INTEGRATION OF ARTIFICIAL NEURAL NETWORKS (ANNS) AND KNOWLEDGE-BASED EXPERT SYSTEMS (KBS) IS BEING OPTIMIZED TO PROVIDE INTELLIGENT RECOMMENDATIONS FOR SAFE NUCLEAR PLANT OPERATIONS. THE INTEGRATION OF ANNS AND KBS IS AN IDEAL STEP IN THE DEVELOPMENT OF INTELLIGENT SYSTEMS. IN GENERAL, ANNS PROVIDE "SOFT" CONSTRAINTS, WHILE KBS ALLOW "HARD" CONSTRAINTS. THE TWO METHODS COMPLEMENT EACH OTHER IN A MANNER SIMILAR TO THE HEMISPHERES THAT CONTAIN THE HUMAN MIND. ANNS, CORRESPONDING TO THE RIGHT HEMISPHERE OF THE BRAIN, PROVIDE ADAPTABLE PATTERN RECOGNITION AND CLASSIFICATION CAPABILITIES THAT CONTINUE TO PERFORM EVEN IF SOME PROCESSING UNITS FAIL; KBS, WHICH RELATE TO THE LEFTHEMISPHERE, INVOLVE A STRUCTURED, RULE-BASED REPRESENTATION OF EXPERT KNOWLEDGE AND STEP-BY-STEP DECISION STRATEGIES. THIS PROJECT TAKES ADVANTAGE OF THE COMPLEMENTARY STRENGTHS AND WEAKNESSES OF EACH METHOD TO CREATE A HYBRID SENSOR MONITORING SYSTEM THAT CAN OUTPERFORM EITHER METHOD ALONE. IN THIS PHASE I STUDY, NEURAL NETWORK AND HYBRID NEURAL-NETWORK/KNOWLEDGE-BASED STRATEGIES ARE BEING DEVELOPED FOR INTELLIGENT SENSOR MONITORING CAPABILITIES WITHIN NUCLEAR POWERPLANT APPLICATIONS. THE ANN FOCUS INVOLVES DETERMINING PLANT OPERATING STATES, INCLUDING NORMAL AND OFF-NORMAL CONDITIONS, AND INVESTIGATING THE TEMPORAL SIGNATURES OF THE VARIOUS SENSOR MEASUREMENTS. THEHYBRID STRATEGIES FOCUS ON INTEGRATING THE ANN RESULTS WITH EXPERT ASSESSMENT AND PREDICTION KNOWLEDGE IN ORDER TO PROVIDE THE PLANT OPERATOR WITH INTELLIGENT RECOMMENDATIONS.