A BAYESIAN DIAGNOSTIC SYSTEM

THE BAYESIAN DIAGNOSTIC SYSTEM (BDS) IS PROPOSED AS THE CENTRAL INFERENCE ENGINE OF AN ARTIFICIAL INTELLIGENCE SYSTEM (EXPERT SYSTEM) FOR REACTOR PLANT CONTROL AND MAINTENANCE. THE EXPERT SYSTEM, COPILOT, IS ALREADY UNDER DEVELOPMENT AT PLG UNDER LIMITED IN-HOUSE FUNDING. THE EXISTING WORK INVOLVES DEVELOPMENT OF THE "DOMAIN KNOWLEDGE BASE" OF POWER PLANT SYSTEMS, PROCEDURES, AND ANALYSIS RESULTS USING THE OBJECT ORIENTED, FRAME BASED AI SHEN, LOOPS. THE BDS REPRESENTS A UNIQUE APPROACH TO THE DIAGNOSIS MODULE AND DIAGNOSTIC KNOWLEDGE BASE. THIS APPROACH IS BASED ON OUR EXTENSION EXPERIENCE IN DEALING WITH RISK AND UNCERTAINTY AND OUR UNDERSTANDING OF THE DEEP MEANING OF PROBABILITY AND ESPECIALLY OF BAYES' THEOREM. THIS THEOREM GOVERNS AND DESCRIBES THE THINKING PROCESS THAT AN IDEAL HUMAN EXPERT WOULD GO THROUGH AS HE INTERPRETSTHE INFORMATION PATTERNS COMING THOUGH HIS SENSORS. WE THEREFORE BELIEVE THAT THIS THEROEM PROVIDES THE BEST WAY TO STRUCTURE THE HUMAN EXPERT'S KNOWLEDGE INTO AN ORGANIZED KNOWLEDGE BASE SO THAT IT CAN BE "TAUGHT" TO A MACHINE. FORTHE SAME SEASON, THIS STRUCTURE PROVIDES THE BEST FRAMEWORK WITHIN WHICH THE MACHINE CAN BE TAUGHT TO LEARN FROM ITS EXPERIENCE. THIS FRAMEWORK ALSO PROVIDES THE BEST OPPORTUNITIES FOR DEVELOPING FAST, REAL-TIME ALGORITHMS. WHAT WE ARE PROPOSING, THEREFORE, IS TO DEVELOP AN EXPERT SYSTEM FOR REAL-TIME INTERPRETATION AND RECOGNITION, "DIAGNOSIS" IF YOU WILL, OF PATTERSN OF IMFORMATION AVAILABLE AT THE PLANT. GIVEN THE INFORMATION, HOWEVER INCOMPLETE, THE SYSTEM WILL RETURN A SET OF POSSIBLE INTERPRETATION, EACH WITH A PROBABILITY. STATED IN OTHER WORDS, IT RETURNS A PROBABILITY DISTRIBUTION OVER THE SPACE OF POSSIBLE INTERPRETATION. THE "WIDTH" OF THIS DISTRIBUTION EXPRESSES OUR UNCERTAINTY ABOUT THE STATE OF THE PLANT BASED ON THE INFORMATION AT HAND.