This investigation will assess and demonstrate the feasibility of using artificial intelligence (AI) techniques to develop an expert system serving as an Accident Diagnosis and Prognosis (AD&P) Aide. The need for such a system is based on industry and regulatory commitment to improved nuclear plant performance and safety. AI technology is just now making its way out of the laboratory and into the real world; as such, the time is ripe for investigating practical, value-added AI applications. The contractor will investigate and demonstrate the feasibility of using (1) Critical Safety Function Status Tree technology to diagnose nuclear plant accident situations, (2) Probabilistic Risk Assessment Event Sequence Analysis technology to predict the various directions an accident-in-progress might take, and (3) Functional Restoration Guideline technology to generate corrective action recommendations for stabilizing the reactor plant. During the course of the project, a working scale model, referred to as the AD&P Aide Demonstration Prototype, will be developed. It will provide an experimentation environment for providing rapid feedback on the concepts under investigation, including technical feasibility, practicality, and usefulness. The Final Report will document (1} the results of the feasibility studies, (2) a description and evaluation of the demonstration prototype, and (3) conclusions and recommendations.Anticipated Results/Potential Commercial Applications as described by the awardee:Successful completion of the AD&P Aide Demonstration Prototype is anticipated, as is development of a full prototype in Phase II and commercialization in Phase III. The ability of an expert system to quickly and accurately diagnose an accident situation, provide insights into how the accident might progress, and provide concrete, reliable recommendations on how to stabilize the plant will be of value to the nuclear industry and to the Nation as a whole. Phase II may address more advanced implementation topics such as confidence levels, accurate processing in light of missing or ambiguous sensor readings, and identification of false alarms. Delivery systems and portability will also be addressed at that time.
