ABSTRACT
DIAPO is an AI diagnostic system for reactor coolant pump sets in EDF nuclear power plants. About 200 pump and plant state parameters can be used to describe the phenomenon to be diagnosed. Basically, diagnosis relies on the cooperation of a set of fault models used by an abduction procedure in order to explain observations which are elaborated from monitoring data or provided by plant staff. Diagnosis of a dysfunction thus consists in the conjunction of identified initial causes, recognized faults, recognized abnormal situations and locations of faulty components. In addition, prediction of future states of the pump can be provided. Heuristic knowledge is also used in order to focus diagnosis. A full scale prototype of DIAPO has been jointly developed by EDF and Jeumont-Industrie and validated. An industrial version is currently under development.
