Expert Systems for Fault Diagnosis in Nuclear Reactor Control

An expert system for accident analysis and fault diagnosis for the Loss of Fluid Test (LOFT) reactor, a small scale pressurised water reactor, has been developed for a personal computer. The knowledge of the system is represented using a production rule approach with a backward chaining inference engine. The data base of the system includes simulated time dependant state variables of the LOFT reactor model. Another system is designed to assist the operator in choosing the appropriate cooling mode and to diagnose the fault in the selected cooling system. The response tree, which is used to provide the link between a list of very specific accident sequences and a set of generic emergency procedures which help the operator in monitoring system status and to differentiate between different accident sequences and select the correct procedures, is used to build the system knowledge base. Both systems are written in Turbo Prolog language and can be run on an IBM PC compatible with 640k RAM, 40 Mbyte hard disk and colour graphics.