Development of an on-line early fault detection expert system using an advanced man-machine interface

The paper describes the principal features of an expert system currently under development for fault diagnosis on an Italian nuclear power plant. The system development starts with the definition of a methodology to acquire from a plant simulator the basic information that is needed to build a prototype knowledge base that will be subsequently improved by plant experts. The main advantages of this approach are the rapid creation of a first working version of the system and the possibility to submit to the experts a well defined formal structure in which changes or implementations of new items into the knowledge base are feasible without much programming effort. The system is developed in the OPS5 production rule system, in which an extension of the basic OPS5 functions is generated, allowing a backward chaining search strategy. An advanced man-machine interface is also under development based on CRT colour graphic displays in which a mimic background depicts the plant areas under investigation and the foreground is used to highlight all the current fault candidates and to give a graphic presentation of the search trees investigated during the inferential process. (author). 6 refs, 1 fig., 1 tab