Criteria of Diversity Evaluation for Intelligent Diagnosis of Nuclear Power Plants

One of important problems of a current operation support system for a nuclear power plant is that the credibility of its resultant suggestions is not always high sufficiently. The authors have proposed an efficient remedy called "Diversity Criteria" for this issue in the previous works. It employes a variety of information resources and reasoning mechanisms for the system to enhance its entire credibility. Within this framework, a complementary combination of the resources and mechanisms is desired. The work presented here proposes systematic and quantitative measures determining the appropriate combinations. First, concrete and systematic guidelines are proposed for the detailed criteria of "Information Diversity" and "Methodology Diversity". Next, two concepts of "Orthogonality of Identified Result" and "Orthogonality of Utilized Symptom" are presented together with their quantitative measures. These guidelines and measures have been applied to an example of failure diagnosis of a nuclear power plant, and their efficiency has been clearly confirmed.