| During the course of nuclear power evolution, accidents have occurred. However, in the western world, none of them had a severe impact on the public because of the design features of nuclear plants. In nuclear reactors, barriers constitute physical obstacles to uncontrolled fission product releases. These barriers are an important factor in safety analysis. During an accident, reactor safety systems become actuated to prevent the barriers from been breached. In addition, operators are required to take specified actions, meticulously depicted in emergency response procedures.; In an accident, on-the-spot knowledge regarding the condition of the core is necessary. In order to make the right decisions toward mitigating the accident severity and its consequences, we need to know the status of the core [1, 3]. However, power plant instrumentation that can provide a direct indication of the status of the core during the time when core damage is a potential outcome, does not exist. Moreover, the information from instruments may have large uncertainty of various types. Thus, a very strong potential for misinterpreting incoming information exists. This research endeavor addresses the problem of evaluating the core damage severity of a Pressurized Water Reactor during a transient or an accident.; An expert system has been constructed, that incorporates knowledge and reasoning of human experts. The expert system's inference engine receives incoming plant data that originate in the plethora of core-related instruments. Its knowledge base relies on several massive, multivariate fuzzy logic rule-sets, coupled with several artificial neural networks. These mathematical models have encoded information that defines possible core states, based on correlations of parameter values. The inference process classifies the core as intact, or as experiencing clad damage and/or core melting. If the system detects a form of core damage, a quantification procedure will provide a numerical assessment of the extent of the damage. The inference procedure is the Generalized Modus Ponens, which has its origin in the field of Approximate Reasoning. In addition, the use of neural networks enhances the accuracy of the quantification procedure.; The model was tested for accuracy of assessment under severe accident conditions that compromised the reliability of instrumentation. The accuracy of the results established that the engagement of fuzzy logic in core state diagnosis constitutes a very promising method. Valid assessments were achieved in the vast majority of the test cases, in spite of troubling data deficiencies, which included inaccurate, distorted, or missing data. |
