Nuclear reactor condition monitoring by adaptive resonance theory

The authors present an evaluation of the performance and a comparison between various paradigms of the ART family of artificial neural networks in nuclear reactor signal analysis for development of a diagnostic monitoring system. To closely represent reactor operational data, reactor pump signals from the Experimental Breeder Reactor (EBR-II) are analyzed. The signals were measured signals collected by the data acquisition system as well as simulated signals. ART2, ART2A, fuzzy ART, and fuzzy ARTMAP were applied. Several simulators were built, and the study indicated that, although all ART paradigms are appropriate for application in reactor signal analysis, each has its own unique characteristics and features which can be utilized whenever needed and applicable.<>