Results in the Application of Pattern Recognition Methods to Nuclear Reactor Core Component Surveillance

Abstract

Pattern recognition methods were applied to analyze and interpret neutron noise data from the High Flux Isotope Reactor (HFIR) at ORNL. The results show that it is feasible to detect some core component failures by means of machine-discemible differences in the time-dependent noise power spectra. These neutron spectra (signatures) were analyzed by using a cluster-seeking algorithm to derive a set of templates for automatic computer evaluation of the reactor's mechanical integrity and soundness.