Investigation of neural network paradigms for the development of automatic noise diagnostic/reactor surveillance systems

The use of artificial intelligence (AI) techniques as an aid in the maintenance and operation of nuclear power plant systems has been recognized for the past several years, and several applications using expert systems technology currently exist. The authors investigated the backpropagation paradigm for the recognition of neutron noise power spectral density (PSD) signatures as a possible alternative to current methods based on statistical techniques. The goal is to advance the state of the art in the application of noise analysis techniques to monitor nuclear reactor internals. Continuous surveillance of reactor systems for structural degradation can be quite cost-effective because (1) the loss of mechanical integrity of the reactor internal components can be detected at an early stage before severe damage occurs, (2) unnecessary periodic maintenance can be avoided, (3) plant downtime can be reduced to a minimum, (4) a high level of plant safety can be maintained, and (5) it can be used to help justify the extension of a plant's operating license. The initial objectives were to use neutron noise PSD data from a pressurized water reactor, acquired over a period of {approximately}2 years by the Oak Ridge National Laboratory (ORNL) Power Spectral Density RECognition (PSDREC) system tomore » develop networks that can (1) differentiate between normal neutron spectral data and anomalous spectral data (e.g., malfunctioning instrumentation); and (2) detect significant shifts in the positions of spectral resonances while reducing the effect of small, random shifts (in neutron noise analysis, shifts in the resonance(s) present in a neutron PSD spectrum are the primary means for diagnosing degradation of reactor internals). 11 refs, 8 figs.« less