Stochastic models, numerical algorithms and optimal computerised graphics and decision making processes for nuclear plant surveillance and anomaly detection

Abstract

The identification, on-line and in near to real-time, of the resonant frequencies, modes and amplitudes of selected key reactor components and the visual monitoring of these phenomena by nuclear power plant operating staff will serve to further the safety and control philosophy of nuclear systems and lead to design optimisation. The rapid recognition of the onset of unusual or accidental conditions in reactor sub-channels by means of the optimal filtering of noise-like sources will lead to the development of new safety systems. In this paper a computer based nuclear power plant surveillance system utilising pattern recognition techniques is described and the results of preliminary tests and experiments carried out with it are discussed. Computer based algorithms utilising correlation and spectral analysis and linear and non-linear auto-regression techniques are assessed and their optimal utilisation described.