Analysis of time-varying characteristics of signals from borssele reactor

Abstract

A method of random data analysis, based on least squares autoregressive (AR) modeling, was applied to signals from Borssele reactor. In this method, the model parameters are determined directly from the data without estimating correlation functions, enabling to avoid the modeling errors often encountered when the number of samples is small. From the estimated AR model, auto power spectral density (APSD) and secondary noise signatures are derived easily. The following secondary signatures were used to quantify the time-variation of statistical properties of the signals: • - sample variance of the signal, • - sample variance of the residual, • - resonance frequencies of peaks in the APSD, • - amplitudes of the peaks, • - magnitude of the APSD at the lowest frequency. The purpose of this paper is to demonstrate the usefulness of the method as a practical tool for surveillance and diagnostics of a nuclear power plant. The method was implemented into the PDP-11 system of the physics department o ECN, where signals are available taking the advantage of on-line link with Borssele reactor instrumentation. The method was applied to several sets of data in which non-stationary was recognized. Typical examples of them are: the signals obtained while the reactor was undergoing a power level change, the response of vibration sensors to artificial disturbance imposed under cold shut-down conditions and pressure standing wave phenomena observed during the cool-down procedure. Through the analysis, it was confirmed that the method is highly efficient in detecting changes of the statistical properties of the signals. In addition to the detection, information relevant to the cause and extent of the changes was also extractable. Development of diagnosis systems covering not only stationary but also time-varying operation modes of nuclear reactors can be accomplished by proper use of the present method.