Development of methods for analyzing time-varying characteristics of power reactor noise

Abstract

A method of signal processing for analyzing small sample data was developed, aiming to establish a tool for diagnosis of nuclear power plants under non-steady state operations. Least squares fitting of parametric models is the main part of the present methodology. The method of Householder transformation was employed in determining the model parameters to circumvent the problem of numerical difficulty often encountered in the least squares modeling. The resultant models were used as estimators of physical parameters or noise signatures relevant to integrity of the plant or its components. The auto power spectral density(APSD) is estimated by transforming the fitted autoregressive(AR) model. Prediction values of a signal is obtained by using the single-input single-output(SISO) model as a predictor. Estimates of delay times between two variables are derived from so-called simplified SISO model. Time-dependence of statistical characteristics of the measured signals are clearly represented by the variations of these quantities from case to case. Advantages of the present approach over the conventional methods were demonstrated through analysis of several examples for simulation data and reactor noise.