Empirical Modeling Approach to Fault Detection and Identification in Nuclear Power Plant

Abstract The increasing emphasis on the operational safety of nuclear power plants necessitates the development of dependable methods for incipient fault detection and identification (FDI). This paper presents some of the recent developments in the FDI methods based on empirical modeling. The empirical modeling approach was taken to alleviate the difficulty in determining physical models for wide operational conditions. The least-squares modeling principle was adopted to enhance the applicability of the method to nonstationary time series data. The parametric models such as univariate and multivariate autoregressive(UAR and MAR) models, single-input single-output(SISO) model and multiple-paths delay(MPD) model are efficiently obtained by using the same modeling algorithm. These models are utilized as estimators of (1) process variables, (2) frequency spectra, and of (3) physical parameters. The estimated quantities provide us with redundant information to be utilized in FOI schemes. Applicability of the present methods is demonstrated through simulation and actual experiments at a nuclear power station at Borrsele, the Netherlands.