A wavelet approach for development and application of a stochastic parameter simulation system

In this research a Stochastic Parameter Simulation System (SPSS) computer program employing wavelet techniques was developed. The SPSS was designed to fulfill two key functional requirements: (1) To be able to analyze any steady state plant signal, decompose it into its deterministic and stochastic components, and then reconstruct a new, simulated signal that possesses exactly the same statistical noise characteristics as the actual signal; and (2) To be able to filter out the principal serially-correlated, deterministic components from the analyzed signal so that the remaining stochastic signal can be analyzed with signal validation tools that are designed for signals drawn from independent random distributions.; The results obtained using SPSS were compared to those obtained using the Argonne National Laboratory Reactor Parameter Simulation System (RPSS) which uses a Fourier transform methodology to achieve the same objectives. RPSS and SPSS results were compared for three sets of stationary signals, representing sensor readings independently recorded at three nuclear power plants. For all of the recorded signals, the wavelet technique provided a better approximation of the original signal than the Fourier procedure. For each signal, many wavelet-based decompositions were found by the SPSS methodology, all of which produced white and normally distributed signal residuals. In most cases, the Fourier-based analysis failed to completely eliminate the original signal serial-correlation from the residuals. The reconstructed signals produced by SPSS are also statistically closer to the original signal than the RPSS reconstructed signal.; Another phase of the research demonstrated that SPSS could be used to enhance the reliability of the Multivariate Sensor Estimation Technique (MSET). MSET uses the Sequential Probability Ratio Test (SPRT) for its fault detection algorithm. By eliminating the MSET residual serial-correlation in the MSET training phase, the SPRT user-defined false alarm rates can be met, even for signals which contain serially-correlated components.