Signal processing and neural network applications in pressurized water reactors

This report discusses feasibility of applying neural networks for signal validation, plant-wide monitoring, and diagnostic applications. Data used in this study were acquired from Borssele Nuclear Power Plant using on-line monitoring system implemented at ECN Petten, Three major applications have been completed during this study: signal validation applications, plant-wide monitoring applications, sensor failure applications. The signal validation application involves the estimation of generated electric power using 2 different sets of data as input for the network models. Three networks were created for estimating the generated electric power signal. First, stretch-out data have been used as a training data set. A second network was created with the data obtained at normal operation conditions. A third network was created using the secondary system signals from a stretch-out data set. Details of these applications will be given in the report. Plant-wide monitoring is a type of signal validation which is performed by tracking multi-sensor outputs simultaneously. Instead of estimating only one sensor output at each time, multi-sensor outputs can be estimated by using the plant-wide monitoring system methodology. Finally, a sensor validation application has been completed for estimating the sensor readings during a failure of a sensor. (author). 8 refs.; 27 figs