On-line reactor surveillance algorithm based on multivariate analysis of noise

The development and demonstration of a computer-oriented approach for on-line reactor surveillance are described. A characterization of the reactor state is developed from the analysis of noise signals, and the surveillance algorithm forms a statistical description of normal behavior from this characterization. Hyperellipsoids constructed from this description are used to enclose normal regions of behavior in the multi-dimensional measurements space which represents all possible reactor states. When measurements falling outside of normal regions are detected, the status of the reactor is suspect. This algorithm has been implemented on a minicomputer and has demonstrated its effectiveness in on-line tests. An experiment at the High Flux Isotope Reactor demonstrated the ability of the algorithm to sense changes in reactor conditions that the plant instrumentation could not detect