A multivariate correlation method as a tool for fault detection and its application to feedwater system in a nuclear power plant

A new method based on multivariate correlation has been developed for surveillance and anomaly detection in nuclear power plants. Natural signal noises observed under normal plant operations are applied in the method, thus the method is suitable for online surveillance. In the method, correlation of signals and noise properties in the process are estimated. Eigenvalues of normalized covariance matrix and correlation function among signals are used for surveillance of process dynamics and operational condition. The effectiveness of the method has been demonstrated with an application to feedwater system of a BWR plant. (author)