Instrumentation Monitoring Using Multivariate Statistical Projection Methods

Abstract

The objective of this research was to investigate the feasibility of monitoring sensor accuracy using Principal Component Analysis (PCA). The specific project was to develop and test a monitoring scheme for the pressure-based transmitters associated with an operational CANDU nuclear generating station. The results of the research indicate that individual groups of redundant sensors can be adequately monitored using PCA models with one to four principal components. These models can generate warning alarms for errors of small magnitude and action alarms for large magnitude errors.