Canonical Correlation Analysis-based Fault Detection Methods

Abstract

Statistical FD methods consist of two major procedures: off-line training and on-line monitoring. The differences between PCA- and PLS-based methods are that the PCAbased methods only consider the process variables in both procedures and detect changes in the condition of the process, sensors and actuators, while PLS-based methods are applied to process variables and output variables (quality variables) or key performance indicators, which are on-line unmeasurable or measurable only with a large time delay.