Multivariate statistical methods for monitoring continuous processes: assessment of discrimination power of disturbance models and diagnosis of multiple disturbances

A new methodology was reported [1,2] for integrated use of principal components analysis (PCA) and discriminant analysis in order to determine out-of-control status of a continuous process and to diagnose the source causes for abnormal behavior. Most of the disturbances were identified with good rates of success, with a higher success rate for step or ramp type of disturbances. tative tools that evaluate overlap and similarity between high-dimensional PCA models are proposed in this communication, and their implications on determining the discrimination power of PCA models of processes operating under disturbances are discussed. Diagnosis of several disturbances occurring simultaneously is also investigated. The criterion developed provide upper limits of discrimination power of various single and multiple process disturbances. The techniques developed are illustrated by assessing the process described by the Tennessee Eastman Control Challenge problem [3].