Dynamic multivariate statistical process control using partial least squares and canonical variate analysis

Multivariate statistical techniques have been shown to be useful tools for multivariate statistical process control (MSPC) and process modelling. However, these approaches have been mainly applied to static systems. In the present work, a well known system representation, the state space model, is developed to deal with dynamic situations. The states of the system are approximated using two multivariate statistical projection techniques, Partial Least Squares (PLS) and Canonical Variate Analysis (CVA). These two model representations are compared both in terms of their predictive ability and also their monitoring power using a simulation example. An application to an industrial fluidised bed reactor will be presented at the conference following company approval.