Dynamic multivariate statistical process control using subspace identification

In this article, the monitoring of continuous processes using linear dynamic models is presented. It is outlined that dynamic extensions to conventional multivariate statistical process control (MSPC) models may lead to the inclusion of large numbers of variables in the condition monitor. To prevent this, a new dynamic monitoring scheme, based on subspace identification, is introduced, which can (1) determine a set of state variables for describing process dynamics, (2) produce a reduced set of variables to monitor process performance and (3) offer contribution charts to diagnose anomalous behaviour. This is demonstrated by an application study to a realistic simulation of a chemical process.