Subspace Method Identification for Dynamic Multivariate Statistical Process Control
| Authors: | Treasure Richard, The University of Western Australia, Australia
Kruger Uwe, Queens University Belfast, United Kingdom
Sreeram Victor, The University of Western Australia, Australia |
|---|
| Topic: | 6.1 Chemical Process Control |
|---|
| Session: | Modelling, Estimation and Fault Detection for Process Control |
|---|
| Keywords: | fault detection, model reduction, multivariate quality control, process control, identification, subspace methods. |
|---|
Abstract
Subspace method identification (SMI) and model reduction for Multivariate Statistical Process Control has been proposed as an improvement to dynamic principal component analysis (DPCA). The linear parametric model structure captures both static and dynamic information from the system. In this paper, an analysis of the dimension reduction capabilities of the subspace approach is provided. It is proven that the SMI method yields a parsimonious model structure that requires fewer latent variables and uses fewer process measurements than DPCA. These findings are illustrated by an industrial application study.
