Data-Based Latent Variable Methods for Process Analysis, Monitoring and Control
Prof. John MacGregor, McMaster University
This paper gives an overview of methods for utilizing the
large amounts of highly correlated data available in process databases.
These data matrices are almost always of less than full statistical rank,
and therefore latent variable methods are well suited to obtaining useful
subspace models for treating a variety of important industrial problems.
The following problems are discussed and illustrated with industrial
examples: (i) the analysis of historical databases and trouble-shooting
process problems; (ii) process monitoring; (iii) using of multivariate
information from novel sensors; and (iv) process control in reduced
dimensional subspaces. In each of these problems latent variable models
provide the framework on which the solutions are based.
John MacGregor
received his PhD degree in Statistics, his MSc degrees in Statistics and
in Chemical Engineering from the University of Wisconsin, Madison, and his
Bachelor of Engineering degree from McMaster University, Hamilton, ON,
Canada. After working in industry for several years as a process
specialist with Monsanto Company in Texas, he joined McMaster University
in 1972, where he is currently a Professor in the Department of Chemical
Engineering. His research interests have spanned a wide range of areas,
from polymer reaction engineering to process systems engineering, control
theory, and statistical methods. In recent years he has concentrated on
the development of multivariate statistical methods for use in process
monitoring, fault detection, and control using the very large multivariate
data-bases available from industrial processes. This multivariate research
includes problems in both continuous and batch processes as well as image
analysis methods.