Development of inferential sensors for chemical processes using partial least squares
- Authors
- Steven D. Roney
- Published
- 1998
- External link
- http://library.mcmaster.ca/catalogue/Record/1107665
- Citation
- Roney: "Development of inferential sensors for chemical processes using partial least squares", Masters thesis, McMaster University, 1998.
- Abstract
- This thesis focuses on the development of inferential sensors for chemical processes using partial least squares (PLS). Inferential sensors are used in many areas of chem ical engineering where desired variables are difficult or costly to measure, or where industrial instruments to measure these variables simply don't exist. In these cases, it is often possible to build models relating the desired variable to readily available process measurements, thus being able to "infer" the state of the desired variable.
The objective of this work is to present a systematic approach to building in ferential sensors. Partial Least Squares (PLS) methods are used as the model building tool because of their ability to deal with highly correlated process measurements in a meaningful way, and their ability to handle missing data. The fundamentals of the PLS algorithm are presented along with discussions of the important areas of variable selection and scaling, and determining the dimension of the model.
Approaches to detecting and modelling nonlinearities and process dynamics are presented and compared. Both simulated and industrial data sets are used to illustrate these concepts and methodologies.
The importance of using data collected under appropriate conditions is dis cussed. In particular, the effect of the presence of feedback control in the process is examined as an issue that needs to considered in order to implement an empirical inferential sensor successfully in an inferential control scheme.
This work is intended as a guide to aid in the development of industrial inferential sensors using PLS. Important issues that need to be kept in mind through the model building process are brought forward, and diagnostic plots and model building procedures for the important areas of modelling dynamics and nonlinearities are presented. All methods are illustrated with industrial data sets.
- Tags
Related items
- Development of vision-based inferential sensors for process monitoring and control (2003)
- Improvement of processes and product quality through multivariate data analysis (2000)
- Process systems approaches to diagnostic imaging and property prediction (2006)
- Development of inferential process models using PLS (1994)
- Learning from data using latent variable methods (2006)