Date of Award
Doctor of Philosophy (PhD)
Professor F. MacGregor
This thesis develops inferential sensors for on-line process monitoring and control based on multivariate image analysis. Several methodologies based on multivariate statistical methods, such as Principal Component Analysis (PCA) and Partial Least Squares (PLS), are developed to efficiently extract information in real-time from time varying images and to predict process and product properties. The inferential sensors developed based on these methodologies are showen to be sufficient for on-line monitoring and feedback control as illustrated through two industrial applications: snack food processes and flame monitoring. These methods can be easily extended and used for a wide variety of other on-line monitoring and control problems.
In the snack food applications features are extracted from RGB color images and used in predicting the average coating concentration on the product and the coating coverage distribution over the product pieces. Data collected using both on-line and off-line imaging from several different snack food product lines are used to develop and evaluate the approaches. Results of successful monitoring and control from the on-line applications to the industrial processes are shown. Several robustness issues such as detection of model inadequacy and on-line correction are also discussed. Preliminary results on the prediction of organileptic properties (taste and texture of snack foods) are shown as well.
In the flame application, an on-line digital imaging system is developed for monitoring a turbulent nonpremixed flame in an industrial boiler. By using PCA score plots stable information can be obtained for highly fluctuating flame images. A feature extraction approach is proposed to extract the information from the flame color images. The information extracted from the images is then used to successfully predict the performance of the boiler system, such as the energy content of the fuel, and the concentration of NOx and SO2 emissions in the off-gas using PLS regression. Results show that flame images contain a large amount of information that is useful for monitoring the performance of the boiler system. The approach is very general and can be applied to a wide range of combustion processes.
A general framework for building vision-based inferential sensors for monitoring and control of process and product properties using multivariate image analysis is presented. The applications to the snack food processes and the flame monitoring system are shown to fit into this general framework.
Yu, Honglu, "Development of Vision-Based Inferential Sensors for Process Monitoring and Control" (2003). Open Access Dissertations and Theses. Paper 866.