Date of Award
Doctor of Philosophy (PhD)
Professor John F. MacGregor
The main goal in many processes is to obtain consistent and reproducible operation and end-quality properties. In this thesis the problem of product quality control in batch and semi-batch processes is addressed. Unlike from much of the published literature that uses first principles models, this thesis studies the end-quality feedback control problem using only empirical Partial Least Squares (PLS) models. Several simple, practical and effective regulatory control strategies are proposed.
The thesis consist of four main chapters: i) On-line control of a distributed end quality property (particle size distribution, PSD) using mid-course correction strategies (MCC), ii) an inferential-adaptive control approach that combines on-line and batch-to-batch control, iii) a novel reduced dimensional space control algorithm to obtain complete manipulated variable trajectories (MVT) consistent with past operation, and iv) incorporation of prior batch-to-batch information for batch analysis and monitoring.
In the first section, three on-line empirical MCC strategies are proposed for the control of bimodal PSDs in emulsion polymerization systems. The performance of the control strategies is evaluated using a detailed theoretical simulator. Control is applied only when the predicted properties falls outside a statistically defined "no-control" region. Each control strategy corresponds to a control objective: i) Control of second mode of the distribution, ii) control of the full bimodal PSDs and iii) control of relative distributions. Advantages and disadvantages of each one of the control strategies are discussed.
In the second part a combined on-line and batch-to-batch control strategy is presented. The approach extends MCC strategies used before to include multiple decision and correction points, batch-to-batch information to reject batch-wise correlated disturbances, and an adaptive PLS approach to update the models from batch-to-batch to overcome model error, changing process conditions and unknown disturbances. The methodology is also illustrated with the control of PSD in emulsion polymerization. The problem of regulation about a fixed set-point PSD in the face of disturbances, and the problem of achieving new set-point PSDs are both illustrated.
In the third part a novel strategy for controlling end-product quality properties by solving on-line for complete MV trajectories, for the remainder of the batch, is presented. Control through the optimal solution for complete trajectories using empirical models is achieved by performing the model inversion and the MVT reconstruction in the reduce space of a latent variable model. The approach is illustrated with a condensation polymerization example for the production of nylon and with data gathered from an industrial emulsion polymerization process.
In the last section an extension of the multi-block multiway Principal Component Analysis (MPCA) and MPLS approaches is introduced to explicitly incorporate batch-to-batch trajectory information. It is shown that the advantage of using information on prior batches for analysis and monitoring is often small. However it can be useful for detecting problems when monitoring new batches in the early stages of their operation. The approach is illustrated using condensation polymerization and emulsion polymerization systems as examples.
Flores-Cerrillo, Jesus, "Quality Control for Batch Processes Using Multivariate Latent Variable Methods" (2003). Open Access Dissertations and Theses. Paper 831.