On-line monitoring and control of product quality for batch and semi-batch processes with applications to nylon 6,6 productio

The objective of this dissertation is the development of effective approaches for improving the control of product quality for batch and semi-batch processes subjected to a complete lack of on-line measurements of product quality and frequent random variations in feed conditions and process operating parameters. For this work, quality variables are defined as the engineering variables that are most closely related to the end-use characteristics of the product. This category typically includes difficult to measure variables such as polymer molecular weight distribution parameters and pulp Kappa number. Primarily, the stated objective will be achieved using a strategy which can be described as inferential quality monitoring and control. The inferential method is essentially a model-based approach in which the quality variables of interest are not measured but are "inferred" using on-line secondary measurements and process models.;The two types of inferential monitoring and control approaches that will be developed in this work are classified according to the form of the model that will be used. In the fundamental model approach, some form of a fundamental process model is available to relate on-line measurements and process inputs to the final product quality. The primary contribution in this area will be improving fundamental model-based batch quality monitoring by developing nonlinear smoothing techniques. The smoothing algorithms presented in this work are designed to reduce the effects that uncertain initial conditions resulting from frequent feed disturbances have on state estimation and quality monitoring for batch processes. In the data-based approach, empirical input-output type models are built using only the historical process measurement data of past batches. The main contribution in this area will be the development of novel recursive quality prediction and control techniques using the data-based empirical models (usually partial least squares or principal component regression models). Finally, both the fundamental model-based and data-based approaches are applied to an industrially relevant case study involving quality control for the batch production of nylon 6,6 in vaporizing autoclave reactors.;The dissertation is organized as follows: First, a general introduction to the research contained in this work is discussed in Chapter 1. The fundamental model-based monitoring approach is developed in Chapter 2. In Chapter 3, the data-based quality prediction and control approach is formulated. Chapter 4 examines an industrially relevant case study by applying both of the approaches in Chapters 2 and 3 to quality control for the batch production of nylon 6,6. Finally, some concluding remarks are made in Chapter 5.