&&ReWrAp:HEADERFOOTER:0:ReWrAp&&

Date of Award

5-1998

Degree Type

Thesis

Degree Name

Doctor of Philosophy (PhD)

Department

Chemical Engineering

Supervisor

Professor John F. MacGregor

Abstract

This thesis considers the utilization of historical process data for three different process engineering problems. Latent variable methods such as Principal Components Analysis (PCA) and Partial Least Squares (PLS) are shown to be key tools for dealing with the highly correlated variables typically found in process operating data and for ensuring feasibility of the solutions. The restrictions and limitations encountered by any databased approach are recognized and discussed. The problem of designing process conditions that yield a new product grade quality within the range of already existing grades is addressed by latent variable regression models and their inversion. Latent variable techniques allow for models that not only describe the relationship between process conditions. This leads to the design of new conditions that are consistent with the plant operations from the past. Feasibility issues of both the new quality specifications as well as the predicted operating conditions are addressed. The approach is illustrated on a simulated process and three industrial processes. The related topic of moving the production of a particular product grade from a plant A to another plant B when both plants have already produced a similar range of grades is treated as well. Since the two plants may differ in size, configuration etc. the process conditions for one grade may be very different in the two plants. A latent variable method is proposed which uses data from both plants to predict process conditions for plant B for a grade previously produced only in plant A. The issue of feasibility is again addressed for both product quality and process conditions. The last part of the thesis explores the problematic issue of utilizing large process operating databases for process performance improvement. The major problem found with using normal operating data for this purpose is correlation among manipulated variables and disturbances as caused by feedback operation or other operating strategies. Such correlation is shown to impede the extraction of causal information from the database, which is necessary in order to infer better process conditions. This implies that only in exceptional circumstances would it be feasible to use empirical databased methods for process optimization.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS