Doctoral Dissertations

Nitin Kaistha

Date of Award

12-1999

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Chemical Engineering

Major Professor

C. F. Moore

Committee Members

George C. Frazier, Mary G. Leinaker

Abstract

This work develops a generic Statistical Process Control (SPC) framework for characterizing the systematic variability seen in a historical database of batch profiles in terms of meaningful scale parameters and studying the correlation with the final product quality. The complete framework, in contrast to existing methods, is geared towards giving meaningful results that can be easily connected to the actual process.The variability in the profiles is partitioned into two parts - consistent and inconsistent. The consistent variability is characterized using scale parameters. The Consistent variability is further partitioned as along the time and measurement axes leading to time and magnitude scale parameters for the same, respectively. Time Scaling refers to the alignment of events in a batch while magnitude scaling refers to the use of projection methods for extracting the systematic variability along the measurement axis. The framework thus integrates the time synchronization of thebatch trajectories and the subsequent application of multivariate techniques for monitoring and quality correlation purposes.Tools for time and magnitude scaling are developed. The use of DynamicTime Warping (DTW) for time scaling is studied. A novel technique based on translation of feature vectors is also developed and is shown to be especially suited for time scaling batch profiles. Magnitude scaling is based on the application of projection methods such as Principal Component Analysis (PGA) for extracting directions of systematic measurement axis variability in the profiles. PGA is modified to an evolving factor type method, for extracting easily interpretable factors.IllMultivariate SPC charts for process monitoring on the scale parameters and the residuals remaining after scaling are developed. The framework presented is an offline analysis and can be readily adapted for online monitoring purposes.A polymethyl methacrylate batch polymerization simulation is used in order to demonstrate the application of the SPC framework. The importance of time scaling in explaining a significant amount of the overall variability in the profiles and also significant correlation with the final batch quality is established. Comparisons with multiway principal component analysis (MPCA), the existing framework, show the better sensitivity of the proposed technique to special cause disturbances. The ease in the engineering interpretation of the results as contrasted to MPCA is also established.

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

Share

COinS