Master of Science in Chemical Engineering (MSChE)
High-purity distillation columns are highly nonlinear systems. Nonlinear Model Predictive Control of these columns is challenging. The complexity of a first-principles model of a multi-stage distillation column, which involves a large number of differential and algebraic equations, makes Model Predictive Control computationally expensive. This thesis has focused on Nonlinear Model Predictive Control based on a simplified wave model of a distillation column; the simplified wave model consists of only one ordinary differential equation along with several algebraic equations. The predictive capability of the nonlinear wave model can be improved significantly through the application of nonlinear state estimation. Off-line estimation by the least-squares method is explored to determine what parameters should be estimated to well replicate the composition profiles. The Principal Component Analysis technique is utilized to determine the number and locations of measurements that can best represent the changes of the estimated parameters over a wide range of operating conditions. The state estimator is combined with a nonlinear model predictive controller that regulates the overhead composition by adjusting the wave position via manipulation of the feed flow rate. The differential equation for the wave position includes parameters such as the wave slope and the upper limit of the vapor composition profiles that vary according to the operating conditions. These two parameters and the wave position are estimated with an extended Kalman filter designed from a simplified wave model in which the dynamics of the combined reboiler/condenser are neglected. Satisfactory reconstruction of the composition profiles produced by the rigorous dynamic simulator is achieved if the overhead vapor composition, bottoms liquid composition, and two intermediate tray liquid compositions (chosen for their sensitivity) are selected as measured outputs. A nonlinear model predictive controller is designed from the simplified wave model by utilizing the EKF state and parameter estimates to obtain an accurate estimate of the current wave parameters. The effectiveness of the NMPC strategy is evaluated for output set point changes where a rigorous dynamic simulator is used as to represent the nitrogen column.
Document Availability at the Time of Submission
Release the entire work immediately for access worldwide.
Bian, Shoujun, "Nonlinear state and parameter estimation and Model Predictive Control of nitrogen purification columns" (2002). LSU Master's Theses. 2444.
F. Carl Knopf