One-day Workshop: Monday M-1

MODEL PREDICTIVE CONTROL

Jay H. Lee, Georgia Institute of Technology
Joseph Lu, Honeywell Process Solutions
Richard D. Braatz, University of Illinois at Urbana-Champaign

8:30am - 5:00pm
Adams

Abstract
This one-day short-course describes the most popular advanced control techniques applied in the process industries. The course describes the systematic "best practices" approaches for model identification and the use of the model in predictive control algorithms developed for linear and nonlinear, continuous and batch processes. Many company applications are described in some detail, to provide guidance on how to address the issues that commonly arise during controller design for industrial processes, including sensor calibration, model uncertainties, and constraints on the actuators and states. This course is suitable for practicing engineers, students, instructors, and researchers interested in control engineering practice.

The workshop begins with a description of model identification that includes optimal techniques for sensor calibration, experimental design, parameter estimation, and model selection. Application to a complex pharmaceutical crystallization process at Merck is used to illustrate the "best practices" approaches. Especially interesting in this example is how to best exploit information from a wide variety of sensors to construct a first-principles model with accurate kinetic parameters. This is followed by a discussion of state estimation techniques including Luenberger observers, Kalman filters, extended Kalman filters, and moving horizon estimation. Application to an industrial polymerization reactor illustrates how to deal with nonidealities when constructing a state estimator for a highly nonlinear process.

The next section on model predictive control (MPC) describes algorithms for linear and nonlinear model predictive control for continuous processes that are applied to industrial processes, including the coupling of state estimation with state feedback to derive output feedback controllers. Tuning guidelines including selection of control and output horizons are llustrated through applications of linear MPC to a high-speed adhesive coating process at Avery-Dennison, and of nonlinear MPC to the control of an industrial ethylene plant. The final section of the workshop discusses model predictive control for batch and semibatch processes, including how to achieve low sensitivity to within-batch and batch-to-batch disturbances.