| Many chemical engineering processes require a rapid and controlled transition without overshoot from one operating condition to another during plant startup or batch operation. For nonlinear, poorly-modeled processes with several large, interacting capacities, achieving the desired transition rapidly and without overshoot may pose a difficult control problem. Previous approaches have included control techniques such as time-optimal control that are not directly applicable for poorly understood, nonlinear, or varying processes. This thesis demonstrates the potential of adaptive control techniques, such as long-range prediction and self-tuning control, for achieving a desired trajectory during changes in operating conditions.;Because overshoot is often an undesirable feature of transitions between operating conditions, two new theoretical results are presented that provide a necessary and sufficient condition for the occurrence of overshoot, and a sufficient condition for no overshoot.;A new adaptive control policy is developed for startup control of nonlinear processes with interacting process capacitances. During the transition a discrete-time dynamic model is identified, providing the basis for the long-range predictions used in the control calculations. In addition, the identified model is used to determine preliminary controller settings for a self-tuning PID controller implemented after the desired operating condition is reached.;The new adaptive control strategy is evaluated in simulations and experiments on an industrial pilot-plant fluidized sandbath. The sandbath is a nonlinear process characterized by large thermal capacities that interact. Since the sandbath is operated at different nominal temperatures, the nonlinearities and process uncertainty are important considerations and, thus, adaptive control is an attractive approach. A physically-based dynamic model of the fluidized sandbath is developed and compared to pilot-plant data. The proposed adaptive control strategy for process transitions is investigated in theoretical analyses, computer simulations, and experimental studies.;The amount of process information available during a single change in plant operating conditions and the resulting limitations for on-line parameter estimation are explored. Confidence limits are calculated for the estimated model parameters, the output predictions, and the corresponding model parameters in a continuous-time model. The influence on the confidence intervals of user-specified design parameters and the signal-to-noise ratio is investigated. (Abstract shortened with permission of author.)... |
