| This thesis develops a new class of feedback controllers for poorly modeled processes. The calculation of the control action is based on a set of dynamic models, as opposed to a single dynamic model used in standard model-based control. The advantage of the proposed set-based controllers is that they reduce the need for "detuning", which is a standard practice in model-based control for improved robustness to modeling errors.;A general notion of set-based control is formulated and its theoretical properties are investigated. Set-based control is then applied to two types of control problems. The first is the control of processes with either a large and imprecisely known deadtime or an imprecisely known time constant and deadtime. These types of processes are difficult to control with traditional model-based controllers; even a small error in the model dead time may result in a large loss of performance. A set-based control strategy, which uses a range of deadtimes, instead of a single value of deadtime is then developed and compared to standard model-based control.;The second type of control problem is for processes with linear dynamics and nonlinear output map; a classical example is pH processes, which are characterized by the presence of a strong and uncertain static output nonlinearity, which is related to the titration curve of the system. A set-based control strategy is developed using the set of possible titration curves for the system. This controller is equivalent to a model-based controller with the most conservative nonlinearity. Robust stability of this model-based controller is proven and case studies are used to evaluate its performance. Then, the model-based controller is compared with an adaptive controller, in which the titration curve is estimated on-line. Experiments are performed to demonstrate the practical application of the adaptive pH control algorithms and evaluate their performance. |
