Doctoral Dissertations

Gaiye Zhou

Date of Award

8-1993

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Electrical Engineering

Major Professor

J. D. Birdwell

Committee Members

J. M. Bailey, P. B. Crilly, T. W. Wang

Abstract

This dissertation describes three new methods to automatically design autotuners for control systems: the decision tree (DT) method, the simulated annealing and decision tree (SA-DT) method, and the simulated annealing and fuzzy logic (SA-FL) method. The DT method uses representatives from a class of systems to construct a decision tree autotuner. The autotuner constructed is therefore to be applied to the systems within that trained class, particularly lower order systems. The SA-DT method uses a nominal process model but allows the parameters of the model to change within a certain range. A simulated annealing optimization method is used to guide the modifications of controller parameters for extracted example systems, and an inductive inference method is used to construct a decision tree autotuner. This method is intended to be applied to more complex systems or higher order systems. The SA-FL method uses a simulated annealing optimization method to construct a fuzzy logic autotuning (FLA) rule base. This approach produces a FLA rule base automatically by making tests to the process without any a priori information about the process or human expertise about the tuning procedures. The main advantages of the three methods over the existing methods are that (1) the autotuner is constructed automatically using machine learning and/or simulated annealing optimization; (2) the design procedure is easily repeated or modified to construct autotuners for different types of systems with particular performance requirements; (3) methods such as Ziegler-Nichols' and of minimizing integral of time-weighted absolute error (ITAE) or integral of squared error (ISE) are not used to tune the controller or select an initial controller; (4) neither human experience about the process nor particular specifications on the system dynamics is required to design an autotuner; (5) controller structures are not limited; (6) the tuning process Is adaptive to the process changes, and (7) practitioners (users) can specify their own controller objectives and ways to modify the controllers.

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

Share

COinS