(Dr. Nader Sadegh, advisor)
"Output Control of Nonlinear Systems: A Formulation Based on State Trajectory Learning"
A nonlinear dynamical plant of arbitrary relative degree and possibly non-minimum phase is considered. The control objective is asymptotic tracking of a desired output trajectory without knowledge of the desired state trajectory, as is often the case in practice. An approximate plant model of lower order is assumed known but only the actual plant's output is considered available for measurement.
This research consists of the formulation and analysis of an output controller for the above objective. Block input-output (BIO) techniques are applied to obtain a square MIMO discrete model of the plant. This BIO model is controlled via an inner feedback loop to achieve discrete-time tracking of a desired state trajectory which, in turn, is estimated via an outer learning control loop. Robustness of the controller to disturbance and unmodeled dynamics are investigated in simulations of a flexible arm.
The controller is implemented numerically but formulated as a static functional mapping for: a) implementation feasibility since closed-form solutions will not exist in general, b) extension to the output control of systems with more outputs than inputs (appears to have no continuous-time solution), and c) possible application of Neural Networks for on-line identification and parallel computation of the controller.
Application areas include output control of under-actuated, non-minimum phase, and/or non feedback-linearizable systems. These applications include flexible and free-flying manipulators, satellite structures, and land vehicle steering and navigation.