Robust adaptive control design techniques

This thesis develops and applies methods for designing robust adaptive identification and control algorithms. Algorithms for modal identification, pointing control, and active vibration damping are tested in a flexible beam control experiment. This experiment is analogous to many practical aerospace systems because it is lightly damped, has noncollocated sensors and actuators, and an accelerometer as the primary sensor.; Methods are developed for designing identification techniques to be robust to unmodeled dynamics, disturbances, and sensor noise. The viewpoint of analyzing prediction error identification methods (PEM) in terms of frequency domain criteria is adopted here. Frequency domain methods are used to design an on-line transfer function identification approach for the laboratory setup which is based on modal decomposition. It is shown that a PEM cost criterion based on relative transfer function bias error is a better predictor of identification performance than the bias criterion often used. This relative fit criterion includes the effect of system zeros, which makes it valuable for evaluating sensor dynamics and placement. Another contribution is the formulation of conditions for the exact equivalence of multivariable autoregressive moving-average exogenous input (ARMAX) and autoregressive exogenous input (ARX) models. This result shows that systems governed by multivariable ARMAX models can be identified exactly using linear least square methods based on ARX models.; Design rules for constructing digital controllers for the laboratory model are developed using classical frequency domain techniques. These rules utilize system parameter estimates and the resulting controllers are more robust to unmodeled dynamics than pole placement techniques typically used in adaptive control designs.; The identification and control designs are integrated into an adaptive control approach. A robust controller which provides stability for all a priori plant variations is used initially. When the uncertainty in the system parameters is reduced, the controller is updated to a design with better performance. This approach avoids start-up transients that conventional adaptive controllers may experience. The basic feasibility of this approach is demonstrated.