| This thesis presents the results of an independent implementation of two control design techniques from the literature. The Maximum Entropy Method, due to Hyland, is based on the use of a time-varying plant with a minimal set of known stochastic parameters (modal decorrelation times or modal frequency variances). Robustness is achieved by minimizing the output 2-norm in a probabalistic sense, but is not guaranteed. The Multiple Model Method uses a set of N specific plants and minimizes the output 2-norm over the ensemble. Both approaches are implemented with improvements for usability and efficiency. A variant of Hinfinity design is used and extended as a basis for comparison. All methods are for Linear Time-Invariant systems. The methods are applied to three control problems of increasing complexity---a mass-spring system; a single, cantilevered gyrodynamic beam; and a linear model of the DemoDICE system, a ground-based prototype of a free-flying control research facility planned for the Space Shuttle Mid-deck. Controllers are designed for each plant using the three methods, then tested for robust performance. Results indicate that the Multiple Model design technique is very promising, while the Maximum Entropy technique is useful also, though not performing as well. The Hinfinity technique was only a good competitor when the control architecture had the same input for disturbance and actuation, which is the case on the DemoDICE system.; In addition to the main work on control, a number of ancillary pieces of analysis are added to the DICE/gyrodynamics canon, including a transformation to velocity control for Control Moment Gyroscopes and a Jordan Canonical Form for the DemoDICE linear model. |
