CONTROL SYSTEM DESIGN FOR LIGHTLY COUPLED LARGE SPACE STRUCTURES (MODEL REDUCTION, SINGULAR VALUE)

Concepts for future spacecraft include large structures with many sensors and actuators and a large number of modes in the control bandwidth. Control system design for systems of such large dimension can tax the ability of control design packages to generate candidate designs as well as the ability of the control designer to understand those designs and make improvements.; This research is concerned with techniques of reducing the control problem to manageable size and making it convenient to use time and frequency domain techniques together in the analysis and improvement of control system designs.; Model reduction and system decoupling are complementary methods of reducing a system to manageable size. A model reduction algorithm is presented which extends the class of systems which may be reduced by the balanced realization technique. For system decoupling, an algorithm is presented which approximates a lightly coupled system by a set of decoupled subsystems. This algorithm maximizes the number of subsystems given the acceptable decoupling error.; Control design and analysis insights are available in both the time and frequency domains. For the multi-input/multi-output (MIMO) case, simple, numerically well conditioned algorithms are not readily available to transform frequency domain system descriptions into time domain (state space) form. An efficient, reliable algorithm has been developed to transform the partial fraction expansion of the vast majority of MIMO systems of engineering interest into minimal state space form. This algorithm has been combined with others to produce a user-friendly controls package capable of transforming systems descriptions from state space, partial fraction, pole-zero, or numerator-denominator polynomial form to any one of the others. Other capabilities of the package include LQG design, model reduction, and singular value analysis. Another application of this package is to analyze and perhaps modify (perturb) an LQG compensator in either the time or frequency domain. A solution to the neighboring optimal control problem has been found and included in the package. This finds the weighting matrices associated with a perturbed LQG compensator.; Applications of these algorithms and techniques are given to a helicopter autopilot as well as to flexible spacecraft attitude and vibration control.