Statistical Techniques for Designing a Decoupled Controller to be Robust to Model and Sensor Noise.

摘要

This thesis developed and tested statistical design methods for configuring a robust decoupled controller on a lightly damped structure. The test model used was a lumped mass finite element representation of the Charles Stark draper Laboratory Model I (CSDL-I). The decoupled control methods consisted of a system of three individual subcontrollers designed on the basis of a subset of plant dynamics with control authority enacted through the total system of structural actuators. Transforms were applied to the subcontrollers to insure that dynamic coupling, called observation spillover and control spillover, didn't destabilize the global system. The decoupled controller system was designed based on only the first eight of the twelve natural frequencies (modes) for the CSDL-I. The remaining four high frequency modes were modeled as residuals. This investigation used signal-to-noise ratios, orthogonal numerical experiments, and linear regression to efficiently probe the design space and to produce a robust control system. The measure of system performance was the structural alignment of a particular node referred to a line-of-sight (LOS). Keywords: Flexible structures. (kr)