H2 Optimal Sensing Architecture with Model Uncertainty

Abstract

In this thesis, I shall present an integrated approach to control and sensing design. The framework assumes sensor noise as a design variable along with the controller and determines l1 regularized optimal sensing precision. This design satisfies a given closed-loop performance in the presence of model uncertainty. Two methods will be proposed to achieve this.

The first method designs a controller for an open loop uncertain system, which is scaled in order to have a finite H2 norm. Within this, two approaches have been pursued. In the first approach, uncertainty has been represented as polytopic and, in the second formulation, modelled using integral quadratic constraints (IQC). These two approaches have been applied to an active suspension control and sensing design problem and demonstrate that the IQC based approach provides better results and is able to incorporate larger system uncertainty.

The second method finds an appropriate scaling to bound the H2 norm of an uncertain controlled system. The sensor precision is found as the minimal solution to an optimization problem. The design is tested for stability and robustness on a tensegrity robot arm model.