Robust optimal sensor placement for optimal estimation

Abstract

The goal of this thesis is to present a framework which can be used to optimize the observer and controller performance by placement of sensors and actuators on a continuous domain. The framework provides choices of model structure, cost functions and optimization methods to perform the placement. Systems which would be suitable for the use of the framework are for example adaptive optics systems, active dampening of windmill blades or climate control systems in buildings. To demonstrate the use of the framework, two applications are presented. Both applications are sensor placement for an adaptive optics system. The first implementation uses a Linear Time Invariant (LTI) state space model and uses a Vector Autoregressive (VAR) approach for identification. The second implementation uses a Linear Parameter Varying (LPV) state space model with VAR identification approach. The second implementation is meant to demonstrate how the placement can be made more robust to changing conditions. Finally we show how parametrization of the sensor locations can be used to gain insight into what properties an optimal placement exhibits.