Model free subspace based H(infinity) control

Plant knowledge is essential to the development of feedback control laws for dynamic systems. In cases where the plant is difficult or expensive to model from first principles, experimental data are often used to obtain plant knowledge. There are two major approaches for control design incorporating experimental data: model identification/model based control design, and model free (direct) control design. This work addresses the direct control design problem.; The general model free control design problem requires the engineer to collect experimental data, choose a performance objective, and choose a noise and/or uncertainty model. With these design choices, it is then possible to calculate a control law that optimizes expected future performance. Recently, there has been significant interest in developing a direct control design methodology that explicitly accounts for the uncertainty present in the experiments data, thereby producing a more reliable and automated control design technique.; This research exploits subspace prediction methods in order to develop a novel dire control design technique which explicitly allows the inclusion of plant uncertainty. The control design technique is known as model free subspace based $H\infty$ control. The new control law can be viewed as a method of “predictive control” similar to the $H2$ based “generalized predictive control” (GPC), or the new control law can be viewed as an extension of model free subspace based linear quadratic Gaussian (LQG) control. The ability to easily include plant uncertainty differentiates the model free subspace based $H\infty$ technique.; A computationally efficient method of updating model free subspace based controllers derived, thereby enabling on-line adaptation. This implementation method is particularly effective because the computational effort required to incorporate new data is invariant respect to the total amount of data collected.; The $H\infty$ design technique is demonstrated through a number of laboratory experiments utilizing a flexible structure. High performance control is experimentally demonstrated for a non-collocated control problem using a very short “identification” data set. In simulation, the adaptive model free subspace based $H\infty$ technique is found to rapidly develop an excellent control law after just a few seconds of system operation.