Survival of the Fittest Controller

Abstract

This thesis presents computational approaches to controller tuning. It addresses the hypothesis that standard hard disk drive track following control designs do not take full advantage of actuator capability. In the first experiment a controller is synthesized with Matlab’s hinfstruct for H∞ robust optimal performance on a training set modeled on real drive actuator response data. Compared to standard robust synthesis the sensitivity bandwidth of the closed loop system increases, at the expense of robustness to hardware uncertainty. In the second experiment the scalar parameters of a classical control law are tuned with Matlab’s fmincon to increase the simulated closed loop bandwidth of the drives in a training set. The results of both experiments support the hypothesis while providing avenues along which these modification can be developed and applied to other applications.