Robust tuning of fixed-structure controllers for hard disk drives.

摘要

This dissertation is concerned with the tuning of fixed-structure controllers and its application in the design of track-following controller for hard disk drives.; Instead of building the controller on an over-simplified nominal plant, a comprehensive method of statistically modeling a large number of drives is considered. The model is built based on the decomposition of PESs (position error signals) collected from multiple drives. It can be used to predict the time-domain performance of a population of drives with a given controller in terms of the mean value of variance of PES and the variance of variance of PES, without tedious time-domain simulations.; The parameter optimization of fixed-structure controller is of great interest in control practice. Due to the structure and order limitations, the problem cannot be parameterized as a convex optimization problem. A large number of approaches focus on making the optimization a convex one through appropriate parameterization and approximation. The motivation is that there are effective and powerful algorithms to solve the convex optimization problem. This dissertation develops a MOGA (multi-objective genetic algorithm) to directly solve the multi-objective non-convex optimization problem. The population-based nature of the MOGA enables the evolution of a set of Pareto-optimal solutions without requiring weights before optimization. Furthermore, due to the stochastic nature of search mechanism, the MOGA is more likely to find the global optimum than conventional optimization methods in a non-convex search space. As shown by simulations and experiments, the proposed method is capable of optimizing the controller in a large range in which gradient-based methods generally fail.; While the gradient-based techniques lack robustness over global non-convex optimization problems and are sensitive to initial starting points, they are more efficient than the MOGA in local fine-tuning search. Therefore this dissertation proposes a two-phase algorithm combining the advantage of the MOGA and the gradient-based techniques to further improve the solution quality and computational efficiency.; This dissertation also provides a systematic analysis of the state truncation errors associated with the digital implementation of the track following controller.; Although the methods presented in this dissertation are devised to be applied in the design of HDD track following controllers, the mathematical treatment employed is general and applicable to other engineering applications.