A Statistical Methodology of Designing Controllers for Minimum Sensitivity of Parameter Variations

Abstract

A statistical methodology is presented for designing controllers in problems where analytical solutions are unobtainable. The methodology is applicable to many complicated systems containing, for example, nonlinearities, uncertainty, and multiple inputs and multiple outputs. Because the design technique is a simulation based approach, no specific restrictions are placed on either the plant or the controller structure. A Monte Carlo technique is used to map the parameter space onto the indices of performance. The system performance either passes or fails the performance index. The objective in systems with uncertain parameters is to select (controller) parameter values which maximize the probability of passing the performance criterion. In deterministic systems, the goal is to find parameter values in the pass region that are as insensitive as possible, that is, parameter values that allow for the maximum amount of parameter variation without causing the system response to leave the pass region.