Quality engineering approach for parameter design of dynamic systems

Abstract

Parameter design, proposed by Taguchi as a method for improving product quality, has been the subject of a substantial amount of interest and debate among statisticians and quality engineers. The objective of parameter design is to reduce performance variation (i.e., to achieve robustness) of a system by determining an optimal setting of its design parameters. To achieve the stated objective, Taguchi suggested to use the SN ratio as a measure of robustness and recommended that experiments using orthogonal arrays be conducted with the goal of finding an optimum setting that maximizes the SN ratio. Although there exists no disagreement on Taguchi``s idea for quality improvement, there has been a considerable amount of debate and criticism on his specific techniques, and various alternative approaches have been recently proposed. Parameter design problems can be broadly classified into static and dynamic ones depending on absence or presence of a signal parameter, respectively. Most previous studies on parameter design focused on the static case, and relatively little work has been done for the dynamic one despite its importance in practice. In addition, the applicability of the existing approaches for dynamic parameter design is limited to simple cases and a consistent framework to encompass a wide class of dynamic parameter design problems is lacking. This thesis is concerned with parameter design problems in dynamic systems. Based on the expected loss as a measure of performance variation and the concept of adjustment, a consistent framework for parameter design is developed for dynamic systems. For the case where both input signal and output characteristic are binary, we show that an optimal design setting found by the Taguchi approach is approximately optimal in terms of minimizing the expected loss. This thesis also develops a parameter design method for the case where the variances of feature variables are not equal, which has never been considered by Tag...