Estimation of SN ratios for dynamic parameter design problems

Abstract

Parameter design, proposed by Taguchi, has received much attention from engineers in industry as an efficient method for designing products or processes. The basic idea of parameter design is to determine settings of design parameters at which the performance characteristic of a product is robust to such noises as environmental disturbances, deterioration, imperfect production, etc. In parameter design, it is very important to obtain statistically valid estimators of SN ratios at each experimental setting. For parameter design of dynamic systems, Taguchi proposed several estimators of SN rations. The purpose of this thesis is to identify the problems with the Taguchi SN ratios and to develop statistically valid alternatives. This thesis deals with two classes of dynamic parameter design problems. The first includes the cases where noise factors are not compounded and the second considers the cases where a compound noise factor is used. For each class of problems, the ideal relationship between the performance characteristic and the signal is assumed to be linear, while the actual relationship between them is modelled by orthogonal polynomials. For each class of problems, we develop statiscally more valid estimators of SN ratios than the ones Taguchi proposed.