Abstract: Robust parameter design, originally proposed by Taguchi, is a very useful tool for reducing variation and improving product/process quality. By exploiting the relationships between control parameters (design factors) and noise variables, it reduces the effect of uncontrollable variations on the response. This is done by using statistically designed experiments in off-line situations where the settings of the noise variables are controlled and systematically introduced and their relationships with design factors studied.
Frequently, however, it is too expensive or not practically feasibl e to control the noise variables, even in off-line experiments. Variations in the noise variables can then invalidate the usual methods of analysis. In this paper, we develop alternative methods of analysis for situations where the noise variables are uncontrolled but can be observed. Our approach involves treating the noise variables as covariates and modeling both the location parameters and the regression coefficients as functions of the design factors. These coefficients can be viewed as the equivalent of Taguchi's S/N ratios for reducing variability induced by the observed noise variables. We propose a general data-analysis strategy for determining various dispersion and location effects and improving performance under this framework. The approach is also applicable to experiments where there are covariates, and one must remove the effects of these nuisance variables before identifying the location and dispersion effects of the design factors. The ideas are illustrated by applying them to an experiment for thermal design of cabinets for telecommunications switching systems.
Key words and phrases: Design of experiments, dispersion effects, location effects, quality improvement, S/N ratios.