An investigation of parameter design for off-line quality control

Abstract

In off-line quality experimentation, most engineers are concerned almost exclusively with the performance mean of product quality. In contrast, the Taguchi method is concerned not only with the performance mean but also with the performance variability of product quality, through the use of orthogonal arrays. However, both approaches typically assume that the underlying populations of experimental responses are known and normally distributed. Therefore, they rely heavily on ANOVA methods with F tests to identify important factors or variables in screening experiments. This underlying assumption may be invalid in many cases.

This dissertation develops a parameter design analysis method for off-line quality control, with the use of orthogonal arrays for experimental layouts, in situations where the underlying populations are unknown. It examines the shape characteristic, kurtosis, of a common Taguchi performance statistic. In addition, it develops nonparametric counterparts to the Taguchi experimental analysis methods for experiments involving factors with two levels and with three levels.