Using Canonical Correlation to Optimize Taguchi’s Multiresponse Problem

Abstract

A conventional and efficient approach, the Taguchi method, is applied to an off-line quality control so that the experimental design is combined with quality loss in order to optimize a single response problem. Three stages, system design, parameter design, and tolerance design, are included in the Taguchi method. In the real world, more than one quality characteristic should be considered for most of the industrial products; i.e., most applications customers are concerned with multiresponse problems. Therefore, the Taguchi method is not appropriate to optimize a multiresponse problem because engineering judgment and experience are the main optimization scenario in Taguchi’s method. In order to overcome this gap, this study proposes the effective procedure, canonical correlation, to optimize the multiresponse problem. Using canonical correlation, multiple responses in each experiment are transformed into a performance index. Therefore, the optimal factors/levels combination for the multiresponses can be determined. Two case studies in Tarng et al. [7] and Lin et al. [9] are resolved by the proposed procedure. The result indicates that canonical correlation can yield a satisfactory solution for multiresponse problems.