A methodology advocated by Taguchi, often called parameter design, has received much attention from practitioners working for quality improvement in industry. Recently, several alternatives to Taguchi's design and analysis strategy for parameter design have been proposed.
In this thesis, we first review the basic idea of Taguchi method and compare various alternative methods in terms of their assumptions, required information for implementating, cost of experimentation, etc. Our findings may be summarized as follows. Leon et al. proposed a performance measure independent of adjustment as an alternative to the Taguchi SN ratio. However, if we do not have enough information on the transfer function, loss function, and adjustment factor prior to experimentation, it cannot be used. The Box transformation approach for achieving independence of the variance from the mean is not valid when the noise factors are controlled. Joint modeling of the mean and dispersion is of theoretical value, but determining an appropriate variance function seems to be a dufficult task in practice. Finally, response modeling approach depends heavily on the choice of data-analytic techniques for model identification and is likely to be sensitive to the target constraint.
Although the proposed alternative methods have the above shortcomings, such ideas as combined array and examination of control by noise factor interactions are of great value. A better parameter design strategy can be developed if these ideas are combined with the Taguchi method.