The objective of Taguchi's parameter design is to improve the performance of a product or a process by determining the design parameter setting at which the performance characteristics are robust against the various causes of variation. Parameter design problems are broadly divided into two cases, static and dynamic ones, depending on whether a signal parameter exists or not. Most existing studies of parameter design are concerned with the static case, although the dynamic one is more important in practice.
This thesis deals with parameter design problems in dynamic systems in which both the performance characteristic and the signal parameter are continuous. First, we summarize the previous studies concerning the principles of selecting a performance characteristic and propose a more systematic set of guidelines. Next, based on the classification scheme for static characteristic problems, we also classify dynamic characteristic problems into several cases and propose proper SN ratios for each case. Further, we show the validity of the SN ratios proposed by Taguchi for both the cases where the data consist of complex numbers and where errors are introduced in the measurement of the signal parameter. Finally, assuming that the actual relationship between the performance characteristic and the signal parameter is modeled by orthogonal polynomials, we propose statistically more valid estimators of SN ratios than the ones proposed by Taguchi.