Development of SN ratio and interaction graphs for dynamic parameter design problem

Abstract

The Taguchi method has received much attention from engineers and scientists as an innovative and effective method for improving product quality. Parameter design is the core of the method, and its purpose is to determine settings of the control factors at which the performance characteristic of a product becomes robust to various noise. Most of previous studies are concerned with a static parameter design problem in which the target value of a performance characteristic is fixed, and relatively little work has appeared for the dynamic problem despite its importance in practice. In this thesis, we consider the case where controlled noise factors are introduced in a dynamic parameter design experiment, and develop an appropriate model and analysis methods. We first analyze the Taguchi SN ratio under the proposed model and show that it tends to underestimate the true SN ratio. We then propose two separate performance measures called linearity and robustness measures. Finally, we develop graphical analysis methods for investigating control by noise and control by signal interactions. These interaction graphs as well as the above separate measures can provide valuable information on the underlying mechanism of the system under investigation.