Abstract
Robust design is an experiment-based method for reducing the dispersion of a system's quality characteristic through sensitivity reduction and, if necessary, noise source control. Sensitivity reduction is to select the values of the control factors so that the quality characteristic should become as insensitive, or robust, as possible to the variations of the noise factors of the system. Noise source control is the costly effort of reducing the variations of some noise factors themselves within a certain limits. In addition to a traditional overall dispersion measure like SN ratio, this paper first introduces multiple dispersion measures each of which evaluates the robustness of the quality characteristic to each noise factor. Then it presents a new approach to robust design using these multiple dispersion measures. When there are more than two noise factors, manipulating the values of some control factors so as to reduce the sensitivity of the quality characteristic to a certain noise factor will not always lower the sensitivities to the other noise factors, but may increase some of them on the contrary. That is, there can be trade-off relationships among sensitivity reductions of the quality characteristic to various noise factors. The proposed approach enables the designer not only to capture but also properly deal with these trade-off relationships. As a result, the noise source control required for reducing the dispersion of the quality characteristic down to a satisfactory level will be less expensive than the case where the traditional approach is used. The power of the proposed approach is illustrated with an example.
