ABSTRACT
Reducing variation in key product features is an important goal in process improvement. Finding and controlling the cause(s) of variation is one way to reduce variability but may not be cost effective or even possible in some situations. Alternatively, we can reduce variation in a critical output by reducing the sensitivity of the process to the main sources of variation rather than controlling these sources directly. This approach is called robust parameter design and exploits interaction between the causes of output variation and control factors in the process. In the literature, a variety of experimental plans have been proposed to help implement robust parameter design. We compare two classes of plans that we call desensitization and robustness experiments. With a desensitization experiment, we need knowledge of a dominant cause and the ability to set its level in the experiment. With a robustness experiment, we use time or location (Shoemaker et al. Citation1991) to indirectly generate the effect of the dominant causes of output variation. In this article, we explore qualitatively and quantitatively the differences between robustness and desensitization experiments. We argue that for an existing process, desensitization is the preferred choice.
Notes
S = 3.72021; R 2 = 84.67%; R 2 (adj) = 82.92%.