The identification of factors that affect the location and dispersion of the probability distribution of quality characteristics is essential in the optimization of production processes and product design. For this purpose, factorial experiments are typically used. The analysis of dispersion effects, however, usually requires replications of the experiment, which can be expensive. Therefore, several methods have been proposed for the identification of the dispersion effects with unreplicated factorial experiments. In this article, we analyze some of these methods concluding by the superiority of an interactive method based on generalized linear models and which provides a model for the mean together with a model for the dispersion of the quality characteristic. Finally, the use of the resulting model in an optimization process is presented trough an example.
Factorial experiments; Generalized linear models; Process optimization
