Abstract
In the highly competitive environment in which companies work, quality has become a key tool of survival. It is now accepted that quality must be achieved from the design of both products and processes. Taguchi proposed a methodology for robust design of parameters, to design products less sensitive to random factors or noise that cause variability in the parameters that define its quality. This methodology has been commonly used in industry despite the great controversies aroused from its inception.
This paper provides a complete and statistically robust alternative for improving parameter's design, considering firstly experiments with a single quality characteristic as response variable and subsequently it has been generalized to cases with multiple quality characteristics. This proposal, termed as Forest-Genetic Method, combines data mining tools and metaheuristics in 3 phases: normalization, modeling and optimization.
For both univariate and multivariate cases, results are numerically compared with the most recent contributions in the literature using 4 different case studies. We verified that our proposed methodology is focused on the most important variables of the product modeling process, allowing us to achieve significant improvements in the quality objective considered in each case.