No abstract

No abstract

This research presents a new method to generate near-orthogonal balanced mixed-level fractional designs. The proposed method showed that it is possible to create near-orthogonal balanced fractions of economic size. The method is based in the analysis of the behavior of the genetic algorithm used to generate the efficient arrays (EAs) developed by Guo; a pattern was detected, and this led to the generation of an algorithm capable of constructing fractions in a simple way. These fractions were called near-orthogonal balanced arrays (NOBAs). To analyze the properties of the NOBAs and the capabilities of the proposed method, a series of performance indicators were defined. The NOBAs were compared with the EAs developed by Guo; results are provided.

This article presents a more efficient method for sequential augmentation of mixed-level designs. The proposed approach reduces the optimal foldover plan of a mixed-level design to a semifold plan by selecting half of the treatment combinations of the foldover fraction using exhaustive search and the criterion of general balance metric. The resulting design is a more economic run size augmented fraction that possesses good balance and orthogonality properties for main effects and two-factor interactions. Three efficient arrays consisting of 20, 24 and 30 runs were selected for the analysis. Efficient arrays composed of a higher number of runs can be semifolded in a similar manner.

The Taguchi approach for robust design has been a common practice in industrial experimentation for many years. However, these designs possess serious disadvantages such as the inability to estimate control × control interactions. In this article, we propose the application of the R3 algorithm as an augmentation tool for Taguchi experiments. The augmented Taguchi designs were compared with its competitors, mixed resolution designs, and D‐optimal augmentation, using performance indicators. The results showed that Taguchi designs augmented with the R3 algorithm are capable of estimating control × control interactions, possess similar values for performance indicators when compared with other techniques, and in most cases require less runs.