Construction of marginally coupled designs with mixed-level qualitative factors

Abstract

Marginally coupled designs (MCDs) with more economical run sizes than sliced Latin hypercube designs were widely used in computer experiments with both quantitative and qualitative factors. However, the construction of MCDs with mixed-level qualitative factors was still very challenging. We developed five algorithms to generate MCDs with mixed-level qualitative factors, which were very easy to implement. In some of the MCDs constructed in this paper, the quantitative factor designs have two- or higher-dimensional space-filling properties compared to the existing MCDs, where the qualitative factors were mixed-level. Moreover, the resulting MCDs had more flexible run sizes than the existing MCDs with mixed-level qualitative factors.

Keywords

• resolvable orthogonal array
• sliced latin hypercube design
• space-filling
• stratification
• difference scheme