| Screening designs are used in the early stages of industrial and computer experiments to find the most important input factors affecting a system's output. They provide an economical way to remove unimportant factors from further, potentially costly, experimentation. However, when an experiment has a large number of control factors and limited number of available runs, it is infeasible to run a traditional screening design. In these situations, experimenters can use supersaturated designs. A supersaturated design is a factional factorial design that can screen a set of k factors in n runs, where k > n − 1. Unfortunately, they do not always provide definitive results. Improper and incomplete analysis of supersaturated designs can cause an experimenter to misclassify active factors and waste resources in subsequent experiments. In light of these concerns, this research investigates how to construct efficient and effective supersaturated designs, how to analyze such designs, and how to strategically plan follow-up runs to designs. |
