Efficiency of Screening Designs with a Small Number of Runs

Abstract

Abstract: Two level non-regular designs are widely used in industrial experimentation for factor screening. At the initial stage of experimentation, it is not possible to know in advance which factors are active so it is necessary to choose the screening design that can accommodate as many models as possible. In this thesis, we have used D-efficiency criteria to evaluate projections onto three dimensions with and without blocking and onto four dimensions without blocking for some screening design with small number of runs. The investigated designs are the 12 run Plackett-Burman designs (PB), the Minimum Resolution IV 10 and 12 runs design, and the John s ¾ design in 12 run. Overall, 12 run PB designs came out as the preferable choice of design. Only in a few cases, it was out performed by the 12 run Minimum Resolution IV design.