Selection of active effects based on a ridge regression in supersaturated response surface designs

Abstract

Response surface method has been a commonly used statistical method for quality improvement. As the importance of short-term and low-cost product development becomes higher nowadays, conventional response surface designs may need too large number of experimental runs to be accepted. Recently, supersaturated response surface designs were constructed by using several columns of Hadamard matrices (two-level orthogonal arrays), which considerably reduce the number of experimental runs, and a stepwise regression based on -tests was applied to their data analysis. In this paper, a criterion is introduced for constructing supersaturated response surface designs, and a ridge regression is used for the data analysis of supersaturated response surface designs. Several comparisons through numerical simulations are also presented.