Presentations

This paper supposes that the ‘effect sparsity’ assumption, which implies only a few control and noise factors actually affect the response, holds and proposes a method for reducing the number of experimental runs by applying a supersaturated design for RSM to robust parameter design. In this method, a supersaturated design is constructed by using a (8m+4)×(8m+4) Hadamard matrix (m is a natural number) in place of a two-level orthogonal resolution V design in CCD, and a fitted response surface model is estimated using a stepwise variable selection. An illustrative example describes the proposed procedure. Type I and II error rates are evaluated through some numerical experiments. The results suggest that the proposed method may be useful under the ‘effect sparsity’ assumption.