Constructing Improved Supersaturated Designs with Genetic Algorithms

• Activity Number: 101
• Type: Invited
• Date/Time: Monday, August 4, 2003 : 10:30 AM to 12:20 PM
• Sponsor: Section on Quality & Productivity
• Author(s): James R. Simpson*+ and Adnan Bashir
• Companies: Florida State University and Florida State University

Abstract

Often in industrial experimentation there are a large number of factors to be considered even though only a small subset are important. Several methods have been proposed to develop two-level supersaturated designs for screening many factors, but these designs are not necessarily the best in terms of the criteria used for selection. We propose an adaptive screening technique using a genetic algorithm to construct improved supersaturated designs from various originating matrices. The criteria for developing designs is E(s2), a metric commonly used in this area of research, which reflects the associated correlation among design columns. The originating matrices include the Hadamard matrix, a general balanced matrix containing all balanced combinations of the -1's and +1's, and a combined matrix formed from a Hadamard and balanced incomplete block design. The results are compared using the E(s2) with previous published methods. The proposed methods generate designs with smaller (better) E(s2), and often achieve the theoretical lower limit of E(s2).