Nonparametric Approaches to Response Surface Methodology

• Activity Number: 125
• Type: Topic Contributed
• Date/Time: Monday, August 8, 2005 : 10:30 AM to 12:20 PM
• Sponsor: Section on Quality and Productivity
• Author(s): Stephanie Pickle*+ and Jeffrey B. Birch and Timothy J. Robinson
• Companies: Virginia Polytechnic Institute and State University and Virginia Polytechnic Institute and State University and University of Wyoming
• Keywords: Nonparametric Regression ; Model Rubust Regression ; Genetic Algorithm ; Response Surface Methodology

Abstract

Industrial statisticians, engineers, and other researchers often employ the techniques of response surface methodology (RSM), a sequential experimental strategy originally proposed by Box and Wilson (1951). Historically, RSM involves running a series of small experiments and modeling the data parametrically to find the operating conditions for the design variables that will optimize the response(s). In many industrial settings, however, parametric models may not adequately represent the true relationships between the variables. For this reason, Vining and Bohn (1998) first propose the use of nonparametric smoothing in an RSM setting. While their work is innovative, several improvements and extensions can be made to it. We propose the use of methods that extend the ideas of classic RSM to include new advances in regression and optimization, such as local polynomial regression, model robust regression, and genetic algorithms. These proposed methods will offer greater flexibility, robustness, and efficiency. Furthermore, they may provide a better understanding of the process being studied as well as superior optimization solutions.