Identification of Dispersion Effects in Robust Parameter Design Experiments

• Activity Number: 420
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Quality and Productivity Section
• Author(s): James L Rosenberger and Kwame Adu Kankam*+
• Companies: Penn State University and Penn State University
• Keywords: Response surface ; variance effects

Abstract

When performing industrial experiments, an important goal is identifying factors that affect the location and the dispersion of a quality characteristic. Taguchi popularized a methodology called Robust Parameter Design (RPD). We focus on modeling dispersion and identifying which factors affect variability. Two definitions of dispersion effects persist. Firstly, a factor has a dispersion effect if it interacts with a noise variable. This approach requires determining which variables are noise variables and including them in the experiment. The second approach determines that a variable has a dispersion effect if it has a significant coefficient term in the variance model. We compared the two approaches theoretically and evaluate their performance with a simulation study. RPD can be studied within a Response Surface Model (RSM) framework and also within the Generalized Linear model (GLIM) framework. We apply penalized model selection techniques to these approaches with the aim of obtaining the most parsimonious model.