The Practical Study on Response Surface Methodology

• Activity Number: 331
• Type: Topic Contributed
• Date/Time: Wednesday, August 11, 2004 : 10:30 AM to 12:20 PM
• Sponsor: Section on Quality and Productivity
• Author(s): Chosei Kaseda*+
• Companies: Yamatake Corporation
• Keywords: response surface methodology ; black-box modeling ; radial basis function ; thin plate spline ; artificial neural network

Abstract

Response Surface Methodology is one of effective statistical approaches for optimum design using experimental data. But conventional RSM has some practical problems on generating response surface. Generally, polynomial model is applied to experimental data based on Design-of-Experiment. But it sometimes becomes difficult to easily generate the surface when data is not ideal DOE data due to experimental limitations. Also it becomes more difficult in the case of complex nonlinear targets. Neural network is also sometimes used. But it requires model-tuning by trial-and-error. So the conventional RSM does not always improve the efficiency of product design works because of requirement of much labor and time for surface-generation. To solve this problem, we apply multivariate spline based on radial basis function to surface-generation. This RSM can contribute to making more efficient design on practical use. Practical problems of conventional RSM shall be discussed first. Then, as a means of solving those problems, our RSM and developed software shall be introduced. Lastly, we'll show results of applications to pharmaceutical design and electrical device design, etc.