A KNOWLEDGE-BASED SYSTEM FOR NUMERICAL DESIGN OF EXPERIMENTS
DS 84: Proceedings of the DESIGN 2016 14th International Design Conference
Year: 2016
Editor: Marjanovic Dorian, Storga Mario, Pavkovic Neven, Bojcetic Nenad, Skec Stanko
Author: Blondet, G.; Le Duigou, J.; Boudaoud, N.
Series: DESIGN
Section: DESIGN INFORMATION AND KNOWLEDGE
Page(s): 1997-2006
Abstract
Numerical Designs of Experiments (DoE) can be used in a simulation process for optimization or metamodelling. A DoE may be costly, and methods are used to reduce its computational cost, as adaptive DoE. They are efficient but complex to be configured and controlled. The time saved by using these methods may be lost for the configuration step. A knowledge based-system is proposed to capitalize and reuse each DoE process configuration. An inference methodology, combining bayesian network and artificial neural network, is proposed. This system proposes improved configurations to the designer.
Keywords: design of experiments, knowledge-based system, simulation-based design, bayesian network, artificial neural network