Intelligent Experimental Design Using an Artificial Neural Network Meta Model and Information Theory

Authors:
Shi-Shang Jang;David Shan-Hill Wong;Junghui Chen
Affiliations:
Chemical Engineering Department, National Tsing-Hua University, Hsin Chu, Taiwan 300;Chemical Engineering Department, National Tsing-Hua University, Hsin Chu, Taiwan 300;Department of Chemical Engineering, Chung-Yuan Christian University, Chung-Li, Taiwan 320
Venue:
Proceedings of the 2006 conference on Integrated Intelligent Systems for Engineering Design
Year:
2006

Citing 10
Cited 0

Empirical model-building and response surface

Empirical model-building and response surface
Convergence theory for fuzzy c-means: counterexamples and repairs

IEEE Transactions on Systems, Man and Cybernetics
Multilayer feedforward networks are universal approximators

Neural Networks
Universal approximation of an unknown mapping and its derivatives using multilayer feedforward networks

Neural Networks
Introduction to statistical pattern recognition (2nd ed.)

Introduction to statistical pattern recognition (2nd ed.)
Introduction to the theory of neural computation

Introduction to the theory of neural computation
Universal alignment probabilities and subset selection for ordinal optimization

Journal of Optimization Theory and Applications
Genetic Algorithms in Search, Optimization and Machine Learning

Genetic Algorithms in Search, Optimization and Machine Learning
Convergence Properties of the Nelder--Mead Simplex Method in Low Dimensions

SIAM Journal on Optimization
Design and Analysis of Experiments

Design and Analysis of Experiments

Quantified Score

Hi-index	0.00

Visualization

Abstract

Ability to rapidly design products and their manufacturing process is a key to being competitive in a dynamic market environment. Traditional methods of design of experiment development are unsatisfactory when applied to design problems with large number of input variables and nonlinear input-output relation. A meta-model driven experimental design scheme is developed. The approach uses artificial neural network as the meta-model, and a combination of random-search, fuzzy classification, and information theory as the design tool. An information free energy index is developed which balances the needs for resolving the uncertainty of the model and the relevance to finding the optimal design. The procedure involves iterative steps of meta-model construction, designing new experiments using meta-model and actual execution of designed experiments. The effectiveness of this approach is benchmarked using a simple optimization problem. Three industrial examples are presented to illustrate its applicability to a variety of design problem.