期刊名称：Journal of Computational Science and Technology
电子版ISSN：1881-6894
出版年度：2009
卷号：3
期号：1
页码：287-302
DOI：10.1299/jcst.3.287
出版社：The Japan Society of Mechanical Engineers
摘要：We propose a new multi-objective parameter design method that uses the combination of the following data mining techniques: analysis of variance, self-organizing map, decision tree analysis, rough set theory, and association rule. This method first aims to improve multiple objective functions simultaneously using as much predominant main effects of different design variables as possible. Then it resolves the remaining conflictions between the objective functions using predominant interaction effects of design variables. The key to realizing this method is the obtaining of various design rules that quantitatively relate levels of design variables to levels of objective functions. Based on comparative studies of data mining techniques, the systematic processes for obtaining these design rules have been clarified, and the points of combining data mining techniques have also been summarized. This method has been applied to a multi-objective robust optimization problem of an industrial fan, and the results show its superior capabilities for controlling parameters to traditional single-objective parameter design methods like the Taguchi method.
关键词：Parameter Design;Multi-Objective Optimization;Taguchi Method;Analysis of Variance;Self-Organizing Map;Decision Tree;Rough Set;Association Rule
