We applied data mining techniques (Decision Tree Analysis and Rough Set Theory) to an optimal design database in order to extract useful design rules. The database was created using data obtained by solving of a four-objective aerodynamic optimization problem for a centrifugal fan, in which three types of aerodynamic efficiencies and the turbulence noise level were optimized. The extracted rules are expressed in "if…, then…" form and can be used by designers to make decisions. We demonstrated that these data mining techniques can be used to select important design variables applicable to various design objectives. While Decision Tree Analysis derives a single statistical rule, to which the most sensitive design variables are related, Rough Set Theory derives multiple rules, with which even non-linear interactions among design variables can be taken into account. We confirmed that using several data mining techniques is necessary for correct understanding of the design problem.
