One of the difficulties in multi-disciplinary design optimization lies in the complicated interactions between large numbers of objective functions, design variables, and constraints. This difficulty often leads to an unsuitable formulation of design problems. Data mining is often used to address these challenges. Data mining provides insight into the design of complicated systems. The information obtained from data mining can be used to support (a) formulation of design problems, (b) decision making, and (c) design steering. This report presents a review of recent developments and applications of data mining techniques in the engineering design field, and introduces real-world examples of state-of-the-art data mining techniques.

The problem of designing a centrifugal blower was explored using multi-objective genetic algorithm and data mining techniques. Blade-to-blade regions of an impeller and a diffuser were modeled and time-averaged non-uniform inflow to the diffuser was considered. The design objectives were blower efficiency and uniformity of the inflow to the diffuser. The impeller’s shape was represented by NURBS curves and then optimized. The obtained non-dominated solutions showed a trade-off relationship and the design variables controlling the trade-off were found to be related to the dimensions of the vane-less diffuser and the load balance of the impeller. We also applied Decision Tree Analysis and Rough Set Theory to reveal design rules, which led to good performance. Although the design rules derived from the optimization result and data mining results partly agreed with each other, we also clarified that there were some differences due to the characteristics of the data mining methods we used.