Exploring Automated Concept Generator Output Through Principal Component Analysis

During conceptual design it is desirable to produce many potential solutions. Recently, computational tools have emerged to help designers more fully explore possible solutions. These automated concept generators use knowledge from existing products and the desired functionality of the new design to suggest solutions. While research has shown these tools can increase the variety of solutions developed, they often provide unmanageably large sets of poorly differentiated results. This work proceeds from the hypothesis that automated concept generator output includes many permutations of a relatively few principal solution variants. A method to discover these underlying solution types from the initial concept generator output is proposed. The proposed method employs principal component analysis for variable reduction followed by cluster analysis for classification. The method is applied to the automatically generated solutions of three sample design problems. Preliminary evidence of the utility and efficiency of the proposed method is presented based upon those sample problems. Finally, a method for extending the proposed technique to much larger solution sets is discussed.