Data mining of non-dominated solutions using proper orthogonal decomposition

A new approach to extract useful design information from non-dominated solutions of real-world multiobjective optimization problems is proposed. The proposed approach enables an analysis of line, face, or volume data that Pareto-optimal solutions have such as flow field and stress distribution by decomposing the data into principal modes using proper orthogonal decomposition. Analysis of the shape and surface pressure data of the non-dominated solutions of an aerodynamic transonic airfoil shape optimization problem shows capability of the proposed approach for design knowledge extraction for real-world design optimization problems.