適応的累積関数近似によるロバスト最適設計における計算効率の改善法

Abstract

This paper proposes a new scheme for saving the computation cost of robust optimal design by using cumulative function approximation. Since robust optimal design seeks the optimal solution considering the variation of functions over distribution region, it requires much more computation cost than nominal optimal design because evaluation of each tentative solution requires iteration of system analysis at a certain number of reference points within a distribution region. The proposed scheme dramatically saves such computational cost by substituting system analysis at respective reference points with a cumulative function approximation that has an ability of gradually improving fidelity of approximation through gradual, adaptive and cumulative addition of sample points. In this paper, the outline of the scheme is discussed, the procedures of mini-max type robust optimal design and Voronoi diagram based cumulative function approximation are reviewed, and their hybridization for the purpose is described. Finally, numerical examples are demonstrated to ascertain its effectiveness.