2019 年 85 巻 874 号 p. 19-00027
This study proposes a modified response surface single loop (RSSL) method which improves both approximation accuracy of an optimal solution and the computational efficiency from an original RSSL method. The original RSSL method was proposed recently as one of the single loop methods for a reliability-based design optimization (RBDO) problem that evaluates the failure probability of the second-order limit state function directly by using Hermite polynomials without searching for the most probable point (MPP). When a limit state function is not described as the second-order polynomials, the function is approximated around the deterministic optimum point and then the single-loop searching for the RBDO is carried out. Though the method has sufficient accuracy regardless of the approximation point, two optimization runs; the deterministic and the reliability-based optimizations are required. In order to improve the computational efficiency, the concept of the single loop single vector (SLSV) method that approximate the limit state function around the approximated MPP every optimization searching step is introduced. Through numerical examples, it is demonstrated that the proposed method improves the approximation accuracy of the limit state function and also the searching efficiency.