Response surface method using vector projected sampling points

Abstract

In this study an improved sequential response surface method is proposed. Using the gradient projection method, the sampling points for response surface approximation are selected to be close to the original failure surface. The proposed method also suggests the use of linear type response surface functions in conjunction with Rackwitz-Fiessler algorithm rather than quadratic functions, for which advanced Monte Carlo algorithms are commonly used to evaluate the reliability estimates. Also, a method controlling the selection range of sampling points considering the nonlinearity of the limit states is proposed to reduce the error produced by approximating the original nonlinear limit state with a linear response surface. With various examples including both structural and non-structural problems, the proposed method is found to yield quite accurate results and to be efficient.