Active-set sequential quadratic programming with variable probabilistic constraint evaluations for optimization problems under non-Gaussian uncertainties

Abstract

Design optimization under random uncertainties are formulated as problems with probabilistic constraints. Calculating these constraints presents a major challenge in the optimization. While most research concentrates on uncertainties that are Gaussian, a great number of uncertainties in the environment are non-Gaussian. In this work, various reliability analyses for non-Gaussian uncertainties within a sequential quadratic programming framework are integrated. An analytical reliability contour (RC) is first constructed in the design space to indicate the minimal distance from the feasible boundary of a design at a desired reliability level. A safe zone contour is then created so as to provide a quick estimate of the RC. At each design iteration reliability analyses of different accuracies are selected based on the level needed, depending on the activity of a constraint. For problems with a large number of constraints and relatively few design variables, such as structural problems, the active set strategies significantly improve efficiency, as demonstrated in the examples.