Active learning-based optimization of structures under stochastic excitations with first-passage probability constraints

Abstract

In various efforts to manage the risk of structural failures caused by stochastic excitations such as wind and earthquake loads, the first-passage probability often needs to be calculated as a reliability constraint. Estimating the first-passage probability with low computational costs is essential, especially in structural optimization requiring iterative reliability calculations for design alternatives. This paper presents a new active learning-based framework for reliability-based design optimization (RBDO) of structures under stochastic excitations. A mixture-distribution-based formulation of the first-passage probability is utilized to handle the high-dimensional sequences of stochastic excitations during the optimization. The design parameter sensitivity of the first-passage probability is introduced to use a gradient-based optimizer in the RBDO iterations. These procedures employ heteroscedastic Gaussian process-based surrogates of the logarithmic responses. An active-learning scheme identifies the best training point to reduce the computational costs in the first-passage probability calculations and optimization. The numerical examples dealing with the optimal design of an eight-story building system and a tower structure subjected to stochastic wind loads demonstrate the accuracy and efficiency of the proposed method.