An efficient focal element classification strategy for evidence theory-based reliability analysis using Kriging model

Abstract

To address the high computational cost of joint focal element (JFE) classification in evidence theory under the surrogate model framework, this paper proposes an efficient classification method that improves the efficiency of JFE classification through a three-level collaborative strategy: a sparse grid strategy is adopted to construct discrete points; drastically reducing the computation for JFE category judgment; a focal element freezing mechanism is introduced to lock stably classified JFEs, thus eliminating redundant computations; and a dual-point active learning method is employed to accelerate iterative convergence. Verification results from three numerical examples with different dimensions show that this method can effectively reduce the computational burden at the surrogate-model level, and maintain high classification accuracy without increasing the number of true performance function evaluations.