Structural time-variant reliability analysis by single-loop active learning Kriging

Time-variant reliability analysis (TRA) is crucial in assessing the safety of engineering structures subjected to stochastic and time-variant loads. However, existing methods often suffer from excessive computational cost when applied to complex engineering problems, especially when involving high-fidelity models. This study presents a new method, term as single-loop active learning Kriging combined with error-based convergence condition, for efficient TRA. In the proposed method, an estimation error measure of time-variant failure probability is proposed, which forms the foundation for designing an efficient criterion to end the active learning procedure. Meanwhile, a learning function is designed to choose new point to update the Kriging model. Comparative studies on mathematical benchmarks and engineering black-box cases indicate that the proposed approach achieves markedly higher computational efficiency while maintaining comparable accuracy, showing strong potential for practical applications in complex time-variant reliability problems.