AK–TSAGL: A two-stage hybrid algorithm combining global exploration and local exploitation based on active learning for structural reliability analysis

The active learning Kriging methodology is widely employed in structural reliability analyses, but there are disadvantages in existing studies concerning computational efficiency and prediction accuracy. In response, a two-stage hybrid algorithm, AK-TSAGL, is proposed in this paper by combining global exploration and local exploitation for enhanced structural reliability analysis. Initially, it identifies the global optimal point by considering predicted variance nearing the limit state surface and candidate point distribution characteristics. Then, it defines local regions around the global optimal point for further analysis. High-quality learning points are identified using a synthetic local evaluation indicator to update the Kriging model. A termination criterion based on maximum relative error ensures convergence with shorter time and higher prediction accuracy. Results from four numerical examples and comparisons with conventional methods demonstrate that AK-TSAGL maintains prediction accuracy and conducts efficient reliability analysis with fewer sample points, offering dynamic accuracy adjustment through various threshold settings. Practical application cases to the reliability analysis of a slider-crank mechanism and a bridge truss structure show its advantages in engineering with more performance factors. This study contributes to advancing Kriging modeling techniques for structural reliability analysis.