Structural reliability analysis method based on adaptive parallel Kriging model

With the advancement of computer performance and parallel computing resources, conducting multiple performance function evaluations has become increasingly feasible in structural reliability analysis. To address the efficiency limitations of the classical active learning reliability method, which combines adaptive Kriging and Monte Carlo simulation (AK-MCS), a novel adaptive parallel learning framework with two key innovations is proposed: i) A new parallel learning method under the balanced point-adding strategy is developed, combining the U learning function and the model improvement degree function. ii) A convergence criterion based on K-fold cross-validation is introduced. The first innovation identifies multiple valuable samples through a dual screening mechanism, while the second effectively overcomes the conservatism inherent in traditional convergence criteria. The efficiency and accuracy of the proposed method are validated through two numerical examples and a practical engineering fatigue example. The results demonstrate significant improvements over conventional methods.