A dynamic-switching active-learning Kriging considering epistemic uncertainty for structural reliability analysis

This paper proposes a dynamic-switching active-learning Kriging (DS-AK) method considering epistemic uncertainty in modeling and active learning processes to enhance the efficiency and accuracy of structural reliability analysis. To address the epistemic uncertainty associated with modeling and active learning, the method dynamically selects the optimal Kriging model and learning function from pools of candidates based on the maximum relative error of failure probability and the Kriging believer criterion. A hybrid stopping criterion that considers the stability of the failure probability estimation is also developed. The proposed method is validated through three numerical examples and a practical engineering application, demonstrating its superior performance over some existing methods.