This research proposes a novel hybrid reliability analysis method for rare failure events, which integrates the coupled Adaptive Kriging model and Generalized Subset Simulation (AK-GSS). In the proposed method, the adaptive Kriging model is applied to approximate the actual Performance Function (PF) to reduce the number of PF calls. A newly updated strategy is proposed to look for samples on the limit state surface to achieve active learning of the Kriging model. This updated strategy avoids the limitations of most current learning functions based on the prediction variance of Kriging models. The advantages of AK-GSS are illustrated through five examples, including two engineering applications of aircraft wings and hydraulic turbine rotor brackets. The results show that the proposed method is more efficient and accurate for rare failure events.
