Adaptive Kriging-assisted system reliability method for implicit limit state surfaces and its application in landslide runout risk assessment

It is still a challenging task to implement efficient methods for reliability analysis, especially for complex engineering systems that have implicit limit state surfaces and multiple failure modes. In this paper, an adaptive Kriging-assisted system reliability method is proposed to solve system reliability assessment problems. The proposed method uses an adaptive Kriging-based trust region to obtain the design point of each component failure mode based on the first-order reliability method in the first stage. By adding barrier functions to the potential limit state functions (LSFs) around the candidate design point, a new limit state function is obtained and used for the next failure mode identification in the second stage. After identifying all the potential failure modes, the system reliability assessment can be initially approximated by direct integration or binomial bounds theory. The accuracy of approximating failure probability can be enhanced by importance sampling or active learning in the third stage. Two case studies with quantitative landslide risk assessment are used to show the efficiency and accuracy of the proposed method. With the help of the smoothed-particle hydrodynamics method (SPH), the large deformation behaviours which may induce a strong nonlinear limit state surface can be used as the indicators of the failure consequences. The results show that the method proposed in this paper is capable of solving high non-linearity implicit LSFs. It also exhibits potential in handling problems with small failure probabilities.