ABSTRACT
The first applications of Kriging to reliability are rather recent (Romero et al., 2004, Kaymaz 2005). At first, Kriging was not performed in fully active learning methods. The methods were in several iterations however Kriging variance was not used to determine the next most interesting point to evaluate. Furthermore, Kriging is used as an approximation metamodel. Following this, Bichon et al. (2008) propose a first active learning method inspired by EGO and the Kriging contour estimation method by Ranjan et al. (2008). The method is found to be extremely efficient as it enables to focus on the evaluation of points in the vicinity of the limit state. However, it still uses Kriging as an approximation metamodel. Indeed, the limit state is approximated in the whole design space and therefore, even in regions where configurations show very weak densities of probabilities and have negligible effects on the probability of failure.
