^{1,2,a}, Sylvain Dubreuil

^{2}, Jerome Morio

^{2}, Nathalie Bartoli

^{2}and Christian Gogu

^{1}

^{1}Université de Toulouse, UPS, CNRS, INSA, Mines Albi, ISAE, Institut Clément Ader (ICA), 3 rue Caroline Aigle, 31400 Toulouse, France

^{2}ONERA/DTIS, Université de Toulouse, F-31055 Toulouse, France

^{a}morgane.menz@univ-tlse3.fr

Running a reliability analysis on complex models can be very expensive, requiring advanced simulation methods to reduce the overall numerical cost. Kriging-based adaptive sampling methods for reliability analysis such as AK-MCS Echard et al. (2016a) represent a promising way for reducing this computational cost. The learning phase of these methods consists in building a Kriging surrogate model (Gaussian process interpolation) of the performance function G and using the uncertainty structure of Kriging to enrich iteratively this surrogate model. For that purpose a learning criterion has to be defined to estimate the failure limit state. Then, the estimation of the probability of failure is typically obtained by a classification of a set of Monte Carlo samples evaluated on the final surrogate model. Hence, the estimator of the probability of failure holds two different uncertainty sources related to the surrogate model approximation 𝒢_{n} of G and to the Monte Carlo (X_{i})_{i}𝑖 based integration technique.