Kriging based reliability and sensitivity analysis – Application to the stability of an earth dam

Abstract

Abstract This article presents a Kriging-based probabilistic analysis of an earth dam. The dam failure probability with respect to the sliding stability is investigated by considering the influence of various factors: the filter drain length, the full reservoir water level location and the correlation between the input parameters. A procedure which combines the Kriging surrogate model with the Monte Carlo Simulation (MCS), the Global Sensitivity Analysis (GSA) and the First Order Reliability Method (FORM) is proposed. It aims at benefiting from the computational efficiency of a Kriging surrogate model to provide as much as possible results such as the failure probability, the sensitivity index of each input parameter and the design point. Having more useful results in a probabilistic analysis can help engineers to make more rational decisions. The proposed procedure is compared with the direct MCS, GSA and FORM, and shows a good accuracy and efficiency. In addition, two commonly used slope stability analysis methods (strength reduction method (SRM) and limit equilibrium method (LEM)) are compared in a probabilistic framework. The comparison shows that the two methods can lead to similar estimates of the failure probability for most cases, except when the pore water pressure is important for the determination of the critical slip surface. This kind of results can help engineers to judge when LEM is accurate enough and when SRM is required for a probabilistic analysis.