System reliability analysis of slopes considering spatial variability of soil properties

The paper aims to propose an effective method for evaluating the system reliability of soil slopes based on the multiple stochastic response surface method considering the spatial variability of soil properties. A new approach for identifying the representative slip surfaces is also suggested. The Karhunen-Loève expansion method is employed to discretize the two-dimensional cross-correlated non-Gaussian random fields of soil properties. The factor of safety of each representative slip surface is explicitly expressed as function of input uncertain parameters using the Hermite polynomial chaos expansion. The direct Monte-Carlo simulation is then employed to calculate the system probability of slope failure based on these representative slip surfaces. Two examples of system reliability analysis of slopes with a consideration of spatially varying soil properties are finally presented to demonstrate the validity and capability of the proposed method. The results indicate that the proposed method can provide an effective tool for evaluating the system reliability of slopes considering the spatial variability of soil properties. The proposed approach for determining representative slip surfaces can not only avoid tedious procedures of calculating the correlations between the factors of safety of any two potential slip surfaces, but also effectively evaluate system reliability of slopes at relatively small probability levels. The system probability of slope failure increases with increasing the variability and vertical autocorrelation distance, while decreases as the negative correlation of shear strength parameters becomes strong. In addition, it will lead to overestimating the system probability of slope failure if the spatial variability of soil properties is ignored.