ABSTRACT
The factor of safety of a geotechnical structure is often estimated by the ratio of resistance and load. For the slope stability problem, the safety factor of a slope can be calculated by the ratio of shear resistances to shear stresses along a given slip surface using a limit equilibrium method. However, the resistance and load are subjected to uncertainties, making the factor of safety of a slope also uncertain. To apply the same nominal value of safety factor to the conditions that involves widely varying degrees of uncertainties may not be logical. Probabilistic approach and reliability analysis methods provide a rational way to account for the uncertainties from different sources and to estimate the probability of satisfactory performance in a systematic way. Extensive studies have been conducted on the probabilistic assessment of slope stability. The mean first-order reliability method (MFORM) is mainly used in early studies on slope reliability (e.g., Wu and Kraft, 1970; Tang et al., 1976; Vanmarcke, 1977; Li and Lumb, 1987; Liang et al., 1999; Christian et al., 1994). The advanced first-order reliability method (AFORM) is used to study both the location of the most critical slip surface and the reliability index of the most critical slip surfaces (e.g., Hassan and Wolff, 1999; Bhattacharya et al., 2003; Xue and Gavin, 2007). Sampling methods have been used for reliability analysis for a given slip surface (e.g., El-Ramly et al., 2002), or system reliability of slopes considering multiple slip surfaces (e.g., Ching et al., 2009; Zhang et al., 2011, 2013a), and reliability analysis of slopes considering the spatial variability of soil properties (e.g., Griffiths et al., 2009; Santoso et al., 2011; Li et al., 2015). The traditional response surface method (RSM), which can replace the computationally intensive deterministic models with an approximated less expensive surrogate model, has been developed to achieve computational efficiency (Li et al., 2011, 2015; Zhang et al., 2013b). In most of previous reliability studies of slope stability, limit equilibrium methods are generally used as a deterministic model. Complex failure mechanisms due to soil behavior such as progressive failure (Chowdhury et al., 1987; Gilbert and Tang, 1989; Liu et al., 2001) are usually considered by numerical models.
