System reliability analysis of layered soil slopes using fully specified slip surfaces and genetic algorithms

This paper presents a new approach to identify the fully specified representative slip surfaces (RSSs) of layered soil slopes and to compute their system probability of failure, Pf,s. Spencer's method is used to compute the factors of safety of trial slip surfaces, and the first order reliability method (FORM) is employed to efficiently evaluate their reliability. A custom-designed genetic algorithm (GA) is developed to search all the RSSs in only one GA optimization. Taking advantage of the system aspects of the problem, such RSSs are then employed to estimate the reliability of the slope system, and a proposed linearization approach-based on first or second order (FORM or SORM) reliability information about the identified RSSs-is used to efficiently estimate Pf,s. Three typical benchmark-slopes with layered soils are adopted to demonstrate the efficiency, accuracy and robustness of the suggested procedure, and advantages of the proposed method with respect to alternative methods are discussed. Results show that the proposed approach provides reliability estimates that improve previously published results, emphasizing the importance of finding good RSSs-and, especially, good (probabilistic) critical slip surfaces that might be circular or non-circular-to obtain good estimations of the reliability of soil slope systems.