A stochastic slope stability method based on a total stress approach is presented with an application to a well - documented experimental excavation site. Undrained strength values are model led as outcomes of a stochastic process. This permits to reproduce the random nature of measured soil properties and the associated spatial variability structure. Variogram analyses indicate that a Stationary Random Function of the second order yields a satisfactory model for spatial variability. Using this model, virtual undrained strength fields that are conditioned to available field measurements are generated through a sequential gaussian simulation technique. Stability analysis for a given slope geometry is then performed for each generated virtual field of a series. This creates a set of factors of safety (Fs) which are realisations of the Fs random variable (RV). These results are then compared to actual field conditions where slope failures have or not been observed. Results show how statistical parameters such as average factor of safety and failure probability are dependent of the geometry of the slope and of the neighbouring conditioning undrained strength profiles.
The common approach in designing a slope against slide failure is to adopt a set of design parameters and compute a factor of safety (Fs). Selection of a target value for the factor of safety is made in view of the understanding and knowledge of subsoil nature and in particular its variability, problem geometry, and consequences of failure. This design approach is said to be deterministic since it is based on a fixed set of input variables, yielding a unique output for the factor of safety. Although as can be seen, the way design target values are selected is strongly influenced by the stochastic nature of many input variables.
