Use of random fields in slope stability analysis and application to a case history

Abstract

A numerical investigation is carried out to study the stability of homogeneous slopes accounting for the uncertainty of some material properties such as cohesion and internal friction angle under self-weight loads. For such scope, random fields are generated for these material properties using the Local Average Subdivision (LAS) Method, which is combined with the Finite Element Method within a Monte Carlo simulation to compute failure probabilities. Verification of the employed numerical tool is provided by computing failure probabilities versus safety factors diagrams for the studied examples. For instance, it is shown that for a typical undrained slope, obtained outcomes indicate that a minimum deterministic design safety factor of 1.65, suggesting a stable slope as recommended by the Brazilian regulation NBR11682, could be actually associated to significant failure probabilities larger than 15%, being on the unsafe side. This finding is important and should call the attention of practitioners. Finally, a study case of a real slope located in Brazil with available measured field data was also analysed. The effect of some key parameters such as the influence of correlation length, effect of uncertainty of material properties and cross-correlation between soil strength parameters on the final failure probability were studied. In general, it was shown that situations defined with apparently adequate safety factors can still be associated with considerable failure probabilities. Indeed, this will depend upon the value of employed coefficient of variation and spatial correlation distance assumed for the stochastic properties in question.