Probabilistic analysis of compacted embankments using kriging surrogates

Abstract

The natural variability of material properties in both soil deposits and earth structures is often disregarded in geotechnical design. To address this gap, this study proposes a probabilistic approach for evaluating the impact of soil variability on embankment response, utilizing a probabilistic finite difference model that employs kriging surrogates with randomly varying values of preconsolidation stress, porosity, and suction. To enhance computational efficiency, an intelligent surrogate model was developed. The study uses a coupled hydro-mechanical analysis to calculate crest settlement at various stages of rainfall. The anisotropic Barcelona basic model characterizes soil mechanical behavior, while van Genuchten relationships describe water retention and permeability. By using unsaturated soil mechanics and a probabilistic approach, the study consistently analyzes rainfall-induced displacements. The results of the study indicate that displacements can be vastly underestimated if the variability of parameters and partial saturation are not taken into account at the same time. The variability of settlements and their statistical characteristics during rainfall are dependent on soil variability statistics. The intelligent surrogate model accurately predicts embankment failure under infiltration, with superior computational efficiency.