Application of conditional random fields for the reliability assessment of slope stability
Abstract
This study employs conditional random fields to integrate different sources of information on soil properties in the reliability assessment of slope stability in Rissa, Norway. The reliability assessment is conducted to quantify the effects of uncertainties in soil properties and the simulation model on the safety assessment of the slope. The uncertainties in soil properties are modeled with normal and lognormal random fields. The random fields are conditioned on the values of soil properties interpreted from Cone Penetration Test (CPT) measurements at several locations along the slope. The interpreted values of soil properties are subjected to uncertainty due to the inherent variability of CPT measurements, measurement errors, and transformation uncertainty. The application of conditional random fields enables a consistent treatment and integration of different sources of information on soil properties with varying degree of certainty. The reliability assessment of the slope stability is conducted within the Monte Carlo framework by propagating the uncertainties to the slope response. The response of the slope is simulated by a 2D finite element model with an error term accounting for the uncertainty of the simulation model.