HYDROMECHANICAL SLOPE STABILITY ANALYSIS: MODELLING, MONITORING AND PREDICTION USING BP-FF ARTIFICIAL NEURAL NETWORKS

Abstract

Sloping regions of the earth’s crust have formed part of human settlements from time immemorial. However, these regions are prone to geological hazards especially mass wasting processes such as debris flows and landslips. While literature on soil mass movements is available, precise mathematical models, versatile instrumentation systems, and the utilization of intelligent artificial neural network (ANN) models for forecasting these events in space and time is limited. This study was aimed at undertaking explicit characterization of convex configuration slopes under varying hydrological conditions. Specifically, it involved formulation of numerical models based on spherical-cap-shaped slip zones as well as development of ANN and hydromechanical landslide model. Computational results from the models were calibrated using experimental findings based on a solar powered data acquisition system which comprised of a laboratory flume, sensor array and data broadcasting scheme. Finally, a (Back-Propagation Feed-Forward) BP-FF ANN model was developed for purposes of predicting the slope stability status by way of numerical values of the factor of safety (FS). Results from quantitative analysis indicated that the mode of failure and configuration of the slip zone is a function of the volumetric water content (VWC), location of the apparent phreatic surface, magnitude of cohesive strength, orientation of weak planes and existence of discontinuities. Consequently, progressive translational displacement is the most dominant mechanism of failure and the slip zones assume the shape of single or double spherical caps, depending on the morphology of the potential failure plane (planar or curvilinear), location of the phreatic surface and flaws. Furthermore, results from numerical models demonstrated that geotechnical, geophysical and hydrological parameters and by extension the FS can be defined as empirical functions of the VWC. Additionally, results showed that the amount of VWC at the interfaces between adjacent lithostratigraphic units is the principal trigger of soil mass movements. The calibrated BP-FF ANN model was used to predict the FS values of slopes. In conclusion, since an improved effective wetness index has been derived taking into consideration the moist unit weight, threshold VWC extracted directly from the hydromechanical model and a BP-FF ANN model has been trained, an early warning system can be developed based on this information for purposes of prediction and disaster mitigation. Inferences derived from the study will provide precise constitutive computations as well as baseline geophysical and hydrological information to the public on the stability status of slopes.