On the use of Bayesian networks as a meta-modelling approach to analyse uncertainties in slope stability analysis

Abstract

ABSTRACTIn this paper, we report on the use of Bayesian networks, BNs, learnt from data generated by physical and numerical models, to overcome to a certain degree a number of complications in traditional slope stability analyses that jointly consider the mechanical and hydraulic properties of soils. Discrete Bayesian networks resulted to be useful and efficient to acquire knowledge from simulated data and to identify significant factors by the combined use of backward inference and global sensitivity analysis. Further, BNs enable decision thresholds to be estimated quickly. Along with this, backward inference and global sensitivity analysis are performed in BNs at low computation costs. Moreover, under conditions in which knowledge is scarce, we show how a practitioner can be better informed using the proposed approach. All these previously under-reported modelling features in the specialised literature encourage the further application of the proposed approach to enhance slope stability analysis.