Probabilistic Analysis of Ground Deformation Induced by Excavation based on Hypoplastic Constitutive Models

Abstract

Empirical model and finite element method are two commonly-used methods for prediction of ground deformation induced by excavation. Compared with the former, the finite element method can not only predict the deformation of different modes, but also predict the distributed deformation of the whole site. However, results of finite element analysis depends on the constitutive model used in the analysis. This paper uses an advanced hypoplastic constitutive model and its improved edition, which considers the small-strain effect of soil, to represent the soil behavior. Uncertainties are unavoidable in excavation engineering, such as those in soil parameters, loads, and models, etc. These uncertainties have profound effects on the prediction of deformation induced by excavation obtained from the finite element analysis. In order to consider the effect of parameter uncertainty on the prediction results, random variables are used to characterize the parameter uncertainty. Direct Monte Carlo simulation (MCS) method was used to incorporate the parameter uncertainty into reliability analysis of the deformation induced by excavation. The computational costs and convergence issues of finite element method in together with advanced constitutive model result in significant computational challenges in MCS-based reliability analysis. In order to improve the computing efficiency and robustness, artificial neural network (ANN) is adopted as a surrogate model of the finite element method to compute the soil deformation for a given set of uncertain parameters. Results show that responses predicted by the improved hypoplastic model fit the real response better.