UNCERTAINTY OF EMPIRICAL PREDICTION MODEL FOR WALL DEFLECTION OF DEEP EXCAVATION IN SHANGHAI SOILS

Empirical and semiempirical methods are simple models for estimating the maximum wall deflection induced by an excavation by practicing engineers for preliminary design. Various factors, such as excavation geometry, wall stiffness, strut spacing, ground condition, dewatering, etc, may affect deformation behavior of an excavation. It is impossible and not practical to incorporate all these factors in a prediction model for excavation-induced wall deflection. Hence, the prediction model of wall deflection is subject to model uncertainty, which is necessary to be quantified. In this paper, a database of 25 well-documented case histories of braced excavations in Shanghai is established. The model uncertainties of two semiempirical models for wall deflection, i.e., the KJHH model (Kung et al. 2007) and the C&O method (Clough and O'Rourke 1990) are quantified using the Bayesian updating approach. A model bias factor is defined as the ratio of the observed maximum wall deflection over the estimated value by the prediction model. With the information of the case histories, the uncertainty of the model bias factor is reduced. It is found that the posterior mean of the bias factor of the KJHH model is closer to 1.0 than that of C&O method and the uncertainty of the KJHH model is smaller than that of C&O method.