Updating model parameters and predictions in SEM tunnelling using a surrogate-based Bayesian approach

Abstract

This paper presents a surrogate-based Bayesian approach for updating the ground parameters within an application of the observational method in sequential excavation method (SEM) construction. A three-dimensional (3D) finite-difference model is used in the forward analysis to simulate SEM construction explicitly considering 3D multi-face excavation effects and ground–structure interaction. The polynomial-chaos Kriging (PCK) method was employed to provide a surrogate for the 3D finite-difference model to alleviate the cost of probabilistic analysis. The uncertain geotechnical parameters are updated during SEM construction through a progressive Bayesian updating procedure. Time-series observations of multiple types of measurements are used to form the likelihood function. The posterior distributions of the uncertain parameters are derived from the affine invariant ensemble sampling (AIES) algorithm. The proposed framework is illustrated through application to data from the Regional Connector Transit Corridor (RCTC) crossover cavern project constructed in downtown Los Angeles. The uncertainties of the geotechnical parameters were substantially reduced. The posterior estimations indicate higher elastic modulus and cohesion of the Fernando formation than what was assumed before the construction. The updated predictions of the ground surface, subsurface and structural deformations showed improvement in agreement with the field measurements through the continuous updating process.