This paper introduces an approach that utilizes field measurements to update the parameters characterizing spatial variability of soil properties and model bias, leading to refined predictions for subsequent construction stages. It incorporates random field simulations and a surrogate modeling technique into the Bayesian updating framework, while the spatial and stage-dependent correlations of model bias can also be considered. The approach is illustrated using two cases of multi-stage braced excavations, one being a hypothetical scenario and the other from a case study in Hong Kong. Making use of all the deflection measurements along an inclinometer, the principal components of the random field and model bias factors can be updated efficiently as the instrumentation data become available. These various sources of uncertainty do not only cause discrepancies between prior predictions and actual performance, but can also lead to response mechanisms that cannot be captured by deterministic approaches, such as distortion of the wall along the longitudinal direction of the excavation. The proposed approach addresses these issues in an efficient manner, producing prediction intervals that reasonably encapsulate the response uncertainty as shown in the two cases. The capability to continuously refine the response estimates and prediction intervals can help support the decision-making process as the construction progresses.
Bayesian updating of subsurface spatial variability for improved prediction of braced excavation response
