Bayesian inference for predicting the long-term deflection of prestressed concrete bridges by on-site measurements

Abstract

Predicting the long-term deflection of prestressed concrete bridges (PSCB) by the deterministic models is subject to significant epistemic uncertainties, and the Bayesian statistics provide a practical way to make more reliable prediction by use of the long-term measurements. In this paper, a comprehensive Bayesian inference frame was built, in which the model parameters associated with creep, shrinkage, dead load and prestress level were updated by the Bayesian inference of the deflection measurements. A 3-D time-dependent finite element model was constructed, based on which a surrogate model was built to enhance computation efficiency. The Markov Chain Monte Carlo (MCMC) algorithm was adopted to approximate the manifold of the posterior distribution. In the case study, two sets of deflection of the case bridge were utilized to perform the Bayesian inference in a sequential manner. The results revealed that the updated models could produced better fit to the measurements than the deterministic model, while the variance of the prediction was significantly reduced. The updated model revealed changing trends of the time-dependent deflection, which characterizes a later-age acceleration from the initial moderate trend. Based on the updated models, the time-dependent reliability associated with the limit states defined by deflection limits was computed. For such a task, the MCMC-based pre-sampling and the quasi-optimal importance sampling were adopted to reduce the variance of the computation.