Model updating using hierarchical Bayesian strategy and error scale factor employing B-WIM calibration data

Resumo

Data collected during the calibration process of bridge weigh-in-motion (BWIM) systems can be applied
both for evaluating the behavior of the structure and for updating models and parameters. These model update
techniques aim to adjust parameters of a structural model making predicted responses closer to the experimental
behavior. Bayesian modeling is well applied to the present problem, as it makes possible the combination of
previous knowledge and experimental data, allowing better parameter estimates. However, in some civil
engineering applications the updated parameters may contain inherent variability during the experimental process,
due to external factors such as environmental conditions, and may have considerable changes during the process.
To consider this inherent variability, a hierarchical Bayesian model was adopted. Sampling from Markov Chain
Monte Carlo (MCMC) methods is applied. It was also observed that there is an increase in variability with
increasing vehicle weight. The introduction of this effect to the model was then studied, comparing 2 ways of
considering this variation, both as a linear function of the expected signals for a given vehicle, and using the area
under this predicted signal. Results for both numerical simulations and real bridge calibration data indicate that
the hierarchical Bayesian approach proposed for the model update, including the scale factor according to vehicle
weight, is able to perform properly, providing confidence intervals for predicted signals by unseen vehicles that
best fit within the observed strains.