RELIABILITY ASSESSMENT OF AN EXISTING STEEL PLATE GIRDER BRIDGE USING POSTERIOR DISTRIBUTIONS OF MODEL PARAMETERS

<p> This study shows the applicability of using the posterior distributions of model parameters on the structural reliability analysis of existing bridges. The target here was a steel plate-girder bridge that showed corrosions and deterioration in its girders, bearings, and concrete slab. The mean and variance of resonant frequencies were acquired through multiple measurements of traffic vibration for use as observation data in the Bayesian inference. A nominal finite element (FE) model of the target bridge was then constructed. The prior distributions of model parameters that represented the uncertainties due to the deteriorations were then determined based on the inspection data, and the parameters that had sensitivities to resonant frequencies were extracted by the analysis of variance (ANOVA) to prevent ill-posedness. The posterior distributions of the spring stiffness at the ends of steel girders, which were given to represent performances of the corroded bearings in the FE model and the density of the concrete slab, were estimated by applying surrogate modeling and the Markov Chain Monte Carlo (MCMC). It was then recognized that the uncertainties of those model parameters could be reduced, and those posterior distributions could explain the actual deteriorations. The posterior distributions were used to calculate the structural reliability index β of the target bridge. The probability density functions of maximum Mises-stress on the steel girders under the design load were first obtained by the Monte Carlo FE model calculations using the prior and posterior distributions, respectively. In the results, the reliability index against the variability of steel yield stress in the case of posterior distributions became higher than that in the case of the prior. It was then concluded that the procedure presented in this study could provide an understanding of the structural reliability of existing bridges quantitatively.</p>