SEQUENTIAL ESTIMATION OF POSTERIOR DISTRIBUTIONS OF MODEL PARAMETERS OF BRIDGE BEARNG UNDER CORROSION PROCESS

Abstract

The validated modeling of existing bridges will take important roles; however, there are a lot of uncertainties in the properties of existing structures due to deteriorations, seismic loading histories, and the other factors during their operations. Authors have been working on the model V&V of existing structures that is the systematic methodology to deal with those uncertainties in the modeling process. The V&V basically consists of (1) verification process of the numerical model, (2) data acquisition, (3) difinition and sensitivity analysis of prior uncertainties, (4) estimation of posterior distributions by the Bayesian inference with the Marcov Chain Monte Carlo (MCMC), and (5) discussion of quantified parameter uncertainties for obtaining a validated model. In this paper, we especially focused on the time-variant uncertainties due to the detoriaration process; the corrosion process in the bridge bearings that was one of the most significant issues in the maintenance of existing bridges. The experimental verification was conducted by using a beam structure with the steel bearings. Here, the corrosion was promoted in one of two bearings by squirting salt solution, and the strain data were acquired by the loading tests several times during the corrosion process. The Phenomenon Identification and Ranking Table (PIRT) was then applied to consider uncertainties and sensitivities of the FE model parameters. It was then shown that the the systematic decision-makings for updating uncertainties was realized by using the PIRTs in each corrosion process. The posterior distributions indicated that: as the corrosion was promoted, the range of high probability was moved to high coefficient of friction value only in the corroded bearing. It was then concluded that the uncertainties of model parameters that were influenced by the corrosion process could be systematically and appropriately quantified by the V&V process.