Bayesian updating methodology for probabilistic model of bridge traffic loads using in-service data of traffic environment

Abstract

The traffic environment of a bridge generally varies over its lifetime and can be affected by unexpected changes in the surroundings such as the construction of new roads. Therefore, for accurate estimation of traffic loads, changes in the traffic environment need to be continuously monitored and incorporated into traffic load predictions. To this end, this study first further develops the comprehensive probabilistic model of bridge traffic loads by introducing micro-simulation models to describe accurately congestion state. Next, a Bayesian methodology is proposed to update the parameters of the distributions in the probabilistic model of bridge traffic loads based on in-service data representing the traffic environment. Three Bayesian inference methods are used: conjugate prior distributions, Bayesian linear regression, and Gibbs sampling. Hyper-parameters of the prior model are set up appropriately based on the measurement accuracy and the degrees of belief in the prior model. The proposed Bayesian updating methodology is demonstrated by numerical examples with various scenarios of traffic environment changes and in-service weigh-in-motion (WIM) data measured on a real bridge. The results confirm that the proposed methodology can successfully incorporate changes of the traffic environment into the estimation of traffic load effects on bridges in operation.