Bayesian Networks for post-earthquake assessment of bridges

We propose a probabilistic framework to estimate the condition state of a bridge stock in a post-earthquake situation based on the knowledge of the state of individual bridges, appraised through visual inspection. The capacity of each bridge is modeled using Hazus' fragility curve approach, while the geographical distribution of the seismic demand is calculated using an attenuation function. Uncertainty terms are considered both in capacity and demand, and the logical correlation between capacity and demand inherent to different bridges is handled through Bayesian Networks. In this paper this framework is developed in the simplest case of two bridges. The correlation between two bridges is first analyzed, and their prior probabilities of being in a damage state are calculated during the initialization procedure. When an earthquake occurs, if data such as magnitude and epicenter are known, or the damage state of one of the bridges is observed, then the probability of the companion bridge being in a damage state can be predicted and updated. Two bridges of similar characteristics located in the Italian Trentino region, are used to illustrate this approach.