| Seismic damage of bridges may pose a severe threat to motorway users, and preventive closure until post-seismic inspection may be viewed as the only safe option. However, such an action may incur pronounced losses by obstructing the rescue teams to reach their destination in due time. On the other hand, allowing traffic on severely damaged bridges is a difficult decision with potentially dire consequences. Hence, the main dilemma for the motorway administrator will be whether to interrupt completely or partially the operation of the network, calling for timely development and implementation of a RApid REsponse (RARE) system. Such a RARE system requires accurate and real time forecast of the seismic damage of the motorway structures.;This thesis proposes an inter-disciplinary approach which combines nonlinear finite element (FE) simulations with advanced econometric modeling. Based on established classification schemes, representative bridge classes are selected and examined. Simulating the response of a bridge employing rigorous 3D FE modeling of the bridge--foundation--abutment--soil system is currently feasible, but the computational effort would be quite substantial. Moreover, a large number of nonlinear dynamic time history analyses (of the order of 300) are required to cover a wide range of strong motion characteristics and to generate a statistically significant dataset.;To that end, an established simplified modeling methodology is employed to develop simplified finite element models of the examined bridge classes, accounting for key structural components and soil--structure interaction. Using as input the results of the finite element analyses, the seismic damage of the bridge -- using the maximum and residual drift ratio, and the ratio of ductility demand over ductility capacity as damage indices -- is assessed as a function of statistically significant intensity measures, through the use of non-linear three-stage least squares (3SLS) models.;The proposed 3SLS approach accounts for: (i) both simultaneous equation bias and cross-equation contemporaneous correlation of the errors, caused by shared unobserved effects across the damage indices; (ii) endogeneity among the damage indices with the use of instrumental variables; and (iii) unobserved heterogeneity and panel effects, through the use of fixed effects. The 3SLS approach is counter-imposed to a traditional ordinary least squares regression, and the comparison illustrates the benefits of 3SLS in terms of explanatory power and forecasting accuracy. |
