Probabilistic Seismic Assessment of RC Bridges: Part I — Uncertainty Models

Abstract The seismic response of structures depends on a large number of aleatory and epistemic uncertainties surrounding the estimation of the structural demand and capacity, both usually featuring considerable dispersion levels, particularly when reinforced concrete structures are being assessed. When bridges are considered, the complexity level increases, given that most of those behave irregularly in the transverse direction. Several procedures may be used for the assessment of the seismic safety of bridges, namely the ones used to estimate the demand, investigated in a companion paper, ranging from linear or nonlinear static procedures to more accurate ones, based on nonlinear dynamic analysis. This work makes use of the latter, commonly seen as more accurate, to compute the failure probability of existing bridges using a relatively simple framework. Different variables typically considered in a seismic assessment procedure (geometry, material properties, earthquake records, intensity level) are statistically characterised, enabling a global simulation process, where each iteration step is associated to an independent structural nonlinear dynamic analysis. Failure probability is then obtained through the probabilistic analysis of a safety indicator, defined as the difference between capacity and demand. An alternative uncertainty model, given by the convolution between the capacity and demand distributions, obtained independently, is also applied. A case study of seven bridge configurations, with different (ir)regularity levels, is considered together with a relatively large set of real earthquake records. The simulation process is carried out using the Latin Hypercube sampling algorithm, expected to considerably reduce the number of realisations with no reliability loss. Conclusions have allowed the identification of vulnerable configurations and shown the importance of the variable detail level when considering different uncertainty models.