Data-driven probabilistic curvature capacity modeling of circular RC columns facilitating seismic fragility analyses of highway bridges

Abstract

The availability of reliable probabilistic capacity models of reinforced concrete (RC) columns is a cornerstone for high-confidence seismic fragility and risk analyses of highway bridges. Existing studies often perform physics-based pushover or moment–curvature analyses for the capacity modeling of RC columns, which may encounter nonconvergent problems under high levels of nonlinearities in structural material constitutive models and elements, and become computationally inefficient especially when the analysis model contains plenty of cases involving multisource uncertainties. To mitigate the nonconvergent issues as well as release the computational burden of RC column capacity estimates, this study explores the potency of artificial neural network for data-driven probabilistic curvature capacity modeling of circular RC columns, which can facilitate seismic fragility assessment of highway bridges. To this end, a large database is developed by fiber-section-based moment–curvature analyses covering major ranges of concrete and steel strengths, reinforcement ratios, vertical loads, and geometries of RC columns in engineering practices. To obtain an accurate data-driven model, a fivefold cross-validation training and test process is performed to optimize the neural network architecture. The optimized neural network leads to a reliable data-driven model for estimating multilevel curvature capacity indices with percentage errors less than 15%. Finally, a typical highway bridge is taken as a case study to demonstrate the applicability of the developed data-driven capacity model for the expediency of seismic fragility analysis. For ease of implementation, the database and associated codes are available at https://bit.ly/3A1dh1V.