Abstract
This study presents a comprehensive probabilistic evaluation of seismic response prediction methods for horizontally curved reinforced concrete (RC) bridges under bidirectional earthquake excitations with varying incident angles. A total of 14 bridge models-comprising both straight and curved configurations with different abutment conditions-were subjected to over 4,000 nonlinear time history analyses using 22 far-field ground motion records rotated across 13 angles from 0° to 180°. Seismic responses, including column drifts and abutment displacements, were assessed in global, local, and vectorial directions. A definition for "real" responses, representing resultant displacements independent of orientation, was proposed to capture maximum structural demands. The influence of horizontal curvature, abutment boundary conditions, and ground motion directionality on seismic performance was systematically examined. Results show that neglecting incident angle variability leads to underestimation of displacement demands by up to 25%. The study further evaluates the reliability of conventional combination rules-100/30, 100/40, and SRSS-in predicting maximum seismic responses. A new probabilistic framework was adopted to evaluate the likelihood of exceedance associated with the predicted structural responses under each combination rule. Findings indicate that while the 100/30 rule may be suitable for straight bridges, it underperforms for highly curved configurations. The SRSS rule consistently offers more accurate estimates, particularly when real responses are used as the benchmark. The study highlights critical limitations in existing design practices and provides targeted recommendations for selecting appropriate combination rules based on bridge geometry and abutment type, contributing to more reliable seismic design and assessment of irregular RC bridges.