Quantifying post-earthquake residual vertical load-carrying capacity (VLCC) of RC bridge bents: Parametric study and development of interpretable machine learning models

Abstract

The post-earthquake residual vertical load-carrying capacity (VLCC) of bridge bents serves as a measure of bridge functionality to carry traffic loads following earthquake events, which is directly associated with the seismic resilience of bridges. To facilitate the next-generation resilience-based seismic design of bridges, this study systematically quantifies the post-earthquake residual VLCC of widely adopted reinforced concrete (RC) single-column and double-column bridge bents. An analytical framework comprising two-phase cyclic pushover analyses followed by pushdown analyses (i.e., pushover in the vertical-downward direction) is applied to evaluate the post-earthquake residual VLCC of bridge bents. An in-depth parametric study is conducted using validated numerical modeling techniques to quantify the VLCC degradation following earthquakes, and meanwhile, to explore how it is affected by bent structural parameters. Additionally, a dataset containing a total number of 860 post-earthquake residual VLCC results is gathered, statistically analyzed, and applied to develop interpretable machine learning (ML) predictive models. It is found that the degradation of VLCC can be attributed to a combination effect of physical damage to RC materials during the seismic loading and post-earthquake residual deformation-associated P-Δ effect during the vertical loading. At the design damage state that half of the inelastic deformation capacity of bents is mobilized, the mean residual VLCC of single-column and double-column bents reaches 89.1 % and 95.7 % of the original level, respectively. The developed ML models can provide reasonable predictions of the post-earthquake residual VLCC for bridge bents with errors mostly within 20 % and provide interpretation results align with the findings from the parametric study and statistical analysis of the dataset.