The influence of aging on seismic bridge vulnerability: A machine learning and probabilistic model integration

Document Type

Article

Publication Date

12-1-2025

Abstract

This study investigates the seismic vulnerability of multi-span simply supported concrete girder bridges in seismic-prone regions leveraging machine learning (ML), with focus on the interaction effects of aging and deterioration among bridge components. While prior research has explored aging effects using ML techniques, this study emphasizes the synergistic interactions between column and bearing deterioration (e.g., rebar corrosion and oxidation-induced stiffness loss in bearing pads) and their impact on seismic performance. These degradation processes incorporate physical deterioration parameters (e.g., reduced rebar diameters, modified bearing stiffness, dowel strength degradation) as input features in ML-based probabilistic seismic demand models. These features serve as proxies for aging conditions and allow the ML models to predict seismic demands conditional on specific deterioration states, capturing how degradation influences both individual component performance and system-level vulnerability. The methodology integrates probabilistic seismic analysis (PSDA) with diverse ML regression techniques, spanning four algorithm classes: linear, polynomial, decision trees/ensemble, and kernel-based methods, to predict seismic responses of deteriorated components and quantify interaction-driven vulnerability amplification. Among these models, Lasso regression emerged as the top performer, offering an optimal bias-variance trade-off and effective variable selection for structured deterioration-response mapping. The ML-based models demonstrate high accuracy in predicting seismic responses of bridge components across varying seismic intensities, outperforming traditional nonlinear time history analysis in computational efficiency. Sensitivity analysis revealed that column deterioration—particularly rebar corrosion—has a disproportionately large impact on drift ratios compared to aging effects in bearing components, emphasizing the need for targeted maintenance strategies focused on columns. Moreover, results indicate that while new bearing pads improve performance, aging column rebars may still contribute to increased bearing displacements under extreme seismic events. Additionally, aging bearing pads affect active abutment performance even with new column rebars. By accounting for interactive aging effects, the ML framework quantifies vulnerability amplification from coupled deterioration states. This study establishes a refined analytical framework for evaluating interdependent aging mechanisms in bridge infrastructure. The actionable insights can guide planners and engineers in prioritizing maintenance, retrofitting, and replacement strategies to address cross-component vulnerabilities. By leveraging interaction-aware models, this study contributes to managing long-term risks in aging transportation networks while enhancing seismic resilience.

Publication Source (Journal or Book title)

Structures

This document is currently not available here.

Share

COinS