Machine learning enhanced bridge vulnerability quantification under rockfall hazards

Rockfall hazards pose a significant threat to bridge safety in mountainous regions. While existing studies often separate geological hazard analysis from structural vulnerability assessments, leading to inaccurate risk evaluations. Despite advancements in rockfall trajectory modeling and structural impact simulations, limited integration of these disciplines hinders the precise quantification of bridge failure risks under dynamic rockfall scenarios. This leads to inaccurate assessments of structural damage characteristics and failure risks. This study proposes a machine learning (ML)-assisted framework for assessing bridge failure risks that integrates geological disaster analysis. First, a high-dimensional joint distribution model of rockfall impact parameters is constructed using rockfall disaster simulation and ensemble ML approach. Next, a surrogate model for evaluating the residual load-bearing capacity of bridges is developed using the XGBoost algorithm. The dataset for model training is derived from the finite element restart analysis method and Table Generative Adversarial Network (TGAN) augmentation. Monte Carlo sampling (MCs) is employed to accurately quantify bridge risk, using the high-dimensional joint distribution of impact parameters and the surrogate model for residual performance evaluation. The joint geological-structural framework enables rapid risk assessments for site-specific slopes and bridge configurations, providing actionable insights for infrastructure resilience.