Graph Neural Network Surrogate for Seismic Reliability Analysis of Highway Bridge Systems

TL;DR

This work tackles the challenge of estimating node-level seismic reliability for highway bridge networks, which is computationally prohibitive with traditional Monte Carlo approaches. It proposes a Graph Neural Network surrogate that predicts node-level connectivity probabilities P_c under seismic scenarios by integrating edge failure probabilities derived from GK15 ground motion predictions and HAZUS-HM fragility curves, via a two-module pipeline. The contributions include first predicting all origin-destination connectivities on road networks with inductive generalization to unseen graphs, achieving substantial speedups over Monte Carlo while maintaining accuracy, and demonstrating robustness and inductive transfer from smaller to larger networks. The approach enables rapid, scalable reliability assessment for transportation systems, supporting preparedness, risk mitigation, and emergency response planning.

Abstract

Rapid reliability assessment of transportation networks can enhance preparedness, risk mitigation, and response management procedures related to these systems. Network reliability analysis commonly considers network-level performance and does not consider the more detailed node-level responses due to computational cost. In this paper, we propose a rapid seismic reliability assessment approach for bridge networks based on graph neural networks, where node-level connectivities, between points of interest and other nodes, are evaluated under probabilistic seismic scenarios. Via numerical experiments on transportation systems in California, we demonstrate the accuracy, computational efficiency, and robustness of the proposed approach compared to the Monte Carlo approach.

Figures (12)

• Figure 1: Message-passing with both node and edge features.
• Figure 2: Pipeline of connectivity analysis of highway bridge system.
• Figure 3: Architecture of graph neural network model for node-level connectivity analysis.
• Figure 4: Map illustration of the studied regions in different levels.
• Figure 5: Prediction of connectivity probabilities between the target node (red circle) and the other nodes in the Level 1 region. The nodes are color-coded based on the predicted connectivity probability (Figure \ref{['fig:level1_reg_pred']}) and the absolute error between predicted connectivity probabilities from GNN and MCS (Figure \ref{['fig:level1_reg_diff']}).