Damage Localization and Severity Assessment of a Cable-Stayed Bridge Using a Message Passing Neural Network

Area of Science:

• Structural Engineering
• Artificial Intelligence
• Civil Infrastructure

Background:

• Cable-stayed bridges are vulnerable to damage from various factors, including natural disasters and operational loads.
• Damage to essential components like stayed cables can compromise the entire bridge structure.
• Accurate and timely condition assessment of bridge cables is crucial for safety and maintenance.

Purpose of the Study:

• To develop a deep learning model for identifying damaged cables in cable-stayed bridges.
• To estimate the cross-sectional area of damaged cables.
• To provide a technology-based evaluation strategy for bridge cable health monitoring.

Main Methods:

• Simulated tension data from reduced area cables using the Practical Advanced Analysis Program (PAAP).
• Graph representation of sensor data, mapping tension to vertices and sensor relationships to edges.
• Application of a Graph Neural Network (GNN), specifically the Message Passing Neural Network (MPNN) framework.
• Multi-task learning approach for efficient training and optimization.

Main Results:

• The proposed GNN model accurately locates damaged cables.
• The model effectively estimates the cross-sectional area of damaged cables.
• High performance was demonstrated using simulated cable-stayed bridge data.

Conclusions:

• The developed deep learning model offers a robust solution for damaged cable detection and area estimation in cable-stayed bridges.
• Graph Neural Networks provide an effective method for analyzing complex sensor data in structural health monitoring.
• This technology-based strategy enhances the safety and maintenance protocols for critical bridge infrastructure.