Graph Convolutional Networks Model for Fire Damage Assessment of Suspension Bridge Towers

Ensuring fire resilience in critical infrastructure is a vital component of Structural Health Monitoring (SHM), enabling early detection of vulnerabilities and informed decision-making during extreme events. In the context of bridge fires, Computational Fluid Dynamics (CFD) simulations are widely used to evaluate fire conditions and derive thermal responses for assessing potential structural degradation. A key measure in such analyses is the Adiabatic Surface Temperature (AST), which represents the temperature of a perfect insulator under identical heating conditions to a real surface. AST serves as a critical indicator of heat exposure and allows for precise thermal assessments. This study introduces a novel surrogate modelling framework utilizing Graph Convolutional Networks (GCN) as NDT method for post-fire assessment of suspension bridges. The proposed model restructures the temperature records from FDS simulations into a relational network based on shared AST devices. The trained GCN model uses FDS simulation data to predict thermal responses and potential damage for the subsequent time steps. The proposed framework demonstrated high predictive accuracy, enabling rapid estimation of damage condition across multiple positions on steel tower. By offering a computationally efficient and reliable tool for scenario analysis, this approach advances SHM practices, supporting targeted inspections and enhancing structural resilience under fire conditions.