Transportation infrastructure requires continuous assessment due to performance degradation caused by aging and adverse operational and environmental conditions. AI-based approaches support both damage detection and life-cycle assessment by identifying anomalies in structural responses, such as stiffness loss or cracking, enabling early warning and conditionbased maintenance. For life-cycle assessment, while traditional methods rely on costly strain measurements to estimate remaining useful life, AI leverages virtual sensors to infer strain from easily measured acceleration data, offering a practical alternative for structural health monitoring. While purely data-driven models have shown promising results, integrating physics-based information has gained attention for improving both prediction accuracy and model interpretability. Based on a review of advancements in fields where Graph Neural Networks (GNNs) are well established, the authors outline a road map for damage detection and life-cycle assessment. They also propose a graph-based approach for damage detection and life-cycle assessment incorporating physics information. This approach forms a core component of a digital twin, where the GNN represents the structure through nodes (sensor locations) and edges (structural elements). The hybrid model continuously updates itself using physics-based features embedded in the adjacency matrix and is trained on sensor data. This framework demonstrates the potential for adaptive, physics-informed digital twins capable of predictive life-cycle assessment of bridge structures.
