Multi-UAV-aided spatial-temporal traffic load recognition for rapid diagnosis of the long-span bridge and its SHM system

Long-span bridges are widely equipped with structural health monitoring (SHM) systems (SHMS) for real-time health diagnosis. However, existing diagnosis methods typically classify anomalies using only output data, often assuming SHMS data are reliable for bridge assessment or that the bridge is healthy for SHMS evaluation, thus failing to distinguish whether anomalies originate from the structure or the monitoring system. To address this limitation, this study proposes an input-output-based SHM approach for rapid diagnosis of the long-span bridge and its SHMS using multi-unmanned aerial vehicle (UAV)-aided spatial-temporal traffic load recognition. Multiple UAVs capture operational traffic, employing an improved YOLOv11 model for vehicle detection and the ByteTrack algorithm for tracking. Vehicle trajectories from multiple UAV views are stitched via spatial-temporal matching and are correlated to weigh-in-motion records to obtain the full-field spatial-temporal traffic load distributions. These traffic loads are then precisely matched with measured structural responses, establishing a structural verification coefficient for condition assessment. Field application on a long-span cable-stayed bridge demonstrates high effectiveness: vehicle detection achieves a mean Average Precision (mAP) 50 of 98.9% and precision exceeding 97% across all vehicle categories during daytime and nighttime; trajectory tracking achieves mean absolute errors of 1.755 and 0.286 m relative to high-precision real-time kinematic global positioning system data in longitudinal and lateral directions. Traffic-induced static deflections calculated from the spatial-temporal traffic loads closely match the measured deflections in temporal evolution at all monitored locations. Structural verification coefficients range from 0.90 to 0.92 during daytime and 0.76 to 0.82 at nighttime, all within the codified 0.70–1.05 range, confirming that the bridge is in good condition and the SHMS performs reliably. The approach requires only UAVs in addition to the existing SHMS and extends sectional traffic data into full-field spatial-temporal traffic load distributions. By accurately mapping traffic loads to structural responses, it provides a practical input-output-based solution for rapid diagnosis of long-span bridges and their monitoring systems.