Traffic Load Estimation via Time-Frequency Analysis of Dynamic Monitoring Data and Deep Learning

The growing variability of traffic loads over the lifespan of bridges presents significant challenges for structural integrity and safety. Traditional traffic load estimation relies on weigh-in-motion (WIM) systems, which are costly, non-scalable, and difficult to integrate directly into bridge structures. To address these limitations, this study proposes a data-driven approach leveraging existing Structural Health Monitoring (SHM) systems. The methodology is demonstrated on a reinforced concrete girder bridge equipped with WIM sensors 600 m upstream, providing vehicle weight data, and MEMS accelerometers mounted along the bridge to capture vibrations at 100 Hz. The proposed solution involves two key steps: a vehicle detection algorithm optimized for single-sensor data, achieving robust detection even in heavy traffic, and a deep learning-based Temporal Convolutional Network (TCN) for vehicle weight classification using time-frequency data. A dual strategy aligns WIM and accelerometer data for model training and evaluates classifier performance under weak-label conditions. Results show high accuracy in vehicle detection and weight classification. This method enhances traffic load monitoring and structural health assessment, offering scalable, automated solutions for infrastructure management across bridge networks.