Bridge health monitoring via drive-by sensing: A novel CNN-based framework with full-scale field validation

This study introduces a pioneering approach to bridge Structural Health Monitoring (SHM) that integrates drive-by sensing with 1D Convolutional Neural Networks (CNNs). In one of the first known field tests that combined drive-by sensing and controlled damage of an in-service bridge, a full-scale reinforced concrete (RC) bridge in Nebraska was subjected to a series of progressive damage scenarios, with each subsequent damage state building upon the previous structural damage: healthy state (D0), saw-cutting of a parapet (D1), scour (D2), and deck damage (D3). A U-Haul truck, equipped with five strategically placed accelerometers, collected vibration data while traversing the bridge at speeds ranging from 0 to 30 mph (48 km/h), simulating real-world traffic conditions. The acquired time series from accelerometers were processed and input into a custom-designed 1D CNN model for damage detection and classification. The model demonstrated 99.9% training accuracy, 94.44% validation accuracy, and 93% testing accuracy. Detailed analysis of the confusion matrices revealed high classification precision across all damage states, with healthy (D0) and severely damaged (D3) conditions classified with over 95% accuracy, scour damage (D2) with over 92% accuracy, and parapet damage (D1) classified with slightly lower but robust performance exceeding 85%. This research represents the first successful application of 1D CNNs for damage detection of a full-scale, in-situ, bridge using drive-by sensing. These findings serve as an important step towards offering a non-intrusive, cost-effective, and scalable alternative to traditional SHM techniques. The proposed method demonstrated promising accuracy and efficiency in classifying bridge damage states, representing a significant first step toward advancing bridge health assessment practices. This approach has strong potential to enable safer and more resilient transportation infrastructure while optimizing maintenance resources.