AI-based drive-by methodology for early damage detection in a steel truss railway bridge under multiple environmental and operational factors

Railway infrastructure monitoring is fundamental for transportation safety, particularly as networks expand and structures age. While traditional approaches rely on direct sensor installation on bridges, these systems are costly and impractical for widespread implementation. Despite advances in drive-by monitoring techniques, existing methods struggle to handle environmental and operational variations (EOVs) effectively, limiting possible applications in real-world conditions. This study develops an AI-based drive-by automated methodology for detecting structural damage in railway bridges. Using a three-dimensional finite element model of a Warren-type truss bridge and vehicle-track interaction simulation, the methodology was tested through comprehensive numerical simulations for freight railway operations. These simulations, offering essential flexibility for developing the technique, constitute a key step before experimental validation. The proposed damage detection framework encompasses multiple stages: initial preprocessing for noise reduction, feature extraction using Continuous Wavelet Transform, deep learning analysis via Convolutional Autoencoders, and final damage assessment using Structural Similarity Index error maps combined with statistical outlier detection. Results demonstrate successful structural damage detection across various scenarios, even at early stages, sensitivity to damage severity, and robustness against EOVs. This research provides a cost-effective alternative to traditional railway bridge monitoring systems while maintaining robust damage detection capabilities. Future works include experimental validation.