Damage detection of highway bridges is a significant part of structural heath monitoring. Conventional accelerometers or strain gauges utilized for damage detection have many shortcomings, especially their monitoring gauge length being too short, which would result in poor damage detection results. Under this circumstance, long-gauge FBG sensors as a novel optical sensor were developed to measure the macro-strain response of the structure. Based on this sensor, many derived damage detection methods were proposed. These methods exhibit various characteristics and have not been systematically compared. As a result, it is difficult to evaluate the state of the art and also leads to confusion for users to select. Therefore, a strict comparative study on three representative methods using long-gauge FBG was carried out. First, these methods’ theoretical backgrounds and formats were reformulated and unified for better comparison. Then, based on validated vehicle–bridge coupling simulation, these methods’ performances were tested through a series of parametric studies including various damage scenarios, vehicle types, speeds, road roughness and noise levels. The precision and reliability of three methods have been thoroughly studied and compared.

The lateral distribution characteristic of a bridge is one of the important features reflecting its in-service state. Conventional deterministic indicators often struggle to consider this time-varying feature. In this study, a data-driven approach is employed to establish the mapping model among responses at the different lateral positions using the eXtreme Gradient Boosting model optimized by the Bayesian optimization algorithm. The proposed method is validated based on both data from a bridge health monitoring system equipped on an actual bridge and data from numerical simulation. At different lateral positions of the actual bridge, the overall coefficient of determination (R²) of the strain response mapped by the model is above 0.991, and the overall root mean squared error (RMSE) of the strain response mapped by the model are below 1.159 με. The numerical simulation method is used to consider a variety of working conditions with different road surface roughness and different traffic densities. Under different working conditions, the overall R² of the deflection response mapped by the model is still above 0.961, and the overall RMSE of the deflection response mapped by the model is below 0.249 mm. These indicate that the proposed model can consider time-varying mapping relationships among responses at different lateral positions, and has good accuracy and applicability.

The rapid and accurate assessment of structural damage following an earthquake is crucial for effective emergency response and post-disaster recovery. Traditional manual inspection methods are often slow, labor-intensive, and prone to human error. To address these challenges, this study proposes STPEIC (Swin Transformer-based Framework for Interpretable Post-Earthquake Structural Classification), an automated deep learning framework designed for analyzing post-earthquake images. STPEIC performs two key tasks: structural components classification and damage level classification. By leveraging the hierarchical attention mechanisms of the Swin Transformer (Shifted Window Transformer), the model achieves 85.4% accuracy in structural component classification and 85.1% accuracy in damage level classification. To enhance model interpretability, visual explanation heatmaps are incorporated, highlighting semantically relevant regions that the model uses for decision-making. These heatmaps closely align with real-world structural and damage features, confirming that STPEIC learns meaningful representations rather than relying on spurious correlations. Additionally, a graphical user interface (GUI) has been developed to streamline image input, classification, and interpretability visualization, improving the practical usability of the system. Overall, STPEIC provides a reliable, interpretable, and user-friendly solution for rapid post-earthquake structural evaluation.

Internal reinforcement corrosion threatens the durability and safety of concrete structures, yet conventional nondestructive testing (NDT) methods remain largely qualitative and struggle to detect early‐stage corrosion concealed within the concrete cover. Magnetic flux leakage (MFL) is a promising NDT approach for reinforcement corrosion detection. However, variations in corrosion morphology and reinforcement layout induce substantial signal fluctuations and intrinsic uncertainties that deterministic models cannot represent. To address this, this study proposes a probabilistic framework that integrates MFL signals with a Monte Carlo Dropout Uncertainty‐Aware Neural Network (MCD‐UANN) for quantitative corrosion detection. An automated indoor platform acquires 640 MFL samples from reinforced concrete specimens covering 32 corrosion states and four duct positions, each scanned at four angular orientations and summarized by an averaged waveform. The trained MCD‐UANN achieves high accuracy ( R ² = 0.932) and reliable uncertainty quantification, with 92.9% of the measured corrosion degrees falling within two standard deviations of the mean prediction . Benchmarked against well‐optimized Multilayer Perceptron, eXtreme Gradient Boosting, and Random Forest models, it attains equal or better accuracy while additionally separating the predictive uncertainty into aleatoric and epistemic components. Interpretability analysis confirms that the model captures physically meaningful MFL features consistent with magnetic dipole theory, bridging data‐driven learning and domain knowledge. The quantified uncertainty is further shown to support inspection planning and maintenance prioritization so that the framework offers a robust, uncertainty‐aware solution for reinforcement corrosion assessment.