Conventional fatigue assessment of steel railway bridges relies on strain monitoring and cycle counting (e.g., rainflow) to obtain stress spectra. Motivated by accelerometers’ long-term stability and ease of installation, recent studies seek to infer local stress histories from bridge vibrations; however, this remains challenging because acceleration signals are noisy and confounded by structural dynamics, variable loading, resonance, and train–track interactions. To address these challenges, this study introduces Vib2StressNet, a deep-learning architecture that maps multi-channel vertical vibration signals directly to fatigue stress spectra. Unlike sequence-to-sequence models that reconstruct full time histories, this approach bypasses intermediate steps to focus on damage accumulation. The architecture integrates convolutional layers with a multi-head self-attention mechanism that captures information across multiple frequency bands. A critical design feature is the inclusion of train speed as an auxiliary scalar embedding, allowing the network to dynamically adapt to speed-dependent resonance. Vib2StressNet demonstrated strong generalizability across three Swedish railway bridges with diverse structural configurations, train types, and dynamic loading conditions. Under resonance and variable-speed conditions, adding train speed significantly improved prediction accuracy, reducing the mean squared error (MSE) by more than 20%. Relative to a previous sequence-to-sequence baseline model that reconstructs stresses before rainflow counting, Vib2StressNet simplifies the workflow and reduces MSE from 59.8 to 0.34 with more test samples. Interpretability analyses further confirm that preserving high-frequency components associated with axle impacts improves accuracy. These findings establish Vib2StressNet as a cost-effective tool for long-term monitoring that supports predictive maintenance without permanent strain instrumentation.

Modal properties are widely used for structural model updating and damage detection, yet their sensitivity to environmental conditions, particularly temperature can complicate interpretation. This paper investigates how seasonal temperature variations affect structural identification based on modal properties, using a falsification-based approach. A railway bridge in service is used as a case study, with data from both forced vibration testing and long-term monitoring. Freezing temperatures are found to increase overall stiffness by up to 90%, including a significant rise in boundary stiffness, an increase in ballast modulus of up to 10 GPa, and the activation of rail continuity stiffness. The set of plausible physical models shifts between winter and summer, demonstrate that structural behaviour varies seasonally and that identification results are not directly transferable across temperature ranges. Compared to traditional residual minimisation, model falsification provides a more comprehensive and robust set of plausible models and avoids unreliable parameter estimates. These findings underscore the need for temperature-sensitive identification strategies and adaptive models to support accurate interpretation, damage detection, and seasonally informed maintenance decisions in structural health monitoring.

To explore the governing mechanism underlying the tensile fatigue behavior of Ultra-high Performance Fiber Reinforced Cementitious Composites (UHPFRC), this study tested eight specimens using four advanced non-destructive measurement techniques. First, magnetoscopy is conducted on each specimen to determine the local fiber orientation and volume. Afterward, seven specimens are statically preloaded to the tensile strain of 1.5 ‰, identified as the typical maximum strain of UHPFRC in structural applications; while one specimen to the strain of 0.19 ‰, within the tensile elastic domain. During testing, the specimen response is monitored using digital image correlation and acoustic emission, in addition to displacement transducers. All specimens show similar evolution of fatigue deformation, characterized by three development stages. It is found that the local fiber orientation governs the fatigue deformation behavior. Fatigue deformation concentrates in low fiber orientation zones and fatigue fracture always occurs at the zone with lowest fiber orientation coefficients. The acoustic emission measurement, represented by cumulative energy curve and Ib-values, can appropriately characterize specimen damage degree and distinguish cracking patterns.

Effective anomaly detection and classification in bridge monitoring data critically depend on robust feature extraction to ensure reliable performance across diverse sensors, loading scenarios, and anomaly types. This study presents a novel deep metric learning-based framework that automates feature extraction, addressing the limitations of traditional manual methods that often lack generalization. The proposed approach employs continuous wavelet transforms to convert time-series signals into time-frequency representations, followed by a convolutional neural network trained with metric learning to produce low-dimensional feature optimized for anomaly classification. These embeddings capture similarities between anomalies, enabling the model to distinguish normal loading events from anomalous ones while classifying specific anomaly types. The method was validated on bridge monitoring data collected from strain gauges and accelerometers under varying loading scenarios, achieving high detection and classification accuracy. This robust and scalable framework highlights the potential of deep metric learning to advance automated bridge monitoring.

This paper investigates the fracture mechanism of ultra-high-performance fiber-reinforced cementitious composites (UHPFRC) using acoustic emission (AE), digital image correlation (DIC), and magnetoscopy testing. Four specimens undergo uniaxial tensile loading, preceded by magnetoscopy testing to determine local fiber volume and orientation. DIC captures matrix discontinuities, crack initiation, and propagation. Acoustic emission monitors fracture mechanisms at different loading phases. During the elastic phase, matrix discontinuities and fiber debonding are observed to occur. A higher density of matrix discontinuities during this phase enhances hardening behavior and tensile performance. The softening phase of UHPFRC is found to be characterized by three stages based on AE parameters: emergence and competition of multiple fictitious cracks, propagation of a dominant fictitious crack, and real crack formation. The rate of dominant fictitious crack propagation can be determined by analyzing the evolution in AE parameters with stress decrease. Uniform fiber distribution limits the initiation and propagation of fictitious cracks.

In this paper, the performance of a single-span railway bridge with integrated retaining walls during high-speed train passage is investigated by considering different uncertainties originating from modeling assumptions and measurement processes. For this purpose, a single-span railway bridge is equipped with numerous accelerometers and is excited using a hydraulic actuator across different frequency ranges. A comprehensive 3D model of the bridge and the surrounding soils is created in Abaqus. Different sets of material properties for concrete and soil components are derived by converging the frequencies and damping ratios of the first three structural modes, using both the Error-Domain Model Falsification (EDMF) and Residual Minimization (RM) methods. These material properties are subsequently utilized in high-speed train passage analysis, and the results are compared.

Accurate differentiation between damage-related anomalies and data errors is a critical challenge in bridge monitoring. This paper presents a data-driven framework for anomaly detection and classification, addressing the question: How can anomalies be classified in multi-sensor bridge monitoring to distinguish structural changes from noise? The framework combines an adapted anomaly taxonomy with a deep neural network trained on synthetic data. It is validated using long-term monitoring data from a railway bridge, incorporating strain gauges, accelerometers, and an inclinometer. In offline training, the model achieves high precision, recall, and F1-scores, effectively detecting anomaly classes across sensor types. For online prediction, it provides anomaly type percentages and visualizations over daily, weekly, and annual timeframes, distinguishing frequent noise-related anomalies from rare anomalies signaling structural changes. Requiring one month of training data, the framework delivers a scalable solution for bridge monitoring and lays the groundwork for future self-learning anomaly detection in infrastructure management.

The bridge network is progressively aging, with an alarming proportion of bridges over 100 years. This situation engenders substantial risks to the overall reliability of transportation networks, requiring innovative methods for efficient management. Monitoring can provide a direct source of information about structural behavior generating alerts when changes occur. Real-time alerts enable effective infrastructure management and decision-making during damage or anomalous situations. However, monitoring can result in a large amount of data that is often difficult to convert into valuable information in real time. This paper presents an approach for real-time detection of abrupt damage occurrence in bridges using unsupervised anomaly detection algorithms and strain/acceleration measurements. The approach incorporates the separation of measurements into events having the same loading nature and the construction of three feature matrices based on statistical features, time-frequency features, and wavelet spectrum features. It includes the evaluation of five anomaly detection algorithms including Isolation Forest, One-Class Support Vector Machine, Robust Random Cut Forest, Local Outlier Factor, and Mahalanobis Distance. The approach is illustrated with a case study of a steel-bascule-railway bridge, that has experienced a brittle cracking event during monitoring. Results highlight the robustness of One-Class Support Vector Machine, Isolation Forest, and Local Outlier Factor algorithms in promptly detecting abrupt changes across different features. The separation of strain and acceleration data into loading-based events, coupled with the comparison of previous and new event features, provides robust feature matrices for effective damage detection. Enhanced detection and higher scores are particularly attributed to time-frequency domain features during damage occurrence. The presented approach can be used as a base on how to perform real-time anomaly detection within the context of bridge monitoring.

Bridge monitoring is a useful tool for alerting to changes in behavior and supporting informed bridge management decisions. However, the triggering of alerts is often based on predefined thresholds, which results in incomplete anomaly detection, leading to excessive false positives or negatives. This highlights the need for an efficient anomaly detection approach adapted to bridges. This paper presents a supervised classification framework for detecting and classifying anomalies in bridge monitoring data. The framework involves labelling data from existing datasets and training a classification algorithm to detect similar anomalies in new measurements. The focus is on strain measurements labelled using a predefined time series taxonomy, as illustrated by a case study of a bascule railway bridge. The framework demonstrates high accuracy in detecting and classifying anomalies, making it easy to identify their causes. It triggers alerts only when necessary and provides a reliable method for detecting changes in behavior.

Detecting damage in bridges that present signs of deterioration or have exceeded the expected lifespan is critical for ensuring safety in service. This paper suggests an approach for real-time damage detection for such bridges through monitoring and machine learning algorithms, which serve as timely alarms for decision-making and subsequent damage identification. The approach involves five steps: monitoring, data collection, data separation, feature extraction, and anomaly detection in real-time. Monitoring is ensured by strain gauges, accelerometers, and a temperature sensor. Data collection is ensured at high frequency continuously to capture the dynamic effects of loading. Data separation is provided to classify monitoring data according to loading events, which is in the case of the study characterized by the bridge opening, the bridge closing, and train passages. Feature extraction is provided to characterize monitoring data for each loading event. Anomaly detection is performed by the Isolation Forest and the One-Class Support Vector Machine algorithms. The algorithms are implemented in real-time for each new event. The approach is illustrated in a full-scale post-damage case study of a steel-bascule-railway bridge, in service since 1916, with signs of corrosion and fatigue. The results demonstrate the ability of the approach to capture a cracking event in real-time. The Isolation Forest algorithm is found to be more robust for damage detection compared to the One-Class Support Vector Machine. It assigned high scores to the events occurring during and after the cracking, highlighting its ability to capture such incidents promptly. These findings have significant implications for bridge owners as they can identify damage in components in real time, enabling them to take timely measures such as traffic interruption and subsequent repairs.