Continuous monitoring of railway bridges is essential for ensuring safety and operational reliability, considering aging mechanisms, rising traffic, and elevated speeds of railway vehicles. Frequently, traditional vibration-based approaches, including modal identification and data-driven diagnostic strategies, are strongly influenced by environmental and operational variability, requiring labeled damaged datasets or numerical simulations to provide reliable outcomes. However, the acquisition of complete and representative datasets for training neural networks in structural health monitoring remains a challenging task, particularly for large-scale civil structures such as bridges. In these cases, unsupervised learning approaches represent promising solutions. An unsupervised anomaly detection methodology for railway bridge monitoring based on a long short-term memory (LSTM) autoencoder (AE) trained exclusively on bridge accelerations under healthy structural conditions is proposed in the present work. Specifically, the acceleration responses are obtained from simulations made on a calibrated finite element model of the bridge, reproducing realistic train–bridge interaction scenarios. The multi-channel acceleration signals are reconstructed by the proposed LSTM AE to produce the Root Mean Square Error (RMSE) between measured and reconstructed acceleration responses as indicators of potential structural anomalies. A dual-threshold strategy is adopted for damage detection purposes, including a global threshold for identifying anomalies in the overall dynamic response and per-sensor thresholds derived from the healthy-condition RMSE distribution for detecting localized damages. Only healthy-condition data are required for employing the proposed technique, avoiding labeled damaged data for training purposes. The LSTM AE constitutes an effective and computationally efficient tool for anomaly detection and continuous structural health monitoring of railway bridges, as demonstrated by the obtained results, representing a promising alternative to classical modal-based approaches and existing deep learning-based methods.

This work investigates the capability of a permanent monitoring system to detect structural degradation before it compromises bridge operational requirements. The target structure is a multi-span steel truss railway bridge located in Italy, characterized by a complex geometry featuring Gerber connections and dual-deck configuration. Utilizing a high-fidelity finite element model of the entire structure, a comprehensive set of damage scenarios is simulated. The research evaluates the sensitivity of a monitoring setup consisting of inclinometers mounted on all three spans. A data processing pipeline is implemented to purge sensor signals from temperature-induced variations using multiple linear regression and principal component analysis. Damage detectability is then assessed by comparing damage-induced variations from ”virtual sensors” against the residual statistical variance (±2σ) derived from the real-world monitoring data of the healthy structure. The outcomes provide insights into the effectiveness of static monitoring systems for the identification of different damage conditions in complex truss bridges.

Structural health monitoring of bridge infrastructures often faces the challenge of data scarcity, where reliable or labeled data is available for one portion of a structure but not for others. Transfer learning offers a solution by leveraging data from a source domain to improve diagnostic performance in a target domain. This study investigates a transfer learning approach for damage detection between two different spans of the same bridge: the riverbed and the floodplain span. To ensure physical consistency between the domains, four modal frequencies are selected for each span based on high modal assurance criterion values. Normal condition alignment is employed to minimize the distribution discrepancy between the source and target datasets under healthy conditions. To evaluate the effectiveness of the alignment, an artificial damage scenario is simulated in the target span by introducing a 0.1 Hz frequency shift in the final monitoring period. A Gaussian mixture model classifier is trained exclusively on the source data and then tested on the target data. Comparative analysis is performed by evaluating classification accuracy before and after statistic alignment. Normal condition alignment significantly enhances the detectability of damage in the target span by compensating for operational and structural variations between the two spans.