Editorial for the Special Issue “Advances in Structural Health Monitoring and Industry 5.0 Innovations for Bridge Management and Conservation”

1. Introduction

Bridges are critical components of transport infrastructure, supporting the movement of people and goods while contributing to the economic, social, and cultural development of societies. At the same time, bridge networks worldwide are exposed to ageing, deterioration, increasing traffic demands, environmental actions, and limited resources for inspection, maintenance, and rehabilitation [1]. These challenges are particularly significant for bridges with cultural heritage value, for which structural deterioration or failure may result not only in economic and safety consequences but also in the irreversible loss of important cultural assets [2]. Conventional inspection and maintenance practices, which often rely on periodic inspections and fragmented sources of information, are increasingly challenged by the scale and complexity of contemporary bridge networks. Consequently, there is a growing need for more continuous, data-informed, and proactive approaches to structural assessment and asset management [3].

Structural Health Monitoring (SHM) has emerged as an important component of this transformation by enabling the systematic acquisition, processing, and interpretation of information about the condition and performance of structures. Recent advances in sensing technologies, wireless communication, data analytics, Artificial Intelligence (AI), and digital modelling are creating new opportunities for automated damage detection, condition assessment, predictive maintenance, and decision support. In parallel, Digital Twins (DTws), defined as “digital representations of a target entity with data connections that enable convergence between the physical and digital states at an appropriate rate of synchronization” [4], are increasingly being investigated as a means of connecting physical infrastructure with dynamic digital representations throughout the asset life cycle. In the context of bridge management, the integration of numerical structural models, SHM systems, and AI-based anomaly detection offers the potential to generate more comprehensive information about bridge condition and support more informed management decisions.

The idea for this Special Issue emerged from the gaps identified in a systematic literature review of bridge management through DTw-based anomaly detection systems [5]. The review identified several challenges that continued to constrain the development of integrated digital approaches to bridge management, including software interoperability, the performance of anomaly-detection algorithms, and the integration of DTws at the macro scale. The comparatively limited attention given to cultural heritage and conservation within the engineering literature is also highlighted, and the implementation of Industry 5.0 concepts within DTw frameworks is identified as an important direction for future research. These findings provided an important motivation for bringing together research that moves beyond the isolated development of individual technologies and instead considers how sensing, data, computational intelligence, digital representations, and human expertise can be integrated into more effective bridge management and conservation practices.

Industry 5.0 provides a particularly relevant framework for addressing these challenges. Rather than representing a replacement for Industry 4.0, Industry 5.0 builds upon digitalisation and automation while placing greater emphasis on three complementary principles: human-centrism, resilience, and sustainability. Its enabling technologies include AI, DTws, human–machine interaction, cyber-secure data infrastructures, and energy-efficient and trustworthy autonomous systems. In the Architecture, Engineering, Construction, Management, Operation, and Conservation (AECMO&C) sector, the rapid adoption of these principles and technologies could support more human-centred decision-making, enhance the resilience of infrastructure systems, improve resource efficiency, and enable more sustainable approaches to the management and conservation of the built environment [6]. Recent research, however, indicates that Industry 5.0 adoption within built cultural heritage conservation remains limited. Although AI and DTws are among the most extensively investigated enabling technologies, resilience and human-centrism require substantially greater research and implementation efforts to achieve a more holistic transition towards Industry 5.0 [7].

Against this background, this Special Issue of Infrastructures, entitled “Advances in Structural Health Monitoring and Industry 5.0 Innovations for Bridge Management and Conservation”, was conceived to bring together recent advances at the intersection of SHM, bridge engineering, digitalisation, and Industry 5.0. The Special Issue focuses on how emerging technologies and human-centric approaches can transform the ways in which bridges are monitored, assessed, maintained, and conserved. Particular attention was given to sensor placement optimisation; automation of damage detection, localisation, and quantification; data mining and data fusion for bridge SHM; data-driven, model-based, and hybrid SHM strategies; remaining service-life prediction; SHM-aided decision-making; and the application of smart sensors, AI-driven analytics, and DTws.

The Special Issue ultimately brought together 11 contributions, encompassing research and review studies that collectively illustrate the breadth of current developments in this rapidly evolving field. Taken together, these contributions provide insights into the development of more intelligent monitoring systems, advanced data-analysis methods, predictive approaches, and decision-support frameworks, while also highlighting the continuing importance of human expertise, practical implementation, and resilience in the management and conservation of bridge infrastructure. The following sections provide an overview of the contributions and discuss the common themes and emerging research directions identified across the Special Issue.

2. Overview of the Contributions

The Special Issue opens with the contribution of Nguyen Thi Cam Nhung, Hoang Nguyen Bui, and Tran Quang Minh, who address an important practical challenge affecting the reliability of SHM systems: the loss or corruption of sensor data. The authors propose a hybrid Convolutional Neural Network–Gated Recurrent Unit (CNN-GRU) architecture for recovering missing accelerometer data. The approach combines the ability of CNNs to extract local and spatial features from sensor signals with the capacity of GRUs to capture temporal dependencies. Using data collected from the Thang Long Bridge, the study considers both single-channel and multi-channel data-loss scenarios and demonstrates the effectiveness of the proposed architecture in reconstructing missing measurements. The work highlights the importance of ensuring data integrity within SHM systems and demonstrates how hybrid deep-learning architectures can contribute to more reliable and resilient monitoring infrastructures [8].

Plevris and Papazafeiropoulos provide a broader perspective through a review of the rapidly growing application of AI in SHM for infrastructure maintenance and safety. Their study combines bibliometric analysis with a thematic examination of seven key areas in which AI is transforming SHM: (i) data acquisition and sensor networks; (ii) data processing and signal analysis; (iii) anomaly detection and damage identification; (iv) predictive maintenance; (v) reliability and risk assessment; (vi) visual inspection and remote monitoring; and (vii) resilient and adaptive infrastructure. In addition to highlighting the technical potential of Machine Learning (ML) and deep learning, the authors discuss ethical and societal considerations, including data privacy, equity, and transparency. Their review, therefore, provides an important overview of the technological landscape while reinforcing the need to consider AI not simply as an analytical tool, but as part of a broader transformation towards more efficient, safe, sustainable, and resilient infrastructure management [9].

Abdelkader, Al-Sakkaf, Ebrahim, and Elkabalawy turn attention from monitoring technologies to the strategic allocation of resources for bridge maintenance. Their contribution combines the systematic literature and scientometric reviews to examine 380 peer-reviewed publications on bridge maintenance budget allocation published between 1990 and 2025. Using Scopus and Web of Science data together with VOSviewer and the Bibliometrix R package, the authors map the evolution of the field, its principal research communities, and its methodological developments. Six dominant methodological paradigms are identified: (i) multi-criteria decision-making, (ii) life-cycle assessment, (iii) DTw, (iv) inspection planning, (v) AI, and (vi) optimisation. By bringing these strands of research together, the study provides asset managers with a structured view of existing approaches for prioritising maintenance expenditure and highlights the potential of increasingly integrated and data-driven approaches to support sustainable bridge management under constrained resources [10].

Flores Cuenca, Yardim, and Hasan investigate the potential of transformer architectures for forecasting future bridge condition. The authors argue that conventional inspection records are frequently treated as isolated condition ratings rather than as elements of a continuous structural health timeline, thereby limiting their predictive value. To address this limitation, they develop a transformer-based architecture inspired by Natural Language Processing (NLP), in which bridge condition ratings and other relevant features are treated as tokens within temporally ordered inspection “sentences” covering the period from 1993 to 2024. The self-attention mechanism enables the model to capture long-range temporal dependencies, producing reported prediction accuracies of 96.88% for short-term forecasting and 86.97% over a seven-year horizon. The study illustrates how emerging AI architectures can transform historical inspection records into predictive information, supporting a shift from reactive assessment towards data-driven bridge preservation and proactive management [11].

Mehrjoo, Hom, Beigi, and Betti address another important challenge in automated SHM: developing monitoring algorithms that can transfer across structures without requiring asset-specific retraining. Their study presents a real-time, output-only monitoring framework that combines damage-sensitive cepstral features with a streaming Long Short-Term Memory (LSTM) network. Acceleration time histories are transformed into cepstral coefficients and subsequently processed sequentially by a stacked LSTM architecture with state carry-over, allowing temporal dependencies to be retained while enabling low-latency inference. The methodology is evaluated under a strict zero-shot setting using the full-scale Z24 Bridge benchmark, without training or calibration data from the monitored bridge. The results demonstrate the ability of the proposed framework to discriminate damage states and track their progression using vibration measurements alone, suggesting a promising pathway towards scalable monitoring systems capable of transferring learned knowledge between different bridge assets [12].

Bhandari, Jang, Han, and Malla bring the Industry 5.0 perspective more explicitly into the Special Issue by reporting lessons learned from the long-term, real-time field monitoring of three in-service highway bridges. Rather than focusing exclusively on algorithmic development, the study examines the practical requirements associated with deploying SHM systems under real-world conditions. The authors propose a human-centric framework integrating Internet of Things (IoT), DTws, AI, cloud and edge computing, and interactive visualisation. Particular attention is given to sensor selection, wireless communication, modular system development, power consumption, weather exposure, site accessibility, and the durability and cost-effectiveness of monitoring installations. The development of a Docker-based analytical and visualisation platform further illustrates how interactive dashboards can facilitate human–machine collaboration and support engineering decision-making. The contribution therefore provides a valuable bridge between technological innovation and practical implementation, demonstrating how SHM systems can contribute directly to the human-centric, resilient, and scalable objectives of Industry 5.0 [13].

Bacha, Puppio, Zucca, and Sassu provide a structured scoping review of dynamic-response-based bridge monitoring and structural assessment. Their review examines the use of dynamic measurements for identifying damage-sensitive indicators, inferring structural stiffness and capacity, interpreting bridge responses under operational and extreme loading, and supporting assessment through contact and indirect or drive-by sensing. Particular attention is also given to data processing, ML, and DTw integration. The authors develop a traceable evidence inventory linking research studies to methodological families and decision objectives and emphasise the limitations associated with environmental and operational variability. Their synthesis shows that global modal properties can provide effective change screening but are often insufficient for damage localisation and state characterisation, which may require denser sensing and higher-fidelity signal processing. The review consequently advocates an evidence-based and conservative approach in which SHM outputs inform inspection and structural analysis rather than being treated as direct declarations of bridge condition [14].

Bilotta, Pollastro, Di Cristinzi, and Pecce investigate vibration-based damage detection through a comparative study of conventional modal-curvature methods and an AI-based approach using Variational Autoencoders (VAEs). The research focuses on post-tensioned prestressed concrete beams representative of an important class of existing bridge structures and considers defects associated with reduced prestressing force, lack of mortar within ducts, and tendon corrosion. Using acceleration measurements, the authors compare the two approaches for both damage detection and localisation. Their results indicate that both approaches can identify damage despite the challenges associated with detecting defects in prestressed concrete through dynamic response measurements. The AI-based method offers the advantage of providing earlier alarms with relatively few sensors, whereas modal curvature provides greater interpretability of the detected anomaly but requires a denser sensor configuration. The study therefore illustrates a central challenge in the transition towards AI-enabled SHM: balancing computational performance and early detection with transparency and physical interpretability [15].

Ul Reyaz, Wang, and Najm address the characterisation of surface-breaking cracks in concrete using ultrasonic imaging. Recognising that cracks can facilitate the ingress of moisture, chlorides, and other aggressive agents and thereby accelerate deterioration, the authors propose an ultrasonic common-midpoint approach for locating crack tips and estimating crack depth. Measurements are obtained using a pitch-catch configuration at multiple transmitter–receiver separations, from which time-of-arrival information associated with crack-tip diffraction is extracted. These measurements are subsequently incorporated into a travel-time-based localisation framework to generate images and estimate crack-tip coordinates. The methodology is evaluated using concrete slabs containing vertical and inclined cracks of different depths and a reinforced concrete beam containing thin flexural cracks. The close agreement between the ultrasonic localisation results and the observed crack geometries demonstrates the potential of the approach as a non-destructive technique for detailed condition characterisation of concrete bridges [16].

Luszczynska and Nowak shift the focus towards the monitoring and characterisation of traffic-induced live loads. Their study examines site-specific live-load spectra using extensive Weigh-in-Motion (WIM) data collected in Alabama, United States. More than 150 million vehicle records collected during 2023 and 2024 are processed to calculate bridge live-load effects for spans ranging from 9 to 60 m, and the resulting distributions are extrapolated to a 75-year period. The analysis indicates that the current AASHTO provisions for the Strength I Limit State do not fully represent the traffic loads observed at WIM locations on interstate highways, with the results demonstrating substantial site-to-site variation in live-load effects. The study reinforces the value of continuous traffic data collection and reliability-based analysis for understanding actual bridge loading conditions and provides a data-driven basis for more accurate assessment, monitoring, and management of bridge safety and deterioration [17].

Finally, Sinden and Jiménez Rios address one of the fundamental limitations facing the application of ML to bridge SHM: the scarcity of diverse, labelled datasets representing realistic structural damage and environmental variability. The authors propose a framework for generating synthetic modal-frequency data using a calibrated finite-element model of the S101 bridge, incorporating aleatory uncertainties and environmental variability derived from a 20-year climate record. The resulting dataset comprises 7300 synthetic modal-frequency iterations and demonstrates that thermally induced frequency shifts can exceed those associated with minor structural damage, highlighting the importance of environmentally representative training data. The study further evaluates supervised and unsupervised learning approaches, with the supervised multi-layer perceptron achieving high levels of classification accuracy. Importantly, however, the research also exposes the limitations of simplified linear finite-element modelling, which failed to reproduce the experimentally observed response to pier settlement. By making the resulting FAIR-compliant synthetic dataset openly available, the contribution provides a potential benchmark for SHM algorithm development while simultaneously demonstrating the importance of validating digital and numerical representations against physical structural behaviour [18].

3. Common Threads and Future Directions

Read as a whole, the contributions in this Special Issue converge on several themes. First, AI and classical, physics-based methods are increasingly treated as complementary rather than competing: modal curvature sits alongside VAEs, finite element models are calibrated against and then used to generate training data for ML surrogates, and DTws are repeatedly framed as the bridge between the two. Second, environmental and operational variability (i.e., temperature above all) emerges as a recurring confounder that must be explicitly modelled, normalised, or represented in training data, whether the underlying method is a modal indicator, a deep learning model, or a synthetic dataset. Third, data scarcity is addressed from multiple angles across the issue: through data recovery for corrupted sensor streams, zero-shot architectures that avoid asset-specific retraining, and synthetic dataset generation grounded in calibrated numerical models, which all represent different responses to the same underlying shortage of labelled, real-world damage data.

Fourth, the issue reflects a growing appreciation that SHM outputs feed into decisions made by, and with, people: the review of maintenance budget allocation models, the field-deployment lessons from Connecticut, and the evidence-maturity and escalation frameworks proposed in the dynamic-response scoping review all treat monitoring data as an input to human and institutional decision-making rather than as a self-sufficient verdict on structural safety. Finally, several contributions highlight that validation on full-scale, in-service bridges, as opposed to laboratory specimens or numerical benchmarks, remains comparatively rare, and that open, well-documented field datasets are still needed to enable independent comparison of the many methods now being proposed. Overall, the articles published within this Special Issue contribute to improving our understanding and enhancing the resilience of existing bridges to short- and long-term impacts of climate change [19].

On behalf of the Guest Editorship, we would like to thank all the authors for their valuable contributions, the reviewers for their careful and constructive feedback, and the Infrastructures Editorial Office for its continuous support throughout the preparation of this Special Issue. We hope that the collected papers will encourage further interdisciplinary research at the intersection of structural engineering, data science, and human-centric design, and contribute to a more resilient and sustainable approach to bridge management and conservation.