One of the main challenges hindering Structural Health Monitoring (SHM) of bridges is the high cost associated with sensor deployment and other monitoring tools. Ideally, all infrastructure assets should be monitored regularly and efficiently to ensure their safety and longevity. However, financial constraints often limit the feasibility of widespread and frequent monitoring. A promising solution to this issue is mobile crowdsensing, which involves utilizing public participation to monitor bridges using data collected from personal devices. In this approach, there is no need to invest in dedicated sensing devices, nor to monitor or replace them during operation. Crowdsensing transforms everyday users into mobile sensors and dramatically reduces costs while expanding data coverage. Nonetheless, despite the many advantages of using crowdsensing and IoT technologies, their application is still in the early stages, especially in bridge health monitoring. Efficient implementation has not been widely explored, and there is currently no clear roadmap or framework to guide efforts and address existing challenges. As with any emerging technology, crowdsensing and IoT introduce a range of issues that must be carefully considered and resolved, some of which are technically complex. This study aims to present a clear and structured framework for applying mobile crowdsensing in the context of bridge health monitoring. The framework is designed to support scalable and cost-effective SHM systems that complement traditional monitoring approaches. To ensure practical applicability, the study demonstrates key challenges associated with implementing such systems, categorized into three main aspects: data collection, data transmission, and data processing. The flow of information through each category is discussed, and potential challenges are identified. The framework is expected to demonstrate the feasibility of large-scale, user-driven monitoring while identifying practical barriers such as data quality control, incentive mechanisms, cloud management, and privacy concerns. This paper outlines the key elements and offers guidance for future research in mobile crowdsensing.

Large-span balanced cantilever bridges experience significant thermally induced displacements, and insufficient expansion-joint capacity can lead to elevated internal stresses and long-term structural damage. Addressing this problem conventionally requires installing in-situ sensors, which provide high temporal resolution but are costly, intrusive, and limited in spatial coverage. This study presents a multitemporal Interferometric Synthetic Aperture Radar (InSAR)–based Structural Health Monitoring framework designed to capture and interpret these thermal displacements. A tailored spatial-filtering strategy is developed to address the difficulty of identifying coherent scatterers on slender bridge decks. Measurement points are extracted using a spatial buffer calibrated to European Ground Motion Service (EGMS) positional uncertainties, ensuring reliable attribution to bridge elements. These points are subsequently grouped by structural control nodes—piers and expansion joints—to isolate differential thermal responses across spans and cantilever segments. The framework is demonstrated on the Strängnäs Bridge in Sweden, which experiences large annual temperature fluctuations from –18 °C to 30 °C. Because Line-of-Sight (LOS) displacement from InSAR alone is not directly interpretable in the longitudinal direction, a short-term in-situ campaign using a Linear Variable Differential Transformer (LVDT) at the main expansion joint is incorporated. This dataset validates the LOS-to-longitudinal displacement projection, enabling long-term thermal-movement tracking even after the ground-truth measurements end. Importantly, the framework is implemented entirely using open-source tools and processing chains, allowing the monitoring to be continuously updated with each new satellite acquisition rather than being limited to a one-off forensic analysis. The integrated system establishes a robust link between calibrated InSAR displacement measurements and thermal loading history, providing a validated operational method for early detection of abnormal thermal-response patterns and improving the management of thermally vulnerable bridge structures.

Conventional structural health monitoring (SHM) systems rely on centralized servers for data analysis, which results in heavy data communication and high latency. To address these limitations, this study proposes a MEMS-based sensing system with edge computing, utilizing a lab-scaled steel truss bridge instrumented with a lightweight accelerometer. The accelerometer was connected to the STM32 microcontroller (MCU) as an edge node. Both static and dynamic tests were conducted, including artificial damage scenarios at specific truss members. After the bridge was excited by the shaker, dynamic responses were analyzed on the MCU using Fast Fourier Transform (FFT) to extract the natural frequency peaks, while static tests focused on zero-point offset before and after introducing damages. Thus, the MCU performs preliminary signal processing and communication for detection and notification of structural anomalies in real time, reducing raw data transmission to the minimum. Finally, multiple indicators are used to classify different structural damage scenarios for decision-making. Experimental results demonstrate the efficiency and capability of edge computing for intelligent bridge health monitoring, showing great potential for future field deployment.

Visual inspection remains a cornerstone of bridge maintenance and Structural Health Monitoring (SHM), yet progress in automated defect detection is limited by the scarcity and heterogeneity of annotated datasets. Publicly available collections such as CODEBRIM, DACL1K, and DACL10K contain only a few thousand labeled samples, often constrained to specific materials or defect types. As a result, supervised deep learning models trained on these datasets tend to overfit to illumination, texture, or viewpoint variations, leading to poor generalization across different bridge structures and environments. This data bottleneck highlights the need for transferable and data-efficient representations that can serve as universal visual priors for SHM.
Recent advances in vision foundation models (VFMs) have demonstrated the emergence of such representations through large-scale self-supervised pretraining. Among these, DINOv3, a Vision Transformer trained on millions of unlabeled images via self-distillation, represents the current state of the art in universal visual feature learning. DINOv3 surpasses previous self-supervised and supervised models on tasks ranging from classification and segmentation to correspondence and retrieval, showing remarkable adaptability without task-specific fine-tuning. Its dense and semantically rich feature embeddings capture geometry, texture, and contextual information in a unified space, suggesting strong potential for domains characterized by limited labeled data, such as bridge inspection.
This study investigates the applicability of DINOv3 features to visual bridge defect representation. We evaluate whether its pretrained embeddings can effectively encode and separate defect patterns such as cracks, spalling, corrosion, and material delamination when extracted from existing inspection imagery. By analyzing feature distributions and separability across standard SHM datasets, we assess their suitability for lightweight downstream classifiers and detectors. The goal is to verify if a foundation model trained on natural images can generalize to structural materials and deterioration cues without fine-tuning, thereby offering a path toward scalable and data-efficient SHM.
Leveraging the demonstrated robustness and generalization of DINOv3, this work aims to bridge the gap between generic visual pretraining and specialized infrastructure inspection. If successful, this approach could establish a foundation-model-based framework where universal visual features underpin more reliable, transferable, and annotation-efficient defect detection across diverse structural assets.

Integrating structural health monitoring (SHM) data into Bridge Information Modeling (BrIM) is essential for informed bridge asset management and decision-making. However, directly incorporating large volumes of raw monitoring data into Industry Foundation Class (IFC)-based information models is challenging, as such data must first be processed and interpreted to provide meaningful insights. This study addresses this challenge by proposing an IFC schema extension that enables the integration of real-time data with key performance indicators (KPIs) derived from SHM data.

The proposed framework begins with the identification of core SHM information requirements, including monitored bridge elements, measurement types, and relevant KPIs. These are then formalized within an extended IFC schema that supports the linkage of processed monitoring information with corresponding bridge components and sensors. A parameter mapping approach, developed using the xBIM™ Toolkit, ensures IFC-compliant embedding of KPI-based monitoring information into the BrIM model. Additionally, a web-based visualization platform is implemented to access and interact with the enriched model, allowing intuitive presentation of KPIs for condition assessment and maintenance planning.

A case study involving a sensor-instrumented bridge demonstrates the practical applicability of the proposed approach. Results show that the framework provides an effective and scalable solution for SHM–BrIM integration, enabling KPI-driven visualization and supporting data-informed bridge asset management.