The ongoing requirement for lightweight yet high-performance structures within the aerospace industry has resulted in the extensive utilisation of Carbon Fiber Reinforced Polymers (CFRPs). Nevertheless, their vulnerability to barely visible impact damage and internal delamination poses a significant challenge to ensuring structural integrity throughout the entire service life of the product. Reliable Structural Health Monitoring (SHM) systems that are capable of early defect detection and localisation are therefore of crucial importance for ensuring and maintaining safety while reducing maintenance costs in aircraft and spacecraft applications.

In this study, an integrated experimental–numerical methodology is proposed for the detection and localization of delaminations in quasi-isotropic CFRP plates subjected to shear loading. Twelve square panels were manufactured, six of which were to be in pristine condition and six were to be intentionally defective. The defect under investigation was introduced by the insertion of a thin layer of PTFE (Teflon) between two plies prior to the curing process. This simulates a manufacturing-induced delamination or an in-service defect. The panels were subjected to in-plane shear loading in a custom-designed fixture that converts a uniaxial tensile load into a diagonal tension state. The initial deformation of the plate occurred in a pure shear configuration, characterised by the application of tension along one diagonal and compression along the opposing diagonal. As the load was increased, local buckling occurred along the compressive diagonal, resulting in a partial shear state. A possible data fusion approach has also been attempted combining strain fields data measured under shear instability and linear flexural loading in order to understand if a better defect detection precision could be achieved. The structural response was monitored using a combination of full-field and point-wise sensing techniques. Digital Image Correlation (DIC) was utilised to generate displacement and strain fields across the plate surface, while a reference strain gauge was employed for validation and cross-comparison. The synergy of these sensing systems enabled accurate tracking of strain evolution.

A data-driven metric was established to identify and localise delaminations by comparing strain distributions between pristine and defective panels. The experimental findings were then compared with a finite element (FE) model, which had been developed for the purpose of reproducing the shear response and capturing the onset of instability. The model was calibrated against experimental data, demonstrating a high degree of agreement and offering a physical interpretation of the observed mechanical behaviour. The ultimate objective is to devise a preliminary machine learning (ML) framework for composite structures that is capable of identifying and localising delaminations directly from measured strain and displacement fields.

The proposed methodology demonstrates the effective integration of experimental techniques and numerical modelling for the SHM of aerospace composite components. The employ of DIC, validated by strain gauges comparison, enables high-fidelity strain analysis under complex loading conditions, while the ML approach will offer a pathway toward automated, interpretable, and scalable defect assessment in next-generation aerospace structures.

Visual spectrum cameras have become increasingly popular for non-contact full-field structural dynamics measurements, enabling displacement and deformation identification through techniques such as Digital Image Correlation. However, obtaining strain information from kinematic measurements requires spatial differentiation, which significantly amplifies noise and necessitates known analytical relationships between displacement and stress—particularly challenging for complex geometries.
As an alternative, the thermoelastic principle exploits the reversible relationship between cyclic mechanical stress and surface temperature variations under adiabatic conditions. High-speed infrared cameras can directly measure stress-induced temperature oscillations, typically in the millikelvin range, providing full-field stress information without spatial derivatives.
For strain-based experimental modal analysis, hybrid identification methods combining thermoelastic measurements with high-dynamic-range reference sensors successfully extracted full-field strain mode shapes from Euler-Bernoulli beams.
Critical findings established that minimum surface stress amplitudes of approximately 100 kPa for aluminum and 500 kPa for steel, enable accurate modal identification. These thresholds provide essential design criteria for experiment feasibility assessment.
Thermoelasticity was extended to rotating structures by integrating computer vision techniques with infrared imaging. Infrared-compatible ArUco markers enabled motion tracking and compensation, successfully identifying out-of-plane structural dynamics in rotating beams despite in-plane rigid body motion. Sub-pixel transformation accuracy was achieved, with the frames-per-degree ratio emerging as a key performance metric across varying operational conditions.
For vibration-fatigue assessment, a novel thermoelasticity-based multiaxial criterion was developed. Since thermal cameras inherently measure the first stress invariant (sum of normal stresses), this criterion provides a natural multiaxial-to-uniaxial conversion for fatigue analysis. Experimental validation on Y-shaped specimens demonstrated accurate identification of critical damage locations when normal stresses dominate, proving highly effective for bending-dominated loads where normal stresses govern crack initiation. The methodology enables close-to-real-time, full-field fatigue damage assessment, opening new possibilities for structural health monitoring applications.

This paper introduces a vibration-guided virtual sensor network for high-resolution, markerless structural vibration measurement. Building on phase-based full-field motion estimation, we construct virtual sensor grids through vibration-informed superpixel segmentation, enabling efficient and scalable full-field structural health monitoring. Conventional pixel-level approaches often suffer from noise sensitivity and temporal inconsistency; to address these limitations, the proposed method aggregates pixel responses within each virtual sensor using a confidence-weighted strategy based on a phase nonlinearity–derived pixel-wise confidence metric. This approach preserves the spatial coherence of structural vibration patterns while providing robust and interpretable displacement estimates. Experiments on an industrial air compressor demonstrate that the virtual sensor network achieves high displacement accuracy and reliably detects subtle structural anomalies without the use of physical markers or speckle patterns. These results highlight the practicality of combining phase-based optical sensing with vibration-guided virtual sensor abstraction for non-contact, long-term, and field-deployable structural health monitoring.

Structural Health Monitoring (SHM) is fundamental for ensuring the safety, durability, and functionality of bridges, as it enables the assessment of their dynamic behavior and facilitates the early detection of potential damage. Conventional SHM systems commonly employ accelerometers, which, although accurate, suffer from several drawbacks: installation and maintenance can be costly and time-consuming, they often require wired connections, and displacement estimation derived from double integration of acceleration data is prone to numerical drift and error accumulation. To overcome these limitations, vision-based monitoring techniques have recently gained attention as a promising, non-contact, and affordable alternative. These systems allow direct measurement of displacements with minimal installation effort and reduced costs. Nevertheless, their accuracy and robustness, particularly when subjected to small environmental vibrations, and their ability to effectively capture the global dynamic behavior of large-scale structures and infrastructures remain active areas of research and validation. Within this framework, the present study evaluates the performance of a vision-based monitoring methodology designed to address key challenges in large-structure monitoring. The proposed procedure focuses on detecting small-amplitude vibrations, compensating for perspective distortions, reconstructing 3D displacements from single-camera videos, and identifying global mode shapes. The monitoring procedure comprises five stages. First, the monitoring framework is defined according to the specific application scenario and expected structural responses. This includes selecting appropriate cameras, lenses, and frame rates, as well as determining optimal camera positioning. Checkerboard-pattern targets are installed on the structure to convert 2D image displacements into 3D coordinates and to correct for camera movement. Second, camera calibration is performed to obtain intrinsic parameters, such as focal length and lens distortion. Third, feature tracking is conducted to follow distinctive points, typically the corners of checkerboard targets, across video frames. The Harris-Stephens algorithm identifies these corners, while the Kanade-Lucas-Tomasi algorithm tracks their motion, yielding accurate displacement data in pixel units. In the coordinate transformation phase, these 2D displacements are converted into real-world 3D coordinates using the Perspective-Three-Point method. This analytical approach ensures reliable point correspondence between the image and spatial domains. The fifth step involves expressing displacements in the structural reference system, aligning the calculated movements with the principal structural axes through a rotation and translation matrix. This ensures that the measurements accurately represent the physical deformation patterns of the structure. This methodology was tested on a reinforced concrete railway arch bridge. Videos were recorded using two unsynchronized consumer-grade cameras from different viewpoints, and checkerboard targets were attached to a transverse beam and selected arch hangers to monitor the structural response during and after train passages and identify natural modes. A detailed finite element model of the bridge was also developed and calibrated and adopted to simulate the bridge dynamic behavior under the train loads. The comparison between simulated and vision-based displacements demonstrates the effectiveness of the vision-based approach as a reliable tool for the structural health monitoring of structures and infrastructure.

Real-time Structural Health Monitoring (SHM) demands measurement techniques that are not only accurate but also computationally efficient enough to handle continuous data processing. While computer vision-based methods offer the significant advantage of non-contact remote sensing, the commonly used Digital Image Correlation (DIC) method often incurs high computational cost. This study presents a comparative evaluation of Phase-Based Optical Flow (PBOF) against DIC, aiming to validate PBOF as a superior alternative for operational monitoring where processing speed is a critical constraint.

The study follows a comprehensive two-phase approach, from controlled laboratory conditions to complex operational environments. The laboratory phase focused on establishing the algorithms' fundamental sensitivity. A shake table experiment was designed using a decreasing amplitude excitation. This specific methodology allowed testing the limits and signal-to-noise ratios of both PBOF and DIC at near-ambient vibration levels. By benchmarking these results against accelerometer measurements, the study confirmed the theoretical noise floors and accuracy of both methods under ideal conditions.

Crucially, the study progressed to a blind field validation, representing the core of this research. Acknowledging that laboratory success does not always translate to operational reliability, a blind test was conducted on the San Francisco-Oakland Bay Bridge (Western Span). High-resolution footage from an open-source, non-dedicated live-streaming camera was used to estimate the fundamental frequency of the suspension tower. This phase introduced real-world challenges such as atmospheric haze, variable lighting, and camera compression artifacts, which are absent in laboratory settings. The test's blind nature ensured an unbiased assessment of the algorithm's ability to extract meaningful structural dynamics from uncontrolled video sources.

The preliminary results of this comparative analysis reveal a distinct divergence in performance characteristics. While both methods demonstrated comparable accuracy in frequency identification, PBOF exhibited a substantial advantage in computational efficiency. The phase-based approach significantly reduced the processing time required to extract displacement time histories compared to the correlation-based method. This finding highlights the scalability of PBOF for real-time applications. The preliminary study shows that Phase-Based Optical Flow provides a balance of precision and speed required for the continuous, long-term frequency monitoring of large-scale civil infrastructure.