Conference Agenda

Civil infrastructures are increasingly aging, and reliable condition assessment is essential to ensure their safe and continuous operation. Structural Health Monitoring (SHM) systems have been developed to support this task, with automatic damage detection being one of their primary objectives. However, effective detection is often hindered by measurement uncertainties and low sensitivity of the monitored responses to damage. Since installing SHM systems is costly, it is crucial to assess their detection capability before deployment and prior to the occurrence of damage.

To address this need, several approaches have been proposed to predict SHM reliability and estimate the minimum detectable damage using prior information about the structure and the SHM system. A major limitation of existing methods is that they provide accurate predictions only when the damage detection problem is linear or mildly nonlinear.

This contribution presents a solution to overcome this limitation. We introduce a method capable of predicting the minimum detectable damage even for nonlinear problems. The approach is based on a fully probabilistic formulation of the damage detection process, enabling not only the detection but also the quantification of damage. As a proof of concept, the method is applied to a nonlinear case study, and its ability to predict damage detection reliability is evaluated and compared against existing approaches.

The openLAB is a 45 m long, three-span semi-integral research bridge near Bautzen, Germany, constructed from prestressed concrete girders. It serves as a benchmark platform for evaluating and comparing structural health monitoring (SHM) systems through controlled experiments. This paper presents the experimental setup, procedure, and selected results from load tests conducted in May 2025, focusing on static deformation up to the ultimate limit state (ULS) under a concentrated load of 400 kN, inducing a maximum deflection of 60 mm. The structural response was monitored using interdisciplinary methods – e.g., laser triangulation sensors (LTS), tilt sensors, robotic total station (RTS), unmanned aerial vehicle (UAV) photogrammetry, and fiber optic sensing – with strong agreement among the methods. Finite element (FE) models, developed to support test preparation, showed significant variability, highlighting sensitivity to modeling assumptions. All data – comprising FE models, environmental conditions, geodetic measurements, UAV photogrammetry, crack documentation, and fiber optic sensor readings – are openly accessible, providing a rich, multi-source dataset for future SHM research.

Metallic pipes systems are widespread in industry, with applications ranging from long-distance transport (e.g., pipelines) to in-factory installations (e.g., food, chemical, and pharmaceutical industries). These pipes, typically made of steel, are prone to damage — primarily internal corrosion — which can locally reduce pipe thickness to the point of failure, resulting in costly and hazardous leaks. However, maintaining these pipes is highly challenging. They are often difficult to access (e.g., submerged pipelines or pipes in crowded factory environments), and internal corrosion cannot be visually assessed. Current maintenance policies rely primarily on manual, non-destructive inspections, which are not yet automated. In this context, developing reliable Structural Health Monitoring (SHM) solutions capable of continuously assessing the integrity of such structures represents a major step forward toward predictive maintenance and increased operational safety. Automating the inspection of metallic pipes and continuously monitoring damage from its inception would enable significant cost savings, enhanced safety, and improved prognostic capabilities. Therefore, automated SHM of metallic pipes encompassing damage detection, localization, and quantification, is the focus of this study.

To achieve this, ultrasonic guided wave-based technologies are highly promising. These technologies involve transmitting ultrasonic guided waves through the pipe system using a ring of piezoelectric transducers (PZTs) and monitoring the received waves at another location using a second PZT ring. Non-destructive techniques using ultrasonic guided waves have already demonstrated excellent results in damage detection and shape reconstruction, but they remain manual. These techniques require manual installation and removal of PZT rings to inspect specific pipe sections. To move toward fully autonomous SHM systems, pipes must be pre-equipped with PZT rings and associated algorithms capable of meeting defined industrial requirements for damage detection, localization, and quantification.

An experimental campaign was conducted at CETIM, where a 2-meter pipe segment was equipped with two rings of 24 PZT transducers each. Two progressive internal damages were introduced into the pipe, and the pipe was repeatedly interrogated using ultrasonic guided waves. Based on those experimental data, the performances of several deterministic and probabilistic SHM algorithms were evaluated using dedicated metrics where probability of detection, probability of localization, and size estimation error, are function of the number of transducers in the ring. These findings could allow to benchmark ultrasonic-based SHM algorithms together and with other approach based on tomographic reconstruction, to optimize the number of PZT elements needed to reach given industrial requirements, and to facilitate the wider adoption and deployment of SHM applications for metallic pipes.

Structural health monitoring (SHM) methods can be applied to ensure the safety and longevity of engineering structures. The process of SHM can be divided into a hierarchical level approach (detection, localisation, quantification, and prognosis). While the first level (detection) can be achieved using purely unsupervised, data-driven SHM approaches, the subsequent second level (localisation) typically requires the use of structural models or labelled data.
In the case of model-based SHM approaches, certain parameters of the structural model are continuously updated to reflect the dynamic behaviour of the structure under investigation. This is typically achieved by employing an optimisation procedure that minimises the discrepancy between the measured and modelled dynamic responses, e.g., by comparing their modal properties. Local reductions in model stiffness parameters can indicate potential damage locations.
However, not every detected damage can be accurately localised, as uncertainties in both the input data and the structural model, along with ambiguities in the optimisation problem itself, may lead to inconclusive or false localisation results. Therefore, in this work, a post-processing method inspired by the model-assisted probability of localisation is presented, which quantifies the degree of trust. This value is intended to quantify to what extent the localised damage can be reliably localised under the uncertainties present. The proposed method is applied to a laboratory steel cantilever beam with different mass positions and weights for damage localisation purposes. The results show that the proposed trust metric correlates with the quality of the localisation outcome, demonstrating its potential to support decision-making in model-based damage localisation applications.

Advances in robotics have led to autonomous platforms that convey sensor packages used for reconfigurable (mobile) monitoring solutions. In particular, ultrasound has rapidly evolved in this modality, where in-situ advanced autonomous manufacturing processes are outfitted with ultrasonic interrogation for near real-time part qualification. In the application motivating this current work, swarms of simple, low-cost robots are anticipated to monitor the interior of pipes with an ability to communicate to reconfigure the array for adaptive, optimal inspection. This work considers optimality to be the maximum probability of detection (POD) achievable by an array of robots, each outfitted with a single transmission/receive piezoelectric element capable of pulse-echo measurements only to minimize communication bandwidth among the mobile array robots. The fundamental objective is then to develop an initial framework for understanding and quantifying the performance of such reconfigurable arrays in terms of their ability to detect a target defect class characterized by its expected scattering behavior. A further assumption is to study array geometries that tesselate, since such geometries may be readily scaled to large structural areas while maintaining coverage.

The work first derives a generalized POD model for a point defect characterized by its scattering matrix (orientation, size) and distance from a single transducer. The model is then tested and validated on an experiment with a manufactured defect in an aluminum plate. With the validated model, this work proceeds to build basic arrangements of transducers to investigate array-level POD performance by fusing the information provided by each individual transducer in the network. The work studies how topological array parameters (e.g., transducer pitch, overall array shape, and array depth) affect global POD performance. The overall framework, while quite generally applicable, will be discussed within the context of a highly directional scatterer (e.g., a crack) as a “worst case” defect example and may be used to inform reconfigurable array design.

Some conclusions drawn from the study are: (i) highly specular defects (like a larger crack, with characteristic length larger than half the wavelength of the ultrasonic waves used, l/2) demand significant angular diversity, achieved by adding network layers; (ii) array pitch plays an important role, with smaller pitches consistently performing best as the distance dependence dominates angular dependence, but likely at the cost of much larger array sizes; and (iii) array shape (arrangement of transducers) plays a moderate role, e.g., minimal average distance between transducers and defect locations in triangular shapes maximizes POD for smaller defects, but hexagonal shape angular diversity becomes more important for larger defects. This study is fundamentally a design tool, and future work will look at different objective functions, including mapping POD performance to cost in order to include transducer/robot costs and path planning for a global cost optimization.