Open Access

Digital transfer documents that are machine-readable – and ideally, machine-interpretable – offer a promising route for automating processes that require seamless digital data transmission. To ensure interoperability on both, the issuing and receiving sides, it is crucial to adopt harmonized solutions when transitioning from analogue-based to fully digital calibration certificates. This shift necessitates that the metrological communities establish agreed-upon best practices and guidelines for implementing these digital assets. This article describes how the specification of data formats and terminology for digital calibration certificates facilitates machine-interpretability and automation in metrological traceability and how the German Calibration Service (DKD) elaborates and reveals these harmonized solutions in comprehensive committee activities. With the establishment of these specifications, quality assurance in measurement technology will finally become more fast, easy, safe and affordable.

ISO 376 is a globally established calibration standard for uniaxial force transducers. It regulates the handling, measuring procedure, raw data processing, calibration function determination as well as the assessment of measurement uncertainty and item classification. The DCC task group of the DKD’s technical committees for force, acceleration and acoustics and for torque defined good practice rules for DCCs in the scope of ISO 376 to enable interoperable certificates in force metrology. This talk introduces a specific DCC realization for a 1,000 kN tensile force transducer, highlighting several advanced features that are hardly covered by other model implementations so far, i.e. multiple measurement series with varying loading sequences and mounting positions, embedded balancing functions and coefficients, solitary relative measurement uncertainties and load-specific item classifications. Many of those aspects are also relevant for DCCs from other metrological communities, that may adopt the approaches that are recently harmonized for the quantity of force.

Fatigue assessment of welded joints is a key aspect in the design and life-time evaluation of offshore structures such as jacket foundations. Conventional fatigue design based on S-N curves provides a robust basis for structural design but is of limited applicability for remaining life assessment, as the presence and growth of fatigue cracks cannot be explicitly considered. In inspection-based assessments, cracks below the detection limit must often be assumed to exist in the structure, requiring methods that allow the prediction of crack propagation and residual fatigue life. In contrast, approaches based on linear elastic fracture mechanics provide a suitable framework for such assessments, although predicted fatigue life depends strongly on modelling assumptions and input parameters. In this contribution, four analytical fatigue crack propagation approaches are compared using cruciform welded joints with a plate thickness of 25 mm manufactured from S355 structural steel. High-resolution 3D laser scanning is used to capture the as-built weld geometry, providing the geometric input for all approaches considered. The methods include a BS 7910-based procedure using weld toe magnification factors obtained for T-joints, two alternative formulations considering the local weld effect by stress concentration factors or cruciform-specific geometry functions reported in the literature, and a method based on the IIW recommendation for cruciform welded joints. The fatigue experiments are performed on specimens welded in-house, with extensive digital data recorded during the manufacturing process. Crack initiation and propagation are monitored using strain measurements, optical techniques, and beach marks. Strain gauges and crack luminescence provide real-time monitoring of crack initiation and propagation, whereas beach marks serve for post-hoc spatial calibration. The experimental results enable a specimen-specific comparison of predicted crack growth and fatigue life, accounting for both weld geometry and manufacturing-related influences. The results illustrate the influence of modelling assumptions, geometric representation, and manufacturing conditions on fatigue life predictions and highlight the advantages and limitations of different linear elastic fracture-mechanics-based assessment approaches for cruciform welded joints.

Hazardous gas leaks and air pollution are major environmental risks, yet their sources often remain hidden, inaccessible, or outside the spatial coverage of existing monitoring networks. Mobile Robot Olfaction (MRO) – and its airborne extension, Aerial Robot Olfaction (ARO) – addresses this gap by combining robots, gas sensors, and environmental sensing for autonomous, real-time measurement. In the field, however, turbulent and intermittent gas transport together with slow, drifting sensors mean that methods developed under controlled conditions often need substantial adaptation before they can be used operationally. No single platform optimizes data quality, spatial coverage, and temporal coverage at once: fixed networks cover time but are spatially sparse and costly; ground robots provide high-quality local measurements but cannot cover large areas; aerial platforms offer rapid, flexible access but are limited in endurance and localization accuracy; and dense, low-cost sensing trades data quality for spatial reach. Heterogeneous sensing systems respond by combining complementary platforms, compensating for their individual limitations, and fusing the measurements into coherent maps of gas and dust distributions. As an overview talk, this keynote introduces the concepts of MRO and ARO, heterogeneous sensing, gas tomography, and passive smart-dust chemosensing, and explains how they contribute to real-world detection, mapping, and localization tasks. It shows how they fit together through recent work ranging from completed deployments to ongoing validation. A stationary low-cost network in an operating steel hot-rolling mill, combined with mobile reference instruments, was used to assess how representative conventional measurement campaigns are. Building on this, multimodal, AI-supported monitoring of several gases is now being tested at industrial sites, where ground robots and drones validate and recalibrate fixed sensor networks. Ground-based gas tomography for spatially resolved imaging has been validated under controlled conditions, with outdoor methane-release trials and an aerial extension as next steps. Finally, passive, power-free chemosensors read out by standard camera drones have been field-tested for low-cost detection of hazardous regions, while gas-phase detection remains future work. Together, these examples show that added value comes from combining platforms and validating the resulting systems under realistic conditions, rather than from any single robot or sensor. Two challenges currently limit routine use: reliable gas measurement from moving platforms, especially aerial ones, and gas source localization in turbulent flow. These are joined by practical constraints such as positioning in GNSS-denied environments, platform endurance, and long-term field validation. The keynote argues that reproducible operation in real deployments is required before integrated robotic olfaction systems can be used routinely in environmental protection and industrial safety.

Air pollution in industrial environments poses significant risks to human health and the environment. Yet, current monitoring practices rely on regulatory gradeinstruments that are costly and therefore deployed sparsely in space and time, limiting the ability to detect local exposure hotspots and short-term pollution events. This thesis addresses these limitations by integrating a heterogeneous sensor network with deep learning-based gas distribution mapping for high-resolution air quality monitoring in industrial environments. The first part of the thesis investigates the potential of heterogeneous sensor networks, composed of stationary low-cost nodes and mobile robotic platforms. Although low-cost sensing technologies are becoming more popular, their performance under long-term, high-stress conditions remains poorly understood. To fill this gap, three field studies were conducted in industrial environments, including a ferry car deck and a hot rolling mill, spanning up to 19 months of deployment. These deployments were used to analyze pollution event detection, long-term performance, and the complementarity of mobile and stationary sensing. The findings show that low-cost sensors can be highly valuable for pollution monitoring, especially when complementing high-precision instruments within a heterogeneous strategy. Building on this, the second part of the thesis formulates gas distribution mapping as a super-resolution problem: predicting high-resolution pollutant fields from sparse sensor inputs. Prior work has rarely explored how domain knowledge about gas dispersion, learned directly from data, can be embedded into predictive models. In response, this thesis develops two deep learning frameworks capable of inferring gas dispersion patterns from sparse sensor data without relying on additional wind measurements as input. Both models were trained on a synthetic dataset generated from real wind conditions and outperform traditional approaches, achieving over 30% better reconstruction in terms of root mean squared error on synthetic test data. A key contribution lies in demonstrating the transferability of a graph-based model from simulation to real-world deployment. By representing sensor data as a graph, the model architecture enables flexible adaptation across different deployment scenarios. The pre-trained model was fine-tuned using field data from the ferry car deck and hot rolling mill, demonstrating its practical viability under real-world conditions. In sum, this thesis brings together sensing hardware, field deployment, and deep learning to address the challenge of air quality monitoring in industrial settings. It advances the state of the art by demonstrating how heterogeneous sensor networks can be effectively deployed in real-world environments and how deep learning can be utilized for more precise gas distribution mapping.