Voestalpine (Austria)

Eddy current testing is a widely used approach for the non-destructive evaluation of electrically conductive materials. The accuracy of material parameter estimation depends on the sensitivity of the measured signals to the properties of interest, as well as on the impact of noise and other disturbances. Challenges arise in independent and unique parameter estimation when multiple parameters affect the signals in similar ways, reducing identifiability. This paper introduces an efficient computational framework to assess the estimability and identifiability of parameters in eddy current testing, with a focus on metallic coated steel sheets. The approach utilizes stochastic and deterministic analytical models, as well as finite element simulations, to quantify this uncertainty.

In this work a multi-criteria analysis and an optimization tool were developed, which allows the substitution of fossil-based solvents with bio-based alternatives based on Hansen solubility parameters and various physical parameters, such as the boiling point, evaporation rate, viscosity or wetting behavior. The proof of concept was achieved by formulating two different paints used in coil coatings using the bio-based solvents, and they performed equally as well as their fossil-based counterparts. A potential decrease in CO2 emissions was determined by a life cycle assessment and cradle-to-grave analysis of bio- and fossil-based solvents, which showed a large sustainability bonus when using solvents based on biomass. The introduced methodology provides initial insights into substituting currently used solvents systematically. Overall, implementing bio-based solvents is a viable drop-in method to decrease the environmental impact of paints and coatings, while maintaining the same performance.

Electromagnetic acoustic resonance (EMAR) is a well-established non-contact method for ultrasonic thickness measurement, typically operated at frequencies above 1 MHz using an electromagnetic acoustic transducer (EMAT). This study successfully extends EMAR into the sub-MHz range, allowing supply voltages below 60 V and thus offering safer and more cost-effective operation. Experiments were conducted on copper blocks approximately 20 mm thick, where a relative thickness accuracy of better than 0.2% is obtained. Regarding this result, the research identifies a critical design principle: Stable thickness resonances and subsequently accurate thickness measurement are achieved when the ratio of ultrasonic wavelength to EMAT track width (λ/w) falls below 1. This minimizes the excitation and interactions with structural eigenmodes, ensuring consistent measurement reliability. To support this, the study introduces a system-based model to simulate the EMAR method. The model provides detailed insights into how wave propagation affects the accuracy of EMAR measurements. Experimental results align well with the simulation outcome and confirm the feasibility of EMAR in the sub-MHz regime without compromising precision. These findings highlight the potential of low-voltage EMAR as a safer, cost-effective, and highly accurate approach for industrial ultrasonic thickness measurements.

As a fundamental challenge in unsupervised learning, the identification of outliers in homogeneous populations has received considerable attention over the years. In practice, however, the data at hand stem from diverse processes, making it difficult to differentiate an anomaly from a valid signal. In the case of functional data, in particular, the definition of an outlier is no longer trivial. We propose a method using functional depth as an indicator of centrality, applied to a real industry problem in steel production. A set of potential outlier candidates is identified using functional depth, subsequently clustered and labeled. These labels are then used to expand to the entire dataset using semi-supervised learning techniques. Our method is rigorously compared to a manual labeling of the data as well as a classic functional clustering approach using scores and graphical investigation, showing that using functional depth to narrow down the set of potential outliers allows for more concise and easily identified anomaly categories.

Abstract Metal-cored (MCW) and flux-cored (FCW) wires are widely used for welding low-carbon (C) steels in structural applications, offering higher productivity but generating more welding fumes than solid wires. Growing awareness of the health risks associated with respirable particles has intensified the focus on fume composition. While proper fume extraction and personal protective equipment (PPE), such as helmets with breathing apparatus, are essential, modifying wire composition can help reduce harmful emissions. Manganese (Mn), a key alloying element in C-steels, and their corresponding welding consumables have been linked to neurological effects, prompting stricter occupational exposure limits (OELs) to protect welders. This has driven demands for low-Mn-emission welding consumables. This study evaluates two newly developed seamless low-Mn-emission products, one MCW (MCW-LMn) and one FCW (FCW-LMn), by comparing their fume and Mn emission rates (Mnmg/s) to those of standard seamless and folded wires under varying welding parameters and shielding gas compositions. At equivalent parameter settings, the newly developed wires produced comparable fume levels but achieved Mn emission reductions of 25–70% for MCW-LMn and 60–85% for FCW-LMn.

Attributing uncertainties to the input space elevates the trustworthiness and explainability of machine learning applications. This paper proposes a novel method called Smoothness Constrained Attribution (SCA), which uses the uncertainty propagation mechanism to propagate the output uncertainty back to the input space. This input uncertainty attribution relies solely on test-time data, the trained uncertainty-aware Machine Learning (ML) model, and assumes a smooth input space, resulting in an efficient and simple system. SCA is compared to existing input Uncertainty Attribution Mechanisms (iUCAMs) based on eXplainable Artificial Intelligence (XAI) and an oracle reference using heteroscedastic noise in different synthetic datasets. These evaluations demonstrate the robustness and improvements of SCA compared to existing methods.