Fabian Vieltorf’s research while affiliated with Technical University of Munich and other places

Deep learning-based classification models show high potential for automating optical quality monitoring tasks. However, their performance strongly depends on the availability of comprehensive training datasets. If changes in the manufacturing process or the environment lead to defect patterns not represented by the training data, also called data drift, a model’s performance can significantly decrease. Unfortunately, assessing the reliability of model predictions usually requires high manual labeling efforts to generate annotated test data. Therefore, this study investigates the potential of intrinsic confidence calibration approaches (i.e., last-layer dropout, correctness ranking loss, and weight-averaged sharpness-aware minimization (WASAM)) for automatically detecting false model predictions based on these confidence scores. This task is also called model failure prediction and highly depends on meaningful confidence estimates. First, the data drift robustness of these calibration methods combined with three different model architectures is evaluated. Two datasets from the friction stir welding domain containing realistic forms of data drift are introduced for this benchmark. Afterward, the methods’ impact on model failure prediction performance is assessed. Findings confirm the positive influence of well-calibrated models on model failure prediction tasks, highlighting the need to look beyond classification accuracy during model selection. Moreover, transformer-based models and the WASAM technique were found to improve robustness to data drift, regarding the classification performance as well as obtaining useful confidence estimates.

Increasing demands on component properties are leading to the development of high-performance materials for which conventional production methods are reaching their limits from an economic and ecological point of view. In recent years, two technologies have been developed that show great potential compared to conventional machining processes, particularly in machining high-performance materials such as the titanium alloy Ti-6Al-4V. Ultrasonic-assisted machining leads to reduced cutting forces and increased tool life. Cryogenic minimum quantity lubrication prevents the occurrence of high machining temperatures and allows higher material removal rates without a negative impact on tool life. This paper shows the influence of ultrasonic-assisted milling and grinding processes in combination with cryogenic minimum quantity lubrication on the machinability of the high-strength materials Ti-6Al-4V and Zerodur. The investigation addressed cutting forces, tool wear, and surface roughness. The superposition of the technologies resulted in longer tool life and lower tool wear for both milling and grinding. However, the surface roughness was consistently higher due to the ultrasonic superposition. Nevertheless, machining with ultrasonic vibration-assisted cryogenic minimum quantity lubrication has great potential for difficult-to-machine materials, especially due to the reduction in tool wear.

Bei neuartigen Werkstoffen stoßen herkömmliche Fertigungsverfahren aus ökonomischer und ökologischer Sicht an ihre Grenzen. In Forschungsprojekten werden derzeit die Ultraschallbearbeitung und die kryogene Minimalmengenschmierung (kMMS) betrachtet, die sich bei schwer zerspanbaren Werkstoffen gegenüber konventionellen spanabhebenden Verfahren auszeichnen. Im Verbundprojekt „KryoSonic“ wird die Überlagerung beider Technologien bei Fräs- und Schleifprozessen untersucht.

In the work described here, Gaussian process regression was applied to predict the ultimate tensile strength of friction stir welds through data evaluation and to therefore avoid destructive testing. For data generation, a total of 54 welding experiments were conducted in the butt joint configuration using the aluminum alloy EN AW-6082-T6. Four tensile samples were taken from each of the 54 experiments and the resulting ultimate tensile strength of the weld seam segment was modeled as a function of the weld’s surface topography. Further models were created for comparison, which received either the process variables or the process parameters to predict the ultimate tensile strength. It was shown that the ultimate tensile strength can be accurately predicted based on the weld’s surface topography. Especially for low welding speeds, the correlation coefficients between the true and the predicted ultimate tensile strength were high. However, overall, even higher correlation coefficients could be achieved when providing the process variables or the process parameters to the model. In conclusion, it was shown that the developed Gaussian process regression model is a powerful approach for replacing destructive testing and for predicting ultimate tensile strength based solely on data that can be collected non-destructively.

Friction stir welding is a modern pressure welding process, which is particularly suitable for aluminum alloys. Several studies have been conducted to investigate the interrelations between the process parameters, such as the welding speed and the tool rotational speed, and the resulting mechanical properties of the joint. This study explored the connections between the surface topography of the welds, such as the flash height and the seam underfill, and their mechanical properties (ultimate tensile strength; elongation at break; and Vickers hardness). For this purpose, a total of 54 welding experiments at three different welding speeds were conducted using the aluminum alloy EN AW-6082-T6. The welded specimens were examined using visual inspection, topographic analysis, metallography, hardness measurements, and uniaxial tensile tests. Afterward, a statistical analysis was performed in order to determine the correlation coefficients between the surface topography and the mechanical properties of the welds. The strongest correlations were between the surface topography and the ultimate tensile strength. Thereby, the most pronounced relations were found between the seam underfill as well as the arc texture formation of the weld and its ultimate tensile strength. The interrelations between the surface topography and the elongation at break, as well as the hardness of the welds, were less pronounced. The higher the welding speed became, the less pronounced the interrelations were. The results show the potential of a non-destructive monitoring system based on the topography to support the prediction of the acceptability of welded parts.