Recent Submissions
Subpixel resolution analysis of porous media imagery
(2025) Hauf, Nicolas
The fracture network of a porous media image is interesting for a variety of reasons. It is essential for estimating material strength, elasticity, fluid interactions, and other metrics. In this thesis, 3D image data gathered by an X-ray CT scan of porous marble is denoised, and super resolution images are created. The latter was done because cracks in the fracture network are often of sub-pixel width, therefore needing a super resolution to be depicted accurately. Increasing the resolution of the X-ray CT scan itself is very costly, whereas using image processing to estimate higher-resolution images of similar quality is far cheaper.
To create denoised and super resolution images, two computer vision machine learning models based on the U-Net architecture were examined. These types of models have been improving rapidly and have proven capable for many image processing tasks. While other fields, for example, medical imaging, already use 3D image segmentation extensively, combining the analysis of porous media imagery with U-Net models operating on a 3D input space has not been done before. To achieve the best possible results, both model parameters were optimized to fit this specific task. This evaluation of the resulting images shows promising results, yielding highly accurate super resolution images using this machine learning approach.
Keywords: U-Net, XRCT, super resolution, porosity, machine learning
Verbesserung der Wiederverwendbarkeit von Simulationssoftware durch JupyterLab
(2025) Eble, Julian Valentin
Die Reproduzierbarkeit und Wiederverwendbarkeit von Forschungssoftware ist ein zentrales Thema in der heutigen Zeit. Es kann für manche zunächst einfacher erscheinen als bei physikalischen Experimenten [Boe15]. Jedoch können einem die verschiedensten Probleme begegnen. Das können beispielsweise Annahmen, Abhängigkeiten und Konfigurationen beinhalten, die innerhalb von Forschungsprojekten getroffen wurden. Aber auch die verwendeten Plattformen und eine schlechte Dokumentation, können Hürden darstellen. vor allem im Bereich Software Engineering wird innerhalb der Forschung sehr viel mit Algorithmen, verschiedenen Software-Tools und Frameworks gearbeitet. Diese Problematik beschränkt sich nicht nur darauf, sondern beschäftigt auch andere Bereiche, wie beispielsweise die klassischen Ingenieurswissenschaften, wie Bauingenieurswesen. Aus diesem Grund wird in dieser Arbeit untersucht, wie man, anhand von einem konkreten Forschungssoftware (𝐷𝑢𝑀𝑢𝑥1), das Problem lösen kann.
𝐷𝑢𝑀𝑢𝑥 ist eine Simulationssoftware für mehrphasige Strömungs- und Transportprozessen und basiert auf dem DUNE-Framework 2. Viele Forschungsprojekte verwenden das Framework, indem sie sogenannte 𝐷𝑢𝑀𝑢𝑥-Pub Module bauen, welche die jeweilige Simulation des Forschungsprojekt beinhalten und veröffentlicht werden 3.
Ziel dieses Projekts ist es, mit Hilfe von Docker-Images und JupyterLab 𝐷𝑢𝑀𝑢𝑥-Pub Module reproduzierbar und wiederverwendbar zu machen. Konkret bedeutet das, dass ein Prozess entwickelt wurde, welcher jedem Ersteller eines 𝐷𝑢𝑀𝑢𝑥-Pub Moduls zur Verfügung steht. Dies erstellt ein Docker-Image mit JupyterLab in welches das 𝐷𝑢𝑀𝑢𝑥-Pub Modul automatisiert eingebaut wird. Damit haben alle weiteren interessierten Menschen, die Möglichkeit auf einfach Weise die Simulationen zu reproduzieren und gegebenenfalls wiederzuverwenden.
Es wurden insgesamt über 58 𝐷𝑢𝑀𝑢𝑥-Pub Module untersucht, wie man sie reproduzierbar und wiederverwendbar machen kann. Letztendlich konnten von den bestehenden 𝐷𝑢𝑀𝑢𝑥-Pub Modulen 47 reproduzierbar und wiederverwendbar gemacht werden. Des Weiteren wurde ein Workflow entwickelt, der mit Hilfe von einem Bash-Skript und verschiedenen Templates, die Reproduktion und Wiederverwendbarkeit von neuen 𝐷𝑢𝑀𝑢𝑥-Pub Modulen fördert. Als Forscher hat man nun die Möglichkeit aus den neuen Projekten reproduzierbare Versionen in Form eines fertigen Docker Images und darin enthaltener Dokumentation zu erstellen. Zusätzlich existiert auch ein automatisierter Workflow für alle diejenigen, die ein gegebenes 𝐷𝑢𝑀𝑢𝑥-Pub Modul reproduzieren möchten4.
Comparing few-shot learning methods in various application scenarios for quality prediction in ultrasonic wire bonding
(2026) Buchner, Christoph; Riedle, Benjamin; Seidler, Christian T.; Huber, Marco F.; Eigenbrod, Hartmut; von Ribbeck, Hans-Georg; Schlicht, Franz
Few-shot learning refers to the problem of learning underlying patterns in data from just a few training samples, which contrasts with traditional deep learning methods that usually rely on large datasets. The collection of large datasets is typically costly and time-consuming and often requires significant computational resources. In ultrasonic wire bonding production processes, it is of fundamental importance that a high quality of the joints is ensured, while the application scenario often differs depending on the process, e.g. different materials, process sequences or machine settings. In this paper, we compare various few-shot learning approaches for predicting bond qualities in ultrasonic wire bonding. This is done by quantitatively predicting the shear force of the bonding joint across different application scenarios. The prediction is based on key process variables in the form of time-series data, such as deformation, ultrasonic power, frequency and current. These time-series vary in length depending on the bonding process. The few-shot learning approaches are compared in three application scenarios: Changing the bonding program (application scenario 1), changing the transducer (application scenario 2), and changing the bonding machine (application scenario 3). For example in application scenario 1, with just 15 retraining samples-less than 2.3% of the original training data-the shear force as a quality criterion for a 380 μm aluminum wire can be predicted with high accuracy. Using an autoencoder yields a mean absolute error of 90 cN, while meta-learning improves this to just 81 cN (less than approx. 5% of the average shear force).
Investigation of the dynamics of a versatile laser amplifier upon fast modulation of its CW and pulsed seed sources
(2026) Schmittner, Christian; Reiff, Colin; Hoßfeld, Max; Graf, Thomas; Abdou Ahmed, Marwan
This paper presents an investigation of the dynamics of a versatile thin-disk multipass laser amplifier (TDMPA) with two seed sources. The study focused on the system’s transient response, analyzing the amplified instantaneous power during and shortly after fast modulation of the individual seed powers. For switching times shorter than the amplifier’s characteristic response time, changes in the total seed power led to retardation and overshooting effects , which are attributed to the gain-saturation dynamics of the active laser medium. An analytical formula to express the dependence of the characteristic response time on the input and output powers and the specific TDMPA architecture was derived and validated. This provides a practical guideline for choosing seed-power switching times that avoid transient overshooting and retardation effects . For pump powers between 599 and 1446 W, and depending on operation conditions, the TDMPA under consideration was measured to exhibit a gain-saturation response time ranging between approximately 30 and 90 µs.
The geometric memory of quantum wave functions
(2025) Heinsdorf, Niclas; Metzner, Walter (Prof. Dr.)
Quantum geometry - including both the quantum metric and Berry curvature - arises from the non-zero overlap of well-defined eigenstates and their adiabatic evolution across the Brillouin zone. It has revolutionized condensed matter physics and material science by explaining quantum Hall effects, establishing the modern theory of polarization, and enabling a systematic search for topological materials on a large scale. Yet, despite these advances, we lack a framework for leveraging quantum geometry in the strongly interacting regime. Bridging this gap is critical: if we can harness geometric responses in correlated metals, we stand to engineer desirable transport and optical properties using existing material platforms, thus bypassing the complex task of designing materials with tailored quantum geometry from scratch. In this thesis, we take steps toward such a framework. First, we analyze the resilience of topological boundary modes in the presence of electronic correlations, identifying when interactions preserve, diminish, or destroy boundary modes. Second, we reveal geometric fingerprints of fluctuations in magnetically ordered systems, tying the electric quantum metric to the formation of instabilities and chiral quasi-particle excitations. Third, we generalize quantum geometry to describe families of many-body wave functions, providing a new algorithm to compute state-manifold curvatures suited to interacting phases. Our approach combines analytical theory with numerical methods, including density functional theory and tensor-network simulations, and is supported by open-source software developed during the PhD. Together, these results advance the topological classification of interacting phases and extend quantum geometry from single-particle bands to correlation functions, providing tools to design materials and devices with targeted geometric responses.
Exploring strong correlations in the calcium ruthenates using spectroscopic techniques
(2025) Suen, Cissy T.; Keimer, Bernhard (Prof. Dr. rer. nat.) and Damascelli, Andrea (Prof. Dr.)
The rich multiband physics of the calcium ruthenates gives rise to a complex free-energy landscape shaped by the strong interplay among spin-orbit coupling effects, electronic interactions, lattice distortions, and spin correlations. This competition produces a variety of emergent quantum phenomena with promise for future applications. For instance, beyond its Mott-insulating ground state, the single-layer Ca2RuO4 hosts exotic fluctuation modes of the antiferromagnetic state and an unconventional non-equilibrium metallic state induced upon application of an electrical current. Meanwhile, the bilayer Ca3Ru2O7 exhibits pseudogap behaviour and an incommensurate cycloidal magnetic phase arising uniquely from the Dzyaloshinskii–Moriya interaction.
By simultaneously employing transport measurements with angle-resolved photoemission spectroscopy (ARPES), this work reveals how the electronic band structure evolves across the current-induced insulator-to-metal transition in Ca2RuO4. A reduction of the insulating band gap and a modification of the Ru bands are observed in the current-induced L phase. Landau–Ginzburg free-energy analysis indicates that current flow imposes a directional anisotropy that breaks the symmetry between the out-of-plane orbitals. Thus, while the insulating and metallic phases are thermodynamically equivalent, they host distinct orbital populations. Understanding the differences and triggers for each state provides new avenues for Mottronic devices, such as energy-efficient electronic switches.
Historically, ARPES measurements have avoided electromagnetic fields due to their effect on the outgoing trajectory of the emitted photoelectrons. This manifests as a shift in energy and momentum and a broadening of the recorded spectra. While technological developments such as nano-ARPES mitigate the latter issue, transport-ARPES has been confined to the study of materials with clearly defined features in energy, such as the d-wave node in superconducting cuprates or the Dirac point in graphene. By using core level spectra as the energy reference, not only can transport-ARPES be extended to any ARPES suitable material, careful observations of the core level energies and widths can provide further information about the material under current.
Meanwhile, Raman measurements in an external magnetic field reveal an unusually large field-induced energy shift of a phonon mode in Ca3Ru2O7, pointing to exceptionally strong spin–phonon coupling in a regime where the Dzyaloshinskii–Moriya interaction plays a dominant role. While even larger couplings have been observed in some 5d compounds such as iridates and osmates, this is the largest reported value among the ruthenates, and appears to be strongly influenced by the incommensurate cycloidal magnetic phase. Additionally, the strength of this coupling underscores the delicate interplay between the different degrees of freedom in the system, making these materials highly tunable by small perturbations. For instance, substituting 1% Ti switches the magnetic ground state from A-type to G-type antiferromagnetic order and applying < 1% compressive strain can shift the insulator-to-metal transition by more than 70K.
This thesis therefore integrates external electromagnetic fields with advanced spectroscopic probes in order to probe regions of the ruthenate free-energy landscape not previously explored by conventional techniques. A complex landscape is uncovered, illustrating how these materials can be finely tuned with current, field, doping, or strain, offering new insights into the underlying physics of 4d systems and opening opportunities for their incorporation into future quantum devices. Additionally, with the developments in transport-ARPES presented by this work, this thesis builds upon our ever growing capability to look at strongly correlated electron systems in operando.
Assessment of methods and strategies for lunar dust mitigation experiments within a low-fidelity test environment
(2025) Gewehr, Moritz; Al-Barwani, Amran; Bölke, Daniel; Klinkner, Sabine
Lunar regolith is identified as one of the greatest challenges for future lunar surface missions. With a rising number of planned surface exploration missions within the next decades, the significance of technologically robust, dust-tolerant solutions and lunar dust mitigation strategies have never been more present. Within this context, the University of Stuttgart’s Institute of Space System is investigating technical solutions and methods for active and passive dust mitigation strategies. A low-fidelity test environment was set up in order to provide the ability to characterise adhesion forces and abrasive effects of lunar simulant particles on multiple technical surfaces. Microscopic imaging is used in order to quantify remaining particle depositions after several experiment series. The test campaigns presented in this paper include three specific experiments: dust particle adhesion characterisation on a specific surface type by centripetal force measurements (1), dust-surface cleaning and abrasion tests with varying types of brushes (2), as well as magnetic cleaning methods (3). A developed optical particle detection algorithm is being used in situ to the experiment to resource efficiently specify the amount and size of remaining particles to derive data on the respective adhesiveness. The low-fidelity test environment is seen as a very efficient solution for first precursor tests and experiment series with varying boundary conditions. The control and variation of sample surface substrates, lunar analogue types, grain sizes, and some environmental conditions allow a high variation of different testing scenarios in order to characterise first impact factor dependencies. This paper describes the general setup of the developed test environment, the specific experiments, methods and results, as well as the lessons learned.
Development of a capability-based scheduling heuristic for matrix production systems
(2026) Dechant, Stefanie; Möhring, Hans-Christian
The growing diversity of product variants, accelerated technology shifts, and shorter life cycles demand highly flexible and efficient production systems in the automotive industry. Matrix-based assembly systems offer a promising response by enabling modular workstation configurations and variant-specific task execution. A key challenge lies in designing such systems for optimized task-to-resource scheduling under fluctuating workloads and varying capability requirements. This paper presents a capability-based scheduling logic tailored for matrix production environments, which incorporates heuristic rules for task assignment under constraint-based planning conditions. The algorithm is formally described through a mathematical model that defines the objective function, constraints, and decision variables, supported by prioritization heuristics. To ensure transparency and reproducibility, the scheduling logic is also visualized in a Unified Modeling Language (UML) activity diagram. A case study using real-world data from the electric vehicle final assembly of a leading OEM validates the method. Results show notable improvements in station utilization and system adaptability compared to conventional line-based configurations. The proposed scheduling framework contributes to the development of agile and human-centered production control architectures, supporting the transition toward flexible, high-variant manufacturing.
Self-consistent transformation of first-, second-, and third-order potential gradients among Cartesian, cylindrical, and spherical coordinates
(2026) Deng, Xiao-Le; Sneeuw, Nico
Knowing how to transform potential gradients between different coordinate systems is of fundamental importance in potential field theory. For first- and second-order gradients, such transformations are conventionally dealt with in terms of vector-matrix notation. However, matrix notation is not helpful for deriving the expressions for transformation of third-order potential gradients. In this contribution, we derive the general detailed expressions for transformation of first-, second-, and third-order potential gradients between two coordinate systems by using the direct expansion method. As examples of these general expressions, we derive detailed expressions for forward and inverse transformations of physical components of first-, second-, and third-order potential gradients among Cartesian, cylindrical, and spherical coordinates. Laplace’s equation has been applied for a first validation of partial expressions. However, to validate all newly derived expressions in a systematic way, we propose the closed-loop transformation cycle method that presents a full-fledged commutative diagram of forward and backward transformations among all three coordinate systems, i.e., potential gradients can become themselves after the closed round-trip transformation cycle among Cartesian, cylindrical, and spherical coordinates. Results reveal that this transformation cycle method confirms the correctness of all derived expressions. These general expressions for transformation of first-, second-, and third-order potential gradients can be applied under arbitrary two coordinate systems, and their detailed expressions can be systematically validated by the proposed transformation cycle method.
Which spaces can be embedded in reproducing kernel Hilbert spaces?
(2025) Schölpple, Max; Steinwart, Ingo
Given a Banach space E consisting of functions, we ask whether there exists a reproducing kernel Hilbert space H with bounded kernel such that E⊂H. More generally, we consider the question, whether for a given Banach space consisting of functions F with E⊂F, there exists an intermediate reproducing kernel Hilbert space E⊂H⊂F. We provide both sufficient and necessary conditions for this to hold. Moreover, we show that for typical classes of function spaces described by smoothness there is a strong dependence on the underlying dimension: the smoothness s required for the space E needs to grow proportional to the dimension d in order to allow for an intermediate reproducing kernel Hilbert space H .