Eldorado - Repository of the TU Dortmund
Resources for and from Research, Teaching and Studying
This is the institutional repository of the TU Dortmund. Ressources for Research, Study and Teaching are archived and made publicly available.
Communities in Eldorado
Select a community to browse its collections.
Recent Submissions
Explainable and interpretable time series forecasting and predictive maintenance
(2025) Jakobs, Matthias; Liebig, Thomas; Gama, João
Time-series data is ubiquitous in numerous application domains today, including safety-critical settings such as medical and industrial scenarios.
However, real-world data can change its characteristics with time, which is referred to as drifts in concept.
The phenomenon of concept drifts is particularly troubling for time-series forecasting tasks, where the goal is to predict future time-series values given known, past values.
Traditional forecasting methods assume that these characteristics are fixed to achieve a high predicting performance.
In this thesis, we will inspect methods how to adapt to changing concepts for time-series forecasting using both complex, deep learning methods and simple, interpretable models.
To do that, we utilize the online model selection and online ensemble pruning frameworks for selecting between pools of different models in an online manner.
Specifically, we will focus on aspects of explainability and interpretability throughout as these aspects are crucial for applying forecasting methods in practice.
Additionally, we investigate a real-world scenario for predictive maintenance, one area of application where explainable and interpretable time-series analysis methods are crucial to gain insights for practitioners and technicians alike.
Towards understanding and avoiding limitations of convolutions on graphs
(2025) Roth, Andreas; Liebig, Thomas; Kriege, Nils Morten
Many challenging tasks involve data with an inherent graph structure. Examples of such graphs include molecules, road networks, and transaction records. As such data becomes increasingly available, the need for methods that can effectively leverage graphstructured data grows accordingly. While message-passing neural networks (MPNNs) have shown promising results, their impact on real-world applications remains limited. Although various phenomena have been identified as possible causes that limit the performance of MPNNs, their theoretical foundations remain poorly understood, leading to fragmented research efforts. In this thesis, we provide an in-depth theoretical analysis and identify several key properties limiting the performance of MPNNs. Building on these findings, we propose several frameworks that address these shortcomings.
We identify and analyze two properties exhibited by many MPNNs: shared component amplification (SCA), where each message-passing iteration amplifies the same components across all feature channels, and component dominance (CD), where a single component gets increasingly dominantly amplified as more message-passing steps are applied. These properties lead to the observable phenomenon of rank collapse of node
representations, which we identify as a generalization of the established over-smoothing phenomenon. By generalizing and decomposing over-smoothing, we enable a deeper understanding of MPNNs, more targeted solutions, the identification of the relevance of each property, and more precise communication within the field.
To avoid SCA, we show that utilizing multiple computational graphs or edge relations is necessary. We propose the multi-relational split (MRS) framework, which transforms any existing MPNN into a variant that employs multiple edge relations. We identify edge splitting properties that enable the resulting MPNN to avoid SCA. Additionally, we define the spectral graph convolution for multiple feature channels (MIMO-GC), which naturally uses multiple computational graphs. We propose localized MIMO-GCs (LMGC) as a framework for MPNNs that approximate the MIMO-GC and inherit its beneficial properties.We show that LMGCs can avoid SCA and be injective on multisets.
To address CD, we demonstrate the close connection between MPNNs and the Page-Rank algorithm. This connection allows us to transfer insights and modifications from PageRank to MPNNs. Based on personalized PageRank, we propose a variant of MPNNs that similarly allows for infinitely many message-passing iterations, while preserving initial node features.
Collectively, these results contribute to a more complete theoretical understanding of MPNNs, allowing future research to be more targeted. Our findings also establish frameworks for constructing MPNNs that exhibit favorable properties.
Silver-based grating structures for label-free biomolecule detection
(2025) Sarapukdee, Pongsak; Palzer, Stefan; Wöllenstein, Jürgen
Addressing global healthcare challenges, including emerging diseases, antibiotic resistance, and environmental threats, requires innovative diagnostic tools. Biosensors based on label-free detection have gained attention for their rapid, sensitive, and cost-effective analysis. Among them, Surface Plasmon Resonance (SPR) stands out as a powerful technique for real-time biomolecular interaction monitoring. However, traditional prism-based SPR systems face limitations in integration and portability, spurring interest in grating-based alternatives.
This thesis explores the theoretical, computational, fabrication, and experimental aspects of grating-based SPR systems for label-free biosensing. It introduces key principles including surface plasmon polaritons (SPPs), and grating-coupling mechanisms. A literature review covers recent advances in plasmonic materials, microfabrication methods, and performance optimization.
Finite-Difference Time-Domain (FDTD) simulations are used to design grating structures, evaluate refractive index sensitivity, and validate results experimentally. Gratings are fabricated using electron beam lithography and deposition on Complementary Metal-Oxide-Semiconductor (CMOS)-compatible substrates, with microchannel integration enabling real-time biosensing. Experimental results show high sensitivity and strong agreement with simulation.
The study also investigates two-dimensional (2D) and gradient grating period (GGP) structures for multiplexed sensing and evaluates durability enhancements like protective coatings. Inverted gratings on glass offer environmental resilience but exhibit reduced coupling efficiency.
In summary, this work advances grating-based SPR sensor technology through modeling, fabrication, and experimental validation. The findings support their potential for compact, sensitive, and versatile biosensing, with future integration into semiconductor-based systems promising further miniaturization.
Versuchstechnische und rechnerische Untersuchungen einer kreislaufgerechten Stahl-Holz-Verbunddecke im Kontext einer nachhaltigen Tragwerksplanung
(2025) Krinitzki, Christopher; Hartz, Christian; Schmid, Volker
Die vorliegende Arbeit umfasst die Konzeption und systematischen experimentellen, numerischen und umweltbezogenen Untersuchungen einer Stahl-Holz-Verbunddecke aus Stahltrapezprofilen mit darauf befestigten Brettsperrholzplatten, um deren Potenzial im Hinblick auf eine nachhaltige Tragwerksplanung sowie die erforderliche Dekarbonisierung des Bausektors zur Erreichung der Klimaschutzziele einordnen zu können.
In diesem Zusammenhang wird zunächst die Eignung unterschiedlicher reversibler Verbindungsmittel zur Fügung der beiden Teilquerschnitte auf Basis des Herstellungsaufwandes, der Verbundsteifigkeit sowie des Versagensmechanismus mithilfe von Abscherversuchen evaluiert. Dabei stellen sich für die Ausbildung der Stahl-Holz-Verbundfuge senkrecht eingebrachte Sechskant-Holzschrauben als adäquater Kompromiss zwischen einem geringen Herstellungsaufwand, einer hohen Anfangssteifigkeit und einem duktilen Versagensverhalten sowie vorteilhaft gegenüber Blechbohrschrauben und geneigt eingebrachten Sechskant-Holzschrauben heraus.
Aufbauend auf der Evaluation geeigneter Verbindungsmittel wird das Biegetragverhalten der Stahl-Holz-Verbunddecke auf Grundlage von 4-Punkt-Biegeversuchen mit 5,0m langen Probekörpern für zwei verschiedene Auflagersituationen untersucht. Sowohl die auf dem Stahltrapezprofil aufgestellte Auflagersituation als auch die ausgeklinkte Auflagerung auf der BSP-Platte ermöglichen die Einhaltung von üblichen Durchbiegungsbegrenzungen auf Gebrauchstauglichkeitsniveau. Lokale Versagensphänomene kündigen sich durch große Verformungen an und treten erst bei Flächenlasten von über 13 kN/m2 auf. Zudem sind die Spannungen und Durchbiegungen des Verbundquerschnittes analytisch mit dem im Holzbau verbreiteten .-Verfahren in guter Näherung quantifizierbar.
Darüber hinaus wird das Tragverhalten der Stahl-Holz-Verbunddecke mithilfe numerischer Simulationen in Ansys abgebildet und die Wirksamkeit der Verbundtragwirkung analysiert, wobei die Implementierung der Verbundsteifigkeit auf den Ergebnissen der Abscherversuche fußt. Eine Gegenüberstellung mit den experimentellen Erkenntnissen bezeugt eine hohe Repräsentativität des FE-Modells, welches demnach für eine Parameterstudie mit anderen Spannweiten und Konfigurationen der Stahl-Holz-Verbunddecke verwendet wird, um die erforderlichen Stahl- und Holzmengen im Hinblick auf den Ressourcenbedarf zu ermitteln.
Eine abschließende vergleichende Lebenszyklusanalyse mit einer Stahlbetonflachdecke, einer Holz-Beton-Verbundflachdecke und einer Brettsperrholzdecke zeigt, dass die Stahl-Holz-Verbunddecke die Grundprinzipien einer nachhaltigen Konstruktionsweise durch den Einsatz von nachwachsenden und zirkulären Materialien, einer ressourceneffizienten und demontagegerechten Ausbildung sowie einer hohen umweltbezogenen Qualität repräsentiert. Hierbei sind insbesondere das hohe Wiederverwendungs- und Recyclingpotenzial der einzelnen Bauteilkomponenten der Stahl-Holz-Verbunddecke hervorzuheben.
Insgesamt ist die Stahl-Holz-Verbunddecke als leistungsstarke, flexibel planbare und sinnvolle Erweiterung bisher verwendeter Deckensysteme mit zahlreichen Stärken im Kontext einer nachhaltigen Tragwerksplanung einzuordnen.
System-wide investigation of hydroxymethylcytosine recognition and distribution in the mammalian genome
(2025) Engelhard, Lena; Summerer, Daniel; Pfander, Boris
The epigenetic cytosine modifications 5-methylcytosine (mC) and 5-hydroxymethylcytosine (hmC) are key regulatory elements of mammalian genomes, occurring within CpG dinucleotides in either strand- symmetric or strand-asymmetric combinations. While reader proteins for symmetrically and hemi- modified CpG dyads have been identified, the recognition of asymmetrically modified CpGs and the potential regulatory implications of their symmetry remain unexplored. In this work, mammalian nu- clear proteins binding to hmC-containing CpG dyads were identified and characterized. A protocol for generating asymmetrically modified DNA probes was established and applied in pull-down assays coupled with mass spectrometry-based proteomics. Comparative enrichment studies were performed using promoter probes bearing symmetric or asymmetric C, mC, and hmC modifications, allowing for direct assessment of reader profiles in the same sequence, tissue and experimental contexts. In hu- man and mouse nuclear extracts, numerous tissue-specific readers of hmC-modified sequences were identified, falling into distinct, probe-specific subgroups. These include transcription factors and chro- matin regulators such as MYC and MAX, which read hmC in a sequence-dependent manner, and RFX5, which selectively discriminated between hmC symmetry states in CpG dyads. The presumably preva- lent asymmetric CpG dyad hmC/mC exhibited a distinct reader protein profile, supporting the hy- pothesis that hmC symmetry information provides unique regulatory outputs. To complement these proteomics analyses, selective enrichment and sequencing of native double-stranded DNA fragments containing hmC/mC CpG dyads was implemented, based on an evolved methyl-CpG-binding domain protein with enhanced binding specificity. This approach enabled genome-wide mapping of hmC/mC sites, offering a valuable tool for investigating their roles in chromatin biology. Together, these find- ings provide a comprehensive framework for the study of symmetric and asymmetric hmC-containing CpG dyads and lay the groundwork for elucidating their functional contributions to transcriptional regulation, development and disease.