KLUEDO Dokumente/documents https://kluedo.ub.rptu.de/index/index/ Fri, 16 Oct 2026 14:14:43 +0200 Fri, 16 Oct 2026 14:14:43 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13728 The integration of deep learning and algebraic topology opens new perspectives for the analysis of complex data. Autoencoders, a central method of unsupervised learning, can map high-dimensional data spaces to lower dimensions in a nonlinear way, yet it remains unclear which topological invariants are preserved in the process. This dissertation investigates autoencoders through the lens of algebraic topology. Persistent homology is employed to systematically analyze the representations learned by autoencoders. The central question is whether autoencoders are capable of inducing isomorphisms between the persistent homology groups of the relevant spaces. Recent advances from fiber theory, together with newly developed bounds on the interleaving distance, enable a precise mathematical characterization of the behavior of such models. Both the computation of persistent homology over field coefficients and the application of global persistent homology in the deep learning context encounter fundamental limitations. One of these limitations is examined through a combined mathematical and computational–experimental study of torsion phenomena in the autoencoder setting. As a further solution perspective, it is proposed to embed local persistent homology into neural architectures via the persis tent Mayer–Vietoris sequence. In this way, the dissertation contributes to the mathematical foundations of topological deep learning. Maria Walch doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13728 Fri, 16 Oct 2026 14:14:43 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13727 Since the performance of thermoplastics depends significantly on temperature, their tribological behavior also depends in particular on the temperature at the point of contact with a friction partner. However, measuring this local temperature is vir-tually impossible; therefore, it can currently only be determined through simulations based on simplified model assumptions and experimentally determined data. Fur-thermore, tribological experiments are very time-consuming and costly. This study aims to validate the effectiveness of a novel intelligent testing methodology devel-oped within our research group in improving the efficiency of tribological investiga-tion, while also determining the interfacial temperature using finite element (FE) simulations. Tribological experiments were conducted using two polymer-based composite/steel sliding pairs on three testing rigs. Results show that, within the studied load range, tribological behavior mainly depends on material properties and the load conditions, while initial load and load sequence have negligible influence on tribological performance. The maximum contact surface temperature depends on the combination of surface pressure and sliding velocity. At low pv products, wear is dominated by localized abrasion induced by high surface pressure. Under high pv products, however, the elevated contact temperatures at high sliding speeds lead to material softening and reduced wear resistance. The proposed intelligent testing methodology and FE simulations enable efficient data acquisition and provide a deeper, data-driven understanding of tribological behavior, offering valuable in-sights for the design and selection of tribological materials. Chi Hua doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13727 Fri, 16 Oct 2026 09:11:20 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13703 Brazilian extractive reserves constitute a model for the territorial governance of protected areas. With a focus on the territorial rights of traditional populations, they can serve as a catalyst for transformative change that reconciles the conservation of nature with the sustainable use of natural resources. However, the sustainability of land use in the Chico Mendes Extractive Reserve (CMER) is increasingly under pressure. Located in the western Amazon, sustainable forest extractivism has traditionally constituted the main livelihood of inhabitants. Yet, over the last 20 years, deforestation rates have increased significantly. This increase has been driven by the expansion of cattle ranching, illegal land sales, and population growth. While the causes of deforestation in the CMER have been widely documented, this study aims to understand how different actors and institutions interact to shape territorial governance and land use in the CMER, and how these governance arrangements can serve as a catalyst for transformative change. The research is based on a qualitative case study design, involving qualitative interviews, participant observation, and a photobased approach, applied during three fieldwork visits to the research region between March 2024 and April 2025. The findings indicate that governance interactions are not only influenced by state institutions. They are shaped by economic incentives provided by the market and are contingent upon the social organization of civil society. To realize its potential for transformative change, territorial governance should adopt a multilevel governance approach, consider the local context, be based on cooperation between actors, and facilitate social learning. Axel Prestes Dürrnagel; Nell Dana Richter; Lissy Flörchinger; Vanessa Libanio Alemão; Janpeter Schilling article https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13703 Thu, 15 Oct 2026 17:01:57 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13721 Runtime safety monitors are a key safety measure for autonomous systems, but they are often designed for worst-case behavior of other agents and the environment. This keeps safety arguments simple, yet it can make systems overly conservative, with unnecessary stops, large distances, and reduced throughput. The core problem addressed in this dissertation is how to increase monitor permissiveness without weakening safety justification: when a monitor relaxes an environment assumption, engineers must be able to argue why this relaxation is safe in the relevant hazardous situations and within the stated operational design domain. This dissertation proposes Situation-Aware Dynamic Risk Assessment (SINADRA), an engineering method that extends hazard analysis and risk assessment (HARA) work products so that they support monitorable low-risk situation hypotheses for adaptive constraint relaxation. SINADRA targets risk variability that stems from external agent behavior like human workers in warehouses or human drivers in traffic. The method has three main elements. First, Structured Environment Models (SEM) scope each hazardous situation into an explicit, risk-driven environment view, including initiator scenarios and environment factors that matter for risk. Second, Critical Behavior Cause Trees (CBCT) refine hazardous situations into behavior risk scenarios and relate these scenarios to risk implausibility hypotheses – monitorable situation conditions indicating low risk – and other safety measures from the safety concept like system behavior restrictions. CBCTs form a Boolean model that supports minimal cut set analysis as reviewable verification evidence that, within the modeled scope, risk scenarios are sufficiently blocked by hypotheses and safety measures. Third, SINADRA provides a runtime model transformation that turns argued hypotheses into executable inference components, using either Boolean logic or Bayesian networks (BNs). SINADRA is evaluated in two case studies. A warehouse robotics study executes the full design-time chain, integrates Boolean runtime models into a ROS-based safety monitoring system, and demonstrates hypothesis-driven mode switching with worst-case as the default. An analytic comparison reports estimated corridor travel-time gains of about 2% for passing and 21% for overtaking under stated assumptions. An urban automated-driving study demonstrates component-level runtime feasibility in CARLA by inferring behavior intent likelihoods with BNs and computing an intent-weighted collision-risk signal that differs from a worst-case baseline in uncritical situations. Overall, SINADRA provides a traceable path from HARA artifacts to runtime adaptation logic for safer, situation-dependent permissiveness. Jan Reich doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13721 Thu, 15 Oct 2026 09:14:47 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13707 Im Rahmen dieses Dissertationsvorhabens wurde mittels fünf wissenschaftlicher Artikel untersucht, welche Mechanismen die Erwartungsveränderung bei Menschen mit depressiven Störungen und Symptomen begünstigen beziehungsweise hemmen. Es wurde ein experimentelles Paradigma entwickelt, in welchem positive Erfahrungsberichte über Psychotherapie genutzt wurden, um die Erwartungen bezüglich positiver versus negativer Lebensereignisse und die Behandlungserwartungen zu modifizieren. Es wurde an einer kleinen klinischen Stichprobe (N = 17) auf Machbarkeit überprüft (Artikel 1) und an einer großen klinischen Stichprobe (N = 156) angewendet. Die Erfahrungsberichte verbesserten die Erwartungen signifikant. Der Einfluss von kognitiver Immunisierung auf die Erwartungsveränderung konnte nicht nachgewiesen werden (Artikel 2). Menschen mit persistierender depressiver Störung (n = 54) verbesserten ihre Therapieerwartungen weniger stark als Menschen mit episodischer depressiver Störung (n = 102), besonders wenn kognitive Immunisierung experimentell gefördert wurde (Artikel 3). Zudem wurde an einer klinischen Stichprobe (N = 80) der Einfluss von State Affekt auf die Erwartungsveränderung nach positivem versus negativem Leistungsfeedback untersucht. Affekt hatte keinen signifikanten Einfluss auf die Erwartungsveränderung (Artikel 4). Abschließend wurde an zwei Stichproben (N = 149 und N = 212) untersucht, wie der Schweregrad depressiver Symptome mit der Inkongruenz zwischen Selbst- und erwarteter Fremdwahrnehmung (Metawahrnehmung) und dem Feedbacksuchverhalten zusammenhängt. Mit steigendem Schweregrad wich die Selbstwahrnehmung zunehmend negativ von der Metawahrnehmung ab. Außerdem zeigte sich ein negativer Zusammenhang zwischen dem Schweregrad und dem Feedbacksuchverhalten, der jedoch nicht durch Valenz des Feedbacks oder Inkongruenz zwischen Selbst- und Metawahrnehmung erklärt werden konnte (Artikel 5). Edith Rapo doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13707 Wed, 14 Oct 2026 17:40:38 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13705 Reliable calibration of interest-rate models is essential for market-consistent valuation and Economic Scenario Generators (ESGs). In many applications, however, calibration is still understood primarily as fitting model prices to observed market prices by choosing suitable model parameters. This view is too narrow for long-horizon actuarial applications: a low calibration error alone does not show whether the estimated parameters are locally identifiable, stable over time, and suitable for subsequent simulations. This work therefore analyses calibration as a multi-stage quality and validation problem. The main model is the G2++ model, a stochastic two-factor short-rate model that combines analytical tractability with sufficient flexibility to represent term-structure movements. The analysis starts with the construction of an application-adequate initial term structure. Tools from nonlinear regression diagnostics are then transferred to calibration under the risk-neutral measure \(\mathbb{Q}\), including the weighted hat matrix, leverage, influence diagnostics, effective degrees of freedom, and local uncertainty quantification. For at-the-money (ATM) caps, the analytical structure of G2++ cap prices is used to evaluate the local geometry efficiently. In addition, cap and swaption calibrations are compared, and the G2++ model is contrasted with a one-factor Hull-White benchmark to assess empirically the role of instrument class and model structure. Building on this, the analysis turns to the shift from the risk-neutral measure \(\mathbb{Q}\) to the physical measure \(\mathbb{P}\). Different risk-premium specifications are constructed and evaluated with respect to the resulting \(\mathbb{P}\)-dynamics, long-run centres, and suitability for actuarial scenarios. A case study and a structured ex-post validation show how local calibration diagnostics, real-world dynamics, and subsequent simulation outputs can be assessed jointly. The empirical focus is on Euro interest-rate market data from 2016 to 2025, including term-structure and derivative-market data, and on the controlled transfer of diagnostic statements across instrument classes. The main conclusion is that calibration is reliable for its intended use only if the market-data basis is coherent and market fit, local identifiability, real-world dynamics, and ex-post validation are jointly plausible. Philipp Mahler doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13705 Wed, 14 Oct 2026 10:37:32 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13660 Increasing productivity and flexibility to meet the growing demand for chemical and special chemical products is a key area in the industry. At the same time, due to rising demand, protecting chemical reactors against highly exothermic reactions is a major challenge for public safety. To address these challenges within the scope of this work, the core components of new, intelligent safety devices, so-called “Smart overpressure Protection Devices” (SmOPs), have been developed and evaluated. A SmOP consists of sensors, a safety-related process logic controller (sPLC), which examines models cyclically, and a mechanical pressure relief device that can be adjusted according to the reaction's hazard potential. The model, implemented in the sPLC, supports two modes of operation: the zero-emission mode, which completely avoids pressure relief from the system, and the smart-relief mode, which allows controlled pressure relief when necessary. The core elements developed and evaluated in this research are: (1) A reaction-independent online runaway consequence model to determine the hazard potential (2) An early warning detection model for starting dangerous runaway reactions (3) The design and control of an smart safety pressure relief device The work was structured scientifically in accordance with the design-science research approach. All core elements were first developed, research hypotheses were reviewed using simulations and laboratory testing, and the functionality was demonstrated at suitable test rigs. Results prove that the online runaway consequence model can increase the productivity and flexibility of chemical reactors, with special focus on the uncertainty of input parameters. The adapted early warning detection model was capable to detect runaway reactions early in two different test reactors (RC1 and the pilot plant). A combined model prediction and control and was used to control the mechanical pressure relief device. The precise and robust functionality of the control was demonstrated through flow tests on the CSE-High pressure loop. Carsten Schmidt doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13660 Wed, 14 Oct 2026 08:19:10 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13697 Riparian areas serve as integrative landscapes that connect riverine and terrestrial ecosystems, shaping hydrological, geomorphological, and biological interactions. Further, longitudinal changes in hydrology and aquatic communities characterise rivers and streams. Consequently, aquatic insects can function as important subsidies that alter the diet composition of riparian consumers. In addition, riparian communities might also show characteristic changes along the longitudinal gradient based on their strong connection towards the waterbody. However, little is known about the effect of riparian area characteristics on the arthropod communities and their diet composition along the river course. In this dissertation, I first systematically reviewed the importance of aquatic subsidies for riparian arthropod predators. Furthermore, I examine potential variations in community characteristics and the dietary contributions of carabid beetles in three river sections: upstream, midstream and downstream. The literature review (Chapter 2) revealed that aquatic subsidies contribute substantially to the diet of riparian arthropod communities. Factors such as seasonality, riverine characteristics, riparian habitat structure, distance from the water, predator body size and life stage influence the extent to which aquatic subsidies are incorporated into their diets. However, most of the studies conducted have focused on spiders and mainly the horizontal web-building spider Tetragnatha. This reveals a research gap for other riparian arthropods than spiders. Based on this finding, a field study was conducted. Chapter 3 evaluates the effect of the river section to the carabid community. This showed significant differences in the carabid beetle community between upstream, midstream and downstream sections with the highest richness, individual number and functional trait diversity in the downstream section. Chapter 4 assesses the dietary spectrum using DNA metabarcoding for a subset of carabid beetles collected in the field. A dominance of terrestrial prey items was found which differs between carabid beetle tribes and riparian habitat specialists and generalists. Despite the minimal use of aquatic subsidies, the proportion of this food source was highest in the downstream section. The results indicate that carabid beetles serve as reliable bioindicators of riparian habitat characteristics along streams, despite exhibiting a largely generalist diet. Additionally, the use of DNA metabarcoding revealed unexpected predator-prey species interactions. This knowledge should be considered for further conservation and restoration projects. Franziska Middendorf doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13697 Mon, 12 Oct 2026 18:50:19 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13671 Since the early 2000s, there has been an enormous amount of research on awe and its links with other phenomena. One might therefore assume that there is a clear conceptualization of awe. However, based on the review of 168 publications on awe, I argue in this dissertation that this is not the case. The conceptual unclarity of awe makes it difficult to interpret study results and to develop a sound theory of awe. With this dissertation, I contribute to advancing the conceptualization of awe and expanding culturally sensitive research on awe and its equivalents. Based on the exploration of ehrfurcht in German context and its conceptual comparison with the influential definition of awe according to Keltner and Haidt (2003), I highlight the scope and limitations of this definition in the German context. First, I review ideas of awe and ehrfurcht in and beyond psychological research. In Chapter 2, I draw on the analysis of 168 publications on awe to show that the influential conceptualization of awe is unclear and suggest methodological approaches that can overcome problems in the research field. In Chapter 3, I explore the characteristics of ehrfurcht in the German context based on the meaning of the term, the facets of perception, and the associated contexts. In Chapter 4, I explore the influential conceptualization of awe according to Keltner and Haidt (2003) in the German context with a focus on the assigned term, the associated perceptions and contexts. Furthermore, I compare the German meaning of Keltner and Haidt’s (2003) conceptualization of awe with that of German ehrfurcht and compare the German meaning the US-American one. In Chapter 5, I discuss the implications of this dissertation and identify avenues for future research. In this dissertation, I identified fundamental problems in psychological research on awe. The findings indicate to revisit the conceptualization of awe and advocate for methodological pluralism in awe research. The contextualized investigation of awe that also incorporates the perspectives of multiple disciplines is suggested to pave the way for a sound theory building. The exploration of German ehrfurcht revealed the close link between ehrfurcht in nature and sustainable behavior. This suggests the potential for application for example in mitigating the climate and biodiversity crisis as soon as the concept ehrfurcht will be sufficiently clear in the future. Tini (Laura Christin) Katz doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13671 Mon, 12 Oct 2026 16:27:11 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13699 Metallische Mikrostrukturen erfreuen sich einem großen Interesse für ein weites Anwendungsfeld, welches sich von der Plasmonik über die Luftfahrt- und Automobilindrustrie bis zur Biotechnologie und Medizin spannt. Die Herstellung solcher Strukturen stellt jedoch nach wie vor eine erhebliche Herausforderung für Wissenschaft und Technik dar. Besonders für ferromagnetische Mikrostrukturen, die für mikroelektronische Bauteile wie Sensoren und Aktuatoren bis hin zu vollständig magnetisch fernsteuerbaren Mikrorobotern relevant sind, gibt es bisher nur wenige Ansätze. In dieser Arbeit wird ein Konzept für die additive Fertigung metallischer Mikrostrukturen vorgestellt und exemplarisch anhand von Nickelstrukturen demonstriert. Dazu wird die Technik des kommerziellen Direkten Laserschreibens (DLW), die üblicherweise zur additiven Fertigung von dreidimensionalen Polymerstrukturen auf der Mikroskala verwendet wird, mithilfe eines neu entwickelten Ansatzes auf die Herstellung metallischer Strukturen übertragen. Das Konzept basiert auf einem neu entwickelten Photolack, in dem drei eng miteinander verknüpfte photochemische Prozesse simultan und unter Belichtung kontinuierlich ablaufen. Ihr kombiniertes Zusammenwirken ermöglicht eine räumlich präzise Reduktion von Metallionen direkt im Fokusvolumen des Schreiblasers. Die photochemische Deoxygenierung erzeugt zunächst ein lokal sauerstofffreies Volumen, wodurch die Bedingungen für die beiden weiteren Prozesse geschaffen werden und der Einsatz des Lackes unter atmosphärischen Bedingungen möglich ist. Der zweite Prozess, eine sensibilisiert Triplett-Triplett Annihilation Upconversion (sTTA-UC), begrenzt den Reaktionsbereich aufgrund ihrer quadratischen Intensitätsabhängigkeit auf das definierte Fokusvolumen. Beim dritten Prozess handelt es sich um eine Photoredoxreaktion, in deren Verlauf Metallionen aus der Lösung reduziert werden, um Metallstrukturen zu bilden. Die experimentelle Umsetzung des Konzepts wird anhand eines speziell entwickelten Photolacks für die Herstellung von Nickelstrukturen demonstriert. Für die photochemische Deoxygenierung wird 1,3-Dimethyl-2-imidazolidinon (DMI) als deoxygenierendes Lösungsmittel eingesetzt. Erythrosin dient dabei als Sensibilisator sowohl für die Deoxygenierung als auch für die sTTA-UC. Als Annihilator der sTTA-UC kommt Perylen zum Einsatz, das gleichzeitig als Photokatalysator für die Photoredoxreaktion wirkt. Für die Photoredoxkatalyse werden Diisopropylethylamin (DIPEA) als Elektronendonor und Nickel(II)chlorid Hexahydrat \((\text{NiCl}_2\cdot 6\text{H}_2\text{O})\) als Metallvorstufe verwendet. Mittels eines kontinuierlich betriebenen Lasers mit 532 nm Wellenlänge in einem DLW-Aufbau kann die Herstellung von Nickel-Mikrostrukturen erfolgreich demonstriert werden. Untersuchungen der Strukturen hinsichtlich ihrer morphologischen Eigenschaften zeigen klar definierte 2,5D-Geometrien mit hoher Materialdichte und nur wenigen Poren mit charakteristischen Größen unterhalb von 1 \(\mu\)m. Materialanalytische Untersuchungen mittels energiedispersiver Röntgenspektroskopie bestätigen zudem eine hohe Nickeldichte in den Strukturen. Die magnetischen Eigenschaften werden mithilfe von Vibrationsprobenmagnetometrie (VSM) und Stickstoff-Fehlstellen (NV) Magnetometrie untersucht. Dabei kann für die gedruckten Strukturen ein ferromagnetisches Verhalten nachgewiesen werden, das dem von Nickel ähnlich ist. Abweichungen von den Literaturwerten lassen sich durch Materialanisotropien erklären, die aufgrund der geringen Strukturgröße einen signifikanten Einfluss auf die magnetischen Eigenschaften haben. Die lokale Feldverteilung der Proben folgt überwiegend der Geometrie der Strukturen. Untergeordnete Domänen konnten mit den in dieser Arbeit durchgeführten Messungen nicht nachgewiesen werden. Insgesamt stellt diese Arbeit einen wichtigen Schritt dar, um die Möglichkeiten des Direkten Laserschreibens auf metallische und insbesondere ferromagnetische Materialien auszuweiten und legt damit ein Fundament für die Entwicklung von Mikrostrukturen mit auf die jeweilige Anwendung zugeschnittenen Struktureigenschaften. Kristin E. J. Kühl doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13699 Mon, 12 Oct 2026 08:20:20 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13700 Soil texture, the most widely used soil property to describe single grain size distribution, is different from soil structure. Setting them equal neglects the impact of actual particle arrangements under natural conditions. Particularly in deeply weathered soils, the mismatch between soil texture and structure has widely been described. It is therefore not surprising, that biogeochemical models built on texture-based assumptions systematically diverge from field observations, for example, often overestimating nitrous oxide emission rates. Water-stable aggregates like pseudosands, often described for deeply weathered soils, might be responsible for this mismatch. In the field they are perceived as sand, but in laboratory analysis for texture determination measured as clay and silt. This thesis develops pseudosands as a tractable entry point into studying aggregate stability and pore space as complementary aspects of soil structure. An initial literature review revealed a persistent lack of consistent classification for pseudosands, hindering comparison across studies. However, when looking for water-stable aggregation in literature, dynamic water balances emerge as the most determining pattern for their global distribution. Within this Goldilocks zone, spanning humid to perhumid climates in tropical to cold temperate latitudes, drying-wetting cycles act as stabilizing and selective forces through chemical weathering and physical comminuting. Although pseudosands have been described (at least) since the 1960s, this thesis further establishes them as a distinct textural class in their own right, as, unlike sand, they display mechanical stability largely independent of size. Modeling pseudosands as sand holds only moderately for hydrological purposes and fails for biogeochemical processes such as nitrous oxide emission rates for deeply weathered soils. Actually, pseudosands show distinct water retention and porosity, while their quantity alone does not predict functional hydraulic impact. Pseudosands are therefore not sand but neither the plain sum of their clay- and silt-sized units. While most aggregates remain volatile across time and space, constantly forming, dissolving, and transforming, pseudosands stand out as an exemplarily stable aggregate. If such stability is indeed a function of pedogenic age, it stands to reason that all sufficiently aged soils could eventually develop water-stable aggregates of this kind. As a rare anchor point within an otherwise highly diverse and opaque soil system, they offer remarkable potential for study. Simone Kilian Salas doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13700 Thu, 08 Oct 2026 16:38:29 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13695 Gesellschaftliche Veränderungsprozesse setzen die Ev. Gliedkirchen der EKD unter Druck. Die damit einhergehende Ressourcenverknappung durch Rückgang von Kirchensteuereinnahmen und Personalknappheit führen zu schwindender organisationaler Resilienz. Eine Antwort auf die großen Herausforderungen ist das Konzept der Exnovation. Es wird gezeigt, wie diese mit notwendingen Innovationen einhergehen kann und welche Herausforderungen der Organisationsentwicklung sich dadurch stellen. Neben einer Reflexion von Praxisfeldern werden weiterführende Handlungsempfehlungen erarbeitet. Konrad Köhnlein masterthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13695 Tue, 06 Oct 2026 14:33:54 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13696 Predicting diffusion coefficients in mixtures is crucial for many applications, as experimental data remain scarce, and machine learning (ML) offers promising alternatives to established semi-empirical models. Among ML models, matrix completion methods (MCMs) have proven effective in predicting thermophysical properties, including diffusion coefficients in binary mixtures. However, MCMs are restricted to single-temperature predictions, and their accuracy depends strongly on the availability of high-quality experimental data for each temperature of interest. In this work, we address this challenge by presenting a hybrid tensor completion method (TCM) for predicting temperature-dependent diffusion coefficients at infinite dilution in binary mixtures. The TCM employs a Tucker decomposition and is jointly trained on experimental data for diffusion coefficients at infinite dilution in binary systems at 298 K, 313 K, and 333 K. Predictions from the semi-empirical SEGWE model serve as prior knowledge within a Bayesian training framework. The TCM then extrapolates linearly to any temperature between 268 K and 378 K, achieving markedly improved prediction accuracy compared to established models across all studied temperatures. To further enhance predictive performance, the experimental database was expanded using active learning (AL) strategies for targeted acquisition of new diffusion data by pulsed-field gradient (PFG) NMR measurements. Diffusion coefficients at infinite dilution in 19 solute + solvent systems were measured at 298 K, 313 K, and 333 K. Incorporating these results yields a substantial improvement in the TCM’s predictive accuracy. These findings highlight the potential of combining data-efficient ML methods with adaptive experimentation to advance predictive modeling of transport properties. Zeno Romero; Kerstin Münnemann; Hans Hasse; Fabian Jirasek article https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13696 Tue, 06 Oct 2026 12:06:21 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13694 The preservation of the German bridge inventory faces major structural and societal challenges. An ageing substance, increasing traffic loads and political mandates for model-based infrastructure management necessitate a fundamental rethinking of lifecycle management. Focusing on reinforced and prestressed concrete bridges, this work develops automated approaches for representing structures within digital building models. The objective is to provide a planning basis founded on the Industry Foundation Classes standard (IFC) for renovation and construction measures, supporting the scan-to-BIM process for converting point clouds into as-built inventory models. To this end, comprehensive annotation rules and object classifications for the segmentation of bridge structures were designed and elaborately developed. A freely accessible overview of the established structural taxonomies and guidelines is provided in an open-source repository. These fundamentals serve as a basis for future machine learning approaches. To validate this framework, a total of 27 complex point cloud scans of existing bridges with approximately 2.6 billion points were manually segmented during the research work. This dataset functions as training data for research and as an evaluation basis for the developed reconstruction approaches. Furthermore, heuristic methodologies for the geometric reconstruction of bridge piers and superstructure configurations were independently developed. Based on real laser scans, the superstructures are algorithmically partitioned into geometric objects using profound domain knowledge. Thereby, loadbearing concrete slabs and girders, bridge caps as well as roadways are geometrically reconstructed and subsequently converted into an IfcBridge-compatible IFC model. This methodology was implemented in a Python-based software pipeline. The source code is freely available in an open-source repository to support further developments and to ensure long-term computational reproducibility. The validation is performed separately for piers and superstructures using synthetic model data to determine reconstruction limits and real scans for accuracy investigations via the closest point-to-mesh distance. The pier evaluation demonstrated a point coverage of up to 100% within a narrow tolerance limit. High precision was also achieved for the more complex real superstructure point clouds, where the control threshold was methodically adjusted due to scan artefacts. The analysis of the results illustrates that the acquisition quality and the exact registration of the scans significantly influence the precision of the automated reconstruction. In conclusion, the work provides methodical approaches and practical contributions for the automated geometric reconstruction of bridge structures as well as for the generation of reliable IFC as-built models. Marius Schellen doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13694 Mon, 05 Oct 2026 10:36:33 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13680 Biotechnological processes provide a unique platform for the production of bulk and fine chemicals from renewable feedstocks but are often constrained by low yields and limited efficiency. Process intensification aims to fundamentally improve bioprocesses through more integrated and efficient process concepts, thereby enhancing their economic competitiveness compared to conventional petrochemical processes. This work presents and explores selected process intensification strategies at different process levels, including biocatalyst engineering, innovative cultivation strategies and integrated process concepts. Valuable insights can be derived from ABE fermentation with clostridia, a historically established and extensively studied bioprocess. The process intensification strategies developed in this context range from genetic modifications and optimized cultivation strategies to highly integrated continuous process concepts, providing important impulses for the efficient utilization of complex substrates. A key approach to process intensification is process integration, which is discussed in this work through the combination of microbial cultivation and enzymatic biocatalysis. Based on these insights, an integrated process for the production of ethyl butyrate was developed. Here, a microbial co-culture was combined with enzymatic esterification, with the produced ester being continuously removed via in-situ extraction. The limited acid tolerance of the yeast was identified as a key bottleneck in this process. Adaptive laboratory evolution emerged as a promising strategy to overcome this limitation, enabling the rapid development of robust biocatalysts tailored to specific process conditions. Overall, the present results highlight both the potential and the limitations of process intensification and underline the need for a holistic approach to the design of future bioprocesses. In this context, this work aims to provide new impetus for bioprocess intensification and to further strengthen the industrial relevance of biotechnological processes. Katharina Oehlenschläger doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13680 Thu, 01 Oct 2026 10:16:13 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13678 Neural networks are nowadays fundamental for modeling complex dynamical systems and are widely used for system identification and feedback control. However, as model capacity increases, deploying these networks on embedded edge devices is challenging due to limited computational and memory resources. Additionally, control applications impose stringent real-time constraints and require that closed-loop stability be preserved after model compression. Model compression is therefore essential. However, most existing methods are designed for general-purpose, monolithic networks and do not translate well to neural-network controllers. Modern controllers are often implemented as Multi-Component Neural Architectures (MCNAs), in which specialized subnetworks interact through coupling interfaces and often process latent representations derived from high-dimensional observations. Uniform, structure-agnostic sparsification may overlook these dependencies, thereby degrading closed-loop performance. This dissertation introduces a component-aware structured pruning framework tailored for multi-component neural architectures. The proposed approach decomposes an MCNA into module-specific dependency graphs, distinguishing component-internal parameter groups from inter-component coupling groups to facilitate targeted pruning. Group-wise pruning rates are determined by casting sparsity allocation as a constrained search problem, solved via grid search and continuous optimization to achieve compression goals while minimizing task performance degradation. Furthermore, both post-training and training-aware group-importance estimation methods are developed. The training-aware method computes online importance signals using gradient accumulation, diagonal Fisher information, and an empirical Bayesian mechanism to capture both immediate sensitivity and long-term learning dynamics. The framework is evaluated on representative architectures, including an MNIST autoencoder and a TD-MPC agent. Results demonstrate that integrating application-specific metrics into the compression objective enables more precise control over the pruning distribution and achieves greater reductions in model size and inference latency than baseline pruning methods while preserving closed-loop performance and stability requirements in the evaluated scenarios. Ganesh Sundaram doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13678 Thu, 01 Oct 2026 09:57:05 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13676 The number of security incidents in recent years shows a considerable increase in cyber attacks. The targets of these attacks are diverse, but also increasingly safety-critical systems like airplanes or power plants fall victim to attackers. These attacks show that even embedded systems cannot be considered inaccessible to malevolent actors anymore. As a consequence, security has become a primary concern for embedded systems and thus also for real-time systems which are often deployed as embedded systems in a safety-critical environment. Real-time systems are not only constrained by limitations typical for embedded systems, like power or memory, but they must also fulfill timing constraints. Thus, any security measure employed on a real-time system must adhere to timing constraints. However, these constraints allow for new attack vectors targeting the timing of the system. For example, a single deadline miss in a hard real-time system can already threaten system safety and have catastrophic consequences for the system or its environment. This inherent vulnerability of real-time systems to attacks in the time domain poses new and real-time specific challenges to security. This thesis analyses threats to and vulnerabilities of real-time systems with a focus on time-triggered real-time systems. It shows how attackers may threaten fundamental system assumptions. The fault model of a system usually considers failures to be independent from each other, but an attacker violates this independence assumption when provoking failures. Moreover, an attacker may cause the system to reach a state that was not considered when worst case execution times were derived. As a consequence, these worst case execution time bounds may prove to be unsafe in a system under attack. Another vulnerability this thesis presents lies within a desirable system property itself. Time-triggered real-time systems provide deterministic scheduling behaviour. Deterministic behaviour is desirable for safety-critical systems as it allows for easy derivation of safe bounds on worst case execution time and simplifies system certification. However, the predictable schedules of time-triggered systems also ease attacks which exploit timing information for precisely timed attacks. In this thesis, we detail these threats, present attack vectors for single core as well as multicore time-triggered real-time systems and introduce mitigation strategies tailored to these systems. The mitigation strategies employ randomization to prevent precisely timed attacks while also guaranteeing that all tasks meet their deadlines. We implement our mitigation strategies on FreeRTOS and demonstrate their low runtime overheads and small memory requirements. Kristin Krüger doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13676 Wed, 30 Sep 2026 14:12:25 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13675 Die Praxis systemischer Beratung kontextualisiert Verhaltensweisen, Interaktionen und Entscheidungen der Ratsuchenden im Hinblick auf deren Umwelt, regt zur Selbstbeobachtung und zur Einnahme neuer Perspektiven auf ein Problem an. Doch welche Effekte haben menschliche Gefühle und Emotionen auf kognitive Vorgänge und wo sind intrapsychische Prozesse innerhalb der systemischen Beratung zu verorten? Begibt man sich auf die Suche nach einer Antwort auf diese Frage, fällt auf, dass sich die Begriffe Emotionen, Gefühle und Grundbedürfnisse in den meisten Lehrbüchern zur systemischen Beratung und Therapie sowie in den oft zitierten Grundlagenwerken nur selten wiederfinden und ihnen, wenn überhaupt, nur eine untergeordnete Rolle beigemessen wird. Gleichzeitig wird im Kontext der Extremismusprävention nicht nur deutlich, dass systemischer Beratung an vielen Stellen eine gewichtige Bedeutung zugeschrieben wird, sondern legen auch zahlreiche Studien nahe, dass unbewältigte adoleszente Krisen und verletzte Grundgefühle als Hauptantreiber von Radikalisierung ausgemacht werden können. Die vorliegende Arbeit trägt dazu bei, diese Lücken zu schließen. Vor dem Hintergrund der Wechselwirkungen von Denken und Fühlen, die anhand der Theorie der Affektlogik des Schweizer Psychiaters Luc Ciompi sowie deren systemtheoretischen Referenzpunkten analysiert werden, wird der Zusammenhang von Radikalisierungsprozessen und menschlichen Grundbedürfnissen erörtert. Dabei werden psychodynamische Aspekte integriert, in ein Verhältnis zum systemischen Denken gebracht und praktische Zugänge im Umgang mit Emotionen und Gefühlen im Rahmen der Extremismusprävention aufgezeigt. Benedikt Büns masterthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13675 Wed, 30 Sep 2026 14:07:08 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13674 Diese kumulative Dissertation untersucht, wie individuelle KI-Nutzung und organisationale KI-Adoption zusammenhängen und unter welchen Bedingungen KI nachhaltig in Organisationen verankert werden kann. Ausgangspunkt ist die Beobachtung, dass KI im privaten Alltag zunehmend niedrigschwellig und experimentell genutzt wird, während KI-Nutzung im Arbeitskontext häufig situativ bleibt und organisationale KI-Adoption trotz technischer Reife oft nicht über Pilotphasen hinauskommt. Die Arbeit argumentiert, dass diese Unterschiede nicht mit einer einheitlichen Adoptionslogik erklärt werden können. Stattdessen wird nachhaltige KI-Verankerung als Mehrebenenprozess verstanden, in dem subjektive Wahrnehmung, Nutzungspraxis und organisationale Einbettung systematisch zusammenwirken. Die Dissertation unterscheidet drei Analyseebenen: individuelle KI-Nutzung im Alltag, individuelle KI-Nutzung im Arbeitskontext und organisationale KI-Adoption. Diese Ebenen werden in drei empirischen Studien untersucht und in einer vierten Studie zusammenführend ausgewertet. Studie 1 zeigt, dass KI-Aversion wesentlich durch die Fehlkalibrierung zwischen subjektivem und objektivem KI-Wissen geprägt ist. Studie 2 rekonstruiert anhand von zwölf Interviews den Übergang von Erprobung zu routinisierter KI-Nutzung im Arbeitskontext und zeigt, dass dieser Übergang an motivationalen, situativen und organisationalen Schwellen scheitern kann. Studie 3 macht deutlich, dass organisationale KI-Adoption trotz technischer Machbarkeit durch organisationale Rollenkonflikte, Rollenunklarheiten und fehlende bereichsübergreifende Koordination blockiert werden kann. Studie 4 führt diese Befunde zu einem kontextübergreifenden Erklärungsrahmen zusammen. Sie legt nahe, dass private KI-Erfahrungen den Einstieg in KI-Nutzung im Arbeitskontext erleichtern können, aber weder automatisch in stabile Nutzung im Arbeitskontext übergehen noch unmittelbar zu organisationaler KI-Adoption führen. Entscheidend sind Brücken- und Anschlussbedingungen wie Aufgabenwert, freigegebene Tools, Verifikationsroutinen, Rollenklärung und Governance, unter denen sich KI-Nutzung im Arbeitskontext stabilisieren und in tragfähige organisationale Adoptionsprozesse überführen lässt. Der zentrale Beitrag der Dissertation liegt darin, individuelle KI-Nutzung im Alltag, individuelle KI-Nutzung im Arbeitskontext und organisationale KI-Adoption nicht als getrennte Phänomene, sondern als miteinander verbundenen Verankerungsprozess zu verstehen. Nachhaltige KI-Verankerung entsteht demnach durch das Zusammenspiel subjektiver Wahrnehmung, routinisierter Nutzungspraxis und organisationaler Einbettung. Özgür Kartal doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13674 Wed, 30 Sep 2026 12:19:31 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13665 Gallium (Ga) is a critical metal, indispensable for semiconductor industry, used in versatile products such as in light-emitting diodes (LEDs), liquid crystal displays (LCDs), back-lightning applications, integrated circuits (ICs), and so on. In light of technological advancements, demand for these aforementioned products is higher than ever, pressuring the electronic industry to meet the increasing demands. Current supply line for Ga is dependent on primary reserves, predominantly located in China. Ga is irreplaceable, owing to its unique features in semiconductors. Therefore, European Union (EU) commission enlisted Ga as a critical raw material (CRM) to urge the circularity for this particular metal. Urban mining, a secondary production route for metal extraction, involves the recycling of metals from end-of-life (EoL) devices, aiming to address the growing built-in obsolescence related electronic waste (e-waste) conundrum as well as the scarcity of technologically-critical metals such as Ga, rare earth elements (REEs) and so on. Since EoL devices contain a higher concentration of Ga in the waste matrix than known primary ore reserves, attempts have been made to develop processes for the targeted recovery of Ga. This Ga is present in a dispersed electronic waste matrix that also contains base metals, precious metals, and other critical metals. However, current recycling technologies are based on processes that require high energy input and involve the use of aggressive inorganic reagents such as hydrochloric acid (HCl), which poses a significant burden on the environment. Despite the promising results of preliminary studies, these methods achieve low selectivity, as other metals such as Cu, Fe, and Al are also extracted from the waste matrix. In this context, organic acids and microbial systems represent a promising alternative. Studies have shown that organic acids, due to their unique dissociation and ligand-forming properties, can selectively dissolve and complex certain metals within the same waste matrix. Another novel technology, bioleaching, is based on a microbial approach in which microorganisms secrete metabolites that serve as suitable extractants for targeted metal recovery. This work addresses the efficacy and solubility of gallium in environmentally-friendly, selective organic reagents as well as fungal-based bioleaching systems. The aim of this work is to propose a bioprocessing strategy as the basis for a bottom-up approach that is applicable not only to Ga but also to other critical elements in end-of-life (EoL) processing. Thus, not only is the aspect of the circular economy for critical metals taken into account, but the proposed recovery scheme also aligns with the EU’s sustainability goals for 2050. Aylin Nur Erkmen doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13665 Mon, 28 Sep 2026 15:55:29 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13659 Amtliche Bekanntmachung der RPTU Nr.8/25.09.2026 periodicalpart https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13659 Thu, 24 Sep 2026 12:00:16 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13658 Die vorliegende Dissertation untersucht den Beitrag ländlicher Coworking Spaces auf die kommunale Entwicklung unter besonderer Berücksichtigung der Daseinsvorsorge. Durch Migrationsbewegungen und veränderten Bedarfen an die Infrastruktur im Zuge des demografischen Wandels sind Coworking Spaces in ländlichen Räumen als innovatives Instrument zur Stärkung lokaler Strukturen in der Diskussion. Obwohl zahlreiche wirtschaftliche, arbeitsmarktbezogene, soziale sowie siedlungsbezogene Erwartungen an sie gestellt werden, ist ihr Beitrag zur kommunalen Entwicklung in der Praxis bislang wenig erforscht. Ziel der Arbeit ist es, diese Forschungslücke zu schließen und zu untersuchen, welchen konkreten Beitrag ländliche Coworking Spaces in Bayern leisten. Um dieses Thema vertieft zu analysieren, wird ein exploratives Studiendesign auf Basis der Grounded-Theory-Methodolgie verwendet. Als Datengrundlage dienen qualitative, teilstrukturierte Interviews mit Stakeholdern ländlicher Coworking Spaces, worunter sowohl Betreiber als auch kommunale Akteure fallen. Zusätzlich wurde eine Strukturanalyse der jeweiligen Standorte sowie eine Bestandsaufnahme der entsprechenden Coworking Spaces durchgeführt. Hierdurch können Bedingungen für deren Beiträge auf die kommunale Entwicklung abgeleitet werden. Durch dieses methodische Vorgehen ist eine belastbare und aussagekräftige Analyse des Untersuchungsgegenstandes sichergestellt. Im Rahmen der Datenauswertung wird deutlich, dass ländliche Coworking Spaces in unterschiedlicher Intensität auf kommunaler Ebene wirken. Abhängig von der Betreuungsqualität und den Vernetzungsmöglichkeiten, können sie durch Gründungen und Innovationen zur Schaffung von Arbeitsplätzen und Generierung zukünftiger Gewerbesteuereinnahmen führen. Innerhalb des kommunalen Arbeitsmarkts unterstützten sie dabei, zukünftige Fachkräfte in der Region zu halten, insbesondere bei Hochschulnähe. Zusätzlich können sie in der Funktion eines Begegnungs- und Lernorts auftreten und dabei sowohl soziale als auch bildungsbezogene Teilhabe fördern, sofern ein ganzheitliches Konzept und ein niedrigschwelliger Zugang gegeben sind. Als weiterer Aspekt leisten Coworking Spaces einen Beitrag bei der Revitalisierung von Ortskernen im Hinblick auf die Nutzung von Leerständen, der Belebung des Zentrums sowie der Unterstützung lokaler Infrastrukturen. Hierbei spielen die strategische Lage sowie die baulichen Gegebenheiten vor Ort eine entscheidende Rolle. Neben den genannten Beiträgen ländlicher Coworking Spaces für die kommunale Entwicklung existieren dagegen auch Einschränkungen. So zeigt sich, dass die Möglichkeit überregional Fachkräfte anzuziehen sowie die Auspendlerquote zu reduzieren in der Praxis eingeschränkt ist. Darüber hinaus wird der Betrieb durch die Unbekanntheit des Konzepts, die geringe wirtschaftliche Rentabilität der Coworking Spaces und die notwendige Bereitstellung von Personal erschwert. Damit ein förderlicher Beitrag in der Kommune entstehen kann, sind Angebote, die sich an den Bedürfnissen der Kommune orientieren, eine frühzeitige Vernetzung mit lokalen Akteuren und eine koordinierende Instanz notwendig. Für zukünftige Forschung ist vor allem die Analyse finanzieller Konzepte, damit ländliche Coworking Spaces auch langfristig wirtschaftlich rentabel sind, essentiell. Außerdem sollten Strategien zur Anziehung überregionaler Arbeits- und Fachkräfte und von Pendlern vertieft untersucht werden. Abschließend ist weitere Forschung notwendig, um zu prüfen, ob die Ergebnisse sich mit weiteren Standorten vergleichen lassen und damit eine Übertragbarkeit gewährleistet ist. Francis Helen Linkens doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13658 Wed, 23 Sep 2026 16:04:38 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13606 The Time-Triggered Architecture (TTA) has long served as a foundational model for safety-critical distributed real-time systems in domains such as avionics and automotive control, offering strong guarantees regarding extra-functional properties such as determinism, temporal predictability, fault containment, temporal composability and the ability to satisfy complex timing constraints. Historically, these systems relied on a closed-world assumption where they functioned in fully controlled environments and interfaced exclusively with design-time-known and trusted components. However, the evolving requirements of modern safety-critical real-time systems are driving a shift away from closed architectures toward open, integrated platforms characterized by high connectivity, increased complexity, adaptive resource demands, interaction with non-real-time components, and extended autonomous mission durations. This shift exposes two structural limitations of TTA. First, its static design restricts flexibility when systems must handle dynamic workloads. Second, its original fault model, which focuses solely on accidental faults, fails to address intelligent adversaries who exploit predictable temporal structures and extended mission times for time-domain and resource-exhaustion attacks. In this thesis, we examine the combined challenges of flexibility and intrusion resilience in open TTA-based systems and propose solutions that preserve TTA's extra-functional properties. As a first step, we investigate and classify the dependencies between TTA's core dependability services and its extra-functional properties, thereby establishing a foundation that enables architects to identify which aspects of TTA are essential and which can be modified or replaced while preserving the desired system properties. Building on this foundation, the thesis advances two central objectives. In the first objective, we advance flexibility within time-triggered systems. We introduce dynamic time-triggered schedulers that allocate computation slots at runtime while preserving the timing guarantees established through offline scheduling. The Dynamic Task Set Scheduler (DTSS) represents our solution within this class and supports operation with static and dynamic granularity. We enable DTSS to admit, remove and modify aperiodic tasks at runtime under time-triggered operation by integrating the Remaining Processor Capacity Analysis (RPCA) into its mechanisms. We design RPCA as part of this work to provide the analysis required for these runtime adaptations. We further examine the integration of periodic and sporadic tasks into time-triggered systems with dynamic schedulers and develop exact feasibility tests for this purpose. For sporadic tasks, we also develop runtime exact tests, the (Quick) Processor Demand Analysis ((Q)PDA*), which use periodic-demand preprocessing and quick convergence to determine feasibility within practical computational limits (achieving two to three orders of magnitude better runtime performance compared to state-of-the-art methods and our exact tests without these techniques). Our scheduler and exact tests improve flexibility under time-triggered operation while preserving TTA's extra-functional properties. In the second objective, we focus on analyzing the vulnerability of TTA-based systems to time-domain attacks, designing resource- and synchrony-exhaustion attacks targeting TTA nodes and their local bus guardians, and developing corresponding countermeasures. We establish an impossibility result for synchronizing network-agnostic guardians in the presence of malicious nodes and design our countermeasures to circumvent this limitation, thereby strengthening the resilience of time-triggered systems against intruders. Our countermeasures rely on a replica-aware rejuvenation service integrated into a trusted hardware component named SyncGuard. This component operates as a local network-agnostic guardian that coordinates network access with proactive and reactive node rejuvenation. It maintains synchrony, defends the system against resource-exhaustion attacks, and enables collision-free communication across heterogeneous network topologies without incurring the substantial overhead associated with classical network-aware variants. The contributions in this thesis establish a basis for the next generation of intrusion-tolerant and flexible distributed real-time systems, preserving the stringent dependability guarantees of the time-triggered architecture while meeting the flexibility requirements of modern open architectures. Mohammad Ibrahim Alkoudsi doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13606 Wed, 23 Sep 2026 10:20:45 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13608 Akustische Rückkopplung ist mindestens mit der Erfindung der elektroakustischen Beschallungssysteme und spätestens mit der fortschreitenden Entwicklung von Hörgeräten und Konferenzsystemen ein unumgängliches Thema geworden. Wird ein System durch die Postionierung von Lautsprecher(n) und Mikrofon(en) im gleichen akustischen Raum instabil, äußert sich das durch unangenehmes Pfeifen („Howling“) und will unter allen Umständen vermieden werden. Diese Instabilitäten treten erst ab bestimmten Verstärkungen auf, sodass das System in seiner maximalen Verstärkung deutlich beschränkt ist. In der Literatur vorgestellte Methoden versuchen entweder das Pfeifen als Symptom zu unterdrücken, oder die akustische Rückkopplung selbst aufzuheben und damit die maximal mögliche Verstärkung, bei der das System noch stabil ist, zu erhöhen. Übliche Methoden können dabei in Gruppen kategorisiert werden: Phasenmodulation, Verstärkungsreduktion, räumliche Filterung und Raummodellierung. Da die Methoden der adaptiven Rückkopplungsunterdrückung aus der Gruppe der Raummodellierung theoretisch in der Lage sind, die akustische Rückkopplung vollständig aufzuheben, haben sich diese mit der Zeit durchgesetzt. Hier wurden verschiedene Ansätze vorgeschlagen, die auf verschiedenen Optimierungsmethoden wie Least Squares, (Normalised) Least Mean Squares, oder auch Kalman-Filtern im Zeit-, Frequenz-, Teilband-, oder anderen Transformationsbereichen basieren. Die in der Literatur betrachteten Szenarien sind in der Regel einkanalig mit einem Lautsprecher und einem Mikrofon. Da moderne Konferenztechnikkomponenten meist aus einem Lautsprecher und einem Mikrofonarray, also mehreren Mikrofonkapseln bestehen, liegt es nahe, dies auch bei der akustischen Rückkopplungsunterdrückung auszunutzen. Vereinzelte Ansätze mit einem Lautsprecher und mehreren Mikrofonen (meist zwei) kombinieren die adaptive Rückkopplungsunterdrückung mit weiteren Methoden. Das Mikrofonarray wird dabei beispielsweise durch Nullsteering-Beamforming oder mehrkanaliger Rauschunterdrückung ausgenutzt und im Anschluss eine einkanalige, adaptive Rückkopplungsunterdrückung durchgeführt. Diese Arbeit schlägt vor, die Mikrofonarraysignale direkt im adaptiven Filter auszunutzen, indem der mehrkanalige, akustische Rückkopplungspfad mit Kalman-Filtern im Frequenzbereich geschätzt wird. Manuel Ruben Schalk doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13608 Wed, 23 Sep 2026 10:05:53 +0200 https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13605 Life insurers make long-term promises under financial, biometric, behavioral, and managerial uncertainty. Under Solvency II, the capital required to support these promises is summarized by the Solvency Capital Requirement (SCR), which is derived from the lower tail of the basic own funds, here referred to as Available Capital (AC), one year into the future. To estimate the AC, we need the market-consistent valuation of assets and liabilities that in general require another set of simulations. This leads to the costly nested-simulation problem. We study how modelling, valuation, discretization, and proxy choices affect the estimation of AC distributions and SCR values. Our first contribution is openIRM, an open source internal risk model for an artificial life insurer. It combines an economic scenario generator, asset-liability management, risk-factor modelling, and cash-flow projection, and estimates AC distributions and SCR values under changing market data. This provides a public and reproducible benchmark for comparing actuarial methods and serves as the basis for the numerical studies in the following chapters. We then focus on direct and indirect valuation views of AC in a risk-neutral pricing framework. We see that both methods estimate the same economic value through different cash-flows. A simplified equivalence proof leads to mixed control-variate estimators. Our numerical studies show that, depending on the level of asset-liability coupling, they can significantly reduce the original estimator’s variance. We further examine discretization in the G2++ interest-rate model by comparing exact joint sampling with Euler-Maruyama inside openIRM. The experiments show that discretization errors can accumulate and affect discounted cash flows, AC distributions, and SCR ratios. Finally, we consider uncertainty-aware proxy models for SCR estimation. They replace expensive inner simulations by a fitted machine learning model of the AC functional. We separate aleatoric from epistemic uncertainty and compare heteroskedastic Gaussian process regression with ensemble neural networks under cost constraints from openIRM. The ensemble neural networks baseline gives the strongest estimates in the leading configurations, while sequential design improves performance only in selected cases. Taken together, these results argue for actuarial methods that are aware of the persistent tension between excessive simplification and unnecessary detail. Small modelling and implementation choices can propagate through the simulation pipeline and reach the SCR-relevant tail. Reliable internal-model methodology should make these choices visible and use the underlying economic structure to guide efficient computation. Mark-Oliver Wolf doctoralthesis https://kluedo.ub.rptu.de/frontdoor/index/index/docId/13605 Wed, 23 Sep 2026 09:52:10 +0200