Hochschule Bonn-Rhein-Sieg
Publikationsserver der Hochschule Bonn-Rhein-Sieg - pub H-BRSNot a member yet
7939 research outputs found
Sort by
Enhancing the Buckling Performance of Thin-Walled Plastic Structures Through Material Optimization
No full textReducing material usage in plastic products is a key lever for improving resource efficiency and minimizing environmental impact. In thin-walled structures subjected to mechanical loading, material efficiency must be achieved without compromising structural performance. In particular, resistance to buckling, a critical failure mode, must be taken into account during product development. Due to the large number of design and process variables, many of which are interdependent, optimization approaches are uncommon in the blow-molded packaging industry. This paper presents a sensitivity-based optimization approach to improve buckling resistance by modifying the product’s material distribution. Since the sensitivity is nonlinear and depends on the product’s deformation state, various methods are developed and tested to reduce the frame-wise sensitivity data to a single sensitivity vector suitable for optimization. These methods are then tested on common extrusion blow-molded products, achieving improvements in buckling load of up to 60%. This approach is transferable to other thin-walled structures across various engineering domains, offering a pathway toward lightweight yet load-compliant designs
Additional file 1 of Touching Surfaces – Presence of microorganisms on antimicrobial metal surfaces on the International Space Station and in German schools
No full textAdditional file
Viewpoint-Aware Sampling for Effective Online Domain Incremental Learning
No full textWe investigate the problem of online domain continual learning for image classification. Within an extended series of tasks, continual learning encounters the issue of catastrophic forgetting. To mitigate this challenge, one may employ a memory-replay strategy, a technique involving the re-visitation of stored samples in a buffer when new tasks are introduced. However, the memory budget available to autonomous agents, such as robots, is typically limited, making the selection of representative examples crucial. An effective strategy to ensure representativeness is to select diverse examples. To this end, we propose a novel on-the-fly sampling policy, called Viewpoint-Aware Sampling (VAS), which maintains diversity in the memory buffer by selecting examples from different visual perspectives. We empirically evaluate the effectiveness of VAS across the OpenLORIS-Object and the CORe50-NI benchmark and find that it consistently outperforms state-of-the-art methods in terms of average accuracy, backward transfer, and forward transfer, while requiring fewer computational resources
Development of a predictive control model for grid-integrated photovoltaic-diesel-battery systems in the context of the Ghanaian health sector
No full textDiese Dissertation untersucht und validiert vorhersagebasierte Steuerungsmethoden für PV-Diesel-Hybridsysteme in Mikronetzen, mit besonderem Fokus auf deren Anwendung im ghanaischen Gesundheitssektor. Dabei liegt ein besonderes Augenmerk auf der Weiterentwicklung und Integration neuartiger Prognosetechniken, die im Rahmen dieser Arbeit entwickelt wurden und über den aktuellen Stand der Technik hinausgehen. Vor Ort gesammelte Daten aus verschiedenen Gesundheitseinrichtungen in Ghana bilden die Grundlage für diese Analyse. Insbesondere der ländliche ghanaische Gesundheitssektor bietet eine einzigartige Fallstudie, da er die Bereitstellung eines universellen Energiezugangs mit der Notwendigkeit verbindet, ausreichende Zuverlässigkeit zur Senkung der Sterblichkeitsraten zu gewährleisten, was direkte Auswirkungen auf das Überleben gefährdeter Gruppen in ländlichen Gebieten des Globalen Südens hat.
Derzeit sind Dieselgeneratoren die vorherrschende Lösung zur Stromversorgung isolierter Gesundheitseinrichtungen. Obwohl sie kurzfristig effektiv sind, bringen sie erhebliche wirtschaftliche, ökologische und gesundheitliche Risiken für die umliegenden Gemeinden mit sich. Als Antwort auf diese Herausforderungen untersucht diese Arbeit das Potenzial von auf Deep Learning basierenden Vorhersagemethoden — ein Bereich, der viel Aufmerksamkeit erlangt hat, aber weiterhin mit Skepsis betrachtet wird — schlägt Weiterentwicklungen vor und integriert sie mit traditionellen Ansätzen, um Energiemanagementsysteme zu optimieren. Diese Vorhersagen werden in einem Regler eingesetzt, um den Energiefluss zwischen den Komponenten eines PV-Diesel-Hybridsystems zu optimieren. Ein grundlegendes
Prinzip dieser Arbeit ist die wirtschaftliche Inklusivität, wobei auf freie und Open-Source-Software sowie frei zugängliche Daten zurückgegriffen wird, um sicherzustellen, dass die Ergebnisse leicht reproduzierbar und auf verschiedene Regionen und Sektoren weltweit anpassbar sind.
Die Ergebnisse zeigen, dass Deep-Learning-basierte Vorhersagemethoden, wenn sie in prädiktiven Steuerungssystemen für das Energiemanagement angewendet werden, die wirtschaftlichen Kosten, den ökologischen Fußabdruck und gesundheitliche Auswirkungen erheblich reduzieren können. Gleichzeitig erhöhen sie die Systemzuverlässigkeit, alles unter primärer Nutzung intellektuellen Kapitals bei minimalem zusätzlichen Hardwarebedarf. Diese Arbeit soll als Grundlage für zukünftige Untersuchungen und praxisnahe Anwendungen in verschiedenen Sektoren und Regionen weltweit dienen.
Die Arbeit beginnt mit einer Diskussion der Herausforderungen des Energiezugangs, denen der ghanaische Gesundheitssektor gegenübersteht, und formuliert dabei die in dieser Dissertation behandelten Forschungsfragen. Sie bietet einen Überblick über bestehende Vorhersage- und Steuerungsmethoden sowie die Begründung für die Auswahl spezifischer Modellierungsansätze. Eine vergleichende Analyse von Lastvorhersagemethoden wird präsentiert und durch reale Daten aus ghanaischen Gesundheitseinrichtungen validiert. Um die Variabilität der Solarenergie zu berücksichtigen und die kurzfristige Vorhersagegenauigkeit zu verbessern, wird eine neuartige Nowcasting-Methode unter Verwendung von Wolkenkameras vorgestellt. Diese Vorhersagemethoden werden nahtlos mit einem sektorspezifischen numerischen Wettervorhersagemodell in ein modellprädiktives Steuerungsframework integriert. Die Arbeit schließt mit einer Zusammenfassung der Ergebnisse, betont deren Einfluss auf den ghanaischen Gesundheitssektor und skizziert Möglichkeiten zur Skalierung dieser Lösungen auf andere Sektoren und Regionen weltweit.This thesis investigates and validates forecast-based control methods for microgrid PV-diesel hybrid systems, focusing on their application in the Ghanaian health sector. Data collected on-site at various health facilities in Ghana form the foundation of this analysis. A particular emphasis is placed on advancing and integrating novel forecasting techniques, developed as part of this research, which go beyond the current state of the art. The rural Ghanaian health sector offers a unique case study, linking the goal of providing universal energy access with the critical need for reliable systems to manage mortality rates, directly impacting the survival of vulnerable groups in rural areas of the Global South.
Currently, diesel generators are the predominant solution for supplying power to isolated health facilities. While effective in the short term, they pose significant economic burdens, environmental degradation, and health risks to surrounding communities. In response to these challenges, this thesis explores and proposes novel deep learning-based forecasting methods — an area that has garnered significant attention yet remains met with skepticism — and integrates them with traditional approaches to optimize energy management systems. These forecasts are employed within a controller to optimize the dispatch of a PV-diesel hybrid system. A foundational premise
of this work is economic inclusivity, relying on free and open-source software and freely accessible data to ensure that the results can be easily reproduced and adapted to various regions and sectors worldwide.
The findings demonstrate that deep learning-based forecasting methods, when applied in predictive control systems for energy management, can substantially reduce economic costs, ecological footprints, and health impacts. Simultaneously, they enhance system reliability, all while primarily leveraging intellectual capital with minimal additional hardware requirements. This work aims to serve as a foundation for future exploration and real-world applications across different sectors and regions globally.
The thesis begins with a discussion of the energy access challenges faced by the health sector in Ghana, while formulating the research questions being addressed in this thesis. It provides an overview of existing forecasting and control methods, along with the rationale for selecting specific approaches. A comparative analysis of load forecasting methods is presented, validated through real-world data collected from Ghanaian health facilities. To address solar energy variability and improve short-term forecasting accuracy, a novel now-casting method using all-sky imagery is introduced. These forecasting methods are seamlessly integrated with a sector-specific numerical weather prediction model into a model predictive control framework. The thesis concludes by summarizing the results, highlighting their impact on Ghana’s health sector, and outlining opportunities for scaling these solutions to other sectors and regions globally
Projektabschlussbericht GaN-HighPower: Kosten- und gewichtseffiziente PV- und Batterie-Wechselrichter großer Leistung für internationale Märkte der Zukunft durch Gallium-Nitrid (GaN) Halbleiter; Teilvorhaben des Fraunhofer IEE: Entwicklung eines Demonstrators für PV-Anwendungen auf GaN Basis
No full textZiel des Verbundforschungsvorhabens GaN-HighPower war es, die nächste Generation kostengünstiger, ressourcenschonender und effizienter Stromrichter für Photovoltaik-Anwendungen zu erforschen und zu erproben, wobei der Fokus auf Stringwechselrichtern mit größerer Leistung im Bereich von 150 kVA lag. Hierfür sollten Galliumnitrid (GaN) Halbleitermodule zusammen mit anwendungsorientiert stark verbesserten induktiven Bauelementen und Stromsensoren erforscht und erprobt werden. Die GaN-Technologie ermöglicht aufgrund ihrer großen Bandlücke (Wide Band Gap, WBG) eine Miniaturisierung des Halbleiterchips, was sich wiederum für höhere erreichbare Schaltfrequenzen nutzen lässt. Dadurch sinkt die Energie, die in den passiven Bauteilen (Induktivitäten, Kapazitäten) einer Schaltung kurzfristig gespeichert werden muss, wodurch diese ebenfalls deutlich kleiner und leichter ausfallen können, was wiederum Kosten und Ressourcenverbrauch senkt. Damit die GaN-Halbleiter diese Optionen eröffnen können, braucht es jedoch Treiberschaltungen, die diese Halbleiter auch bei hohen Leistungen zuverlässig und effizient ansteuern können, sowie eine dazu passende Systemregelung. Diese neuen Ansätze für Halbleiter und induktive Bauteile sollten mit einer neuen Stromsensorik kombiniert werden, die in der Lage ist, die hohen Schaltfrequenzen der Halbleiter messtechnisch zu erfassen.
The goal of the collaborative research project GaN-HighPower was to investigate and test the next generation of cost-effective, resource-saving, and efficient power converters for photovoltaic applications, focusing on string inverters with a higher output power in the range of 150 kVA. For this purpose, gallium nitride (GaN) semiconductor modules were to be researched and tested alongside application-oriented, significantly improved inductive components and current sensors. The GaN technology enables miniaturization of the semiconductor chip due to its wide bandgap (Wide Band Gap, WBG), which can be utilized for achieving higher switching frequencies. As a result, the energy that must be temporarily stored in the passive components (inductors, capacitors) of a circuit decreases, allowing these components to be significantly smaller and lighter, thereby reducing costs and resource consumption. However, to enable GaN semiconductors to unlock these options, driver circuits are needed that can reliably and efficiently control these semiconductors at high power levels, along with an appropriate system control. These new approaches for semiconductors and inductive components should be combined with a new current sensing technology capable of accurately measuring the high switching frequencies of the semiconductors
Phishing Susceptibility and the (In-)Effectiveness of Common Anti-Phishing Interventions in a Large University Hospital
No full textPhishing attacks via email remain a major entry point for security and privacy breaches in hospitals. In the European Union, faced with both regulatory pressure to act and limited resources for cybersecurity, hospitals may resort to minimal-effort, off-the-shelf anti-phishing interventions such as warning banners in enterprise email systems. However, their effectiveness remains uncertain, particularly given the highly diverse workforce comprising medical, nursing, functional, administrative, IT, and other staff groups. We conducted a large-scale phishing simulation at a German university hospital, targeting 7,044 email accounts, to analyze how phishing susceptibility varies across staff groups, how email characteristics---such as timing, tone, context, and persuasive framing---influence susceptibility, and how 11 common in-situ anti-phishing interventions affect risky staff behavior. We found that susceptibility but also intervention effectiveness differed markedly across staff groups. Even a small number of phishing emails posed a substantial risk that persisted for about three days. The most effective interventions involved robust technical detection, including spam filtering and in-email phishing warnings. Friction-based measures, such as disabling links and active warning pages, showed mixed but promising effects. In contrast, display name suppression and the widely used method of generic [EXTERNAL] email tagging had no or inconsistent effects. Surveys revealed that some staff reacted with fear, shame, guilt, and hostility, highlighting the ethical challenges of such simulations. Our findings provide actionable guidance for phishing resilience in healthcare and similarly complex organizations
Long-tailed class I myosins rely on tail-mediated phosphoinositide recognition for specific membrane recruitment
No full textBACKGROUND
Class I myosins are essential mediators of membrane-cytoskeleton interactions that support key cellular processes such as endocytosis, secretion, intracellular trafficking, and mitosis. However, the mechanisms driving isoform-specific targeting to membrane domains enriched in signaling lipids as well as their stage-dependent recruitment to mitotic structures during cell division remain poorly defined.
METHODS AND APPROACH
Using Dictyostelium discoideum as a highly phagocytic cell model, we demonstrate that long-tailed myosin-1 isoforms (myosin-1B, -1C, and - 1D) exhibit distinct lipid and cytoskeletal binding profiles shaped by their modular tails and variations within the phosphoinositide binding motif. Homology-based structural modelling of the PH-like lipid binding domain within the TH1 sequence, combined with molecular docking explains their differential lipid affinities. Kinetic equilibrium modelling with quantitative data suggests these differences enable cooperative or competitive isoform localization within cells providing a mechanism for temporally controlled recruitment of the myosins in response to dynamic changes in membrane composition and expression profiles. These biochemical insights are corroborated by confocal live-cell imaging, which reveals phosphoinositides-dependent localization dynamics and isoform-specific targeting of the myosins during vegetative growth and mitotic progression.
RESULTS
Myosin-C exhibits phosphoinositide binding preferences nearly reciprocal to those of myosin-1D, especially between mono- and triple phosphorylated phosphoinositides, and shows the strongest tail-mediated, ATP-independent actin binding. Myosin-1B, in contrast, displays low affinity for monophosphorylated phosphoinositides, intermediate actin binding ability, and no microtubule interaction. The comparable affinities of all three myosins for PI(3,5)P₂ and PI(4,5)P₂, the major PIP species at the cell cortex, facilitate their accumulation at membrane protrusions. Live-cell imaging confirms that myosin-1D preferentially associates with PI(3,4,5)P₃- and PI(3)P-enriched endosomes during macropinocytosis and phagocytosis, consistent with its higher binding affinity for these phosphoinositides. Conversely, myosin-1C localization is governed by both actin and phosphoinositides, enabling a rapid dissociation from early endosomes to retarget the cortex and accumulate at actin-rich phagocytic cup tips. Upon mitotic entry, myosin-1D, similar to myosin-1C, redistributes from endosomal compartments to the mitotic apparatus, where it decorates membrane-enclosed nuclear chromatin masses through its TH1 domain and later associates with spindle pole microtubules. This contrasts with myosin-1C, which selectively targets spindle microtubules throughout mitosis, reflecting its stronger microtubule-binding affinity. Inhibition of PI3-kinase disrupts membrane recruitment of both isoforms, confirming their phosphoinositide-dependent localization. These findings reveal an isoform-specific mechanism underlying myosin-1 targeting during endocytosis and mitosis.
CONCLUSION
Collectively, these findings establish a phosphoinositide- and cytoskeleton-guided mechanism that governs myosin-1 isoform-specific functions, providing new insights into how motor proteins interpret complex lipid and cytoskeletal cues to regulate membrane remodelling and cytoskeletal dynamics across cellular states