18069 research outputs found
Sort by
Mapping process IDs to NFSv4 I/O metrics between computing and storage nodes through Linux kernel inquiry using eBPF
Get PDFDie Verwaltung komplexer wissenschaftlicher Rechen- und Speicheranlagen, wie die am Forschungsinstitut DESY (Deutsches Elektonen-Synchrotron), stellt Systemadministratoren insbesondere bei der effektiven Diagnose und Lösung von Systemstörungen vor große Herausforderungen. In dieser Arbeit wird die Entwicklung benutzerdefinierter eBPF-Programme (extended Berkeley Packet Filter) untersucht, um den Einblick in den Betrieb des Linux-Kernels zu ermöglichen und den Administratoren Überwachungs- und Diagnosefunktionen zur Verfügung zu stellen. Die vorgeschlagenen Programme zielen darauf ab, die Systemadministration insbesondere im Hinblick auf die Handhabung von Problemen zu vereinfachen und zu
beschleunigen. Dabei wird die Fähigkeit von eBPF genutzt, Echtzeiteinblicke in den Kernel mit minimalen Leistungseinbußen zur Verfügung zu stellen.Managing complex scientific computing and storage facilities such as the ones at the research center DESY (Deutsches Elektonen-Synchrotron), presents significant challenges for system administrators, particularly in diagnosing and resolving issues effectively. This thesis explores the development of custom eBPF (extended Berkeley Packet Filter) programs to enhance visibility into Linux kernel operations and provide monitoring and diagnostic capabilities to administrators. By leveraging eBPF’s ability to enable real-time insights into the kernel with minimal performance costs, the proposed programs in particular aim to simplify and accelerate system administration practices with regard to managing issues
Schamdynamiken bei autistischen Kindern
Get PDFZusammenfassend lässt sich sagen, dass Schamdynamiken bei autistischen Kindern ein komplexes und vielschichtiges Thema darstellen, das sowohl individuelle als auch gesellschaftliche Dimensionen umfasst. Die Arbeit hat gezeigt, dass ein tiefes Verständnis für diese Dynamiken unerlässlich ist, um die Lebensqualität autistischer Kinder zu verbessern und ihre Inklusion in der Gesellschaft zu fördern. Pädagogische Einrichtungen spielen eine entscheidende Rolle dabei, ein Umfeld zu schaffen, das autistische Kinder gezielt unterstützt und in dem Kinder lernen können, mit Scham konstruktiv umzugehen.
Zukünftige Forschungen sollten sich darauf konzentrieren, spezifische Interventionsstrategien zu entwickeln, die die Schamdynamiken bei autistischen Kindern gezielt ansprechen und die Resilienz sowie das Selbstmitgefühl fördern. Ein interdisziplinärer Ansatz, der Psychologie, Pädagogik und Soziologie miteinander verbindet, könnte hierbei neue Perspektiven eröffnen und dazu beitragen, das Verständnis für die Bedürfnisse autistischer Kinder weiter zu vertiefen
Die Bedeutung des Habitus als Überwindung einer Schichtgrenze im Kontext von Führungspositionen : Habitus als Vererbung von Karrierechancen?
Get PDFTechnical analysis of the accident : Swiss Flight LX1885 with an Airbus A220 – from uncontained engine failure via smoke on board, a dead flight attendant to alleged cover up at Austrian's federal safety investigation authority (SUB)
Get PDFAir conditioning systems of all large passenger aircraft (except the Boeing 787) take the air unfiltered from the compressor of the engine (bleed air)! Any contamination in normal operation and in failure cases (as in LX1885) goes straight into cockpit and cabin. Such a system design is inadequate as stated for decades in SAE AIR 1168-7A: "... the risk of obtaining contaminated air from the engine may preclude its use for transport aircraft, regardless of other reasons [like financial advantages]." --- Why is smoke on board after a severe engine failure? The air conditioning system takes air from the engine! --- Why a dead flight attendant? Oil fumes are toxic, and aircraft do not offer adequate breathing protection! --- Why is the Austrian Air Accident Investigator (SUB) providing an intermediate report below standards? Why is the Austrian Air Accident Investigator (SUB) prosecuted? Why is the case transferred to the STSB in Switzerland? Aircraft manufacturers / airlines do not have an interest that the answers given above get into an (intermediate) report!NonPeerReviewe
Possibilities and limitations using Flightradar24, ADS-B and mode S data for aircraft performance analysis – an overview
Get PDFAutomatic dependent surveillance broadcast, short ADS-B, is used for surveillance purposes by air traffic control. Besides that, it is used by flight tracking services and gains popularity among private users. In this project, two different sources of flight data are discussed: 1.) Data provided by the flight tracking service Flightradar24: Different cases of possible aircraft performance or flight analysis are presented. It is evaluated whether Flightradar24 data is suitable for this analysis. To demonstrate the value of Flightradar24 data, a study of initial cruise altitude and step climbs with 440 data sets of different aircraft types is conducted and the first findings are derived. 2.) A homebuilt receiver for ADS-B and other Mode-S data is used. The technology behind ADS-B is presented and a decoding script for ADS-B and Comm-B messages, based on the open source python library pyModeS, is developed. The potential of this data source is shown by analyzing flight data from an aircraft using the homebuilt set-up. The different sources are compared, and the results show that the data sources differ significantly in terms of the number of parameters, data rate, and coverage: 1.) offers more coverage, but 2.) offers more parameters.NonPeerReviewe
Inverse test based estimation of equivalent acoustic excitation based on aircraft cabin response measurements
Get PDFNonPeerReviewe
Numerical simulation of multi-rotor diffuser augmented wind turbine system using the actuator line method in PyFR solver
Get PDFThis study develops a high-order numerical model, AL-PyFR(DAWT), to analyze multi-rotor diffuser-augmented wind turbine (MRDAWT) systems using the actuator line method within PyFR. The model is validated against a 5-rotor DAWT experiment, accurately capturing asymmetric rotor power, wake interactions, and diffuser effects. A parametric study on a 9-rotor system shows rotor spacing and tip speed ratio strongly affect power output and efficiency, highlighting the need for system-level optimization.NonPeerReviewe
Beyond single rotors : enhancing performance in multi-rotor wind turbines
Get PDFThis presentation details advanced modeling and real-time control for a 23-rotor, 5 MW fixed-pitch wind turbine. The work leverages modeling of aerodynamic interactions and dynamic inflow to develop strategies that maximize energy capture and manage structural loads through thrust reallocation and failure compensation. A central result is the introduction of a novel, efficient control algorithm that uses intentional yaw misalignment ("furling") to actively reduce structural bending moments, providing a simple, self-optimizing approach for next-generation multi-rotor architectures.NonPeerReviewe
Operation and maintenance of multi-rotor wind turbines : insights from a case study and a startup simulation solution
Get PDFMulti-Rotor Systems (MRS) present promising opportunities for enhanced energy capture, modularity, and redundancy, but they also introduce unique operation and maintenance (O&M) challenges. With increased component counts, interdependent failure modes, and complex maintenance logistics, effective O&M strategies are critical to ensuring system reliability and cost-effectiveness. This presentation shares key insights from a case study on the O&M of multi-rotor wind turbines at FINO-1 site in the North Sea. Quantitative results are presented to illustrate the impact of maintenance strategies on downtime, availability, and overall lifecycle cost.
Building on these findings, the session introduces a startup-developed simulation solution designed to support data-driven decision-making for O&M planning. Developed by Ventarion, the simulation tool (Ventarion Sim) enables scenario analysis and strategy evaluation for both multi-rotor and conventional wind turbine systems. By combining technical analysis, case study results, and applied innovation, the talk bridges the gap between academic research and real-world practice, offering valuable perspectives for researchers, operators, and technology developers seeking to advance wind turbine O&M.NonPeerReviewe
An Explainability Analysis of BERT’s Interpretability with GNN-Generated Knowledge Graph Embeddings Delivered through Soft Prompts
Get PDFTrotz großer Fortschritte von großen Sprachmodellen (LLMs) der letzten Jahre in praktischen Anwendungs-zenarien leiden LLMs an Halluzinationen und Biases. Mit retrieval augmented generation (RAG) werden den Sprachmodellen zur Laufzeit eine Faktenbasis übergeben, auf die sie sich beziehen können. Bei Knowledge Graphen (KG) basiert dieser Prozess auf graph representation learning (GRL), einem Aufgabenfeld, welcher die Strukturen von Knoten und Kanten der Graphen auf niedrigsdimensionale Vektorräume abbildet. Die sogenannten knowledge graph embeddings (KGEs) bilden dabei die strukturelle Einbettung von Knoten und Kanten dieser KGs ab. In dieser Arbeit trainieren wir ein BERT-Modell auf die Klassifikationsaufgabe:
link prediction. Wir weisen nach, dass sich die Performance steigert, wenn dem BERT-Modell nicht nur die natürlich sprachigen Elemente des KG übergeben werden, sondern vorverarbeitete KGEs, die von graph neuronalen Netzen (GNN) generiert worden sind. Wir untersuchen dann mit Hilfe von explainable AI (XAI) Methoden die inneren Zustände und Verhaltensweisen des BERT-Modells. Es stellte sich dabei heraus, dass BERT die aus dem GNN generierten KGEs nur teilweise zu interpretieren vermag. Dabei unterschied das Modell zwischen verschiedenen Strategien, zwischen denen es anhand weniger offensichtlichen Faktoren entschieden hat. Als eine der wichtigsten Einflussfaktoren stellte sich die Konnektivität von Knoten heraus.Despite major advances in large language models (LLMs) in recent years in practical scenarios, LLMs suffer from hallucinations and biases. With retrieval augmented generation (RAG), the language models are given a fact base at runtime to which they can refer. For knowledge graphs (KG), this process is based on graph representation learning (GRL), a task field that maps the structures of nodes and edges of the graphs to low-dimensional vector spaces. The so-called knowledge graph embeddings (KGEs) represent the structural embedding of nodes and edges of these KGs. In this work, we train a BERT model on the classification task: link
prediction. We prove that the performance increases when the BERT model is not only given the natural language elements of the KG, but also preprocessed KGEs generated by graph neural networks (GNN). We then use explainable AI (XAI) methods to examine the internal states and behaviors of the BERT model. It turned out that BERT is only partially able to interpret the KGEs generated from the GNN. The model distinguished between different strategies, between which it decided on the basis of less obvious factors. One of the most important influencing factors turned out to be the connectivity of nodes