149544 research outputs found
Sort by
Generalizability of Concept Knowledge in Machine Learning Using TCAV Scores: A Case Study Using Different Skin-Lesion Datasets
Get PDFIn safety-critical fields, such as skin-lesion classification, interpretability of the decisions of a machine learning model is required. This can be provided through concept-based interpretability methods like testing with concept activation vectors (TCAV). TCAV quantifies how specific human-understandable concepts influence a model's decisions. A further issue affecting the performance of ML models is generalizability, i.e., how well a model generalizes to unseen data from a different domain. It is currently unknown how the interpretability provided by TCAV is affected by domain shifts. Here we show that TCAV-based interpretability is predominantly unaffected by domain shifts. To that end, we introduce concept detection scores (CDS) as aggregated TCAV scores which are directionally unified and thus a suitable evaluation metric. The results show only small differences between CDS within domain and across domain for 48 models trained on three distinct source domains. This increases the viability of TCAV as an interpretability tool since it can be used without additional effort to manage generalizability
Sulfuric Acid Splitting (SAS) reactor of Iodine-Sulfur and Hybrid-Sulfur Thermochemical Water Splitting Cycles for Hydrogen production: Experimental Activity and Modelling
No full textFlex-Console - more than just a furniture for Space Operations - experiences from LUNA project
No full textTraditional operator consoles used in control centers in space operations are often rigid and inflexible, hindering the efficiency and adaptability to evolving mission needs. This paper introduces the Flex-Console, an approach that redefines the concept of a console as an integrated system encompassing not only ergonomic furniture but also a sophisticated project switching backend.
The presented concept of Flex-Console empowers operators with flexibility by enabling them to seamlessly reconfigure the console layout and functionality to accommodate different mission phases, operator roles, and data streams. Furthermore, it allows for the customization of user interfaces and display configurations to individual operator preferences and cognitive styles. Finally, the Flex-Console facilitates communication and data sharing among operators, experts, and remote teams.
The Flex-Console development addresses several key challenges, including integrating hardware components (displays, keyboards, input devices) with software systems (project switching, data distribution, user interface management), ensuring optimal ergonomics and user comfort for extended periods of operation, and designing a scalable and adaptable system that can accommodate the diverse needs of different mission control centers.
The Flex-Console will be initially adopted by the LUNA (Lunar surface analogue simulation facility, joint-venture between DLR and ESA), serving as a pilot implementation for this innovative technology. Subsequently, it will be integrated into the newly created Ground Operations Center at GSOC, encompassing all Multi-Mission projects first and ultimately also human spaceflight. The long-term vision is to extend the Flex-Console to all GSOC control rooms and into newly initiated HECC (Human Exploration Control Center by DLR in Oberpfaffenhofen), revolutionizing the human-machine interface for space operations. Last but not least, we present shortly a concept of Control Room Laboratory which will be set up around Flex-Console and support future concept developments and simulations.
The Flex-Console represents a significant advancement in human-centered design for space operations. By prioritizing flexibility, adaptability, and user experience, it empowers operators to perform at their peak, enhancing mission efficiency, safety, and overall success
Managed Automated Driving
Get PDFThe presentation is about managed automated driving in the context of the project IMoGer and the DLR facilities as the U-Shift and transportation infrastructure
Untersuchung der Thermodiffusion in Wasserstoffflammen
No full textIm Zuge des Klimawandels gewinnt die Energiegewinnung durch Wasserstoffverbrennung zunehmend an Bedeutung. Bei der Wasserstoffverbrennung liegt ein besonderes Augenmerk auf diffusiven Effekten. Die Diffusion infolge eines Temperaturgradienten wird als Soret-Effekt oder Thermodiffusion bezeichnet. Ziel dieser Masterarbeit ist es, den Einfluss des Soret-Effekts auf die laminare und turbulente Wasserstoffverbrennung zu untersuchen. Dabei soll überprüft werden, inwieweit vereinfachte Modelle der Thermodiffusion die Effekte der vollständigen, jedoch rechenintensiven, Multicomponentenmodellierung für den Soret-Effekt in Wasserstoffflammen zuverlässig abbilden können und unter welchen Bedingungen die Grenzen dieser vereinfachten Modellansätze erreicht werden. Hierzu wurden das Mixture-Averaged-Thermodiffusionsmodell nach Hirschfelder-Curtiss-Bird (HCB) [47] sowie das Thermodiffusionsmodell nach Kuo et al. [56] in die Simulationssoftware Cantera implementiert und systematisch mit dem vorhandenen Multicomponent-Soret-Modell validiert und verglichen. Untersuchungen an laminaren, frei propagierenden, vorgemischten eindimensionalen Flammen in Cantera verdeutlichen, dass die Berücksichtigung des Soret-Effekts essenziell ist, um experimentell ermittelte laminare Flammengeschwindigkeiten numerisch reproduzieren zu können. Zudem wird gezeigt, dass das Modell nach Hirschfelder-Curtiss-Bird sowie das Modell von Kuo et al. einen sinnvollen Ersatz für das rechenintensive Multicomponent-Soret-Modell darstellen, wobei die Güte der Approximation von den jeweiligen Randbedingungen abhängt. Insgesamt bieten die vereinfachten Ansätze einen guten Kompromiss zwischen Rechenaufwand und Genauigkeit. Ergänzende LES-Simulationen dreidimensionaler turbulenter Boundary-Layer-Flashback-Flammen bestätigen die Relevanz des Soret-Effekts auch unter realitätsnahen Bedingungen. Dabei beeinflusst der Soret-Effekt zentrale Flammenparameter wie Flammenrückschlagsgeschwindigkeit und Wärmefreisetzung unabhängig von Umgebungstemperatur und Äquivalenzverhältnis. Folglich ist der Soret-Effekt für die Auslegung wasserstoffbasierter Verbrennungssysteme von entscheidender Bedeutung
Quantum algorithms to solve partial differential equations in battery modelling
No full textMathematical models of electrochemical systems such as batteries or fuel cells consist of sets of coupled nonlinear partial differential equations. We describe how variational quantum algorithms (VQAs) can be used to solve these equations on a quantum computer [1]. VQAs are hybrid quantum-classical algorithms, where a cost function is calculated on a quantum computer and its parameters are optimised classically. We present a spacetime representation, inspired by the Feynman--Kitaev Hamiltonian [2], where the solution to a PDE at all times is obtained by minimising just one function.
[1] A.J. Pool, A.D. Somoza, C. Mc Keever, M. Lubasch, B. Horstmann, Phys. Rev. Res. 6 (2024),
3, 033257
[2] S. Barison, F. Vicentini, J.I. Cirac, G. Carleo, Phys. Rev. Res. 4 (2022), 4, 04316
A Comparative Study of Varying Incidence Angle Effects on a Low-Reynolds-Number Compressor Cascade Based on Experiments and Low-Fidelity and High-Fidelity Numerical Simulations
Get PDFThe trend towards higher bypass ratios and downsized cores in modern compressorsleads to locally reduced Reynolds numbers, intensifying flow separation and unsteadiness,which limits the reliability of RANS models and motivates the use of LES as a feasibleand attractive high-fidelity approach for these conditions. In this paper, we assess thecapabilities of low- and high-fidelity numerical tools for predicting the effects of varyingincidence angles for a linear compressor cascade at a Reynolds number of 150,000 anda Mach number of 0.6 based on the inflow conditions. The comparison is supported byexperiments carried out at the Transonic Cascade Wind Tunnel at the DLR in Cologne,which feature an incidence angle variation of plus/minus 5 degrees. Particular emphasisis put on the numerical setup to reproduce the cascade experiment, discussing the effectsof spanwise domain size, axial-velocity density ratio and inflow turbulence. The effectsof the incidence angle variation are studied on the basis of instantaneous and mean flowquantities with a focus on separation, transition and loss mechanisms