335598 research outputs found
Sort by
Efficient Methods for Solving Direct and Inverse Scattering Problems in Locally Perturbed Periodic Structures
No full textEffect of Ionomer-to-Carbon Ratio on PEMFC Carbon Corrosion: An Electrochemical Study
No full textThis study investigates the effect of the ionomer-to-carbon (I/C) weight ratio in polymer electrolyte membrane fuel cell (PEMFC) cathode catalyst layers (CCLs) on carbon corrosion during high-potential accelerated stress tests (ASTs). Membrane electrode assemblies (MEAs) with I/C ratios of 0.5, 0.85, and 1.2 were analyzed using polarization curves, cyclic voltammetry, limiting current measurements, and electrochemical impedance spectroscopy. Impedance data analysis, based on distribution of relaxation times and transmission line modeling, showed that higher I/C ratios (0.85 and 1.2) exhibit superior beginning-of-life (BoL) performance due to lower ionic resistance in the CCL. However, the MEA with a lower I/C ratio (0.5) exhibited a performance improvement of up to 35% during initial AST cycles and enhanced carbon corrosion resistance. Compared to BoL, performance improved significantly due to a 28%–46% reduction in charge transfer resistance and a 91% reduction in ionic resistance. These findings emphasize the trade-off between BoL performance and long-term durability when determining the optimal I/C ratio. They also underscore the need for further investigation into how the I/C ratio influences CCL structure and electrochemistry. Optimizing the I/C ratio has the potential to substantially improve PEMFC electrode performance and durability, guiding the design of more resilient catalyst layers
Spiking Neural Networks for Communication Systems: Encoding Schemes, Learning Algorithms, and Equalization Techniques
No full textKünstliche Intelligenz, insbesondere maschinelles Lernen mit künstlichen neu-
ronalen Netzen (ANNs), bietet Lösungsansätze für die wachsende Komplexität
moderner Kommunikationssysteme und ermöglicht Datenübertragung nahe der
theoretischen Grenze. Die steigende Komplexität geht jedoch mit einem hohen
Energieverbrauch und somit mit energieintensiven Systemen einher.
Gepulste neuronale Netze (SNNs) sind vom menschlichen Gehirn inspirierte Mod-
elle, die durch ereignisgesteuerte Mechanismen energieeffiziente Signalverarbeitung
in Echtzeit ermöglichen. Sie unterscheiden sich von ANNs durch ihre inhärente
zeitliche Dynamik, sowie durch ihre Informationsverarbeitung in Form kurzer
binärer Pulse. Offene Herausforderungen sind insbesondere die Wahl geeigneter
Lernregeln und neuronaler Kodierungen zur Konvertierung realer Signale in Pulse.
Diese Arbeit untersucht den Entwurf SNN-basierter Empfänger für verrauschte
sowie frequenzselektive zeitinvariante Kanäle. Der erste Teil fokussiert sich auf
die Entwicklung eines SNN-basierten Detektors für einen verrauschten Kanal.
Untersucht werden drei Lernregeln: eine biologisch inspirierte und zwei gradien-
tenbasierte, wobei die gradientenbasierte Lernregel “Backpropagation through
time mit Ersatzgradienten” als vielversprechende Lernregel identifiziert wird. Weit-
erhin werden verschiedene neuronale Kodierungen untersucht, wobei drei vielver-
sprechende Kandidaten identifiziert werden, z.B. Quantization encoding (QE).
Der zweite Teil widmet sich SNN-basierten Entzerrern und Demappern. Zwei
Architekturen, mit und ohne Entscheidungsrückkopplung, werden jeweils mit den
drei neuronalen Kodierungen kombiniert. Für das Modell einer nicht-kohärenten
optischen Übertragung werden die Ansätze bezüglich ihrer Leistungsfähigkeit und
der Anzahl generierter Pulse verglichen. Der Einsatz von Entscheidungsrückkop-
plung und QE ermöglicht leistungsfähige Entzerrer mit einer geringen Anzahl an
generierten Pulsen. Bemerkenswerterweise übertreffen SNN-basierte Entzerrer
ANN-basierte deutlich hinsichtlich ihrer Leistungsfähigkeit.
Der dritte Teil nutzt Methoden des verstärkenden Lernens (RL) um eine neue
Lernregel für SNNs sowie der neuronalen Kodierung herzuleiten. Es wird ein
RL-basierter Update-Algorithmus eingeführt, der keine Backpropagation benötigt.
Für den SNN-basierten Entzerrer und Demapper werden mittels des neuen Al-
gorithmus die Parameter der neuronalen Kodierung optimiert, wodurch ohne
Leistungseinbußen die Laufzeit, Komplexität und Anzahl der pro Inferenz gener-
ierter Pulse erheblich reduziert wird.
Diese Arbeit leistet einen Beitrag zum erfolgreichen Entwurf von SNN-basierten
Empfängern. Durch die Diskussion zentraler Herausforderungen erleichtert sie
zukünftige Fortschritte im Entwurf und Einsatz energieeffizienter Empfänger auf
Basis von SNNs
Key Elements of Trust in Building Renovation Data Management: A Usability Study of a Centralized Platform
No full textThe urgent need for building renovation is growing, yet the availability of machine-readable data remains limited, hindering the decision-making process for building owners. This paper addresses this gap by identifying key elements in document digitization, data enrichment, and data analysis that are essential to fostering trust in data processing. To address this, we developed a mock-up based on design thinking principles that aims to consolidate existing building information into a central, accessible location. This platform provides users with a comprehensive overview of building data. We conducted a user study with 44 participants to evaluate the usability of the platform using the System Usability Scale (SUS). The results showed a high SUS score, reflecting strong usability and positive feedback. Participants highlighted the value of centralizing building data, which significantly supports renovation decision-making. The results underscore the platform’s potential to drive digital transformation in the building sectors, marking a critical step forward in renovation planning
Concurrent Self-testing and Uncertainty Estimation of Neural Networks Using Uncertainty Fingerprint
No full textNeural networks (NNs) are increasingly used in always-on safety-critical applications deployed on hardware accelerators employing various memory technologies. Reliable, continuous operation of NN is essential for safety-critical applications. During online operation, NNs are susceptible to (single and multiple) permanent and soft errors due to factors such as radiation, aging, and thermal effects. Explicit testing methods for hardware accelerators cannot detect transient faults during inference, are unsuitable for always-on applications, and require extensive test vector generation and storage. Therefore, in this paper, we propose the uncertainty fingerprint approach that represents the online fault status of NN. Furthermore, we propose a dual-head NN topology specifically designed to produce uncertainty fingerprints and the primary prediction of the NN in a single shot. During the online operation, by matching the uncertainty fingerprint, we can concurrently self-test NNs with up to 100% coverage in the backbone and in the full model (with a modified approach) with a low false positive rate while maintaining the performance of the primary task similar to the baseline. Compared to existing works, memory overhead is reduced by up to 243.7 MB, multiply and accumulate (MAC) operations are reduced by up to 10000×, and false-positive rates are reduced by up to 89%
Quantifying the spatial extent and attenuation of lake thermal regulation at diurnal scales under extreme heat
No full textLakes worldwide are experiencing intensifying extreme heat, with escalating ecological impacts. Despite lakes\u27 role as thermal buffers to modulate air temperature is well-documented, the spatial propagation dynamics of lake effects remain poorly understood due to complex interactions of lake-atmosphere. This study proposes a synergistic WRF modeling and directional buffer analysis framework to investigate the spatial propagation dynamics and underlying physical mechanisms of lake-induced thermal regulation during extreme heat, focusing on Poyang Lake, China\u27s largest freshwater lake. The results demonstrate a pronounced diurnal asymmetry in lake-induced thermal effects, with distinct spatial propagation characteristics between daytime and nighttime periods. Daytime cooling exhibits an intensity of −1.16 °C, with its influence confined within a 40 km radius, showing a relatively rapid attenuation rate of 0.28 °C per 10 km. In contrast, nighttime warming (+0.97 °C) propagates 1.75 times farther than its daytime counterpart, extending up to 70 km downwind while maintaining a slower attenuation rate of 0.13 °C per 10 km. Directional analysis reveals north-oriented propagation of lake thermal effects, influenced by prevailing southerly winds and lake-land breeze. Vertical profile analysis reveals distinct altitudinal penetration of lake-induced thermal effects, with daytime influences confined below 900 hPa while nighttime impacts extend up to 700 hPa. Daytime cooling extent is limited by turbulent mixing, whereas nighttime warming is enhanced by stable air conditioning and advective transport. The study underscores the role of lake-atmosphere interactions in mitigating regional climate extremes, providing critical insights for nature-based heat adaptation strategies in lake-rich regions. These findings advance the understanding of inland water bodies as active climate regulators under anthropogenic warming
Cyclopentadienyl Complexes of Technetium
No full textThe number of structurally investigated cyclopentadienyl (Cp) complexes of technetium is limited in contrast to the situation with its heavier homolog, rhenium. Although this could be attributed to the radioactivity of all isotopes of the radioelement, there are also clear chemical differences to analogous compounds of the other group seven elements, manganese and rhenium. Technetium Cp compounds are known with the metal in the oxidation states “+1” to “+7”, with a clear dominance of Tc(I) carbonyls and nitrosyls. Corresponding carbonyl complexes also play a significant role in the development of Tc-based radiopharmaceuticals with the aromatic ring as an ideal position for the attachment of biomarkers. In this paper, the present status of the synthetic and structural chemistry of technetium with Cp ligands is discussed, together with recent developments in the corresponding Tc labeling chemistry
GeoLaB – the URL for Geothermal Energy
No full textGeoLaB (Geothermal Laboratory in the Crystalline Basement) constitutes a novel underground research infrastructure (URL), presently in its exploration and confirmation of site suitability phase, tailored to investigate coupled thermal, hydraulic, mechanical, and chemical (THMC) processes in fractured crystalline rock and advance understanding and implementation of Enhanced Geothermal Systems (EGS). Located in Germany, GeoLaB aims to bridge the gap between laboratory research and field-scale geothermal applications. This paper presents the scientific exploration progress, strategic development, and project management framework of GeoLaB. As a collaborative initiative between leading Helmholtz Centres and academic partners, GeoLaB represents a cornerstone for Europe’s sustainable heating transition and international geothermal innovation. Through its interdisciplinary approach and integration of digital technologies, the project will contribute to safer, more efficient geothermal development and fosters global partnerships to advance renewable energy science