KITopen
Not a member yet
63711 research outputs found
Sort by
Generative AI for fast and accurate statistical computation of fluids
Get PDFWe present a generative AI algorithm for addressing the pressing task of fast, accurate, and robust statistical computation of three-dimensional turbulent fluid flows. Our algorithm, termed as GenCFD, is based on an end-to-end conditional score-based diffusion model. Through extensive numerical experimentation with a set of challenging fluid flows, we demonstrate that GenCFD provides an accurate approximation of relevant statistical quantities of interest while also efficiently generating high-quality realistic samples of turbulent fluid flows and ensuring excellent spectral resolution. In contrast, ensembles of deterministic ML algorithms, trained to minimize mean square errors, regress to the mean flow. We present rigorous theoretical results uncovering the surprising mechanisms through which diffusion models accurately generate fluid flows. These mechanisms are illustrated with solvable toy models that exhibit the mathematically relevant features of turbulent fluid flows while being amenable to explicit analytical formulae. Our codes are publicly available at https://github.com/camlab-ethz/GenCFD
Novel advanced channel reactor for spatio-temporal activity and catalyst state correlations applied for the reduction of NO by CO over Pt/Al₂O₃
Get PDFThe correlation of space- and time-resolved measurements of catalytic activity with catalyst state is an invaluable tool to advance the understanding and development of complex catalytic systems under conditions relevant to technical applications. Such an approach is employed here to investigate the catalytic deactivation in the reduction of NO by CO over Pt/Al2O3 on freshly reduced catalysts in a channel reactor and at concentrations typical for emission control. Planar laser-induced fluorescence (PLIF) is used to visualize the 2D concentration profiles and to derive space- and time-resolved NO conversion rates for different CO/NO ratios, temperatures and mass flow rates. The changes in catalytic activity are correlated with temporal and spatial changes in oxidation state determined by operando X-ray absorption spectroscopy (XAS) under the same conditions. The time scales of the changes in catalytic activity depend not only on the stoichiometry and temperature, but also on the position along the catalyst channel and differ significantly from the temporal changes of the oxidation state. The different time scales are discussed in the context of the known CO poisoning as well as the formation and storage of isocyanate on the support. Isocyanate formation temporarily counteracting CO poisoning could explain the differences in the observed time scales under different reaction conditions and at different locations on the catalyst
Estimation of melt oxidation kinetics
Get PDFThe analysis of the viscous melt behavior in the CORA and QUENCH bundles and the image analysis of the frozen melt show the formation of ceramic precipitates in the melt even in the molten state. The driving mechanism for the formation of precipitates in the melt is the temperature gradient at the oxide-melt interface. The high temperature Prater-Courtright correlation, used usually in computer codes to simulate the oxidation of Zr melt, was obtained based on the oxidation of very thin samples, which does not allow taking into account processes in the bulk melt. Simulation of the melt oxidation by the mechanistic SVECHA code, verified on the basis of many crucible tests, gives more correct result. This takes into account not only the formation of an external oxide layer around the molten pool, but also the formation of ceramic precipitation inside the melt. A numerical calculation carried out for the case with an operating temperature of 2473 K and a temperature gradient at the melt boundary of 50 K showed that the oxidation process occurs parabolically and three times faster than predicted by the Prater-Courtright correlation. The estimated activation energy for the melt oxidation correlation of Arrhenius-type proposed on the basis of SVECHA calculations is 85.9 kJ/mol, which is noticeably lower than the Prater-Courtright activation energy of about 110 kJ/mol
Sodium-ion battery cost projections and their impact on the global energy system transition until 2050
Get PDFSodium-ion batteries (SIB) have recently emerged as an alternative to current lithium-ion batteries (LIB), using low-cost and abundant raw materials. However, previous assessments have come to controversial results regarding their economic competitiveness, and the potential impacts of SIB on the wider energy system are still unexplored. This study combines a bottom-up cost modelling including future performance developments on material level for SIB with a global energy system model to obtain a comprehensive assessment of the potential impact of SIB on the global energy-industry transition until 2050. The results show that with recent cost developments and learning curves, batteries are no longer a cost-critical component in the energy system with projected utility-scale battery system capex of 28.5–51.9 €/kWh by 2050. SIB potentially outperform LIB on the medium term and are less prone to price spikes and supply shortages. Being a so-called drop-in technology, they could be produced on existing LIB production lines with only minor modifications. Therefore, concerns about supply shortages or price increases can be seen as resolved, since any disturbance in LIB supply would simply trigger a shift to SIB. The overall energy system structure remains virtually unaffected, with similar solar photovoltaic shares, but a shift in power-to-X processes operation. In this sense, electrochemical energy storage is not found to be a limiting factor for the global energy transition. Correspondingly, this work projects the possibly highest stationary battery demand published with a range of 67.9–106.5 TWh by 2050, above those in existing cost-optimised energy-industry system analyses
Electrochemical Hydrogenation of Aza-Arenes Using HO as H Source
Get PDFElectrochemical hydrogenation of aza-arenes is an appealing strategy to gain access to privileged saturated heterocycles for drug discovery, overcoming the limitations of classical hydrogenations that often suffer from energy-intensive conditions and safety hazards. Herein, we demonstrate an operationally simple, sustainable, and general electrochemical hydrogenation of aza-arenes with commercialized Ni foam electrodes and setup. With water as the hydrogen donor under acidic conditions, the reaction proceeds at ambient temperature and pressure to deliver broad substrate generality, excellent functional group tolerance, and excellent selectivity. The method tolerates a wide range of aza-arenes─including (iso)quinolines, quinoxalines, pyridines, and their nium salts─highlighting its generality and robustness. Synthetic utility was showcased through the preparation of bioactive molecules, while scalability was achieved up to 25 g of product, highlighting the method’s technical applicability with stable 22 h operation without changes in the cell voltage or significant electrode degradation. Extensive mechanistic investigations using a combination of cyclic and RDE linear sweep voltammetry suggest two plausible routes based on the substrate’s redox properties: hydrogenation by chemisorbed hydrogen (H) or initial substrate reduction followed by H transfer. This work sets a clean, practical, and versatile platform for aza-arene dearomatization, bridging academic interest with industrial targets in electrochemical hydrogenation
Characterization of the Regenerative Capacity of Membranes in the Presence of Fouling by Microalgae Using Detergents
Get PDFThe filtration of microalgae generates fouling through algal matter and exopolymer particles with consequences for the flow rate. Therefore, regeneration that is as complete and continuous as possible is necessary. For this purpose, a commercial membrane with a pore size of 0.8 μm is contaminated with the microalgae mixture Haematoccocus Pluvialis and
Tetradesmus obliquus, and then regenerated with mechanical (backwashing), chemical (HCl, NaOH, NaClO, P3-Ultrasil) and biological (dishwashing and laundry detergents) cleaning methods. The filtration time of the individual experiments is compared with a new membrane, and the increase is determined. Backwashing cleans the pores, but the biofilm sticks to the membrane surface and blocks the pores shortly after a new cycle. It was
shown that the biofilm can only be removed chemically through oxidative effects or anionic surfactants. Hydrolysis does not remove the biofilm, and it can actually make the blockage worse. Bigger cellular residues can only be removed with enzymes. This improves cleaning performance by 61% compared to commercial cleaning agents for membranes and 42% compared to backwashing
Data-driven artificial intelligence applications for tyre-road-noise prediction and road condition monitoring: A review and future directions
Get PDFNoise is an important environmental issue that affects quality of life and health, especially in urban areas. With the widespread adoption of electric vehicles, engine noise inside the car has decreased significantly, making tyreroad noise the main noise source, which also accounts for a large proportion of traffic noise. The powerful tool that is artificial intelligence (AI) has emerged in recent years for noise management and monitoring. AI-based systems can classify noise sources, create noise maps and develop control strategies. As a result, some studies
have focused on improving road, vehicle mechanics, and tyre textures and improving the sound quality of tyreroad noise. However, research specifically on tyre-road noise prediction is quite limited. Studies in the literature have generally focused on predicting road damage, surface quality and weather conditions, with less emphasis on tyre-road noise prediction. Many of these studies estimate tyre-road noise by modeling. However, it is not possible for modeling to capture real environment data. Therefore, more data-based studies on tyre-road noise
optimization, monitoring and prediction are needed in this area. This paper focuses on data-based studies and is a discussion of techniques such as data acquisition, feature extraction and selection, and artificial intelligence
algorithms that have been or could be used in this area. Data-driven artificial intelligence methods, such as deep learning, are highlighted for their significant potential in tyre-road noise monitoring and prediction. As a result, future research is expected to focus more on deep learning applications, opening new perspectives for further development in this field.
Effizientes Fuzzing von IoT-Geräten
Get PDFFuzzing ist eine bewährte Methode zur Identifizierung von Sicherheitslücken in Software. In dieser Dissertation wird untersucht, wie sich die Effizienz des Fuzzings für IoT-Geräte, am Beispiel des ESP32-Mikrocontrollers, steigern lässt. Während Fuzzing in klassischen Softwareumgebungen etabliert ist, fehlen speziell angepasste Verfahren für ressourcenarme IoT-Hardware. Das Ziel besteht in der Entwicklung eines konzeptionellen Frameworks, das eine umfassende und effiziente Testung trotz begrenzter IoT-Ressourcen ermöglicht. Zu diesem Zweck werden vier Ansätze zur Effizienzsteigerung sowie ein Konzept zur flexiblen Kombination dieser Ansätze vorgestellt.
Beim Binary Rewriting wird Binärcode so modifiziert, dass die Funktionalität erhalten bleibt. Für viele gängige Architekturen existieren bereits entsprechende Verfahren. Für die Xtensa-Architektur des ESP32 gab es jedoch bislang keine Lösung. In dieser Dissertation wird gezeigt, wie sich Binary Rewriting auf dem ESP32 umsetzen lässt, um Fuzzing-Instrumentierungen direkt in die Firmware zu integrieren und Laufzeitinformationen an den Fuzzer zurückzumelden.
Zudem wurde die Emulationsumgebung des ESP32 erheblich erweitert. Dadurch ist nun Fuzzing von beliebiger Firmware möglich, auch von Firmware mit zuvor nicht unterstützten Hardwarekomponenten. Im Vergleich zur realen Hardware arbeitet die Emulation deutlich effizienter. Während hardwarebasiertes Fuzzing vier bis 40 Anfragen pro Sekunde verarbeitet, sind es in der Emulation bis zu 320 Anfragen pro Sekunde.
Zur Generierung valider Eingaben wurde ein Verfahren zum automatisierten Protocol Reverse Engineering (PRE) entwickelt, das Künstliche Neuronale Netze (KNNs) verwendet. Während PRE bislang manuell erfolgen musste, können nun Protokollstrukturen automatisch abgeleitet und damit syntaktisch korrekte Netzwerkpakete erzeugt werden. In den Tests waren 67,6 % der erzeugten HTTP-Pakete und 100 % der FTP-Pakete gültig.
Für das grammatikbasierte Fuzzing wird ein Ansatz mittels Large Language Models (LLMs) vorgestellt. Die zentrale Herausforderung bestand in der effizienten Integration des LLM in den Fuzzing-Prozess. Mithilfe der entwickelten Methode lassen sich syntaktisch und semantisch korrekte XML-Dateien generieren. Dies steigert die Programmflussabdeckung um den Faktor sechs gegenüber einer Ausführung ohne LLM und erreicht eine um 50 % höhere Abdeckung als klassische grammatikbasierte Fuzzer.
Abschließend wird ein Integrationskonzept präsentiert, das eine flexible Kombination der Ansätze ermöglicht und deren Verbesserungen additiv nutzbar macht. Dadurch trägt das Framework zur Effizienzsteigerung des Fuzzings von IoT-Geräten bei. Die Dissertation leistet somit einen wichtigen Beitrag zur praxisnahen Absicherung von IoT-Geräten
Orbital Ordering in the Charge Density Wave Phases of BaNi 2 (As₁₋ₓPₓ)₂
Get PDFWe use resonant x-ray scattering at the nickel L2;3 edges to investigate the interplay between orbital
degrees of freedom and charge density waves (CDWs) in the superconductor BaNi2ðAs1-xPxÞ2. Both the
incommensurate and commensurate CDWs in this system exhibit strong resonant enhancement with
distinct energy and polarization dependencies, indicative of orbital ordering. Azimuthal-angle-dependent
measurements reveal a lowering of the local Ni site symmetry, consistent with monoclinic or lower point
group symmetry. The scattering signatures of both CDWs are dominated by contributions from Ni d xz;yz
orbitals, with similar orbital character despite their distinct wave vectors. These findings point to a shared
orbital-driven formation mechanism and provide new insight into the symmetry breaking and orbital and
nematic fluctuations in the high-temperature regime of the superconductor BaNi2ðAs1-xPxÞ2