13979 research outputs found
Sort by
Additive Manufacturing of Copper — A Survey on Current Needs and Challenges
Get PDFAdditive manufacturing (AM) of copper is subject to dynamic development regarding available processes and the quality of produced parts. While challenging, AM processes for copper provide parts with a quality comparable to other metallic material groups like steels. The reasons for the lower prevalence of additive manufacturing of copper components in industrial applications are currently not sufficiently researched, especially in light of the significant progress made in the maturity of this technology. A survey is used to investigate the assessments of protagonists in the field of copper AM. The needs of current and potential users of copper AM are analyzed and outlined. This study reveals that the most relevant technical limitation for users is the reduced surface quality of parts, while overall processes need to become less costly and more reliable to find broader use. Answers given hint to a higher degree of automation, the possibility of multi-material processing, and the upscaling of machine and part sizes as relevant future trends in the copper AM sector
Probabilistic Circuits: Going Bayesian and Spectral with Densities and Time Series
Get PDFWhen using a machine learning model for a specific task, users typically want to understand the reliability of the output from the model. Therefore, estimating the uncertainty of the output is a crucial task. A key approach to capturing uncertainty is probabilistic modeling. Traditional probabilistic models have evolved significantly over time: with the current increase in the amount of data, the complexity of data types, and diverse inference demands, new probabilistic models are constantly being proposed to address more complex scenarios. However, despite these advances, probabilistic modeling still faces challenges when dealing with data types such as time series or mixed tabular data. They often struggle to efficiently encode time dependencies or fail to provide a unified view for discrete and continuous random variables. In addition, they do not naturally integrate with deep neural networks limiting their application to more challenging tasks.
In this thesis, we investigate modeling challenging data types, including time series and mixed tabular data with probabilistic circuits, which allow for efficient and flexible probabilistic inference and can also be vectorized to work jointly with deep neural networks. First, we model the time series into the leaf nodes of a probabilistic circuit by utilizing Gaussian processes, and use product nodes to sequentially encode both the output dimensions and the covariate space, resulting in multi-output mixture of Gaussian processes (MOMoGPs). This results in the Bayesian case, enabling efficient computation for multi-input, multi-output regression tasks, and we then show its application in a real-world energy production use case. Secondly, to model the joint distribution of the entire time series, we leverage the Whittle assumption and model the time series in the spectral domain with its Fourier coefficients, resulting in Whittle sum-product networks (WSPNs), one of our spectral cases. This method not only preserves the time series dependencies but also enables efficient and flexible inference for, e.g., anomaly detection via density estimation and forecasting via conditional sampling. It is further extended to work jointly with other deep neural networks to provide useful uncertainty estimates in autoencoding and time series prediction. Lastly, we go one step further in the spectral domain by leveraging the structure of probabilistic circuits to model the characteristic function of probability distributions, resulting in characteristic circuits (CCs). By modeling densities in the spectral domain, characteristic circuits provide a unified view for discrete and continuous random variables, and can represent distributions that do not have closed-form probability density functions. We also show that characteristic circuits can be easily adapted and extended for causal inference in hybrid domains. We validate the proposed MOMoGPs, WSPNs, and CCs with both synthetic and real-world data sets. At the end of the thesis, we highlight interesting directions for future research on probabilistic models for challenging data types
Multiple Chaperone DnaK–FliC Flagellin Interactions are Required for Pseudomonas aeruginosa Flagellum Assembly and Indicate a New Function for DnaK
Get PDFThe DnaK (Hsp70) protein is an essential ATP‐dependent chaperone foldase and holdase found in most organisms. In this study, combining multiple experimental approaches we determined FliC as major interaction partner of DnaK in the opportunistic bacterial pathogen Pseudomonas aeruginosa. Implementing immunofluorescence microscopy and electron microscopy techniques DnaK was found extracellularly associated to the assembled filament in a regular pattern. dnaK repression led to intracellular FliC accumulation and motility impairment, highlighting DnaK essentiality for FliC export and flagellum assembly. SPOT–membrane peptide arrays coupled with artificial intelligence analyses suggested a highly dynamic DnaK–FliC interaction landscape involving multiple domains and transient complexes formation. Remarkably, in vitro fast relaxation imaging (FReI) experiments mimicking ATP‐deprived extracellular environment conditions exhibited DnaK ATP‐independent holdase activity, regardless of its co‐chaperone DnaJ and its nucleotide exchange factor GrpE. We present a model for the DnaK‐FliC interactions involving dynamic states throughout the flagellum assembly stages. These results expand the classical view of DnaK chaperone functioning and introduce a new participant in the Pseudomonas flagellar system, an important trait for bacterial colonisation and virulence
Two Optimization Approaches for a Small‐Scale Power‐to‐Ammonia Cycle
Get PDFAmmonia is a promising carbon‐free energy vector. Small‐scale renewable power‐to‐ammonia (P2A) is particularly suited for isolated agricultural areas where ammonia can be used as fuel and fertilizer. This work compares two approaches to simulate and optimize the steady‐state behavior of a novel small‐scale P2A process: Aspen Plus® and MOSAIC®. Aspen Plus® is a commercial flow sheeting software whereas MOSAIC® is a freeware where equations and thermochemical properties need to be specified by hand. It can be shown that the results of MOSAIC® and Aspen Plus® are qualitatively comparable, but not identical. This suggests that the model in MOSAIC® can be improved further, starting with the implementation of a more accurate numeric reactor kinetics and equation of state
Alumina Supported Iron Catalysts for Selective Acetylene Hydrogenation Under Industrial Front‐End Conditions
Get PDFThe removal of acetylene traces from ethylene streams coming from the steamcracker is carried out in the industry on an annual scale of several million tons using Pd‐Ag/Al₂O₃ catalysts. The substitution of palladium containing catalysts with more abundant, cheap, and nontoxic materials is a first crucial step toward a more sustainable chemical industry. As iron is one of the most abundant metals and can be mined in almost all regions worldwide, it is an ideal catalyst material. In this work, we present the development of α‐alumina supported iron catalysts with 1, 5, and 10 wt% iron loading and their application in the selective acetylene hydrogenation under industrially applied front‐end conditions. The catalysts were prepared via simple incipient wetness‐impregnation and were analyzed via XRD, XRF, TPR, TEM, and N₂‐physisorption. The catalysts were subsequently calcined, reduced, and tested in the selective acetylene hydrogenation. After an activation phase, the catalysts show excellent activity and selectivity in the acetylene hydrogenation at 90 °C without significant ethylene hydrogenation. The excellent catalytic activity underline the great potential of iron based catalysts as an alternative to conventional Pd‐containing materials
Parameterising Local Reactive Power Control Characteristics of Distributed Energy Resources Through Time Series Based Optimal Power Flow Calculations
Get PDFThis paper presents a method for determining local reactive power control characteristics for distributed energy resources. Based on historical time series, an optimal reactive power dispatch is calculated. The optimal reactive power dispatch minimises a multi‐criteria objective function that allows the reduction of grid losses while complying with a desired vertical reactive power exchange with the overlaid grid and with operational constraints. From the resulting optimal operating points, individual reactive power control characteristics can be determined by piecewise linear regression and then be clustered to reduce the complexity of the grid planning process. The method thus allows the integration of the information contained in the historical time series on the volatile power flow behaviour and offers the possibility of systematic parameter determination. In contrast, conventional grid planning uses empirical values or conservative estimates. In contrast to offline or online optimisations during operation; however, no real‐time capable information and communication infrastructure is required. To validate the performance of the method, operation with the cluster control characteristics is compared with offline optimisation and simple parameterisation approaches for local reactive power control characteristics
High Pressure Synthesis of Pr₂O₅ – A Unique Lanthanoid(IV) Oxide Peroxide
Get PDFReacting praseodymium(IV) oxide with oxygen at 27 GPa in a diamond anvil cell yielded the oxide peroxide Pr₂IV(O₂)O₃, which was characterized by single crystal X‐ray diffraction on multi‐grain samples, Raman spectroscopy and quantum theoretical calculations at various pressure points. The presence of tetravalent praseodymium ions is supported by electronic structure calculations, showing a band gap of ca. 1.2 eV, which is consistent with the anticipated chemical model of an ionic solid. Pr₂(O₂)O₃ thus far represents the most oxygen rich phase of any binary compound of a lanthanoid and oxygen and is the first example of a peroxide anion next to Pr⁴⁺. Additionally, these results demonstrate that instead of oxidizing the praseodymium ions past their +IV oxidation state, oxygen undergoes a comproportionation to form peroxide anions. Direct oxidation of the oxide anions by Pr⁴⁺‐ions was ruled out by a control experiment in argon instead of oxygen, where no oxidation of oxide ions was observed
Erfahrungsbasierte visuelle Lokalisierung für intelligente Fahrzeuge
Get PDFIntelligente Fahrzeuge wie ein autonom fahrendes Straßenfahrzeug benötigen für die Ausführung ihrer Bewegungsaufgabe eine Lokalisierung, d.h. sie müssen ihre eigene Position und Lage in der Umgebung ermitteln können. Eine erfahrungsbasierte visuelle Lokalisierung, die Erfahrungswissen aus vorherigen Befahrungen des Gebietes nutzt, ist ein vielversprechender Ansatz, um für diese Fahrzeuge eine Lokalisierung mit der erforderlichen Genauigkeit und Verfügbarkeit bereitzustellen. Der erfahrungsbasierte Ansatz erfordert die Verwendung von visuellen Langzeit-SLAM-Verfahren, die tausende Kameraposen und Landmarken verarbeiten, welche in Zeiträumen von Tagen bis Jahren erfasst wurden. In dieser Arbeit wird als konkrete Umsetzung eines solchen Verfahrens das LLama-SLAM-Verfahren entwickelt, das aus einer Reihe von Fahrtaufzeichnungen mit Stereokamera- und GNSS-Daten eine Karte mit generischen Langzeit-Landmarken aufbaut.
Das LLama-SLAM-Verfahren enthält drei wesentliche Bausteine. Der erste Baustein ist eine Methode zum Vergleich visueller Merkmalsfolgen, um bereits bekannte Landmarken wiederzuerkennen. Für denVergleich von Merkmalsfolgen wurden sieben Vergleichsmethoden entwickelt, die auf binären Deskriptorverfahren basieren. Eine Bewertung dieser Methoden hinsichtlich Güte und Recheneffizienz zeigt, dass die CoMa-Vergleichsmethode, die einen kombinierten Merkmalsfolgendeskriptor verwendet und unzuverlässige Deskriptorbits maskiert, am besten für den Einsatz in einem Langzeit-SLAM-Verfahren geeignet ist. Der zweite Baustein ist ein optimierungs- bzw. graphbasiertes Verfahren zum Lösen großer SLAM-Probleme. Es wird erläutert, wie ein solches optimierungsbasiertes SLAM-Problem formuliert und mit dem Levenberg-Marquardt-Verfahren gelöst wird. Außerdem werden konkrete Graph-SLAM-Elemente beschrieben, die verwendet werden, um Landmarkenbeobachtungen, GNSS-Positionsmessungen und 3D-Odometriemessungen einzubinden. Der dritte Baustein ist ein Verfahren zur Bewertung generischer Langzeit-Landmarken (LLamas), das steuert, welche Landmarken für die Lokalisierung verwendet werden. Hierfür wird ein probabilistisches Qualitätsmodell verwendet, das auch die Beobachtungsposition berücksichtigt und effizient iterativ berechnet werden kann. Durch die Qualitätsbewertung kann das LLama-SLAM-Verfahren beliebige Umgebungsstrukturen als Landmarken erlernen und ist nicht auf die Verwendung spezifischer Objektklassen beschränkt.
Die Lokalisierungsfähigkeit des LLama-SLAM-Verfahrens und der Einfluss verschiedener Parameter wurden experimentell auf einem Langzeitdatensatz mit 24 Fahrtaufzeichnungen untersucht. Für die Versuchsdaten konnte gezeigt werden, dass durch die qualitätsgesteuerte Auswahl von Landmarken mit wenigen Landmarken hoher Qualität ähnliche oder bessere Lokalisierungsergebnisse erzielt werden können als mit vielen Landmarken geringerer Qualität. Bereits 10 bis 20 Landmarken pro Kamerapose führten bei den betrachteten Metriken zu geringen Lokalisierungsfehlern und die Hinzunahme weiterer Landmarken führte nur in einzelnen Fällen zu wesentlichen Verbesserungen. Die angestrebte Lokalisierungsgenauigkeit, die für autonomes Fahren im urbanen Bereich benötigt wird, konnte auf den Versuchsdaten nicht erreicht werden, was zumindest teilweise auf größere Fehler in den verwendeten GPS-Messungen zurückzuführen ist. Es konnte jedoch eine Verbesserung gegenüber einer reinen GPS-Lokalisierung festgestellt werden, insbesondere für den Gierwinkel des Fahrzeugs. In den Experimenten wurde außerdem gezeigt, dass bei Verwendung des LLama-SLAM-Verfahrens zunehmendes Erfahrungswissen in Form von mehr und mehr Fahrtaufzeichnungen zu einer deutlichen Verbesserung der absoluten Lokalisierungsgenauigkeit führt
Wavelet Enhanced LSTM for Accurate Streamflow Modeling Addressing Nonlinearity and Nonstationarity in Hydrological Data
Get PDFFlood prediction is challenging due to increasing climate variability and nonlinear, nonstationary hydrological data. We propose a novel hybrid model, Wavelet Long Short-Term Memory (WLSTM), combining wavelet transforms with LSTM networks. Wavelet decomposition allows separation of frequency components, enhancing LSTM’s ability to learn both short- and long-term dependencies. Applied to streamflow data from 16 stations in Hessen (2000–2017), WLSTM reduced RMSE by 66.43% and MAPE by 45.49%, while increasing R² by 2.06%. These improvements highlight its effectiveness in modeling complex hydrological dynamics. WLSTM offers a robust tool for adaptive flood risk management under climate change
Development of Sn/SnO₂ hard carbon composites via solvothermal synthesis as anode material for sodium-ion and lithium-ion batteries
Get PDFThis research focuses on the development of carbon sphere composites derived from glucose and fructose, encapsulating metallic tin (Sn) via conventional and microwave-assisted solvothermal/hydrothermal synthesis for sodium-ion battery (SIB) anodes. The rationale behind combining Sn with hard carbon lies in Sn's high theoretical capacity through Na alloying, although its major drawback—up to 420% volume expansion—necessitates encapsulation strategies to maintain structural stability. A range of synthesis parameters, including temperature, precursor types, and solvents, were explored. Materials were subjected to pyrolysis at 600 °C, 850 °C, and 1100 °C. Notably, SnCl₂-derived composites showed superior Sn distribution and electrochemical stability compared to those from SnO₂. The best-performing sample, C2_SnCl2_600_1100, achieved a stable sodiation capacity of 418 mAh g⁻¹ over 100 cycles with 59.1% retention. Traditional solvothermal routes revealed issues like incomplete carbonization and Sn agglomeration, while microwave-assisted synthesis offered faster processing and improved stability. Raman, XRD, and SEM analyses confirmed phase evolution, graphitization, and morphology. Although Sn-containing composites showed poor cycling in lithium-ion batteries, they performed well in SIBs. Overall, this study presents a promising route to engineer Sn–carbon composites for sodium storage and highlights the potential of microwave-assisted synthesis to lower energy use while enhancing performance. Further optimization is necessary to advance these materials for next-generation energy storage