University of Kaiserslautern
KLUEDO Publication Server of University of Kaiserslautern-Landau (RPTU)Not a member yet
6017 research outputs found
Sort by
Physics-Constrained Deep Learning for Accelerating Climate Modeling
Get PDFAccurate modeling of weather and climate is critical for taking effective action to combat climate change. Predicted and observed quantities such as precipitation, clouds, aerosols, wind speed, and temperature impact decisions
in sectors such as agriculture, energy, health, and transportation. While
these quantities are often required at a fine geographical and temporal scale
to ensure informed decision-making, most climate and weather models are
extremely computationally expensive to run, resulting in coarse-resolution
predictions. Recent advances in deep learning (DL) make it an attractive
tool for speeding up simulations. The two main ways to decrease computational efforts with DL are downscaling, the increase of the resolution directly
on the predicted climate variables, and emulation, the replacement of model
parts to achieve faster runs initially.
This thesis leverages DL models for accelerated climate forecasting while
making sure the methods are feasible for physical modeling. Standard DL
approaches often violate simple physical constraints such as positivity or conservation properties. We develop novel methodologies to incorporate those
constraints into the training and into architectures of DL. First, we look
into so-called soft-constraining methods that introduce an additional regularisation term. Then several hard-constraining methods that change the
neural networks (NNs) architectures, by adding final constraining layers, are
discussed.
We consider two application cases to test our constraining methodology
and evaluate the potential of DL for speeding up climate modeling. The
first test case is downscaling. We not only show how our hard-constraining
layers guarantee the constraints to be satisfied, but also increase the overall
predictive performance. In the second employment of our constrained DL
models, the aerosol microphysics module in the global ICON climate model
is replaced by a NN. We both investigate offline performance, as well as
implement the NN in Fortran to run it online within ICON, achieving a
stable an accurate coupled simulation. We discuss challenges and choices for
a successful deployment of DL in climate and weather simulations
Beiträge zur Gestaltung und Anwendung von integrierten Multi-Sensor- und Aktuator-Elektroniksystemen mit Selbst-X-Eigenschaften für robuste integrierte intelligente Systeme
Get PDFThe swift progress in smart sensor technologies, Internet of Things, Industrial Internet of Things, and Cyber-Physical Systems has led to evolving the sensor standards to enable Industry 4.0, the industrial domain where adaptability, efficiency, and reliability are essential. The sensor applications of the new industry era necessitate increasingly adaptable sensor electronics and signal processing capabilities based on machine learning (ML) and artificial intelligence (AI) with other cutting-edge technologies.
This thesis presents a literature review of the design and applications of evolvable hardware and reconfigurable/programmable electronics that can be tailored for smart sensory electronics (SSEs) in Industry 4.0. Through an interdisciplinary approach that weaves together elements from bio-inspired systems, evolvable hardware, and advanced signal processing techniques, this work introduces a suite of design methodologies and implementations for analog front-end (AFEX) systems endowed with self-X capabilities, namely self-optimization, self-configuration, and self-calibration.
Central work to the AFEX is the circuit improvement and implementation of the fully-differential current-feedback instrumentation amplifier (CFIA) that demonstrates high performance in terms of input dynamic range, power efficiency, and adaptability and also integrates advanced features like input-offset voltage autozeroing.
A major limitation of hardware in-field optimization is the chip area due to the configurable elements and the assessment unit implementation; both together increase the cost and almost present the optimization approach as possible but not a practical or attractive industrial solution. In this work, the application of indirect measurement for devices under optimization is implemented using simple non-intrusive sensors (NISs) and THD-based power-efficient indirect measurement techniques. Several design metrics are extracted simultaneously in fewer tests that don’t require the addition of new hardware, except for the utilization of the existing sensor’s data acquisition resources. To reduce the chip cost, it is proposed to configure the sensitive elements only in the circuit.
In addition to the CFIA, the thesis proposes an innovative design of a fourth-order fully-differential anti-aliasing and anti-imaging filter, a crucial device for maintaining signal integrity for various signal processing properties ranging from low to high-frequency sensor applications. The key features of the proposed filter are the wide tunable bandwidth range, fine-step frequency resolution per decade, and a high dynamic signal range approached by the application of a programmable and linearized MOS resistor. Furthermore, to account for the complexity of the bandwidth tuning, an indirect measurement approach based on SSIs is proposed with the help of AI and neural networks.
The practical realization of these designs is fabricated on a chip using the CMOS 0.35 µm technology from XFAB. The conducted LAB experiments under various operating conditions demonstrate not only the feasibility of the proposed solutions but also their potential to enhance the performance and energy efficiency, maximize yield, and improve the reliability of SSE in harsh industrial environments. Furthermore, by enabling sensors to autonomously adapt under varying conditions, it reduced the need for manual recalibration, thereby supporting the autonomous operation of industrial systems.
An experimental demonstration using a Tunnel Magnetoresistance (TMR) sensor in the last chapter, showcases the practical application and benefits of the proposed in-field optimization. This demonstration not only serves as a proof of concept but also illustrates the potential of the proposed design approach in real-world industrial scenarios.Der rasche Fortschritt in den Bereichen intelligente Sensortechnologien, Internet der Dinge, industrielles Internet der Dinge und cyber-physische Systeme hat dazu geführt, dass die Sensorstandards weiterentwickelt wurden, um Industrie 4.0 zu ermöglichen, den industriellen Bereich, in dem Anpassungsfähigkeit, Effizienz und Zuverlässigkeit von entscheidender Bedeutung sind. Die Sensoranwendungen der neuen Industrieära erfordern zunehmend anpassungsfähige Sensorelektronik und Signalverarbeitungsfähigkeiten, die auf maschinellem Lernen (ML) und künstlicher Intelligenz (KI) sowie anderen Spitzentechnologien basieren.
In dieser Arbeit wird eine Literaturübersicht über das Design und die Anwendungen von evolvierbarer Hardware und rekonfigurierbarer/programmierbarer Elektronik vorgestellt, die für intelligente sensorische Elektronik (SSEs) in der Industrie 4.0 maßgeschneidert werden können. Durch einen interdisziplinären Ansatz, der Elemente aus bioinspirierten Systemen, evolvierbarer Hardware und fortschrittlichen Signalverarbeitungstechniken miteinander verwebt, stellt diese Arbeit eine Reihe von Designmethoden und Implementierungen für analoge Front-End-Systeme (AFEX) vor, die mit Self-X-Fähigkeiten ausgestattet sind, nämlich Selbstoptimierung, Selbstkonfiguration und Selbstkalibrierung.
Das Kernstück des AFEX ist die Schaltungsverbesserung und Implementierung des volldifferenziellen stromrückgekoppelten Instrumentenverstärkers (CFIA), der eine hohe Leistung in Bezug auf Eingangsdynamik, Leistungseffizienz und Anpassungsfähigkeit aufweist und auch fortschrittliche Funktionen wie die automatische Nullstellung der Eingangsoffsetspannung integriert.
Eine wesentliche Einschränkung der Hardware-Infield-Optimierung ist die Chipfläche aufgrund der konfigurierbaren Elemente und der Implementierung der Bewertungseinheit; beides zusammen erhöht die Kosten und macht den Optimierungsansatz zwar möglich, aber nicht zu einer praktischen oder attraktiven industriellen Lösung. In dieser Arbeit haben wir die Anwendung der indirekten Messung für zu optimierende Bauelemente unter Verwendung einfacher nicht-intrusiver Sensoren (NISs) und THD-basierter leistungseffizienter indirekter Messverfahren implementiert. Mehrere Designmetriken werden gleichzeitig in weniger Tests extrahiert, die keine neue Hardware erfordern, mit Ausnahme der Nutzung der Datenerfassungsressourcen des vorhandenen Sensors. Um die Chipkosten zu senken, haben wir vorgeschlagen, die empfindlichen Elemente nur in der Schaltung zu konfigurieren.
Zusätzlich zum CFIA wird in dieser Arbeit ein innovatives Design eines volldifferentiellen Anti-Aliasing- und Anti-Imaging-Filters vierter Ordnung vorgeschlagen, das für die Aufrechterhaltung der Signalintegrität bei verschiedenen Signalverarbeitungseigenschaften von Nieder- bis Hochfrequenzsensoranwendungen von entscheidender Bedeutung ist. Die Hauptmerkmale des vorgeschlagenen Filters sind der große abstimmbare Bandbreitenbereich, die feinschrittige Frequenzauflösung pro Dekade und ein hoher dynamischer Signalbereich, der durch die Anwendung eines programmierbaren und linearisierten MOS-Widerstands erreicht wird. Um der Komplexität der Bandbreitenabstimmung Rechnung zu tragen, wird außerdem ein indirekter Messansatz auf der Grundlage von SSI mit Hilfe von KI und neuronalen Netzen vorgeschlagen.
Die praktische Umsetzung dieser Entwürfe wird auf einem Chip mit der CMOS 0,35 µm Technologie von XFAB hergestellt. Die durchgeführten LAB-Experimente unter verschiedenen Betriebsbedingungen zeigen nicht nur die Machbarkeit der vorgeschlagenen Lösungen, sondern auch ihr Potenzial, die Leistung und Energieeffizienz zu steigern, den Ertrag zu maximieren und die Zuverlässigkeit von SSE in rauen Industrieumgebungen zu verbessern. Durch die Möglichkeit der autonomen Anpassung der Sensoren an sich ändernde Bedingungen wird außerdem die Notwendigkeit einer manuellen Neukalibrierung verringert, wodurch der autonome Betrieb von Industriesystemen unterstützt wird.
Eine experimentelle Demonstration unter Verwendung eines Tunnelmagnetowiderstandssensors (TMR) im letzten Kapitel zeigt die praktische Anwendung und die Vorteile der vorgeschlagenen Feld-Optimierung. Diese Demonstration dient nicht nur als Konzeptnachweis, sondern veranschaulicht auch das Potenzial des vorgeschlagenen Designansatzes in realen industriellen Szenarien
Optimization and Generative Models for Face Analysis
Get PDFHuman analysis, with an emphasis on the head and face, has been an important subject of study in a wide range of scientific fields. The facial expressions, eye gaze, gestures and head pose provide non-verbal cues about the physical and mental state of individuals, their emotions, consciousness and attentiveness. In computer vision, these cues have been leveraged to study face images, where fields such as emotion recognition, face identification and gaze estimation have emerged as cornerstones in human modeling and understanding.
Research on face analysis has enabled the development of assistance tools, for instance, to identify the fatigue and inattention of a vehicle driver, detect the level of asymmetry in patients with facial paralysis or recognize the emotions of children with autism. Such tools demand a high level of robustness in a wide range of situations, such as varying illumination, large head rotations and partial occlusion, in addition to being able to operate with low computational costs and with real-time capabilities.
This thesis focuses on the analysis of the rigid and non-rigid motions of the head, with the aim of targeting existing challenges in assistance tools and assistive technologies. The proposed approaches are based on intensity and RGB images, and cover the estimation of the head pose and the detection and tracking of facial landmarks. Additionally, facial expression transfer for avatar animation and the generation of sign language images are explored, where the face and upper body play a crucial part in conveying information.
The main contributions on this thesis are grouped in four categories: (i) the study of head pose estimation from a monocular system, to obtain a robust estimate of the rigid head pose in real time. To that end, multiple architectures are proposed, where the head pose is formulated as an optimization problem based on the 3D-2D correspondences of facial features; (ii) the detection and tracking of fiducial facial landmarks, to model the non-rigid facial motion. The sparse set of facial landmarks are computed using a deep-learning-based generative pipeline, extending the detection to a wide range of facial expressions, including faces with substantial asymmetry. Tracking of 3D landmarks is also investigated, where a method to exploit the landmark connectivity and confidence score is introduced; (iii) the development of a real-time framework for performance-driven facial animation using a monocular system. This method aims to transfer the rigid head pose and facial expression from 2D video sequences to a 3D head model with limited resources; and (iv) a rendering pipeline to extend current RGB-based sign language datasets, to include multiple types of annotated data such as segmentation masks, normal and depth maps, and 3D-2D body joints. A sign language image synthesis architecture based on generative models is also investigated, conditioned on pose and appearance. Our architecture is trained and evaluated on synthetic and real data.
The proposed methods achieve outstanding performance compared to related work on publicly available benchmarks, as well as in datasets introduced in this thesis. Furthermore, the presented approaches for head pose estimation and performance-driven facial animation perform in real time on CPU, while the proposed pipelines for face alignment and sign language image synthesis achieve real-time capabilities on GPU
Perennial ryegrass as feedstock for the bio-based production of chemicals
Get PDFThe economic and environmental challenges are driving the search for alternative and sustainable energy resources. In this scenario, lignocellulosic biomasses are gaining more attraction due to their high abundance, renewability, and low cost, and can therefore contribute to the development of a bio-based economy. Grass materials from dedicated crops or crop residues can be utilized as feedstock in green biorefinery platforms. In these platforms, the biomass is firstly fractionated into a press-cake and a press-juice. Then, further technologies are applied including the extraction of valuable compounds from the press-juice, and the utilization of the press-cake in feed, material, pulp and paper, and fuel applications. This thesis outlines the development of an innovative green biorefinery platform using the perennial ryegrass Lolium perenne for the production of chemicals. To accomplish this, all the conversion steps— including biomass pretreatment, enzymatic hydrolysis, and fermentation— were carried out to promote the utilization of both the press-cake and the press-juice. Mechanical pretreatment was identified as the most efficient method to get sugars from enzymatic hydrolysis, and to separate the press-cake and the press-juice. Different bioprocesses were investigated to convert the press-cake and the press-juice into the final chemicals. The press-cake was utilized as substrate in a Consolidated Bioprocess where Aspergillus niger was used as a biocatalyst for citric acid production. The nutrient-rich press-juice was exploited in ethanol and lactic acid fermentation. For ethanol production, short-term adaptation was applied to enhance the fermentation capacities of Saccharomyces cerevisiae in pure press-juice. Lactic acid production was ensured by optimizing the press-juice as a fermentation medium at bench- and reactor-scale. Overall, the main findings of this thesis demonstrate that, despite their limitations, lignocellulosic biomasses— including the perennial ryegrass— are valuable feedstocks for the bio-based production of chemicals, fuels, and materials
Glycinergic transmission at auditory brainstem synapses
Get PDFInhibitory transmission in the auditory brainstem is essential for sound source localization. Glycinergic synapses between the medial nucleus of the trapezoid body (MNTB) and the lateral superior olive (LSO) process interaural level differences and reliably transmit inhibitory signals, even during minutes of high-frequency activity. Recycling of synaptically released glycine is crucial for reliable inhibitory transmission at MNTB LSO synapses. Glycine transporter 2 (GlyT2) mediates glycine (re)uptake into MNTB axon terminals. Glycine is (re)filled into the synaptic vesicles (SV) with the vesicular inhibitory amino acid transporter (VIAAT), driven by a proton gradient generated by vacuolar (H+)-ATPase (V-ATPase). Loss of GlyT2 or V ATPase inhibition impairs transmission but does not cause complete failure, suggesting the presence of additional mechanisms supporting glycine (re)uptake and SV (re)filling. Potential candidates are the alanine serine cysteine 1 transporter (Asc 1) glycine (re)uptake and the Na+/H+ exchanger (NHE) for SV (re)filling. However, their contribution to inhibitory transmission at MNTB-LSO synapses under high-frequency activity remains largely unclear.
In the first part of my thesis, I investigated the contribution of Asc 1 in wildtype (WT) and GlyT2 knock-out (KO) mice. Immunohistochemistry in the auditory brainstem confirmed Asc-1 expression in both genotypes. Whole-cell patch clamp recordings from LSO neurons were performed while stimulating MNTB axons (Protocol1 and 2). In KO mice, synaptic strength and vesicle replenishment were severely impaired. In WT mice, synaptic transmission remained stable and was only affected after sustained activity. In the second part, I examined synaptic strength during very prolonged activity (Protocol 3) under pharmacological inhibition of GlyT2 and/or Asc-1 in WT. While synaptic transmission and SV replenishment remained reliable under GlyT2 or Asc 1 inhibition, combined blockade led to impaired transmission and reduced SV (re)filling and replenishment. In the third part of my thesis, I investigated SV (re)filling by performing and reanalyzing data under V-ATPase and/or NHE inhibition. Transmission, SV (re)filling and replenishment were reduced. In addition, NHE inhibition affected action potential conduction fidelity, suggesting a non-specific inhibition.
Collectively, my results imply that Asc-1 inhibition reduces glycinergic transmission at MNTB LSO synapses. In KO, the remaining transmission is maintained by Asc-1-mediated glycine (re)uptake. This demonstrates that Asc 1, in addition to GlyT2, serves as a key glycine source at MTNB LSO synapses. Inhibition of SV (re)filling reduces glycinergic transmission, highlighting the essential role of V-ATPase in maintaining reliable transmission at MNTB-LSO synapses. The used NHE inhibitor EIPA is not selective for NHE6 and interferes with AP conduction. Due to these unspecific effects, the functional contribution to synaptic performance could not be directly assessed
Landwirtschaft und Wasserknappheit in Marokko
Get PDFDiese Arbeit untersucht die Auswirkungen der exportorientierten Landwirtschaft Marokkos auf die Wasserressourcen am Beispiel der Souss-Massa-Region. Mittels qualitativer Inhaltsanalyse nach Mayring und unter Anwendung des Integrierenden Nachhaltigkeitsdreiecks werden ökologische, ökonomische und soziale Folgen analysiert. Die Ergebnisse zeigen, dass insbesondere wasserintensive Kulturen wie Tomaten und Zitrusfrüchte die Wasserknappheit verschärfen, während Kleinbauern marginalisiert und Arbeitskräfte stark belastet werden. Der wirtschaftliche Nutzen bleibt ungleich verteilt und basiert auf nicht nachhaltiger Ressourcennutzung. Aufbauend auf theoretischen Konzepten wie dem Menschenrecht auf Wasser und den Zielen für nachhaltige Entwicklung (SDGs) werden Handlungsempfehlungen für ein gerechteres Wassermanagement in Marokko und eine verantwortungsvollere Handelspolitik der EU formuliert
Towards Image and Video Understanding in Challenging Scenarios
Get PDFLearning-based solutions have revolutionized the field of Artificial Intelligence (AI), pushing it to new frontiers in a variety of domains.
AI advancements owe much to highly curated datasets that enable the training and testing of complex deep learning models, leading to excellent accuracy in controlled academic environments.
However, it is essential to acknowledge that the efficacy of these models in these lab settings does not fully encapsulate the complexities and challenges encountered in real-life applications.
To ensure the practical applicability of learning-based solutions, it is crucial to understand their limitations and performance under diverse and challenging measurements.
This requires bridging the gap between academic benchmarks and the complexities of the real world.
In this thesis, we take a step toward better understanding the limitations of current deep learning models in handling corner cases and challenging scenarios, with a focus on computer vision tasks across image and video domains.
Through this exploration, we identify areas and ways to improve the robustness of computer vision solutions.
In the image domain, we study image classification and analyze the model behavior when processing images containing background noise and clutter.
We extend this study to investigate salient image classification, a scenario in which multiple objects are present in a scene, and the model is expected to classify the most prominent or salient one.
In the video domain, we explore the fundamental task of spatiotemporal feature correspondence learning. This task has diverse applications, such as video object segmentation and tracking. Our investigation delves into challenging scenarios, including tracking smaller objects, managing occlusion, coping with crowded scenes, and efficiently processing longer videos.
Furthermore, we investigate self-supervised learning methods for spatiotemporal correspondence learning, motivated by the high cost of annotating video data for this task and the challenges that arise when training on small datasets.
Finally, we study the problem of out-of-domain generalization on video data, a critical issue that affects the applicability of learning-based solutions.
To this end, we evaluate several ways for using self-supervised learning to mitigate the adverse effects of domain shift, enabling the model to perform well in new, unseen domains.
We hope this work fosters advancements in the field of AI by providing insights and directions for designing more robust models that deliver enhanced performance in diverse and complex scenarios
Einblicke in die Entwicklung eines Instruments zur Erfassung von metakognitiven konditionalem Schreibstrategiewissen: Ein Bericht zur Konstruktion eines Vignettentests
Get PDFWer erfolgreich schreiben will, braucht dafür Wissen. In diesem Beitrag widmen wir uns daher einer Facette des Wissens, die aktuell noch unterbelichtet wirkt: dem konditionalen Wissen über die Anwendung von geeigneten Schreibstrategien bei hochschulischen Aufgaben. Auf der Basis von Antworten von N = 105 Studierenden und N = 5 Expert:innen haben wir einen Vignettentest mit geschlossenen Aufgaben entwickelt, den wir vorstellen. Wir legen ausführlich den gesam-ten Test dar und wie wir zur finalen Auswahl von Paarvergleichen gelangt sind, die das Planen, Formulieren und Revidieren als Hauptprozesse des Schreibens in den Blick nehmen. Mit diesem Vorgehen möchten wir die Testentwicklung exemplarisch konturieren und Schreibforscher:innen ermutigen, solche Tests selbst zu entwickeln, um sie in Grundlagen- und angewandter Forschung einzusetzen
Query Processing and Cardinality Estimation over Data Summaries
Get PDFIn this thesis, we propose methods and optimizations for efficient query processing and cardinality estimation over data summaries. For this reason, we put forth algorithms that work on a substantial amount of static data and methods tailored for streaming environments, where the data is constantly being generated and arrives over time. When working in streaming environments, we first address the problem of natural join computation over streams of semi-structured documents. We initially propose an efficient and scalable partitioning algorithm that uses the main principles of association analysis to identify co-occurring attribute-value pairs within the documents. To compute the join pairs, we put forth a novel natural join algorithm that comes in two flavors. The join algorithm utilizes a frequent pattern tree (FP-tree) to store a compact representation of the schema-free documents. Differently, we address the problem of data stream summarization on edge devices to reduce network traffic and speed up further processing. Instead of emitting the whole data stream, our SoftSieving approach computes item-based summaries using core items. Core items of a data stream are the items with the highest values for a given monotone submodular utility function. When working with vast static data, we address approximate query processing and cardinality estimation. We propose a versatile approach to lightweight, approximate query processing by learning compact but tunably precise representations of larger quantities of tuples coined bubbles. Bubbles are tunable regarding the compactness of enclosed tuples as well as the granularity of statistics and the way they are instantiated. Our solution models the content of the bubbles using Bayesian networks and deep autoregressive models. We have also identified an optimization for deep autoregressive models for the task of cardinality estimation. We overcome the limitations of such models when handling queries involving range predicates by circumventing the typically applied, time-consuming estimation algorithm. Consequently, we present our hybrid approach, Grid-AR, that fuses deep autoregressive models with a grid-based structure
Process-related Morphology and Mechanical Properties of Semi-crystalline Polymers in Fused Filament Fabrication (FFF)
Get PDFFused filament fabrication (FFF) is an additive manufacturing technique with signif-icant potential for producing customized and functional polymer-based components. Using polypropylene (PP) as a representative system, the morphology, thermal be-havior, and mechanical performance of pure and microfibrillar-filled semi-crystalline polymers fabricated via FFF were explored, emphasizing the influence of tempera-ture control, material properties, and processing parameters on final part perfor-mance. Results demonstrate that contact temperature is the key parameter control-ling interfacial diffusion and bonding strength, a trend consistently validated across various semi-crystalline polymer systems. A contact temperature model was devel-oped and validated, enabling process condition optimization and achieving interfacial properties comparable to those of injection-molded parts. To exploit the anisotropic properties of FFF using microfibrillar composites (MFC), an unexpected phenome-non of in situ fibrillation was observed, wherein PET droplets were fully transformed into aligned microfibers during printing under suitable thermal and flow conditions. These fibers enhance tensile properties and promote crystallization. These findings establish a comprehensive framework and practical strategy for optimizing the FFF process of semi-crystalline polymers and MFC, enabling simplified processing, en-hanced performance, and geometry-aware parameter control for advanced light-weight structural applications