Offenburg University of Applied Sciences
Hochschulschriftenserver der Hochschule OffenburgNot a member yet
6641 research outputs found
Sort by
Towards deep learning for seismic data processing
No full textDeep Learning (DL), a branch of Artificial intelligence (AI), is presented as
it can provide a data driven approach to the conventional seismic processing
workflow, in an attempt to solve some of the conventional issues. Neural
networks are universal approximators that learn the mapping between one
space, the inputs, into the outputs. The learning process consists of an inter-
active optimization of parameters by minimizing a loss function quantifying
the distance inputs and outputs over an entire training data. Convolutional
neural networks (CNNs), a particular architecture of neural networks, were
developed for computer vision tasks and are particularly efficient in image
processing task such as classification, inpainting, and super-resolution. The
success in this field and their similarity to some of the seismic processing task
have created a new venue of research for applying DL to seismic problems.
In this thesis, we first tackle the problem of seismic interpolation. Re-
constructing missing traces is an ideal test for applying new deep learning
methodologies to seismic data. Unlike other processing tasks, it is possible
to generate ground truth labels by decimating field data. We conducted
a benchmark study comparing supervised and unsupervised deep learning
methods based on convolutionnal neural networks (CNNs) for seismic in-
terpolation. Our results highlight the potential of supervised deep learning
for accurately interpolating missing traces when the training data closely
represents the application dataset. However, we also identified challenges
with supervised methods, particularly when interpolating out-of-distribution
data or handling large gaps of consecutive missing traces. The unsupervised
approach proved to be a promising solution for mitigating generalization
issues, though it also faced difficulties with reconstructing large gaps. As
a response to these challenges, in a manuscript ready for submission for
the journal GEOPHYSICS, we proposed the TX-DPI methodology, which
maps space-time coordinates to seismic amplitudes, significantly improving
the interpolation of multiple consecutive missing traces and dip events, even
in a challenging field dataset. This work demonstrates the advantages of
extending CNNs beyond conventional image-to-image tasks, showing their
significant potential in deep learning applications for seismic data.
In the third chapter we tackle the problem of multiple attenuation. This is
a very important step for prestack seismic processing as the goal is to remove
coherent multiples of the primary reflections leading to improved imaging and
interpretation. A common workflow for multiple elimination involves Surface-
Related Multiple Elimination (SRME) and the Radon Transform (RT). This
process is representative of complex procedures where sensitive numerical pa-
rameters must be found for optimal results and human intervention is needed
to discriminate events. Consequently, a deep learning data-driven approach
could bring several advantages. The challenge here is to train a network that
gives results on many different datasets, i.e. a good generalization. In a paper
currently under revision for the journal GEOPHYSICAL PROSPECTING,
a UNet network is trained using a large range of synthetic NMO-corrected
gathers composed of known primaries and multiples. Several learning strate-
gies are also tested: predicting primaries, multiple or both. The study shows
that predicting multiples gives the most robust results for noisy data. Evalua-
tion in four different field datasets shows a very efficient multiple attenuation
compared to conventional SRME and RT methods. Following this study, we
discuss the limitations of the deep learning methodologies and propose pos-
sible solutions for some of them.
In the fourth chapter, we proposed a supervised deep-learning methodol-
ogy to train CNNs to predict time shifts for residual moveout correction. As
in the previous chapter, we create synthetic training data to address this pro-
cessing task. We suggest particular adjustments to the training methodology
as well as to the CNN architecture to map misaligned primary reflectors to
the corresponding time shifts needed for alignment. The proposed method
shows promising generalization in 2 post-migration field datasets. Our DL
approach works in a parameter-free manner during inference time, relieving
the user from any manual task or optimal parameter search. As a result, it
can provide a new powerful tool for residual moveout correction in existing
workflows.
In the last chapter we conclude this thesis and further explore the broader
implications of the investigations presented. We first discuss extending our
benchmark study for interpolation, considering architecture beyond CNNs
and other deep learning methodologies. Additionally, we discuss extending
the TX-DPI methodology to other applications. In particular, we proposed
including spatio-temporal information in supervised deep learning method-
ologies. Finally, we proposed the Physics Informed Convolutional Neural
Networks (PICNNs), allowing the use of CNNs in Physics Informed Neu-
ral Networks (PINNs) paradigm. A wide range of applications that process
seismic signals with CNNs can benefit from including temporal and spatial
locations as we presented in the TX-DPI method
Design Workflow for Layouts of Heating Foils in Automotive Radar Radomes
No full textSnow and ice can impair the functionality of automotive radar sensors. This is a significant concern for advanced driver assistance systems (ADAS) and autonomous driving (AD). As a solution, heating foils are integrated into bumpers or design emblems to prevent snow and ice accumulation. A heating foil consists of wire grids embedded in a dielectric foil material. It can also be implemented using conductive tracks. This paper provides a workflow for designing different layout geometries of heating foils by taking into account the transmission properties. The transmission properties are evaluated using analytical, simulative and experimental methods. An automated software application (App) is developed to generate different heating foil layouts, which can also help in determining a suitable layout based on the transmission values. The App is intended for use in the manufacturing of heating foils, as it enables the easy generation of layouts with different shapes and properties to meet customer needs
CRM in jungen Unternehmen: Wie B2B Hightech-Startups das Potenzial von CRM bestmöglich nutzen können
No full textEine der größten Herausforderungen von Startups ist es Kunden zu gewinnen und zu binden, bevor ihnen die finanziellen Mittel ausgehen. Gerade bei Frühphasen B2B-Hightech-Startups zeigt sich immer wieder, dass brillante Technologien allein nicht ausreichen – ohne eine durchdachte Marketing- und Vertriebsstrategie scheitern selbst die vielversprechendsten Ideen.
Genau hier setzt diese Masterthesis an und untersucht, wie frühphasige B2B-Hightech-Startups das Potenzial von Customer-Relationship-Management (CRM) bestmöglich nutzen können. Durch eine Kombination aus fundierter Theorie-Recherche und ausführlichen, qualitativen Experteninterviews mit Startupgründern, CRM-Coaches und CRM-Tool-Herstellern wurde erarbeitet, was Frühphasen-Startups bei der Implementierung und Nutzung beachten müssen, um CRM nicht nur für ihre kurzfristige Kundenakquise, sondern auch für eine nachhaltige Wachstumsstrategie nutzen zu können.
Die Ergebnisse sind eindeutig: Wer frühzeitig mit CRM startet, klare Prozesse definiert und standardisiert, dazu seine Daten analysiert und die Erkenntnisse daraus nutzt, kann nicht nur schneller, mehr Kunden gewinnen, sondern auch langfristige Beziehungen aufbauen, den Umsatz steigern und so die langfristige Finanzierung des eigenen Startups sichern. Als Ergebnis dieser Arbeit wurden nicht nur Handlungsempfehlungen abgeleitet, sondern auch ein Hilfsmittel, in Form eines CRM-Workshops, spezifisch für frühphasige B2B-Startups entwickelt. Dieser Workshop wurde bereits im Zuge dieser Thesis erfolgreich mit einem Startup durchgeführt und ist nun fester Bestandteil des Venture Models der innoWerft.
So bietet diese Masterthesis Startups eine konkrete Hilfestellung, wie sie CRM nicht nur als ein weiteres Tool, sondern als Gamechanger für ihr Wachstum nutzen können
Die Zukunft des Online Lebensmittelhandels
No full textDie COVID-19 Pandemie hat dem Online-Lebensmittelhandel eine neue Bedeutung verliehen, und auch für die Zukunft ist mit einer zunehmenden Kaufbereitschaft zu rechnen. Doch wie wird sich die Kaufbereitschaft für den Online-Lebensmittelhandel in den Altersgruppen 21–40, 40–65 und 65–85 Jahre in den kommenden fünf Jahren in Deutschland entwickeln, und wie werden sich dabei die Unterschiede zwischen diesen Altersgruppen verändern?
Diese zentrale Forschungsfrage steht im Fokus der vorliegenden Arbeit und liefert somit qualitative Einblicke in ein bislang wenig untersuchtes Feld. Denn derzeit ist die Datenlage zu Studien, die eine breite Altersspanne im direkten Vergleich betrachten, nur begrenzt vorhanden. Darüber hinaus wird die zentrale Forschungsfrage durch drei Unterforschungsfragen ergänzt und näher definiert. Die Unterforschungsfragen beziehen sich auf die Auswirkungen der COVID-19-Pandemie, die Rolle der digitalen Kompetenz sowie auf die Unterschiede in den Präferenzen und Bedürfnissen der verschiedenen Altersgruppen. Zur Beantwortung dieser Fragen wurde sowohl ein theoretischer Rahmen aufgebaut als auch eine empirische Untersuchung durchgeführt. Insgesamt wurden 30 leitfadengestützte, halbstrukturierte Interviews mit Personen aus drei Altersgruppen geführt und anschließend mit der qualitativen Inhaltsanalyse nach Mayring ausgewertet.
Die Ergebnisse zeigen, dass sich die Kaufbereitschaft während der Pandemie zwar deutlich erhöht hat, sich jedoch im Vergleich zum allgemeinen Online-Handel weniger dauerhaft im Alltag verfestigt hat. Die digitale Kompetenz erweist sich als zentraler Einflussfaktor auf die Kaufbereitschaft, sollte jedoch nicht isoliert betrachtet werden, da auch andere Faktoren, wie etwa die wahrgenommene Nützlichkeit, eine wichtige Relevanz spielen. Zwischen den Altersgruppen lassen sich klare Unterschiede in den Beweggründen und Barrieren feststellen. Während Zeitersparnis vor allem für die jüngeren Altersgruppen relevant ist, spielen in der ältesten Gruppe eher Preisvergleichsmöglichkeiten und das Einkaufserlebnis eine zentrale Rolle. Zweifel an Frische und Qualität sowie die fehlende Möglichkeit zur Produktprüfung sind besonders in den beiden älteren Gruppen ausgeprägt, wohingegen Datenschutzaspekte vor allem in der jüngsten Altersgruppe eine größere Bedeutung haben.
Die Arbeit liefert zudem einen Ausblick für zukünftige Forschung, etwa durch die Empfehlung, die vorliegenden qualitativen Ergebnisse im Rahmen einer quantitativen Anschlussstudie zu erweitern
Visualization of Two-Phase Flow Patterns occurring in a 1 mm inner Diameter Channel with Carbon Dioxide as a Working Fluid
No full textThe goal of this thesis is the development of a method for visualizing two phase CO2 flow in a 1 mm inner diameter channel. For this a pipe with a 5 cm long transparent glass section is assembled, installed into the MIRA experiment and observed using a high speed camera. Three different mass fluxes G = 500 kg/m^2s, G = 800 kg/m^2s and G = 1200 kg/m^2s at six heat fluxes each q = 0 kW/m^2, q = 10 kW/m^2, q = 20 kW/m^2, q = 40 kW/m^2, q = 60 kW/m^2 and q = 70 kW/m^2 are observed over a temperature range from T = 15˚C to T = -25˚C. The observed saturation temperatures are T = 15˚C, T = 5˚C, T = 0˚C, T = -5˚C, T = -15˚C and T = -25˚C. The suggestion of the best possible camera parameters for future observations are a frame rate of 13600 fps, a shutter speed of 30 kHz and a light arrangement with 2 lights.
The local HTCs (Heat Transfer Coefficients) as well as the pressure drops obtained as part of this thesis, do align with expectations as well as previous data taken by Pedano and Hellenschmidt. For the maximum heat flux of q = 70 kW/m^2, the local HTCs are in the range from α = 11 kW/m^2K to α = 21 kW/m^2K. 21 kW/m^2K are observed at G = 500 kg/m^2s, T = 15˚C and x = 0.023.
From the visual data obtained, it can be said, that phenomena of confinement are visibly starting to take place in the temperature area of -5˚C. This aligns both with the observations made by Pedano in 2021, as well as the confinement threshold of Co = 0.79 proposed by Ullmann and Brauner.Das Ziel dieser Thesis ist die Entwicklung einer Methodik zur Visualisierung des Flusses von zweiphasigem CO2 in einem Rohr mit 1 mm Innendurchmesser. Hierfür wird eine 5 cm lange Sektion aus transparentem Glas in das MIRA Experiment eingebaut und mittels einer Hochgeschwindigkeitskamera beobachtet. Dabei werden drei verschiedene Massenflüsse G = 500 kg/m^2s, G = 800 kg/m^2s und G = 1200 kg/m^2s bei jeweils sechs verschiedenen Wärmeflüssen q = 0 kW/m^2 ,q = 10 kW/m^2, q = 20 kW/m^2, q = 40 kW/m^2, q = 60 kW/m^2 und q = 70 kW/m^2 über eine Temperaturspanne von T = 15 ◦C bis T = −25 ◦C betrachtet. Die Einzelschritte der Sättigungstemperaturen sind hierbei T = 15 ◦C, T = 5 ◦C,T = 0 ◦C, T = −5 ◦C,T = −15 ◦C und T = −25 ◦C. Die Kameraparameter, welche sich als optimal für die Visualisierung herausgestellt haben, sind hierbei eine Bildaufnahmerate von 13 600 fps, eine Verschlussfrequenz von 30 000 Hz und der Aufbau mit zwei Lichtquellen.
Der lokale Wärmeübertragungskoeffizient sowie der Druckverlust über die Testsektion, welche im Rahmen dieser Thesis aufgenommen wurden, entsprechen den Erwartungen aus vorherigen Arbeiten von Pedano und Hellenschmidt. Für einen maximalen flächenspezifischen Wärmefluss von q = 70 kW/m^2 , werden Wärmeübertragungskoeffizienten von α = 11 mkW2K bis α = 21 mkW2K erzielt. Das Maximum wird bei einem flächenspezifischen Massestrom von G ≈ 500 kg/m^2s , einer Sättigungstemperatur von T = 15◦C und einem Dampfanteil von x ≈ 0.023 gemessen. Aus den visuellen Ergebnissen lässt sich schlussfolgern, dass ein confinement (Einschränkung der Blasen im Zweiphasenfluss) bei einer Temperatur um −5 ◦C einzusetzen beginnt. Dies steht im Einklang mit den Beobachtungen von Pedano, sowie dem confinement Übergangspunkt von Co = 0.79, wie er von Ullmann und Brauner definiert wurde
Youth Anti‐Corruption Potential: Insights From Germany, Lithuania and Spain
No full textCorruption is universally recognised as one of the biggest challenges for modern societies. Its negative impact on economies and institutions, as well as its erosive effect on citizen trust and state stability, pose a significant strain on good governance. Due to its pervasive nature, implementation of anti-corruption policies and education require persistent efforts and dedication. Understandably, young people are identified as the most important cohort within society, which should be well prepared to address all challenges associated with the malpractice. According to the Theory of Planned Behaviour, knowledge must be transformed into perception, followed by the adoption of a suitable attitude, which should then be reflected in future behaviour. As such, it is paramount to ensure that young individuals are able to comprehend the negative impact of corruption, identify the malpractice, and be prepared to inform the relevant authorities when faced with acts of corruption. This set of competences is referred to as anti-corruption potential. It is shaped by the cultural, societal and institutional constraints of the country (region) as well. It consists of three main elements—perception (knowledge), attitude (values) and behaviour. The aim of this paper is to evaluate the current dynamics of youth anti-corruption potential in three European countries—Germany, Lithuania and Spain. For its purposes, a survey was conducted amongst 1,922 young individuals, aged 15–29, who are currently in education. The countries selected represent three main EU regions—Western, Eastern and Central and Southern Europe. The results demonstrate that corruption is universally recognised as an existing challenge. However, Lithuanian and German young people exhibit higher intolerance towards the malpractice, whilst Spanish youth demonstrate the most positive attitude in regard to integrity as a contributing factor to personal success. Moreover, the majority of respondents from all three countries assert that their decision to report suspected or witnessed acts of corruption would be made after a thorough deliberation, taking into account the specific circumstances and the context of the situation. Results further indicate that anti-corruption education programmes should become an indispensable part of the educational process. However, such programmes must be tailored to reflect the cultural specificities of the society and the unique needs of the youth. This research makes a major contribution regarding the anti-corruption potential of young people across diverse European contexts. It further demonstrates how regional and cultural variations shape perceptions, attitudes and behaviour towards corruption. As such, increasing understanding of the social and cultural context in which corruption occurs—both at personal as well as state level—should be considered a priority by policymakers and practitioners
A Novel and Innovative Approach for Capacitive Sensoring of Spherical Joint Coordinates
No full textThis work presents a mathematical framework that facilitates capacitive position sensing by quantifying the varying capacitance of capacitors formed by movable electrodes. The electrode displacement is described in relation to the resulting capacitor surface, which enables the formulation of a polynomial that correlates the displacement angle with the resulting capacity
High electricity price despite expansion in renewables: How market trends shape Germany’s power market in the coming years
Get PDFExpectations about future energy prices are crucial for investment decisions, market reform debates, and public policy. Yet, the recent energy crisis caused dramatic market uncertainty. This study investigates Germany’s near-future wholesale electricity price in the context of evolving market trends. A flexible econometric model is applied to high-frequency, near-time data, spanning January 2015 through May 2023. A potential endogeneity bias of trade is circumvented by an instrumental-variables approach. Results indicate that expanding renewable energy exerts downward pressure on price, countering trends like the nuclear phaseout, a rising carbon price, increased electrification, and a high gas price. The collective impact suggests a considerably higher electricity price in the coming years compared to pre-crisis levels. This finding is corroborated by a fundamental energy system model. The potential rise in renewables’ production volatility may amplify electricity price volatility. A high and volatile near-future electricity price could spur investments in renewables and flexibility technologies but pose challenges for consumers. Our analysis aids evidence-based decision-making amid the post-crisis landscape
Enhancing human–robot collaboration with thermal images and deep neural networks: the unique thermal industrial dataset WLRI-HRC and evaluation of convolutional neural networks
Get PDFThis contribution introduces the use of convolutional neural networks to detect humans and collaborative robots (cobots) in human–robot collaboration (HRC) workspaces based on their thermal radiation fingerprint. The unique data acquisition includes an infrared camera, two cobots, and up to two persons walking and interacting with the cobots in real industrial settings. The dataset also includes different thermal distortions from other heat sources. In contrast to data from the public environment, this data collection addresses the challenges of indoor manufacturing, such as heat distortions from the environment, and allows for it to be applicable in indoor manufacturing. The Work-Life Robotics Institute HRC (WLRI-HRC) dataset contains 6485 images with over 20 000 instances to detect. In this research, the dataset is evaluated for implementation by different convolutional neural networks: first, one-stage methods, i.e., You Only Look Once (YOLO v5, v8, v9 and v10) in different model sizes and, secondly, two-stage methods with Faster R-CNN with three variants of backbone structures (ResNet18, ResNet50 and VGG16). The results indicate promising results with the best mean average precision at an intersection over union (IoU) of 50 (mAP50) value achieved by YOLOv9s (99.4 %), the best mAP50-95 value achieved by YOLOv9s and YOLOv8m (90.2 %), and the fastest prediction time of 2.2 ms achieved by the YOLOv10n model. Further differences in detection precision and time between the one-stage and multi-stage methods are discussed. Finally, this paper examines the possibility of the Clever Hans phenomenon to verify the validity of the training data and the models’ prediction capabilities