Publikationsserver der Technischen Hochschule Augsburg
Not a member yet
2118 research outputs found
Sort by
The Echoes of Ostracism and Peer Opinion in Subjective Tasks: Untangling Conformity Dynamics in Social Media through Experimental Exploration
No full textTowards Conversational Requirements Engineering: Detecting Defects in Requirements Using Large Language Models
No full textImplementing Business-IT Integration in Digital Enterprises Using Organizational Capabilities
No full textDigitalisierungsstrategie im Verteilnetz: Einflussanalyse von Messwerten auf eine Netzzustandsermittlung
No full textAuswirkungen präventiver Steuerungseingriffe des Netzbetreibers auf den elektrischen Leistungsbezug unterschiedlicher Verbrauchseinrichtungen
No full textProcess FEM Simulation of Fiber Patch Placement Laminates with a Flexible Foam-based Gripper
No full textKI-gestützte Optimierung der Werkzeugstandzeit in CNC-Maschinen
No full textThis thesis presents the development of a prototype, locally deployable AI system for predicting tool life in CNC machines. Its goal is to optimize maintenance cycles and reduce downtime through data-driven analysis. The system uses machine learning methods implemented with TensorFlow/Keras and follows a modular architecture separating the data pipeline, model, and interface integration. Historical machine data are preprocessed, encoded, and scaled for training.
The evaluation shows that the prototype is technically functional but fails to meet its predictive goal, reaching an error rate of about 80%. The main reasons are the limited dataset and inconsistent input quality. This underlines the strong dependency of model performance on data completeness and consistency.
The conclusion identifies key improvements: expanding data sources, refining feature engineering, hyperparameter tuning, and integrating real-time production data. The prototype thus provides a foundation for future research toward higher predictive accuracy and industrial applicability.In der vorliegenden Arbeit wird ein prototypisches, lokal betreibbares KI-System zur Vorhersage der Werkzeugstandzeit in CNC-Maschinen entwickelt. Ziel ist es, durch datengetriebene Analysen Wartungszyklen zu optimieren und Ausfallzeiten zu reduzieren. Grundlage bilden maschinelle Lernverfahren mit TensorFlow/Keras sowie eine modulare Softwarearchitektur mit separater Datenpipeline, Modellkomponente und Schnittstellenanbindung. Als Datenquelle dienen historische Maschinendaten, die vorverarbeitet, kodiert und skaliert werden.
Die Evaluation zeigt, dass der Prototyp technisch funktionsfähig ist, seine prognostische Zielgröße jedoch mit einer Fehlerrate von rund 80 % deutlich verfehlt. Hauptursachen sind die begrenzte Datenbasis und die schwankende Eingabedatenqualität, was die starke Abhängigkeit der Modellgüte von Datenvollständigkeit und -konsistenz verdeutlicht.
Im Fazit werden Optimierungspotenziale benannt, darunter erweiterte Datenquellen, verbessertes Feature Engineering, Hyperparameter-Tuning und die Integration von Echtzeitdaten. Zudem werden mögliche Anwendungen in anderen Branchen wie Automobilbau, Luft- und Raumfahrt sowie Predictive Maintenance aufgezeigt
The Sustainable Revitalization of the Former Residential and Industrial Tower on Baumgartnerstraße: A Process Description
No full textThis paper documents the ongoing sustainable revitalization process of a historic residential and industrial tower on Baumgartnerstraße in Augsburg through the "Green-Heritage360°" project. The 1926-built structure, vacant since 2009, serves as a real-world laboratory demonstrating how participatory planning methodologies can transform underutilized heritage into active community assets through interim use concepts, successive restoration strategies, and stakeholder co-creation processes. Operating as a Living Lab, the initiative integrates ecological, social, and economic sustainability dimensions while fostering collaboration between heritage conservation specialists, academic institutions, and municipal partners. The comprehensive stakeholder network analysis reveals how diverse actors contribute essential expertise to the collaborative revitalization effort, challenging traditional top-down urban planning through bottom-up approaches that emphasize equality and local participation. Initial activation phases include establishing a student-run café using recycled materials and reversible construction methods, alongside academic integration through interdisciplinary coursework. The project aims to develop transferable models for climate-friendly revitalization of vacant heritage buildings while advancing novel educational concepts through digital knowledge exchange platforms
Systemic Influence of Macroeconomic Factors on German Used Textile Collectors and Sorters
No full textThe German textile collection and sorting system is facing mounting ecological, economic, and regulatory pressures. This paper utilizes a PESTEL analysis and a Vester Sensitivity Model to examine the sector's primary challenges. The analysis, which is based on 32 influencing factors, identifies textile recycling and infrastructure as critical levers for sustainable transformation. The findings provide a systemic perspective on interdependencies and support strategic decision-making in the context of the NuCollect research project
Experimental Frequency Analysis of an Axial Flux Motor Component
No full textIn this paper, the vibration behavior of a component of an axial flux motor is to be measured. The results obtained from this are to serve as simulation validation for subsequent work to be able to simulate the dynamic behavior of the component with regards to fatigue strength. To evaluate the data, a MATLAB script is developed that uses the Fourier transform to visualize the natural frequencies of the component by means of a representation in the frequency domain and can calculate the transfer function of the measured component by dividing the determined power spectra. A numerical simulation of a two-mass oscillator is used to prove the qualitative accuracy of the MATLAB script. The data from a single series of measurements are analyzed and it is shown that the strong damping of the measurement setup has a strong influence on the data and therefore no causal relationship between the input and output signal can be proven