DSpace Collection: For doctoral theses, habilitation, bachelor's theses or master's theses.
https://doi.org/10.24405/14746
For doctoral theses, habilitation, bachelor's theses or master's theses.2024-03-19T22:35:23ZFault Diagnosis in a Permanent Magnet Synchronous Motor using Deep Learning
https://doi.org/10.24405/15302
Title: Fault Diagnosis in a Permanent Magnet Synchronous Motor using Deep Learning
Authors: Quseiri Darbandeh, Pedram
Abstract: Condition Monitoring (CM) of electrical equipment, especially electrical machines, has an important role in enhancing reliability and preventing disturbance in production centers such as power plants and factories. One of the important steps of CM's process is diagnosing faults at the proper time to prevent the defect from spreading to other components of the system. Nowadays, the Permanent Magnet Synchronous Machine (PMSM) which uses permanent magnets instead of winding in its structure is widely used among other types of electrical machines. This thesis proposes a new algorithm based on Variational Autoencoder (VAE) as an unsupervised deep learning method to detect different faults with different severities. In this research work, current, vibration and sound signals are collected from a 1 kW prototyped PMSM. This motor is simulated in Finite Element Analysis (FEA) software and compared with the prototyped one. Demagnetization, bearing, and eccentricity faults are selected to be implemented on a prototyped PMSM. The performance of VAE is evaluated for different operating conditions consisting of different loads and speeds. Also, the VAE is investigated for classifying data with different complexity. The feature extraction methods which are known as the pre-processing methods can improve the performance of VAE by reducing dimensionality and extracting important data features. Besides, these methods improve the network's training time by reducing input data size. In this thesis, Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are used as feature extraction methods. The performance of VAE-PCA and VAE-LDA is compared, and it is shown that VAE-PCA has a slightly better performance than VAE-LDA in most operating conditions, especially in the high-class number in which the complexity is higher compared to other cases. Also, this method is developed for analyzing different combinations of sensors. Results show that the combination of current and sound signals can be a successful way to diagnose faults with and without pre-processing methods. In addition, the performance of VAE is compared with other machine learning methods such as Naïve Bayes (NB) and K-Nearest Neighbourhood (KNN) with and without pre-processing methods from accuracy and training time aspects. It is demonstrated that the performance of VAE for classifying the low and high number of classes of data is proper from accuracy and training time aspects. However, the performance of machine learning methods in the high number of classes is not proper for training time.2024-03-14T00:00:00ZEntwicklung und Optimierung von Methoden zur thermischen Charakterisierung von Dünnschichten und Bulk-Materialien
https://doi.org/10.24405/15301
Title: Entwicklung und Optimierung von Methoden zur thermischen Charakterisierung von Dünnschichten und Bulk-Materialien
Authors: Metzke, Christoph Thomas
Abstract: In den letzten Jahrzehnten vollzog sich in der Mikro- und Nanoelektronik eine enorme Verringerung der Strukturgrößen bis hin zu aktuell wenigen Nanometern. Damit einhergehend spielt die Ableitung der Wärme von kritischen Strukturen eine immer wichtigere Rolle. Es werden daher thermisch gut leitfähige, aber elektrisch isolierende Dünnschichten als Dielektrika benötigt. Gleichzeitig stoßen bestehende Messmethoden zur thermischen Charakterisierung bei kleinen Schichtdicken an ihre Grenzen. Ziel dieser Arbeit ist daher die Verbesserung vorhandener Methoden und die Entwicklung einer neuen Methode zur thermischen Charakterisierung. Im Verlauf der Arbeit können dadurch Dünnschichten aus Siliziumdioxid (SiO₂), Siliziumnitrid (Si₃N₄), Bornitrid (BN) und Aluminiumnitrid (AlN) sowie einige Bulk-Samples aus Oxiden, Kunststoffen und Materialien aus der Natur thermisch charakterisiert werden. Ein Großteil der Arbeit fokussiert sich dabei auf die Scanning Thermal Microscopy (SThM), welche im Rastersondenmikroskop (AFM für engl. Atomic Force Microscope) angewandt wird. Durch eine detaillierte Analyse von Artefakten und die Herleitung geeigneter Messparameter kann SThM in vielen Submodi zuverlässig eingesetzt werden. Ausführliche Simulationen mittels Finite-Elemente-Methode (FEM) tragen zum Verständnis der Methode bei. Mit der Widerstandsmethode in Kombination mit FEM-Simulationen wird zudem eine neuartige Vorgehensweise präsentiert, welche relativ schnelle thermische Messungen an Dünnschichten mit geringem Budget ermöglicht. Vergleichsmessungen werden mittels der etablierten 3-Omega-Methode präsentiert, wodurch die Ergebnisse der anderen Methoden verifiziert werden können. Zusammenfassend leistet diese Arbeit einen Beitrag zur thermischen Charakterisierung von dünnen Schichten und Bulk-Materialien, indem vorhandene Methoden sukzessive verbessert werden, eine neue Methode entwickelt wird und vielversprechende Dünnschichten thermisch charakterisiert werden.; During the last decades, micro- and nanoelectronics have seen an enormous miniaturization of process sizes, currently down to just a few nanometres. As a result, the heat transfer away from critical structures is playing an increasingly important role. Consequently, thermally well conductive but electrically insulating thin films are needed as dielectrics. At the same time, existing measurement methods for thermal characterization are reaching their limits at small film thicknesses. Therefore, the aim of this work is the improvement of existing methods and the development of a new method for thermal characterization. Within this work, thin films of silicon dioxide (SiO₂), silicon nitride (Si₃N₄), boron nitride (BN) and aluminium nitride (AlN) as well as some bulk samples of oxides, plastics, and natural samples are thermally character-ized. One main part of this work focuses on Scanning Thermal Microscopy (SThM), which is applied in the Atomic Force Microscope (AFM). Through detailed analysis of artifacts and derivation of appropriate measurement parameters, SThM can be applied reliably in many submodes. Extensive simulations using the Finite Element Method (FEM) contribute to the understanding of the method. The “Widerstandsmethode” combined with FEM simulations is also presented as a novel approach that allows relatively fast thermal measurements on thin films with a low budget. Comparative measurements are presented using the established 3-Omega Method, allowing verification of the results of the other methods. In summary, this work contributes to the thermal characterization of thin films and bulk materials by succes-sively improving existing methods, developing a new method, and thermally characterizing promising thin films.2024-03-07T00:00:00ZMaking Strategic Decisions Under Time Pressure - A Process-based Analysis Approach
https://doi.org/10.24405/15292
Title: Making Strategic Decisions Under Time Pressure - A Process-based Analysis Approach
Authors: Kremer, Marco
Abstract: Although many strategic economic decisions are subject to time constraints, the impact of time pressure on the decision-making process of solving non-cooperative games has not been well studied in the field of behavioral game theory. This includes the effects of time pressure on decision-making in non-cooperative games (Ariely and Zakay, 2001; Ordóñez et al., 2015). Lindner and Sutter (2013) conducted the first study investigating personal sophistication in terms of the cognitive hierarchy model. To experimentally determine the distribution of level-k-reasoning types, they utilized Arad & Rubinstein’s (2012) 11-20-Game. Contrary to the findings of Sutter et al. (2003) and Kocher et al. (2006), the authors discovered a shift towards equilibrium play under growing time pressure. They attribute this discrepancy to chance since the decision information is the only factor available for interpretation.
Applying a process-oriented approach in combination with process tracing methods provides valuable insights into people's decision-making behavior (Kühberger et al., 2011). In normal-form games with no time pressure, Costa-Gomes et al. (2001) found evidence of the application of common decision heuristics by scrutinizing lookup patterns in information search, response time, and decision information. However, the use of heuristics may not be stable or complete under time-pressure conditions (Johnson et al., 2008). This raises the question of what more detailed behavior patterns might be identifiable and how such patterns change with increasing time pressure.
Therefore, this work develops a process theoretical framework called the 'Preparation Time Model'. This framework bases decision-making on Elementary Information Processes (EIPs) following Johnson and Payne (1985) and Chase (1978). Production systems for solving normal-form games are constructed for nine common heuristics based on EIPs (following suggestions of Newell et al., 1972). A minimum set of EIPs and how it can be identified in mouse tracking data is derived. The effectiveness and efficiency of heuristics in different games and under various time pressure conditions are determined through simulation. Significant differences in adaptation velocity between strategic and non-strategic heuristics raise the question of the extent to which it is rational to employ certain strategic heuristics under severe time pressure conditions.
This work also reports on an online experiment that was conducted using Mouselabweb (Willemsen and Johnson, 2011) to investigate the influence of time constraints and task complexity on individual decision-making and its patterns in 2-person normal form games. The empirical dataset was analyzed for fifteen behavior patterns from the process categories Information Search, Information Implementation, and Choice, which frequently show sensitivity to time pressure. Data clustering indicates the existence of different types of decision-makers who pursue individual strategies to deal with time pressure: the “Strategist”, the “Adaptist”, and the “Guesser”. The findings confirm the qualitative response schema of individuals acting under time pressure in individual decision situations, as described by Miller (1960), Ben Zur and Breznitz (1981), and Zakay (1993), and specify the schema for the case of normal-form game tasks. However, questions regarding the extent to which heuristics are applied or whether findings are robust for predicting behavior remain unanswered. The empirical dataset holds potential for further scrutiny.2024-03-01T00:00:00ZKI-basiertes Monitoring und Diagnose von elektrischen Maschinen im industriellen Umfeld
https://doi.org/10.24405/15300
Title: KI-basiertes Monitoring und Diagnose von elektrischen Maschinen im industriellen Umfeld
Authors: Benninger, Moritz
Abstract: Diese Arbeit befasst sich mit dem Monitoring und der Diagnose von elektrischen Maschinen im industriellen Umfeld auf der Basis von künstlicher Intelligenz (KI). Das Ziel ist die Entwicklung einer praxisorientierten Methodik, welche flexibel und übertragbar eine Fehlererkennung für eine Vielzahl an elektrischen Maschinen ermöglicht. Der Fokus der Arbeit liegt folglich auf dem Entwurf eines Frameworks mit hoher Praxistauglichkeit, der eine analytische Modellierung mit neuronalen Netzen aus dem Bereich des Machine Learning kombiniert.
Das Framework beinhaltet zudem eine Methodik, welche die Identifizierung passender Werte für die Modellparameter der überwachten elektrischen Maschine erlaubt.
Nach einer initialen Parameteridentifkation erfolgt mit dem parametrisierten Modell die Erzeugung eines Datensatzes mit simulierten Statorströmen der überwachten Maschine in gesunden und fehlerhaften Zuständen. Der simulierte Datensatz wird von den neuronalen
Netzen zum Erlernen von Fehlermerkmalen genutzt, sodass in der Folge auf der Grundlage von gemessenen Statorströmen eine Einschätzung zum aktuellen Zustand der realen Maschine erfolgen kann.2024-02-05T00:00:00Z