Examensarbeten för masterexamen
Länka till denna samling:
Browse
Senast publicerade
Visar 1 - 5 av 653
- PostExploring factors for electrode prototyping for PEM fuel cells(2024) Björklund Larsen, Frederikke; Chalmers tekniska högskola / Institutionen för fysik; Chalmers University of Technology / Department of Physics; Wickman, Björn; Mikaeili, ParinazAs the world push to lower carbon dioxide emissions to limit climate change, the need for new technologies becomes more and more critical. Some emerging technology is hydrogen fuel cells and amongst them are proton exchange membrane fuel cells. With all new technologies there are several factors that need development, and this work takes a closer look at the catalyst layer of the cathode. Setting out to increase the platinum loading in the cathode while at the same time avoiding cracks, several catalyst inks were made with varying dispersion matrices. The matrices explored were a) water with ethanol and 1-propanol (2:2:1 weight ratio), b) water with 1-propanol (2:3 weight ratio), c) water with 2-propanol (2:3 weight ratio), d) water with tert-butanol (2:3 weight ratio), e) water with 1-propanol and tert-butanol (2:1:2 weight ratio), and f) water with 1-propanol and tert-butanol (1:1:3 weight ratio). To get a deeper understanding of the ink’s properties, rheological tests of the inks and visual analysis of produced electrodes were performed. These analyses found improved coating quality and a higher viscosity for dispersions with a lower dielectric constant and for inks containing solvent with a lower vapour pressure. The improved behaviour was attributed to improved interactions between the ink’s compounds and slower drying of the coatings, leading to less stresses in the electrodes. An improved electrode quality was also observed when the inks were left to mature on a magnetic stirrer for several days. The maturation step resulted in lower viscosity of the inks indicating smaller effective volume fraction of particles and less electrostatic repulsion and steric hinderance between compounds. A final factor in the process that was tested was increasing the relative humidity during the drying process. Here an improved cracking behaviour was observed for the ink containing more water while the opposite was seen in the ink containing more tert-butanol. These findings point towards the need for specific drying processes for each individual ink
- PostExploring feasibility of AI-driven insights for decision making in an e-commerce environment(2024) Johansson, Dan; Chalmers tekniska högskola / Institutionen för fysik; Chalmers University of Technology / Department of Physics; Granath, Mats; Bergman, DanThis thesis explores the potential of data-driven decision-making and machine learn ing inferences within an e-commerce context, focusing on sales and campaign per formance modeling at Viskan Systems. The research initiates by dissecting an exist ing database structure, identifying significant potentials for implementing machine learning methodologies despite encountering systemic data management challenges. These challenges include issues with overwriting campaign instances and handling campaign parameters, which could impede accurate data analysis and modeling. The study implements and evaluates two distinct machine learning models: XG Boost and NeuralProphet. The XGBoost model reveals limitations in handling the wide variance in sales data, leading to a general trend of overestimation in smaller campaigns and underestimation in larger ones. The NeuralProphet model, employed for time series forecasting, shows a hierarchical structure in model performance, with the meta model yielding the most accurate results. Despite their limitations, these models highlight the feasibility of advanced data analytics in enhancing decision making processes for Viskan Systems and its customers. The thesis concludes by recommending strategic modifications to Viskan Systems’ data infrastructure to facilitate the integration of data-driven approaches and ma chine learning. Such enhancements are deemed essential for the system’s adaptation to sophisticated analytics, ensuring data integrity while improving compatibility with emerging technologies.
- PostQuantum Routing using Value-Based Reinforcement Learning(2023) OPPERUD, MIKKEL; Chalmers tekniska högskola / Institutionen för fysik; Chalmers University of Technology / Department of Physics; Granath, MatsThis thesis addresses the Quantum routing problem through the implementation of a reinforcement learning algorithm. Quantum routing is the problem of making quantum circuits executable on a quantum computer with limits of connectivity which requires requires swapping information between qubits. A value-based variant of the Q-learning algorithm, coupled with deep convolutional neural networks, was employed to optimize the routing process in a grid topology environment. The environment allowed the agent to place and remove swaps and to "pull back" any immediately executable qubits. The reward scheme was designed to optimize for a shortened circuit depth with the first layers of swaps not counted, thus solving the Quantum routing and placement problem concurrently. The study focused on smaller grid sizes of 3x2, 3x3, and 3x4. Due to time constraints we were not fully able to adequately access the performance of the model and were only able to achieve solutions for smaller models, while the results for the larger ones (3x3 and 4x3) were lackluster. For larger grid sizes our analysis on multiple hyper-parameters revealed a better understanding for the reasons for this, suggesting possible remedies. In conclusion, while the algorithm encountered issues during the experiment, these obstacles present opportunities for future improvement and refinement. This research provides a foundation for future studies in the realm of Quantum routing, highlighting potential avenues for enhanced algorithm performance.
- PostMasked Prediction of Time Series data Using Novel Machine Learning Models(2024) Gopalakrishnan, Dinesh Krishnan; Chalmers tekniska högskola / Institutionen för fysik; Chalmers University of Technology / Department of Physics; Granath, Mats; Ganesan, Deepak GuruWe are living in a time where data is the new oil, where the industries are data driven. This change was expedited by the enormous amount of data that we are producing every second and the increase in computational power. At present the automotive sector is thriving in this data-driven model by their ubiquitous need for common and industrial purposes and the various data they are collecting to improve the sector as a whole. We can now see an automobile as not just a mechanical prod uct but as a robot on wheels. Along with this data-driven model, the electrification of automobiles has revolutionized the industry. In the electrification process, the battery module is one of the key components that power the systems. To do this we must analyse the data from the battery modules for its efficient usage. However, due to certain hardware issues or if the vehicle is out of range and it cannot update the data we might lose data. This loss of data can obstruct the efficient usage of the data in machine learning models to optimize the system. There are several methods to impute missing data, for example, there are statisti cal methods such as the Auto-regressive methods, which are limited by their time and the high cost of their computations. This thesis focuses on this problem and designing a neural network model for masked prediction of the Time series data. In this thesis, a Transformer Network is implemented for the masked prediction of the missing time series data. In this thesis, we have built the machine learning model from scratch after weighing several factors. The data on which the model is trained is generated by the vehicle collected. This was led by pre-processing, later following the selection of the model. The model developed here is a variation of the transformer model, called the Time Series Transformer(TST), which predicts the missing values in the time series data. This model is then evaluated with suitable metrics by the model and the problem statement. The thesis aims to predict the missing values to improve the quality of the data collected and its quality usage to improve the performance of the vehicle.
- PostQuantum Routing using Value-Based Reinforcement Learning(2023) Opperud, Mikkel; Chalmers tekniska högskola / Institutionen för fysik; Chalmers University of Technology / Department of Physics; Granath, MatsThis thesis addresses the Quantum routing problem through the implementation of a reinforcement learning algorithm. Quantum routing is the problem of making quantum circuits executable on a quantum computer with limits of connectivity which requires requires swapping information between qubits. A value-based vari ant of the Q-learning algorithm, coupled with deep convolutional neural networks, was employed to optimize the routing process in a grid topology environment. The environment allowed the agent to place and remove swaps and to "pull back" any immediately executable qubits. The reward scheme was designed to optimize for a shortened circuit depth with the first layers of swaps not counted, thus solving the Quantum routing and placement problem concurrently. The study focused on smaller grid sizes of 3x2, 3x3, and 3x4. Due to time constraints we were not fully able to adequately access the performance of the model and were only able to achieve solutions for smaller models, while the results for the larger ones (3x3 and 4x3) were lackluster. For larger grid sizes our analysis on multiple hyper-parameters revealed a better understanding for the reasons for this, suggesting possible reme dies. In conclusion, while the algorithm encountered issues during the experiment, these obstacles present opportunities for future improvement and refinement. This research provides a foundation for future studies in the realm of Quantum routing, highlighting potential avenues for enhanced algorithm performance.