Examensarbeten för masterexamen
Länka till denna samling:
Browse
Senast publicerade
Visar 1 - 5 av 2056
- PostIndirect Pressure Measurement in Batteries and State Estimation of SoH - Internal Gas Pressure measurement in Li-ion Cells for SoH Estimation using Deep Neural Network(2024) Sunku, Aditya; Chalmers tekniska högskola / Institutionen för elektroteknik; Thiringer, Torbjörn; Bartholdsson Frenander, Bartholdsson Frenander; Abbasi, AbbasiAbstract This thesis presents a new methodology for internal gas pressure measurement in Li-ion batteries to predict the State of Health of the battery. This is achieved by a simulation study, experimental testing of Li-ion batteries and a Deep Neural Network (DNN) model for state estimation. The primary motive of this thesis is to measure the internal gas pressure in Li-ion batteries in a non-invasive approach and to asses the battery health based on the amount of gas generated inside the cell over its life cycle. The traditional methods used in present day BMSs, use electrical parameters like voltage and current as input parameters to asses the states of a battery like State of Health (SoH), State of Charge (SoC). This study aims to take into consideration the mechanical parameters of the battery like internal gas pressure and cell temperature, for state estimation. This Master Thesis deals with an in-operando, non-invasive pressure measurement of batteries. The internal gas pressure of the battery was monitored by capturing/measuring the strain developed on the battery cell lid. Uses this measured/predicted internal gas pressure as an input variable to the battery management system (BMS) for SoH estimation. Extensive cell testing was performed to capture the cell behaviour during cycling and measure the gas pressure developed inside the cell. This data was used as input parameters for the Deep Neural Network (DNN) model, developed to predict the State of Health (SoH) of the battery. Through this study, a correlation between internal gas pressure and health indicators of batteries like Direct Current Internal Resistance (DCIR), Discharge Energy and & Charge Capacity of the battery was established. This correlation further aids in accurately predicting the State of Health (SoH) of the battery.
- PostA machine-learning approach to reduce the risk of collision when changing lanes(2024) Song, Yaochen; Sun, Junyu; Chalmers tekniska högskola / Institutionen för elektroteknik; Fredriksson, Jonas; Dahl, JohnAbstract Advanced driver assistance systems support drivers to handle different complex traffic conditions. The support systems get traffic information from sensors and use algorithms to avoid risks. However, dealing with complex time series data from various sensors is challenging. In this thesis, a machine learning approach is proposed for threat assessment for lane changing. The emphasis is on vehicle state prediction and maneuvers for autonomous emergency steering. The work includes feature selection, model selection, and model validation. Feature selection is performed using the NSGA-II algorithm and correlation analysis to identify the most influencing features. This helps reduce the data dimension while maintaining prediction accuracy. An artificial neural network model structure inspired by ResNet is developed. This network structure is built from blocks, each with a shortcut. Various model configurations, including the number of input features and the network depth, are tested to find a reasonable tradeoff. In addition, driver-state information is also analyzed, and the "most probable gaze zone" data features enhance the model’s performance. The proposed model is validated on real-world data and has good performance.
- PostInvestigation of unintentional island operation during large amounts of inverter based power production(2024) Stenvall, Emma; Brännman, Martin; Chalmers tekniska högskola / Institutionen för elektroteknik; Chen, Peiyuan; Axelsson, Anders; Kallin, Mikael; Torkildsson, ErikAbstract With increasing efforts to utilize more renewable energy sources, it is becoming more and more popular to install solar panels on buildings and in residential areas. This trend is predicted to continue to grow within the foreseeable future, which brings up questions of how an increased amount of local inverter based power production would affect the contemporary grid and its behavior. There has been concern that an increase of local energy sources, such as solar power, could potentially pose a greater risk of unintentionally transition into island operation if the connection to the utility grid is lost. The aim of this report was therefore to investigate different scenarios in the simulation software Powerfactory and discern the behavior and possible risk of unintentional island operation in different grid models with varying amounts of photovoltaic (PV) modules. All PV modules connected to the Swedish electrical power grid must comply with standards and regulations and their inverters must contain the proper unintentional island protections. The inverter protections come in forms of passive, basic monitoring, protections and active ones, which actively tries to perturb the grid in order to detect islanding events. This thesis work was conducted by building four different grid models in Powerfactory. The grid models consisted of low voltage grids with varying amounts of PV modules, loads and cables, connected to an external grid which would be disconnected in order to simulate an islanding event. Each model was tested for different potential islanding events on the low voltage side while monitoring voltage, current and frequency values. The protection activation time to detect the potential island was also monitored in seconds, where protection triggers above five seconds were regarded as insufficient detection. For one of the models, a non-detection zone was established in order to visualize the detection limits. From the different grid model behaviors, it was concluded that detection capabilities of the inverter protection was highly dependant on the load imbalance. The reactive power mismatch had a much lower threshold compared to the active power, resulting in a very narrow nondetection zone in regard of reactive power compared to the much higher threshold values of active power in said non-detection zone. It was also concluded that there is not much readily available information about the active inverter protection schemes and further investigation about how they interact with the passive protections are needed.
- PostMotor Functions Assessment for Parkinson Disease via Radar Sensors Focusing on Finger Tapping Test(2024) Isaksson, Gustav; Mosaddeghi, Seyedehnaghmeh; Chalmers tekniska högskola / Institutionen för elektroteknik; Kjellby Wendt, Gunilla; Zeng, XuezhiAbstract This thesis investigates the assessment of motor functions using a radar sensor, focusing on the finger tapping test. The primary goal is to help healthcare professionals in accurately detecting and analyzing finger tapping test, which is essential for effective treatment and management of Parkinson’s disease (PD). Data of several finger tapping scenarios was collected by mimicking them by healthy individuals using the radar sensor, followed by comprehensive data and signal processing. The developed model achieved a promising accuracy of 93.18% on the dataset collected by students. It successfully identified and scored the cases with interruptions, as well as amplitude and frequency decrements, although it does not provide a severity score for decrement cases. However, the model’s inability to handle cases involving combinations of interruptions and decrements was identified as a limitation. Tests conducted by physiotherapists resulted in lower accuracy, primarily due to the radar sensor’s high sensitivity to motion and distance changes. This thesis explores the potential of radar sensor technology in monitoring motor functions while highlighting the challenges associated with data collection and sensor sensitivity.
- PostAnti-Money Laundering with Unreliable Labels(2024) Bergquist, Jesper; Chalmers tekniska högskola / Institutionen för elektroteknik; Graell I Amat, Alexandre; Östman, JohanThis report examines the effectiveness of Graph Neural Networks (GNNs) in detecting money laundering activities using transaction data with unreliable labels. It explores how weakly supervised learning, specifically with GNNs, manages the challenges posed by missing and inaccurate labels in anti-money laundering (AML) systems. The study utilizes simulated transaction datasets to compare the performance of GNNs against traditional statistical models. Findings indicate that GNNs, due to their ability to process relational data structures, demonstrate superior adaptability and accuracy in scenarios with label deficiencies. This research provides effective strategies for enhancing anti-money laundering systems by employing GNNs to more effectively manage data challenges.