Department of Electrical and Computer Engineering Graduate Theses

Permanent URI for this collection


Recent Submissions

Now showing 1 - 5 of 473
  • Item
    Modular Multilevel Cascaded Converter for Distributed Photovoltaic Energy Conversion Systems
    (2024-07-08) Gao, Weifeng; Electrical and Computer Engineering; Jain​, Praveen; Pan, Shangzhi
    With the global demand for renewable energy constantly growing, the main goal of photovoltaic (PV) systems is to increase the power output of the grid and, at the same time extract maximum power from different solar panels. A modular multilevel inverter is a competitive design that has the advantages of integrating energy generated from distributed microinverters and reducing costs below those of a central inverter used in large-scale PV plants. This thesis presents a two-stage inverter module connected to each solar panel. A small value of DC-link capacitors instead of a large capacitor in a traditional modular multilevel converter is implemented. However, a small DC-link capacitor would introduce issues larger double-line frequency capacitor voltage ripple, which lead to fluctuations in the DC–DC converter input voltage, and consequently deteriorate MPPT (maximum power point tracking) tracking. A three phase coupling transformer structure is proposed to solve the problem of double-line frequency capacitor voltage ripple. The method called modified multilevel PSPWM (Phase-Shift Pulse Width Modulation) scheme is realized without adding an extra communication unit. Since the modular multilevel inverter generates a higher-quality sinusoidal current with a higher module number, the size of the grid-tied filter was also reduced. PV integration into the grid also challenges the power quality and emerging reliability of the whole PV system. Moreover, grid-connected PV inverters should operate not only in normal conditions but also in fault conditions, according to modern grid codes. Thus, the PV inverter is supposed to achieve an accurate and fast response to injecting synchronized grid current of high-power quality and also provide voltage or frequency support immediately when a grid fault occurs. Finally, yet importantly, the thesis addresses the difficulty in controlling the second harmonic ripple in the DC bus voltage due to unbalanced network conditions. It proposes a distributed alpha-beta frame proportional resonant voltage loop controller as an effective solution for these unbalanced scenarios. This research contributes to the improvement of power quality under such conditions, promoting efficient and reliable operation of the system despite these challenges.
  • Item
    LLC Resonant Converters for Energy Efficient Power Electronics Systems
    (2024-07-03) He, Binghui; Electrical and Computer Engineering; Yan-Fei, Liu
    With modern power electronic systems advancing towards higher efficiency, smaller size, lighter weight, and lower noise, Switch-Mode Power Supplies (SMPSs) based on resonant converters have garnered widespread attention. Among various resonant converters, LLC resonant converters are extensively employed in today's most popular industry applications such as laptop/cellphone power adapters, data center power supplies, and EV (Electrical Vehicle) charging applications. This thesis addresses several challenges in LLC resonant converters, including wide voltage variation ranges, bidirectional operation, and Electro-Magnetic Interference (EMI) noise reduction. New technologies have been proposed to address these issues. The LLC resonant converter faces limitations in voltage gain due to its constrained switching frequency, posing challenges in maintaining high efficiency across a wide voltage regulation range. To address this issue, a Power Cycle Modulation (PCM) control scheme is proposed for high-efficiency operation across wide input and output voltage ranges and load variations. Implemented with low-cost Microcontroller Units (MCUs), the PCM control method is also applicable to other resonant topologies and suitable for high-frequency operation. While the forward operation of LLC resonant converters is well-studied, backward operation, crucial for bidirectional applications, lacks sufficient research. A double-loop control strategy is proposed to enhance the isolated LLC bidirectional converter's performance during backward operation. This strategy reduces double line-frequency ripple current and improves overall efficiency without additional costs, significantly lowering the Root Mean Square (RMS) current within the LLC converter system. Reducing EMI emissions from power converters is crucial to enhance power density and minimize filter size. This thesis proposes effective noise reduction techniques for LLC converters, including Common-Mode (CM) noise cancellation capacitors for Half-Bridge (HB) LLC converters, two resonant branches with a sandwich transformer winding structure for Full-Bridge (FB) LLC converters, and the Split Primary Winding Transformer (SPWT) method with complementary couple-turns. These solutions improve CM noise performance, allowing for smaller CM EMI filters and enhancing system power density. This thesis employs mathematical modeling, PSIM-based simulation, Finite Element Analysis (FEA), and experimental tests to verify the proposed theories and solutions.
  • Item
    Integration of Topological Maps with GNSS and Onboard Sensors for Robust Land Vehicle Navigation
    (2024-07-03) Ragab, Hany; Electrical and Computer Engineering; Noureldin, Aboelmagd; Givigi, Sidney
    Accurate automotive navigation systems are foundational to the efficacy and safety of Advanced Driver Assistance Systems (ADAS) and Autonomous Driving (AD) technologies, necessitating continuous advancements to ensure reliability in the face of Global Navigation Satellite System (GNSS) outages and dynamic road conditions. Innovative techniques and algorithms have been developed and critically evaluated to enhance various facets of navigation performance, marking significant strides in vehicular navigation research. The exploration begins with advancements in odometry-based navigation independent of map data. Integration of Structure-from-Motion (SfM) with the reduced inertial sensor system (RISS) and GNSS substantially improves performance, particularly in managing complex road maneuvers. To better handle urban environments, a novel Semantic Segmentation-based Outlier Rejection (SS-OR) technique was developed to enhance the accuracy of visual odometry systems, presenting notable implications for autonomous navigation and mapping. Further investigation into automotive Wheel Speed Sensors (WSS) leads to the creation of a fusion engine that amalgamates WSS data with stereo visual odometry. This approach effectively minimizes forward velocity errors and biases, contributing significantly to the overall enhancement of navigation system capabilities. The discourse extends to map-aided navigation, introducing a Two-Stage Kinematic Update technique for topological map-matching (TMM) algorithms reliant on conventional GNSS/RISS integration. This innovation demonstrates considerable improvements in navigation accuracy during GNSS outages. Nonetheless, the persistence of challenges related to cumulative errors and drift underscores the necessity for expanded integration with additional perception systems. Concluding the research, the integration of topological maps with the proposed forward velocity fusion engine and the dynamic window approach (DWA) culminates in a comprehensive navigation solution. Although the outcomes are promising, opportunities for further enhancements remain, particularly in refining map details and addressing deviations from planned routes, paving the way for future explorations in the domain. The efficacy of the proposed methods is rigorously evaluated through a series of real-road experiments conducted in the Cities of Kingston and Toronto, designed to assess their viability and advantages comprehensively. The outcomes of these tests reveal a marked improvement in performance, showcasing the proposed methods' superiority in comparison to traditional navigation techniques.
  • Item
    Estimating Human Pose from Pressure and Vision Data
    (2024-06-17) Davoodnia, Vandad; Electrical and Computer Engineering; Etemad, Ali
    Estimating human pose has numerous applications, ranging from healthcare and sports analysis to virtual reality and human-computer interaction. In this dissertation, we study two input domains for pose estimation, vlision and pressure. While vision-based pose estimation is often more robust than pressure-based estimation due to its higher resolution and less noisy signals, it suffers from privacy concerns. Hence, each modality is often considered for different applications; vision-based systems are used for animation, entertainment, and sports, among others, while pressure-based systems are often utilized in smart clinics, homes, and vehicles. In this dissertation, we study and propose solutions to address unique challenges in each domain. First, we address the domain gap between image-based pose estimators and pressure data by proposing a learnable pre-processing module called PolishNetU. Our experiments on two publicly available datasets show that combining PolishNetU and re-training pre-existing image-based pose estimators overcomes the issue of highly vague pressure points. Next, we tackle the challenge of body part occlusions from pressure maps when a limb is not in direct contact with the pressure sensors. To this end, we propose T-ViTPose, a temporal pose estimator based on vision transformers, to capture subtle movements on pressure sensors. Furthermore, we show that self-supervised pre-training using a masked auto-encoder approach improves results. In the latter part of this dissertation, we focus on enhancing the robustness, generalization, and scalability of multi-view systems. We introduce UPose3D, a 3D keypoint estimation pipeline scalable to any number of cameras. We propose a training routine based on synthetic data generation to ensure generalization across different poses and viewpoints. By leveraging uncertainty and our novel cross-view fusion strategy, we improve our model's outlier and noise robustness and achieve state-of-the-art performance in out-of-distribution experiments. Next, we present SkelFormer, an inverse-kinematic model that obtains rotational pose and shape parameters of a body model given 3D keypoints. By training this module to reconstruct correct poses from corrupted 3D keypoints, we improve our pipeline's out-of-distribution generalization, as well as robustness to noise and occlusions. Our research significantly enhances the performance of existing systems and paves the way for future advancements in the field.
  • Item
    Affective Computing Through Time-series Representation Learning
    (2024-06-17) Shome, Debaditya; Electrical and Computer Engineering; Etemad, Ali
    Affective computing is a vital area of research with potential applications across health and well-being, education, workplace, and user experience by allowing systems to recognize, incorporate, and respond to human emotions. Despite recent advances, several challenges remain in accurately capturing and processing affective states from various forms of time-series. In this thesis, we address three major problems in this area. First, in Speech Emotion Recognition (SER), accurately distinguishing the linguistic and prosodic components of emotions is challenging. Existing methods often require explicit text transcription, leading to computational overhead and potential errors. To address this, we introduce EmoDistill, a cross-modal knowledge distillation framework that captures both linguistic and prosodic aspects of emotions from speech without needing text transcription, thereby enhancing SER accuracy and efficiency. Second, continuous monitoring of wearable physiological time-series like Photoplethysmogram (PPG) lacks the detailed information provided by Electrocardiogram (ECG), limiting the predictive power in affective state detection. To overcome this, we present the Region Disentangled Diffusion Model (RDDM), a novel diffusion model for high-fidelity PPG-to-ECG translation. RDDM generates detailed ECG signals efficiently, providing more precise physiological data for enhanced affect recognition. Lastly, current affect recognition systems often overlook the influence of external factors, such as sleep-related measures, on mood, reducing their accuracy and robustness. To tackle this, we propose NapTune, an efficient tuning framework that integrates the previous night's sleep-related measures with wearable time-series data. NapTune significantly improves mood classification accuracy by incorporating this additional modality.