Mestrado em Engenharia Elétrica
URI Permanente para esta coleção
Nível: Mestrado Acadêmico
Ano de início:
Conceito atual na CAPES:
Ato normativo:
Periodicidade de seleção:
Área(s) de concentração:
Url do curso:
Navegar
Submissões Recentes
- ItemEstudo de estágio CC-CC bidirecional com processamento parcial de energia de um SST aplicado em carregamento Extrem Fast Charging de veículos elétricos(Universidade Federal do Espírito Santo, 2025-06-06) Raposo, Duarte Manuel do Sacramento; Co-orientador1; Orientador1; https://orcid.org/; http://lattes.cnpq.br/; https://orcid.org/; http://lattes.cnpq.br/; 1º membro da banca; 2º membro da banca; 3º membro da bancaThis dissertation addressed the development of a solution for ultra-fast charging of electric vehicles (XFC), through a structure based on Partial Power Processing (PPP), which integrates two converters: the bidirectional resonant CLLLC and the three-phase interleaver. Model-Based Predictive Control (MPC), combined with the application of the Back-EMF theory, was implemented as a central control strategy in both topologies. The proposed system demonstrated high efficiency, good dynamic performance. The modular and bidirectional architecture, operating at high frequency, proved to be adequate to the demands of the XFC, contributing to the reduction of losses and improvement of the reliability of the system. The proposal is thus configured as a promising and scalable alternative for future electric vehicle charging stations with high power and performance requirements
- ItemGeneralized tonic-clonic seizures detection using deep learning techniques(Universidade Federal do Espírito Santo, 2025-03-31) Mesa, Juan Sebastian Campos; Ciarelli, Patrick Marques; Salles, Evandro Ottoni Teatini; https://orcid.org/0000-0002-8287-3045; http://lattes.cnpq.br/5893731382102675; https://orcid.org/0009-0005-2242-3236; http://lattes.cnpq.br/8653324785414671; Côco, Klaus Fabian; Cavalieri, Daniel CruzGeneralized Tonic-Clonic Seizure (GTCS) poses serious health risks, including an increased likelihood of sudden unexpected death in epilepsy (SUDEP), postictal pulmonary edema (PPE), and traumatic injuries from falls or jerking movements. However, GTCS detection remains challenging due to the complex and variable nature of EEG signals. Traditional methods struggle with these variations, while deep learning remains a gold standard for seizure detection, making it a powerful tool for GTCS detection. This study explores the detection of GTCS using EEG data from the Temple University Seizure (TUSZ) dataset. To achieve this, three deep learning architectures were employed: Diffusion Convolutional Recurrent Neural Network (DCRNN), Long Short-Term Memory (LSTM), and Convolutional Densely Connected Gated Recurrent Neural Network (C- DRNN). The research evaluated the impact of loss functions and data augmentation on model performance. Dice Entropy loss (DE) proved to be the most effective for DCRNN, while Cross-Entropy loss (CE) optimally enhanced LSTM and C-DRNN. Data augmentation played a crucial role in improving generalization and robustness across all models, improving their performance. As a consequence, DCRNN achieved the best performance, with AUC-ROC of 0.781, F1-score of 67.74%, and Sensitivity of 82.84%. The results obtained were close to or better than those found in the literature. This study highlights the potential for further research into alternative data augmentation techniques, deep learning architectures, and advanced methods to detect GTCS in EEG signals.
- ItemReidentificação baseada em filtro de correlação discriminativo para rastreamento de múltiplos objetos em câmeras de videomonitoramento(Universidade Federal do Espírito Santo, 2025-04-01) Abling, Augusto; Vassallo, Raquel Frizera ; https://orcid.org/0000-0002-4762-3219; http://lattes.cnpq.br/9572903915280374; https://orcid.org/0009-0002-7245-5760; http://lattes.cnpq.br/6477900225667920; Nascimento, Thais Pedruzzi do; https://orcid.org/0000-0002-3962-8941; http://lattes.cnpq.br/8698168347146036; Silva, Bruno Légora Souza da; https://orcid.org/0000-0003-1732-977X; http://lattes.cnpq.br/8885770833300316; Almonfrey, Douglas; https://orcid.org/0000-0002-0547-3494; http://lattes.cnpq.br/1291322166628469; Pereira, Flávio Garcia; https://orcid.org/0000-0002-5557-0241; http://lattes.cnpq.br/3794041743196202This study aims to develop, test, and analyze the use of discriminative correlation filter as a module for object re-identification, integrated with multiple object tracking for use in surveillance cameras with a focus on real-time processing. The study is set in the context of smart cities and Intelligent Transportation Systems (ITS), where object re identification and tracking are fundamental for the creation of advanced technologies. The adopted methodology includes the implementation of a modified discriminative correlation filter for the re-identification task, followed by tests to evaluate the algorithm’s performance in challenging scenarios present in widely recognized datasets in computer vision challenges. The results showed that the proposed correlation filter approaches the accuracy of neural network-based approaches without the need for prior training for specific contexts. Therefore, we may conclude that the integration of this re-identification module with multi-object tracking offers a balanced solution to improve tracking accuracy at a lower computational cost compared to neural networks, contributing to the advancement of technologies in smart cities and ITS
- ItemEstudo comparativo de detecção e rastreamento de elementos no trânsito utilizando imagens ominidirecionais(Universidade Federal do Espírito Santo, 2024-11-22) Scarparo, Heisthen Mazzei; Vassallo, Raquel Frizera ; https://orcid.org/0000-0002-4762-3219; http://lattes.cnpq.br/9572903915280374; https://orcid.org/0009-0009-8243-3320; http://lattes.cnpq.br/4545146039699801; Almonfrey, Douglas ; https://orcid.org/0000-0002-0547-3494; http://lattes.cnpq.br/1291322166628469; Cavalieri, Daniel Cruz ; https://orcid.org/0000-0002-4916-1863; http://lattes.cnpq.br/9583314331960942Traffic detection and tracking play an important role in the context of smart cities. These technologies have the potential to alleviate congestion, optimize the use of resources, and improve the quality of life of the population. However, one aspect of this field that has not yet been explored is the use of omnidirectional videos, which provide a 360° field of view. Omnidirectional images offer a large field of view of the road environment, allowing for a more complete analysis of traffic and moving objects. This panoramic view makes it possible to detect vehicles, pedestrians, cyclists, and other elements in all directions, including angles that are difficult to capture with conventional cameras. Using this type of imagery for traffic light control makes it easier to obtain information on the trajectory of vehicles in real time and, therefore, configure traffic lights in a more intelligent and efficient way. In addition, omnidirectional images can be used to monitor areas of high traffic density, identify congestion points, and analyze road user behavior patterns. This information is valuable for urban planning, the development of mobility policies, and the implementation of strategies aimed at improving traffic flow and street safety. Although the use of 360° panoramic images in the context of traffic detection and tracking is still an underexplored f ield, it represents a good tool for the implementation of smart cities through its integration with traffic light control and traffic management systems in cities. In this context, this work presents a database containing 25 panoramic videos, with their respective annotations. This database is available for use by the academic community. It also presents a comparative study between the application of the YOLOv5, YOLOv7, and YOLO-NAS networks, together with the use of the DEEPSORT algorithm, for detection and tracking of traffic objects present in the database. To compare the networks, the metrics of Precision, Recall, F1-Score, mAP@.5, and mAP@.5:.95 were used. In this study, the best result was obtained using the YOLOv7 network with training. Such result shows the feasibility of considering the use of omnidirectional images as a tool in the task of traffic monitoring and helping provide urban mobility
- ItemIdentificação de falhas em motores de indução trifásicos usando rede neural transformer convolucional(Universidade Federal do Espírito Santo, 2025-03-24) Moraes, Vinicius Andrade Nunes de; Silva, Jair Adriano Lima ; http://lattes.cnpq.br/3099010533644898; Rocha, Helder Roberto de Oliveira; https://orcid.org/0000-0001-6215-664X; http://lattes.cnpq.br/8801325729735529; https://orcid.org/0009-0005-3889-354X; http://lattes.cnpq.br/1557334823568448; Encarnação, Lucas Frizera ; https://orcid.org/0000-0002-6162-7697; http://lattes.cnpq.br/5578918284508758; Augusto, Andre Abel; https://orcid.org/0000-0001-7171-3372; http://lattes.cnpq.br/2166068779417190This work presents a hybrid approach using Convolutional Neural Networks (CNN) and Transformers for fault diagnosis in three-phase induction motors, focusing on the detection and classification of the severity of broken bar faults based on current and voltage signals. Electrical Signature Analysis (ESA), widely used in motor monitoring, offers several advantages. However, ESA-based techniques traditionally rely on spectral transformations, which can result in high computational cost and reduced generalization capability. CNNs can extract discriminative features directly from raw data, eliminating the need for preprocessing steps. The proposed study integrates CNNs with the attention mechanism of Transformers, which captures spatiotemporal dependencies in the data. The Convolutional Transformer Neural Network (CTNN) achieved approximately 97% accuracy when using the entire dataset, significantly outperforming classical machine learning algorithms such as Random Forest and k-Nearest Neighbors (KNN), which obtained 90% and 86% accuracy, respectively. The CNN, tested under similar conditions, achieved 96% accuracy. Compared to other methodologies involving multiple preprocessing steps and transformations to the frequency domain, the proposed approach achieves similar results, close to 100% accuracy, while being simpler, more efficient, and with greater generalization capability. Additionally, the methodology employs a reduced sampling rate, approximately six times lower than the original sampling rate of the dataset, contributing to computational cost reduction without compromising performance