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
- ItemDevelopment of an optical force sensor: a novel approach for monitoring physical interaction in robotic walkers(Universidade Federal do Espírito Santo, 2026-02-20) Garcia Alvarez, Daniel Eduardo; Frizera Neto, Anselmo; https://orcid.org/0000-0002-0687-3967; https://lattes.cnpq.br/8928890008799265; Múnera Ramirez, Marcela Cristina; https://orcid.org/0000-0001-6595-5383; https://lattes.cnpq.br/0934278112355648; Rodríguez Díaz, Camilo Arturo; https://orcid.org/0000-0001-9657-5076; https://lattes.cnpq.br/2410092083336272; https://orcid.org/0000-0002-5491-660X; https://lattes.cnpq.br/9190745277681587; Silveira, Mariana Lyra; https://orcid.org/0000-0002-0368-5629; https://lattes.cnpq.br/5307116832176112; Lima, Eduardo Rocon de; https://orcid.org/0000-0001-9618-2176This dissertation presents the design, development, and experimental validation of an optical sensor (OS) for monitoring interaction forces in smart walkers (SWs). The proposed sensing approach integrates light-sensitive photodiodes and addressable RGB light-emitting diodes embedded within a compliant encapsulation material, enabling force estimation by measuring changes in optical signals caused by surface deformations. Compared to conventional force-sensing technologies (i.e., strain gauges, piezoelectric sensors, and high resolution triaxial force cells) and optical Ąber-based alternatives (i.e., polymer optical Ąbers and Ąber Bragg gratings), the proposed OS reduces system complexity while ofering a cost-efective and easily manufacturable design, facilitating its integration into SWs. The Ąrst OS prototype validated the feasibility of the proposed approach, achieving an average force estimation error of 2.4%. Meanwhile, it identiĄed contact zones with an accuracy of 98%. These results demonstrate reliable performance in both force regression and contact localization, as well as the ability to capture the spatial distribution of applied forces. A second development stage focused on a redesigned OS geometry optimized for walker-handle integration, enabling force sensing across multiple interaction zones. An evaluation of the efects of encapsulation materials and illumination wavelengths on OS performance revealed that combination of EcoĆex encapsulation and red light provided the best results, achieving the lowest mean squared error (MSE) (Validation: 4.72 ± 0.31; Test: 4.96), mean absolute error (MAE) (Validation: 1.61 ± 0.04; Test: 1.79), and the highest coeicient of determination (R2) (Validation: 0.98 ± 0.01; Test: 0.97). The optimized conĄguration also demonstrated good generalization to unseen loads, with an average error of 5.56%. To assess repeatability, four new OS units implementing the optimized conĄguration were fabricated and independently calibrated. Among them, the fourth OS achieved the best results, with the lowest prediction errors (MSE validation: 3.11 ± 0.55; test: 3.33; MAE validation: 1.18 ± 0.10; test: 1.24) and the highest correlation values (R2 validation: 0.98 ± 0.02; test: 0.98). Validation against a commercial reference system conĄrmed estimation errors below 5.78% across all four OSs. Finally, the integration of the OSs into a SW and their evaluation during path-following trials with ten healthy participants demonstrated consistent force redistribution patterns across straight and turning maneuvers, highlighting the sensorŠs capability to capture meaningful interaction dynamics in real-world scenarios. Overall, this work demonstrates that waveguide-based OS combined with data-driven models constitutes a robust, scalable, low-complexity, and cost-efective solution for estimating interaction forces in SWs.
- ItemDetecção de transtorno mental via EEG, microestados e redes neurais de grafos(Universidade Federal do Espírito Santo, 2025-05-27) Candeia, Daniel Ribeiro; Ciarelli, Patrick Marques ; https://orcid.org/0000-0003-3177-4028; http://lattes.cnpq.br/1267950518719423; https://orcid.org/0009-0001-4427-7496; http://lattes.cnpq.br/2696632870728316; Côco, Klaus Fabian ; https://orcid.org/0000-0001-7793-0693; http://lattes.cnpq.br/1374499533178055; Tello, Richard Junior Manuel Godinez ; https://orcid.org/0000-0003-1428-0990; http://lattes.cnpq.br/3966230569744918Electroencephalogram (EEG) is a non-invasive and cost-effective technique widely used to study brain activity and diagnose neurological disorders. However, visual analysis of EEG signals is complex and requires expertise, highlighting the need for automated diagnostic support systems. In this context, this study proposes a graph-based neural network model for detecting mental disorders using EEG signals, leveraging microstate analysis. The proposed model integrates graph neural networks (GNNs) with microstate analysis, which captures transient and stable patterns of brain activity. The TUH Abnormal EEG Corpus (TUAB) dataset, containing normal and abnormal EEG signals, was used. The process included the extraction of microstates, the construction of graphs based on Spearman correlation between EEG channels, the extraction of features from EEG signals, and the application of Principal Component Analysis (PCA) to reduce the dimensionality of these features. Three GNNs were trained, each associated with signals from each microstate, and their outputs were combined using an ensemble technique. The final model achieved an accuracy of 97.46% on the test set, outperforming existing results of methods in the literature. The results highlight the effectiveness of the proposed approach, demonstrating the potential of GNNs and microstate analysis for detecting mental disorders from EEG signals
- ItemDevelopment of a mobile service robot system: enhancing localization, guidance, and search tasks in indoor environments(Universidade Federal do Espírito Santo, 2025-08-06) Rodriguez, Elio David Triana; Jiménez-Hernández, Mario Fernando ; https://orcid.org/0000-0003-0965-277X; http://lattes.cnpq.br/6078067029625341; Frizera Neto, Anselmo; https://orcid.org/0000-0002-0687-3967; http://lattes.cnpq.br/8928890008799265; https://orcid.org/0009-0000-4454-987X; ttp://lattes.cnpq.br/3696912600039873; Santos, Thiago Oliveira dos; https://orcid.org/0000-0001-7607-635X; http://lattes.cnpq.br/5117339495064254; Avilés Sanchez, Oscar Fernando ; https://orcid.org/0000-0001-8676-9926; http://lattes.cnpq.br/7481150993160615Advancements in service robotic systems demand robust Human-Robot Interaction (HRI) strategies capable of operating in multilingual and dynamic environments. However, current HRI approaches based on Natural Language Processing (NLP) often face limitations related to scalability, ambiguity in communication, and dif f iculty in linking unstructured input to structured data, thereby reducing robotic adaptability. This study proposes an HRI framework that integrates NLP through Large Language Models, combined with a decision-making algorithm grounded in Generative Artificial Intelligence (Generative AI) and context-aware reasoning. The system adopts a modular architecture comprising request validation, map valida tion, and response generation, enabling the synthesis and association of structured andunstructured data. As a result, the robot is capable of navigating, guiding users, executing adaptive tasks, and responding to user requests through a chat-style in terface. The framework was implemented on a mobile robot that was structurally, electrically, and software-wise adapted, culminating in the development of an au tonomous system able to complete localization, guidance and search tasks in in door environments. The NLP-based interaction modules were then integrated, and the resulting autonomous responses to user requests were evaluated as satisfactory. Usability evaluations conducted with real users, using the System Usability Scale (SUS), yielded consistently high scores ranging from “good” to “excellent.” How ever, participants reported a slightly lower perception of accuracy and increased frustration when operating in fully autonomous LLM mode compared to a pre programmed control mode. On the other hand, validation experiments demon strated a 91% success rate, confirming the system’s capability to process user re quests andexecutetaskstypicalofguiderobots. Theseresultsvalidatethefeasibility of integrating LLMs into multilingual robotic systems, highlighting both the poten tial and current limitations of NLP in HRI. They also highlight the transformative role of LLMs in enhancing natural language understanding and decision-making in real-world scenarios. Future work should focus on improving the handling of ambiguous user requests and refining feedback mechanisms to enhance the overall user experience.
- ItemMPPT for photovoltaic systems using Fractional Power Processing Architecture(Universidade Federal do Espírito Santo, 2025-08-05) Leão, André Salume Lima Ferreira; Santos, Walbermark Marques dos; https://orcid.org/0000-0002-9871-6028; http://lattes.cnpq.br/5558697161842579; https://orcid.org/0009-0008-1649-3088; http://lattes.cnpq.br/0522526374396770; Simonetti, Domingos Sávio Lyrio; https://orcid.org/0000-0001-5920-2932; http://lattes.cnpq.br/1107005171102255; Coelho, Roberto Francisco; https://orcid.org/0000-0002-4672-0885; http://lattes.cnpq.br/9967005468124403This study presents a novel investigation, with conceptual contributions, into how a fractional power processing (FRPP) architecture can be used to track the maximum power point of a photovoltaic (PV) system while using PV modules as the auxiliary source. Two configurations are proposed for fractional power processing within the photovoltaic system. The first, referred to as coupled FRPP (CFRPP), utilizes only photovoltaic modules as the auxiliary energy source, with direct connection to a converter. The second, referred to as decoupled FRPP (DFRPP), incorporates an energy storage system (ESS) alongside the PV modules, allowing the two PV arrays to function independently. Simulations were conducted to validate the theoretical analysis of a PV system under these two FRPP architecture configurations and to compare their performance to different power processing architectures. The results indicate that the proposed model equations are valid. It is concluded that fractional power processing enables converters designed for lower power levels to regulate the generation of higher. When applied to photovoltaic systems, both types of FRPP have demonstrated the ability to achieve maximum power while using low-power, low-complexity converters, since they are non-isolated and can even adopt simple topologies, such as buck or boost. However, each type exhibited significantly different capabilities. The DFRPP configuration demonstrates its viability by delivering superior performance compared to other power processing architectures. In contrast, the CFRPP architecture exhibits performance limitations, making it less advantageous under comparable conditions
- ItemGesture recognition for prosthesis control using electromyography and force myography based on optical fiber sensors(Universidade Federal do Espírito Santo, 2025-09-17) Ramirez Cortés, Felipe; Segatto, Marcelo Eduardo Vieira ; https://orcid.org/0000-0003-4083-992X; http://lattes.cnpq.br/2379169013108798; Díaz, Camilo Arturo Rodríguez ; https://orcid.org/0000-0001-9657-5076; http://lattes.cnpq.br/2410092083336272; https://orcid.org/0009-0008-7465-3887; http://lattes.cnpq.br/0873671321556842; Silveira, Mariana Lyra ; https://orcid.org/0000-0002-0368-5629; http://lattes.cnpq.br/5307116832176112; Silva, Jean Carlos Cardozo da ; https://orcid.org/0000-0003-2310-9159; http://lattes.cnpq.br/9949032159595994Amputation is the partial or total loss of a limb. It is a challenging event that affects people worldwide, with an estimated prevalence of 552.45 million in 2019 and a growing rate. The loss of an upper limb, in particular, strongly affects a person’s ability to perform activities of daily living (ADL), communicate, and interact with their environment. To restore lost functionality, assistive devices known as prostheses have been developed. Modern active prostheses can be controlled by interpreting the user’s movement intention through various biological signals, such as Surface Electromyography (sEMG), which measures the electrical activity of muscles. While sEMG is an established and predominant control method, it has limitations. Forcemyography (FMG) is a technique that measures changes in muscle volume and pressure during contraction. It has emerged as a promising alternative, offering advantages such as greater signal stability and reduced sensitivity to skin conditions like sweat. This master’s dissertation proposes and evaluates a hybrid sensor system combining FMG and sEMG to create a more robust and precise method for hand gesture classification. The system integrates a custom-developed FMG sensor, which uses a Fiber Bragg Grating (FBG) embedded within a flexible 3D-printed structure, with a commercial sEMG sensor. The primary goal is to improve the control of real and virtual prosthetic hands for amputees. The study involved recording signals from able-bodied subjects while they performed tasks involving different hand angles and grip forces. Data from the sEMG, FMG, and the combined hybrid system were used to train and test seven different machine learning algorithms, with the dataset split into 80% for training and 20% for testing. Results showed that the optimal sensing strategy is task-dependent. For angle classification, the hybrid FMG-sEMG sensor achieved the highest accuracy of 85.62% with the K-Nearest Neighbors (KNN) classifier. For force classification, the sEMG sensor alone was superior, reaching an accuracy of 92.53% with a Support Vector Machine (SVM). Furthermore, the hybrid system’s feasibility for real-time application was validated in a Virtual Reality (VR) environment, where it achieved 99.83% accuracy in classifying binary open/close hand gestures. This research demonstrates the complementary nature of FMG and sEMG signals, concluding that a multimodal approach can be used to develop more sophisticated, reliable, and intuitive control systems for upper-limb prostheses by selecting the best sensing modality for the desired task