Monitoramento de para-raios de ZnO com uso de redes neurais convolutivas e processamento de imagem

• Ano de defesa: 2023
• Autor(a) principal: Araújo, Bruno Vinícius Silveira
• Orientador(a): Ferreira, Tarso Vilela
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Programa de Pós-Graduação: Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: Correntes elétricas; Para-raios; Redes neurais (Computação); Sistemas imageadores; Detecção; Monitoramento; Termografia; Visão computacional; Detection; Monitoring; Surge arresters; Thermography; Computer vision
• Área do conhecimento CNPq: ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: https://ri.ufs.br/jspui/handle/riufs/18482

Monitoring and preventive maintenance of high-voltage electrical equipment are vital to prevent failures and ensure the smooth operation of the Power Electrical System. Infrared monitoring stands out for its convenience, safety, and ability to use temperature as a relevant indicator of the structural integrity and components health of the equipment. In this dissertation, a method is proposed for monitoring the operational condition of ZnO lightning arresters based on thermal measurements of the equipment. To achieve this, a convolutional neural network and computer vision processes were used to detect, segment, and extract the thermal profile of these devices. Additionally, an algorithm was employed to align the thermal profiles of the equipment, enabling comparison, identification of similarities, and classification of operational integrity. This allowed for a more accurate and comprehensive evaluation of ZnO lightning arresters, contributing to their continuous monitoring and efficient diagnosis. Thermal imaging of equipment from a 500 kV substation and thermal imaging from laboratory tests were used and analyzed. The laboratory tests included both healthy and intentionally defective equipment. The detection algorithm exhibited good precision rates of 0.861, a recall of 0.855, an mAP50 of 0.903, and an mAP50: 95 of 0.615, enabling an accurate detection and segmentation process. When applying the proposed method, the classification rates were consistent, correctly identifying both equipment in normal operating conditions and faulty equipment, despite variations due to thermal imager measurement errors, fluctuations in distance and angles, and environmental influences. Furthermore, when evaluating healthy equipment and those with applied defects, the proposed method achieved a 100% accuracy in identifying thermal anomalies, along with their localization.