-
Tamires M. Rezende
Fundação de Inovações Tecnológicas (FITec), Av Cristóvão Colombo, 485, 8º Andar – Belo Horizonte, MG
-
Bruno A. S. Oliveira
Fundação de Inovações Tecnológicas (FITec), Av Cristóvão Colombo, 485, 8º Andar – Belo Horizonte, MG
-
Glauco M. V. de Paula
Fundação de Inovações Tecnológicas (FITec), Av Cristóvão Colombo, 485, 8º Andar – Belo Horizonte, MG
-
Gustavo P. de Souza
Fundação de Inovações Tecnológicas (FITec), Av Cristóvão Colombo, 485, 8º Andar – Belo Horizonte, MG
-
Daniel Calvo
Fundação de Inovações Tecnológicas (FITec), Av Cristóvão Colombo, 485, 8º Andar – Belo Horizonte, MG
-
Eugênio L. Daher
Fundação de Inovações Tecnológicas (FITec), Av Cristóvão Colombo, 485, 8º Andar – Belo Horizonte, MG
-
Adriano O. da Silva
Huawei do Brasil, R. Arquiteto Olavo Redig de Campos, 105 - Chácara Santo Antônio (Zona Sul), São Paulo - SP, 04709-000
Keywords:
Computer Vision, Switches Disconnectors, Power Substation, YOLO algorithm, Monitoring, Object Detection
Abstract
Over the years, electric power consumption has been increasing significantly, making it necessary to adopt measures to monitor the process up to the distribution, so that its supply is not interrupted. In that process, among the various devices that are part of the system, the proper monitoring of the operational state of the substation disconnect switches plays an important role. They are responsible for connecting the high voltage to the medium voltage, used to reconfigure the network and isolate the equipment for maintenance. In this scenario, this work developed a methodology for detecting 69kV disconnect switches and determining their operational state: opened or closed. The system applies computer vision techniques, aiming to generate a generalized model for real-time inference of the switches during the regular operation. The methodology incorporates image acquisition, data preparation, training with the tiny- YOLO algorithm, and testing of the model obtained. As a result, in face of the approached scenario, it was verified that the proposed solution can successfully detect correctly the state of the phases, achieving a mAP equals to 97.50%.
