Sistema de identificação de fonte de gás utilizando nanoquadricóptero

• Ano de defesa: 2022
• Autor(a) principal: Souza, Rafael Monteiro Jorge Alves de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Engenharia Elétrica
• Palavras-chave em Português: Localização de fonte de odor; Odor source localization; mapeamento de distribuição de gás; Gas distribution mapping; quadricóptero; quadcopter; redes neurais artificiais; artificial neural network; aprendizagem de máquina; machine learning; estimação de densidade de probabilidade; probability density estimation; Crazyflie 2.0; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA::ELETRONICA INDUSTRIAL, SISTEMAS E CONTROLES ELETRONICOS; Engenharia Elétrica; Ar - Controle de qualidade; Predator (Drone)
• Link de acesso: https://repositorio.ufu.br/handle/123456789/36129; http://doi.org/10.14393/ufu.te.2022.496

Solutions to the odor source localization problem are necessary to allow accomplishment of several tasks such as environmental gas leakage searching, air quality monitoring and indoor detection of dangerous gas, thus allowing to increase safety and prevent accidents in environments where hazardous gases are handle. In this sense, the use of aircraft in this task brings agility, safety and efficiency, contributing to the development of Industry 4.0. The solution presented in this work makes use of an odor source localization strategy by gas distribution mapping approach in a three-dimensional environment using a Crazyflie 2.0 nano quadcopter. The mapping is constructed from a probability density function by Gaussian kernel, which acts over characteristics extracted from the air quality signal of an on-board SGP30 sensor and produces a continuous map, without the need for segmentation of the workspace. The source position declaration subtask is evaluated through different strategies involving artificial neural networks and machine learning techniques. The results show that the proposed algorithm is capable of making accurate estimates of the odor source location with estimation errors means close to 0.46 m in an environment of approximately 15 m3. The algorithm has the possibility of scalability by incorporation other aircraft and thus explore indoor environments with higher volumes.