@PhDThesis{Carrara:1997:ReNeAp,
               author = "Carrara, Valdemir",
                title = "Redes neurais aplicadas ao controle de atitude de sat{\'e}lites 
                         com geometria vari{\'a}vel",
               school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
                 year = "1997",
              address = "Sao Jose dos Campos",
                month = "1997-06-17",
             keywords = "engenharia e tecnologia espacial, controle de atitude, redes 
                         neurais, controle de sistemas, satelites artificiais, space 
                         technology and engineering, attitude control, neural networks, 
                         control systems, artificial satellites.",
             abstract = "Este trabalho investiga o uso de redes neurais em controle de 
                         atitude de sat{\'e}lites artificiais. Com a inten{\c{c}}{\~a}o 
                         de validar esta aplica{\c{c}}{\~a}o, optou-se por um 
                         sat{\'e}lite que n{\~a}o fosse constitu{\'{\i}}do por um corpo 
                         r{\'{\i}}gido, mas que possu{\'{\i}}sse um comportamento 
                         dinamico vari{\'a}vel em fun{\c{c}}{\~a}o de ap{\^e}ndices 
                         articulados. A din{\^a}mica assim gerada tem o car{\'a}ter 
                         n{\~a}o-linear t{\'{\i}}pico para utiliza{\c{c}}{\~a}o de 
                         redes neurais. S{\~a}o apresentados neste trabalho as principais 
                         rela{\c{c}}{\~o}es que permitem a modelagem de um sistema via 
                         rede neural, bem como duas possibilidades de treinamento: 
                         retro-propaga{\c{c}}{\~a}o e m{\'{\i}}nimos quadrados. Foram 
                         obtidas tamb{\'e}m as rela{\c{c}}{\~o}es din{\^a}micas e 
                         cinem{\'a}ticas do movimento de um corpo no espa{\c{c}}o com 
                         ap{\^e}ndices articulados, levando-se em conta a 
                         posi{\c{c}}{\~a}o do centro de massa e a varia{\c{c}}{\~a}o do 
                         momento de in{\'e}rcia do conjunto, necess{\'a}rios para efetuar 
                         a simula{\c{c}}{\~a}o do movimento do sat{\'e}lite. Para 
                         validar o controlador de rede neural, foi utilizado como exemplo a 
                         geometria do sat{\'e}lite de sensoriamento remoto da MECB, 
                         durante a fase de abertura dos pain{\'e}is solares, que fazem as 
                         vezes dos ap{\^e}ndices articulados. Inicialmente obteve-se o 
                         modelo de identifica{\c{c}}{\~a}o, contendo a din{\^a}mica 
                         direta do sat{\'e}lite. Posteriormente testaram-se varias formas 
                         de obten{\c{c}}{\~a}o do modelo din{\^a}mico inverso 
                         atrav{\'e}s da rede neural, sendo que o treinamento com 
                         realimenta{\c{c}}{\~a}o do erro mostrou os melhores resultados. 
                         Para validar o controle, promoveu-se uma varia{\c{c}}{\~a}o de 
                         par{\^a}metros do sat{\'e}lite (momentos de in{\'e}rcia, massa, 
                         empuxo dos motores) e inclu{\'{\i}}ram-se de ru{\'{\i}}dos nos 
                         sensores, sem entretanto refazer o treinamento da rede. 
                         Comprovou-se, assim, que a rede possui capacidade de 
                         compensa{\c{c}}{\~a}o, capaz de assegurar robustez ao controle 
                         proposto. ABSTRACT: The use of neural networks for satellite 
                         attitude control is addressed in this work. In order to validate 
                         this application, a spacecraft with a variable dynamic behavior 
                         due to articulated appendages fixed to the body was chosen. The 
                         differential equations therefore show the nonlinear dynamic 
                         effects to be identified by neural nets. In this work some o f the 
                         main expressions that allow system modeling through neural nets as 
                         well as two different training procedures- back-propagation and 
                         least squares- are presented. A general method for obtaining the 
                         inertia tensor and center of mass of an articulated space device, 
                         is also explained, as well as the dynamic and cinematic 
                         differential equations. These formulations were used in attitude 
                         simulation for neural network system identification and control 
                         training. The solar array deployment o f the MECB' s remote 
                         sensing satellite was used as an example of attitude control by 
                         means of neural nets. Three solar arrays are articulated in the 
                         satellite body and are deployed after orbit injection by a trigger 
                         device. Initially, a direct model ofthe satellite by means of a 
                         neural net was obtained. Afterwards, severa! arrangements and 
                         training procedures were tested in order to achieve the inverse 
                         model of the dynamics. The best results were obtained with the 
                         inverse training through feedback error. In order to validate the 
                         control procedure, a parameter variation method (inertia tensor 
                         and mass) together with sensor noise were employed after 
                         accomplishing the training phase, so as to verify control 
                         robustness against to parameter variation. The results show that 
                         the neural net is tolerant to sensor noise and has a relatively 
                         large capacity to compensate the parameter uncertainty.",
            committee = "Martins Neto, Antonio Felix (presidente) and Rios Neto, Atair 
                         (orientador) and Oliveira e Souza, Marcelo Lopes de and Lopes, 
                         Roberto Vieira da Fonseca and Nascimento Junior, Cairo Lucio and 
                         G{\'o}es, Luiz Carlos Sandoval",
           copyholder = "SID/SCD",
         englishtitle = "Neural networks based control of a satellite attitude with varying 
                         dynamics",
                label = "7832",
             language = "pt",
                pages = "176",
                  ibi = "6qtX3pFwXQZ3r59YCT/GUnVF",
                  url = "http://urlib.net/ibi/6qtX3pFwXQZ3r59YCT/GUnVF",
           targetfile = "publicacao.pdf",
        urlaccessdate = "11 maio 2024"
}