Identificação do comportamento de atuadores ativos com molas LMF utilizando redes neurais artificiais não-linear autorregressivas com entrada exógena (RNA NARX)

• Ano de defesa: 2019
• Autor(a) principal: Araújo, Caio Fernando Lira Correia
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Engenharia Mecânica; Programa de Pós-Graduação em Engenharia Mecânica; UFPB
• Palavras-chave em Português: Atuador LMF; Molas; RNA NARX; LMF Actuator; Spring; NN NARX; CNPQ::ENGENHARIAS::ENGENHARIA MECANICA
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/19003

The Shape Memory Alloy (SMA) is one of the New Materials (NM) that have been increasing their notoriety in the last decades. Due to their new characteristics and behavioral flexibility, that make possible change its internal structure under stress and temperature conditions. Many researches have been using this material type to develop active or passive actuators, however, active actuators have limitations in the application for the predictability of the behavior because of the alloy non-linearity. With the use of smart systems, it is possible to make feasible many applications of these actuators. Many other researches have been using SMA materials with control systems or monitoring wire elements. So, this work proposes to use the neural network nonlinear autoregressive exogenous input (NN type NARX) to predict the hysteretic behavior of SMA springs and actuators, composed of SMA spring associations, based on experimental data to train and validate of the NN. It was used stress and temperature data obtained under conditions of deformation of 150%, 200% and 300% of springs and 100% and 150% of actuators. After the deformation, it was applied an electric current to heat the spring or actuator. It was obtained satisfactory results, where the NN presented the capability to learn the behavior of the test data and obtained an absolute average percentage error around 0.4% in the NN validation.