Controle de um reator de polimerização de propeno utilizando filtro de partículas e rede neural

• Ano de defesa: 2017
• Autor(a) principal: Dias, Ana Carolina Spindola Rangel
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Mestrado em Engenharia Química; UFES; Programa de Pós-Graduação em Engenharia Química
• Palavras-chave em Português: Modeling and simulation; State estimation; Particle filter; Neural network; Process control; Polymerization; Filtro de partículas; Modelagem e simulação; Estimação de estados; Rede neural; Controle de processos; Modelagem de dados; Métodos de simulação; Teoria da estimativa; Redes neurais (Computação); Controle de processo; Polimerização; Engenharia Química; 66.0
• Link de acesso: http://repositorio.ufes.br/handle/10/7816

Polymeric materials are present in several industrial sectors and in the society daily life, presenting advantages such as lower costs and higher durability. Polypropylene, obtained by the formation of propylene monomer long chains, is one of the most important olefins today, having a wide range of applications. Due to strong economic interest it arouses, there is a continuous search for improvements in its production process. Several methods for its obtaining it can be found by combining production technologies and catalysts. To ensure safety and achieve the operations objectives, it becomes necessary to insert structures for the process effective control. However, without a quality monitoring, this is not possible. In actual polymerization plants, the measuring devices are subject to uncertainties and are not always available; or the equipment does not exist or its purchase/maintenance cost makes its use unfeasible. Thus, this work proposes a virtual sensor scheme based on particle filter (PF) and artificial neural network (ANN), which is applied to a simulated propylene polymerization reactor. This structure allows the uncertainty reduction and the latent variables observation by means of PF. In turn, the ANN detects the polypropylene final properties from the improved data. The concern was to provide controllers with more complete and improved information. The results showed that the virtual sensor allowed improvements in process control, providing accurate estimates and consistent action time with industrial sampling intervals, which highlights its potential for practical application.