Desenvolvimento de sensores virtuais para monitoramento de processos não lineares multivariáveis utilizando redes neurais.

Abstract

In general, industrial processes are complex, non-linear, with multiple inputs and multiple outputs. Representing this type of system by linear models (despite its simplicity and ease of implementation) is often inappropriate when a realistic and detailed approach to the process under study is required. Currently, virtual sensors have been used in industries to make the physical system successfully meet the performance specifications previously established, helping in the monitoring and optimization of processes in general. To simplify this complexity of identifying and modeling nonlinear systems, artificial neural networks can be used, which represent an important part of the knowledge in the area of ​​intelligent systems, with efficient results in the identification of complex and nonlinear systems. Focusing on these issues, this thesis proposes the development of virtual sensors for monitoring variables of complex non-linear processes with multiple inputs and multiple outputs, using neural networks in the estimation of the variables. For the validation of the experiments performed, virtual sensors are implemented following the proposed methodology for monitoring an experimental test platform (fluidic transport system). the variables of interest in platform monitoring are pressure and flow values. With monitoring using a virtual sensor, it is possible to obtain processes with better performance and with less difficulty in detecting and solving possible failures.