Controle preditivo multivariável aplicado a processos químicos não lineares

Abstract

In this work, a predictive controller based on non-linear models (NMPC) was developed and tested in a computational environment using MathWorks’® MATLAB-Simulink®. The controller was designed to operate both controlling and manipulating multiple variables in non linear systems, and it can also apply specific weights to define the importance of each variable as well as being programmed to consider both rigid and soft operational constraints. The controller was tested on the simulated control of a jacketed CSTR in which ethylene glycol synthesis was being carried out from the oxidation of the propylene oxide. In this simulated model, several operating conditions and disturbances were applied to the NMPC performance testing system, with it being compared to a conventional PID controller. Those tests included changes in reagent flow rates, changes in reactor inlet temperatures (including situations that led the system to regions of instability), setpoint changes, among others. The NMPC proved to be efficient in these tests, outperforming or being equivalent to the PID according to the evaluation criteria used (ISE, IAE and ITSE), standing out in situations that moved further away from the linear regions of the system, where NMPC performance was considerably better. The controller has also proved to be flexible, with simple alterations in its programming allowing it to react to pre-defined changes in the system, as well as adapt to different operating conditions, like forcing the controller to comply with established safety limits.