Redes neurais artificiais aplicadas ao controle inteligente de colunas extrativas.

Abstract

In high purity distillation columns, the strict quality specifications of the final product require that the distillation columns control system have a high degree of performance. In cases where disturbances occur in the extractive columns feed, it is very difficult to maintain the composition of the product in its reference value, since, after the disturbance, the setpoints of the controllers fail to correspond exactly to the products specifications. The aim of this work is the development and implementation of an inteligent Soft Sensor for control purposes in an extractive column for anhydrous ethanol production, using ethylene glycol as solvent. To forecast the new setpoints before disturbance, the concept of Artificial Neural Networks was usesd, which proved to be a fast and feasible solution. The results showed that for range considered disturbances, the Soft Sensor was able to predict the new system condition, by intelligently determining the new setpoints of the controllers present in the original instrumentation of the column. The control showed satisfactory performance, keeping the products at the top and bottom of the column within specifications.