Desarrollo de una RNA para el control de la destilación extractiva asistida por recompresión de vapor para la separación de N-Hexano y acetato de etilo

Abstract

The n-hexane and ethyl acetate are very required organic solvents in the chemical industry, however, the main problem is when they form azeotropes of minimum boiling, which cannot be separated using a simple distillation, an effective alternative is to use extractive distillation, however, this process consumes large amounts of energy, therefore, the main objective was to implement a steam recompression pump to the extractive distillation process to improve the energy efficiency of the process and allow the separation of these compounds into pure components, using a simulation platform to subsequently predict the mole fractions of N-Hexane and Ethyl Acetate by means of an Artificial Neural Network. As a first step of the methodology, the process was simulated in DWSIM, followed by the generation of a database arranged in four inputs; feed flow temperature, feed and make-up mass flows in the extractive distillation tower (EDC) and reflux ratio in the recovery tower (ERC) and four outputs corresponding to the distilled products: N-hexane (EDC distillate), NMP (EDC background), Ethyl acetate (ERC distillate) and NMP (ERC background) for RNA creation and training using MATLAB, the performances of the algorithms were compared and the most suitable one was chosen according to the statistical tests of quadratic error (MSE) and Linear Regression (R). The simulation results show an average percentage error of 3.18%, and for the Bayesian Regularization (BR) algorithm, results of MSE=2.5696E-07 and R=0.99397 were obtained. In such context it is concluded that the obtained error percentages do not exceed 10%, which indicates that the simulation is adequate for the process and the constructed model is able to predict the mole fractions under study with an average error of 0.93%.