Modelo de predicción de la concentración de acetona durante el proceso de destilación extractiva de una mezcla acetona-metanol

Abstract

The purpose of this research work was to simulate the extractive distillation process to separate the azeotropic acetone-methanol mixture using the open code software DWSIM and validate it with a bibliographic reference, with the purpose of being the basis for the creation of an Artificial Neural Network (ANN) that is capable of predicting fractions and molar fluxes of the process chemical compounds. Regarding the design of the ANN, a database was made from 100 simulations of the process where the manipulable variables were pressure, solvent / feed ratio and molar fraction of acetone in the feed; the output or studied variables are the acetone molar fraction and the molar flow of the distillate from the extraction column, the molar fractions of acetone, methanol, water and the molar flow of distillate from the recovery column. The ANN has a structure of 3-55-6 being inputs, neurons and outputs respectively; the mathematical algorithm "Bayesian Regularization" was used; in addition, an external validation was carried out with 17 data to ensure the reliability of the network with new data. A mean square error of 2.4804x10e07 and a linear regression in the test of R = 0.9994 were obtained. A comparative statistical analysis was performed, resulting in a reliability of 95%, without the presence of atypical values and with maximum values of 3.22% in mean percentage errors in the data predicted by the ANN. The amount of acetone extracted was 76,087 kmol / h while the methanol obtained was 24,913 kmol / h from 100 kmol / h of azeotropic mixture. It is recommended to carry out studies on new operating variables of the process to observe the influence they have on obtaining acetone and methanol of the azeotropic mixture.