Simulación y validación del proceso de producción de ácido fórmico mediante una columna de destilación reactiva en DWSIM

Abstract

The objective of this study was to simulate and validate the improved process of formic acid production using a reactive distillation column, being the basis for the design of an artificial neural network (ANN) capable of predicting molar fluxes and molar fractions of the main compounds of the process. For the design of the ANN, a database was created from the simulation of the process in the free DWSIM program. This database is made up of 101 data with three entries: methanol molar flow, carbon monoxide molar flow, and column pressure C-1, with their four respective outputs: Molar flow and mole fraction of formic acid separated in the separation column C-2 as residue, the mole fraction of methyl format produced in the continuous stirring reactor (CSTR) and the molar flow of feed methanol in the CSTR reactor. The network was designed in MATLAB, using 20 hidden neurons and the mathematical algorithm "Bayesian Regularization", for the training and testing phase, external validation was also carried out with 16 data with their respective inputs and outputs, to ensure the reliability and accuracy of the network in the face of new predicted data. A mean square error of 0.001312 and total linear regression of R = 1 was obtained. The ANN was validated with the help of a comparative statistical analysis in which 95% reliability was obtained, the simulation managed to obtain 68.38 kmol / h of formic acid from 70.30 kmol / h of carbon monoxide. Furthermore, in said simulation and validation, a maximum percentage error of 1.81% was obtained in the variables of interest. It is recommended to expand the database with new experiments or input and output variables that were not considered.