ANN applied to the prediction of acetonitrile-water azeotrope separation

Abstract

The purpose of this research is to simulate and validate a process of separation of the acetonitrile-water mixture by extractive distillation in DWSIM, which serves as a basis for the design of the artificial neural network (ANN) capable of predicting the mole fractions of acetonitrile, ethylene glycol and water. For the elaboration of the neural network, a database was generated from the simulation performed in DWSIM, and the database was composed of 100 pairs, with 4 inputs, feed flow temperature, acetonitrile mole fraction at feed, pre-concentration column pressure (C1) and ethylene glycol mole fraction. The artificial neural network was designed in MATLAB software using 9 hidden neurons and a Bayesian Regularization algorithm for training, with an MSE value of 3.0723e-05 and a total regression coefficient of 0.99996. The network was validated by developing a comparative statistical analysis, obtaining 95% reliability. The simulation allowed obtaining 0.9830 of acetonitrile in the distillate of column 2 (extraction column), in column 3 (recovery) 0.9016 of water in the distillate and 0.9737 of ethylene glycol in the bottom. In addition, it should be mentioned that there was no recirculation of the products from the bottom of column 3 to column 2. This proposal allowed obtaining results close to the reference research. It should be mentioned that the percentage of errors is less than 10%, giving efficient results in the investigation. It is recommended to normalize the input values to the ANN that are greater than 1.