Modelado y simulación de la destilación del Terc-butanol para la predicción de la composición de los productos mediante Redes Neuronales Artificiales

Abstract

The objective of this work was to develop a model based on Artificial Neural Networks (ANN) for the prediction of the composition of the products of the distillation process of the azeotropic mixture terbutanol and water. The separation method was heterogeneous azeotropic distillation using Cyclohexane as the separating agent. To carry out the simulation, the DWSIM process simulator was used, developing a sensitivity analysis function encoded in Python. This function enables simulation results to be obtained by randomly varying the input parameters: feed molar flow, molar fraction of terbutanol in feed, feed temperature, and azeotropic agent molar flow. The results were tabulated and used as data for the Artificial Neuron Network training process, made up of 4 inputs and 10 outputs. The RNAs were developed in MATLAB with the Neural Network Fitting tool and tested with a second set of data generated with the simulator. The result was compared and validated using the one-way ANOVA test for independent data. The value of the linear correlation coefficient for the predictions was 0.9599 with 95% confidence. The results of the ANOVA test show that there is no significant difference between the results obtained from the simulator and those obtained from the RNA, for all the outputs studied. The feedfoward structure of RNA is made up of 25 nodes in the hidden layers and 10 in the output layer, using a backpropagation-type learning algorithm. The application of other RNA structures, as well as learning algorithms, is recommended to extend the study of the terbutanol-water separation process.