Simulación y validación de un sistema de destilación para la separación de azeótropos de diisopropileter – isopropanol – agua en procesos mejorados para la industria química

Abstract

The objective of this study is to simulate and validate a distillation system with pressure variation to separate a mixture of Diisopropyl ether (DIPE) -Isopropanol (IPA) – Water in the chemical industry to run a database to design an artificial neural network (ANN) capable of predicting the main molar fractions in the distillation columns and the recirculation. The development of the ANN was carried out with a database obtained from the open-source chemical process simulator DWSIM. The database has 150 data with six entries, Inlet temperature, the Molar fraction of the mixture at the IPA inlet, Molar fraction of the mixture at the DIPE inlet, Pressure of condenser of distillation column C1, Pressure of condenser of distillation column C2, Pressure of condenser of distillation column C3. Also, five outputs will be generated, the bottom molar flow, bottom IPA mole fraction and DIPE mole fraction at the top of distillation column C1, the mole fraction of DIPE at the bottom and mole fraction of DIPE at the top of the C3 distillation column. The network was designed using the Python programming language in the Jupiter Notebook computing environment. 244 hidden neurons were used in its structure, the training was executed with the Adaptive Moment Estimation Optimization algorithm (Adam), and as activation function, we used TANH function and linear function. The Mean Square Error value in Training was 0.006891, in the Test the value was 0.006667 and in the Validation 0.008051. The validation of the ANN from three comparative analyses executed in the JASP software expresses the acceptance of the null hypothesis since the evaluated values do not differ significantly, which implies that the evaluated values are significantly similar.