Prediction of Non-Catalytic Esterification of Free Fatty Acids Comprising Karanja Oil Using Ann

Abstract

The present study aims to design an Artificial Neural Network (ANN) to estimate the mole fractions resulting from the non-catalytic esterification process of Free Fatty Acids (FFA) that compose Karanja oil. The ANN was designed in the MATLAB program based on 100 pairs of data samples generated by a simulation validated in the free software DWSIM. Through a sensitivity analysis, it was determined that the inputs of the ANN were: the water mole fraction of stream 1C (1C-Xa), the oleic acid mole fraction of stream 3C (3C-Xo), the percentage conversion of the chemical reaction (%C), and the pressure drop in the reactor (-p). The methyl oleate mole fraction (9C-XM-O), methanol mole fraction (9C-Xm), triolein mole fraction (9C-XOOO), and trilinolein mole fraction (9C-XLLLL) from the liquid stream are established as outputs of the ANN. The mole fraction of methanol (10C- Xm) and the mole fraction of water (10C- Xa) from the gaseous stream. The ANN was trained and validated with a Bayesian regularization algorithm (30 hidden neurons) from which a mean squared error (MSE) = 0.00000411 and a total regression coefficient (R) of 0.99 were found. With these results and through the application of an ANOVA-type statistical analysis, it was determined with 95% reliability that the ANN has a good predictive capacity, which can be applied in the prediction of free fatty acids in the production of second-generation biodiesel (FAME).