Optimization of ultrasound-assisted base-catalyzed methanolysis of sunflower oil using response surface and artifical neural network methodologies

Abstract

The sunflower oil transesterification, catalyzed by KOH in the presence of ultrasound, was optimized by combining a 34 full factorial design of experiments with either a back-propagation artificial neural network (ANN) with the topology 4–10–1 or the response surface methodology (RSM). Four input factors, methanol/oil molar ratio, reaction temperature, catalyst loading and time and one output response, FAME yield, were included into the optimization study. The main goals were to test how accurately these two combinations predict and simulate the FAME yield achieved by the base-catalyzed methanolysis of sunflower oil under ultrasonication. Another aim was to compare the performances of the developed two models as a tool assisting decision making during the investigated methanolysis process. The ANN is shown to be a powerful tool for modeling and optimizing FAME production. Its predictions of FAME yield are very good all through the methanolysis process studied in wide ranges of the process factors. This is proved by a low value (±3.4%) of the mean MRPD between the experimental and simulated values of FAME yield, suggesting that they are almost the same. The ANN predictions were much better than those (±24.2%) obtained by the second-order polynomial equation from the RSM. The generalization ability of the developed ANN model for the base-catalyzed methanolysis optimization was well documented for different feedstocks and operational variables in the presence and absence of the ultrasound. The maximum FAME yield of 89.9% predicted by the ANN model could be achieved in 60min at the reaction temperature of 30°C, the initial methanol/oil molar ratio of 7.5:1 and the catalyst loading of 0.7%.