Mathematical modeling and process parameters optimization studies by artificial neural network and response surface methodology: A case of non-edible neem (Azadirachta indica) seed oil biodiesel synthesis

Abstract

This study aimed at using a non-edible NO (neem oil) for biodiesel production by modeling and optimizing the two-step process involved. A significant quadratic regression model (p < 0.05) with R2 = 0.813 was obtained for the reduction of the acid value of the NO with high FFA to 1.22 mgKOH/g under the condition of methanol–oil ratio of 0.55, H2SO4 of 0.45%, time of 36 min and temperature of 60 °C using RSM (response surface methodology). For biodiesel synthesis, ANN (artificial neural networks) coupled with rotation inherit optimization established a better model than RSM. The condition established by ANN was temperature of 48.15 °C, KOH of 1.01%, methanol–oil ratio of 0.200, time of 42.9 min with actual NOB (neem oil biodiesel) yield of 98.7% while RSM quadratic model gave the condition as temperature of 59.91 °C, KOH of 1.01%, methanol–oil ratio of 0.164, time of 45.60 min with actual NOB yield of 99.1%. R2 and absolute average deviations of the models from ANN and RSM were 0.991, 0.983, and 0.288, 0.334%, respectively. The results demonstrated that the models developed adequately represented the processes they described. Properties of NOB produced were within the ASTM D6751 and DIN EN 14214 biodiesel specifications.