Modeling and optimization of African pear seed oil esterification and transesterification using artificial neural network and response surface methodology comparative analysis

This study aimed at modeling and optimum conditions evaluation of the production of fatty acid methyl esters from high free fatty acid (FFA) and oil content African pear seed. The central composite design of response surface methodology (RSM) was employed for investigating the individual and interactive effect of the process variables (methanol/oil molar ratio, catalyst concentration, temperature and reaction time) on the FFA reduction and African pear seed oil methyl esters (APSOME) yield. A feed-forward neural network model with Levenberg-Marquardt backward propagation training algorithm was also employed to predict the responses. Temperature was found to be the most significant parameter with high F-values of 7.1 and 14.15 and low P-values of 0.0177 and 0.0019 for esterification and transesterification processes respectively. The observed optimum FFA reduction and biodiesel yield were 0.29 wt% and 94.55 wt% respectively, under the optimal conditions of 50 and 63.92 °C, 5.0 and 1.88 wt% catalyst concentration, 60 and 58 min reaction time and 13:1 and 6.86:1 methanol to oil molar ratios for esterification and transesterification processes respectively. At these optimal conditions, the validated experimental responses were 0.26 wt% and 95.027 wt% respectively. The performances of both ANN and RSM models showed adequate prediction of the response, with high coefficient of determination (R2) of 0.899 and 0.909 for esterification and 0.9195 and 0.944 for transesterification processes respectively. The fuel properties of the biodiesel produced at the optimum conditions were found to compare with ASTM D6751 and EN 14214 standard) while the GC–MS characterization showed that the biodiesel has high unsaturation. The overall results confirmed that the biodiesel production using African pear seed oil is viable