Modelling and optimization of luffa oil transesterification via acid activated waste marble catalyst

Abstract

This study assessed the performance of response surface methodology (RSM) and artificial neural network (ANN) in modelling the transesterification of luffa oil using acid activated waste marble catalyst. The waste marble was activated with 0.5 molar sulphuric acid at 600 ^oC for 4 hours and was characterized by SEM, FT-IR, XRD, XRF, and BET; the characterization proved that the catalyst was successfully activated. The experiments were conducted at a catalyst dosage (1-5 wt. %), temperature (40-80 ^oC), methanol-oil ratio (4:1-12:1), time (1-3 hours), and agitation speed of (100- 500 rpm) with output as biodiesel yield. ANN was assessed using three back-propagation (BP) procedures, each comprising five neurons (input layer), one (output layer) and ten (hidden layer). The Levenberg Marquardt technique offered the most accurate prediction for luffa oil transesterification. The models were developed based on experimental and algorithm simulations and designs. The models' performance was assessed using the R² and MSE. Regarding R² and MSE, the ANN model (R²=9.9921E-1, MSE=0.06311) has a superior predictive capacity in forecasting the process than the RSM (R²=0.9885, MSE=0.86). At a catalyst concentration (3wt %), time (2 hours), temperature (60 ^oC), agitation speed (100 rpm) and methanol-oil ratio (12:1), the experimental (92.57 %), RSM predicted (94.0487 %) and ANN predicted (91.1768 %) biodiesel yield showed an agreement between the experimental and predicted values. The findings via physicochemical analysis, FT-IR, and GC-MS confirm that the biodiesel was within ASTM specifications.