Optimization of microwave-aided biodiesel production from waste cooking oil using functionalized catalyst derived from bio-waste: experimental investigations, taguchi modelling, and machine learning

Abstract

In this study, the optimisation of microwave-aided biodiesel production from waste cooking oil using heterogeneous bio-waste catalysts derived from clamshells and cocoa pods, and integrating Taguchi modelling and machine learning, was investigated. The heterogeneous catalyst synthesis process included carbonisation, sulphurization, calcination, and impregnation of acid and base precursors. In-depth characterisation was used to elucidate the intrinsic properties of the synthesised bi-functional catalysts utilising SEM, EDXRF, XRD, FTIR, and BET/BJH. Taguchi L16 and machine learning optimisation were used to simulate the process, examine the interactive effects of process input variables, and optimize and statistically characterize the transesterification process in terms of the waste cooking oil yield and acid values with the models evaluated using coefficient of determination (R 2 ), root mean square error (RMSE), mean absolute error (MAE) and mean square error (MSE). The biodiesel yield (BY) and acid values were modelled using response surface methodology (RSM), artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) models. The input variables investigated were: methanol to oil ratio, % catalyst loading, power intensity, reaction time, and stirrer speed. Reusability tests on the catalyst’s viability concerning effects on waste cooking oil yield and acid values were also undertaken. The synthesised bi-functional catalyst had surface area (171.4 m 2 /g to 768.4 m 2 /g), pore volume (0.145 cc/g to 0.1042 cc/g) and pore sizes (2.108 nm to 5.878 nm). Optimal input variables for the process were: 600 Watts (heating power intensity); 15:1 (methanol: waste cooking oil); 5 min (reaction time); 1000 rpm (stirrer speed), and 2 wt% (catalyst loading). Optimal BY and AV at optimum input conditions were 92.737% and 0.408 mg / KOH respectively. The produced waste cooking oil met the standards enunciated in ASTM D 6751 and EN 14214. The close relationship between the statistical data and experimental values confirmed the accuracy and validity of the models investigated. The predictive model performances of RSM, ANN and ANFIS models compared favourably with the experimental BY and acid values data. The best predictive performance was displayed by the ANFIS model (R 2 : 0.99994; MSE: 0.02069 ; MAE: 1.56250 × 10 − 8 ) for waste cooking oil yield and (R 2 : 0.99996; MSE: 4.61250 × 10 − 7 ; and MAE: 0.00043 ) for acid values. Analysis of variance (ANOVA) results indicated that the methanol-to-oil ratio had the greatest impact (SS: 82.01; F-value: 102.84; p-value: 0.0016) on waste cooking oil yield, while acid values was significantly influenced by power intensity (SS: 0.0485; F-value: 92.49; p-value: 0.0019).