Transesterification of castor oil: neuro-fuzzy modelling, uncertainty quantification and optimization study

Abstract

This study conducted experiments used for the development of both the regression model with uncertainty analysis and the adaptive neuro-fuzzy inference system (ANFIS) model for the prediction of the yield of biodiesel (YB) produced from castor oil in the presence of calcium oxide derived from the eggshell. Box Behnken design (BBD) was used to develop the experimental condition for five different variables while YB was the response. Uncertainty analysis was determined from Monte Carlo simulation (MCS). The model was optimized and validated before the generated data was applied in the three ANFIS modelling techniques. Root mean square error (RMSE), coefficient of correlation (R2) and average percentage error (APE) were used to determine the accuracy of the models developed. The result of this modelling shows that the optimum YB (94.29%) was achieved at a methanol to oil ratio of 11.48, catalyst loading of 3.38 wt%, reaction time of 1.84 h, the temperature of 60.2 °C, and agitation of 343.5 rpm. The prediction from BBD, ANFIS and MCS agreed that the methanol to oil ratio was the most important parameter for investigation. The considered ANFIS model technique (subtractive clustering) for the modelling of YB outperformed BBD model. The novelty of this study are the determination of the optimum condition for the transesterification of castor oil in the presence of thermally treated anthill, the establishment of the use of ANFIS in modelling YB, the prediction of the influence of variables on YB using both statistical and, AI techniques and validation of the predictions from the two methods using MCS.