Enhancing oxidative stability of biodiesel using fruit peel waste extracts blend: Comparison of predictive modelling via RSM and ANN techniques

Abstract

This work harnessed the potentials of banana, mango, and palm fruit peels as green sources of natural antioxidants whose liquid extracts were recovered via supercritical fluid extraction (SFE) to improve biodiesel's low oxidative stability (OS). Response surface methodology (RSM) and artificial neural network (ANN) models were explored and compared in predicting the OS of the extracts' stabilized biodiesel. The induction period (IP) for the freshly synthesized (Y1) and 8-weeks stored (Y2) extracts-blended biodiesel was determined using Rancimat analyzer. Optimization of the biodiesel's IP was carried out using central composite design (CCD) for the chosen process variables namely extracts' dosage ratio, mixing speed, and reaction time against the response variable (IP). Accordingly, the optimal conditions that accounted for the maximum IP value of 17.59 h for Y1 were a dosage ratio of 1.24, a mixing speed of 496.66 rpm, and a reaction time of 64.99 min. Whereas, a corresponding dosage ratio of 0.94, a mixing speed of 430.27 rpm, and a reaction time of 46.71 min produced the highest IP (12.82 h) value for the response Y2. The predictive capability of both models was manifested based on the results and found to be within an acceptable range. However, the ANN presented lower percentage errors and a high correlation coefficient (R2) corresponding to 0.9710 and 0.9462 for Y1 and Y2 respectively, demonstrating the model's superior predicting ability compared to 0.9428 and 0.8091 for the same responses presented by RSM. Finally, the limitation of the traditional RSM chemometric approach can be overcome by utilizing the ANN line of optimization which proved to be more accurate in the prediction ability.

Keywords

• Artificial intelligence
• Biodiesel stabilization
• Process optimization
• Green antioxidants