Simultaneous computational modelling and experimental validation for Sterculia urens oil extraction for biodiesel application

Abstract

Sterculia urens (S. urens) seeds are a non-edible feedstock rarely studied for biodiesel applications. Extraction is the initial step for obtaining oil from the feedstock. For this, an appropriate solvent was identified and the total oil content of the feedstock was 38.9 wt% by the Soxhlet extraction method. The optimum oil yield affected by different parameters was modelled by response surface methodology (RSM) and artificial neural network (ANN) computational methods. The optimum oil yield predicted by the computational methods was experimentally validated. The performance was measured using indices such as the correlation regression coefficient (R2), mean square error (MSE), standard error of prediction (SEP%) and absolute average deviation (AAD%). The results from computational modelling tools showed that the oil yield predicted by both tools was close to the optimum values of 60 °C temperature, 0.5 mm meal size and 180 min extraction time. The deviation from the experimental data is lower for ANN and higher for RSM, which shows the ANN prediction is more accurate than the RSM. Finally, the physicochemical properties of the extracted oil were determined and their possible influence on the biodiesel properties was discussed. These results suggest that the oil can be used as a potential second-generation feedstock.