OPTIMIZATION AND MODELLING OF BIO-OIL YIELD FROM THE PYROLYSIS OF JATROPHA CURCAS SEED USING OPTIMAL DESIGN AND ARTIFICIAL NEURAL NETWORKS
Better quality and quantity of pyrolysis products from biomass can be obtained by regulating the input parameters of the pyrolysis process. Pyrolysis of Jatropha Curcas seed mixed with alumina catalyst was carried out in a fixed bed reactor to study the effects of temperature, time, and particle size on bio char and bio-oil yield. The bio-oil yields were optimized using the Optimal Design (OD) under the Combined Methodology of the Design-Expert Software (12.0). The input and output parameters were modeled and validated using Artificial Neural Networks (ANN) based on 40 experimental data generated by the OD. The optimum bio-oil yield of 15.6 wt. % was obtained at 650 °C, 30 min, and 1 mm particle size. The Correlation Coefficient (R2) of the model for the bio-char and bio-oil yield under the OD were 0.998 and 0.996, respectively. The optimized ANN architecture employed the in-built Levenberg-Marquardt training algorithm in MATLAB software. Random division of the data into training, validation and testing sets followed 70:15:15 percentage proportions with 15 hidden layers. This resulted in the minimum Mean Square Error (MSE) of 2.15e-05 and (R2) of 0.96394 for the bio-oil yield. The FTIR spectra indicated that bio-oil contained phenols, esters, and acids compound while its Gas Chromatography analysis showed the presence of pyrrolidine, pyrimidine, and aldehydes. These properties signified the bioenergy and biochemical capabilities of the pyrolytic oil obtained. The prediction accuracy indicates that both the ANN and OD can be deployed for accurate prediction.