Forecasting soybean oil extraction using cyclopentyl methyl ether through soft computing models with a density functional theory study

This work presents a thermo-statistical assessment using soft computing models to describe green soybean oil extraction by cyclopentyl methyl ether (CPME). Experimental data were collected based on an experimental factorial design and modeled by an Artificial Neural Network (ANN) and Adaptive Neuro-Fuzzy Inference System (ANFIS), as the empirical model was unable to accurately predict the experimental results. The ANFIS structure is related to the best statistical metrics, while the ANN achieves the best thermodynamic fit. The results suggest higher yields for higher temperatures and lower solvent-to-solid mass ratios. The extraction temperature can be significantly reduced with CPME to achieve the same yield as n-hexane. The second-order model was the most accurate ($\mathrm{SAE}$ = 0.1266, $\mathrm{MSE}$ = 5.54·10^-5 and $R^2$ = 0.9876) in representing the extraction kinetics, resulting in an extraction rate constant of 1.9782 min⁻¹. It was noticed that small positive induced charges given by the oxygen atom of CPME could contribute to the potential of this solvent to deplete the oil matrix and that its entropy is similar to that of the n-hexane molecule. The extracted oil presented the typical constitution regarding fatty acids composition; free fatty acid, mono, di, and triacylglycerol contents; and infrared spectrum.