Artificial neural network for liquid-liquid equilibrium of glycerol-methanol-safflower biodiesel mixture

Abstract

The determination of the Liquid-Liquid Equilibrium (LLE) in biodiesel production is essential for the design and monitoring of the production process. Therefore, the search for efficient modeling tools is crucial, since classical thermodynamic models have intrinsic limitations. Artificial neural networks (ANNs) are gaining ground due to their ease of use, generalization capacity, accuracy of results, etc. Within this context, the objective of this work was to obtain two ANNs to deal with the LLE of the glycerol-methanol-safflower oil biodiesel mixture at 298.15 K and 101.325 kPa. The aim of the first model was to classify whether the mixture is in the biphasic LLE region or in a homogeneous liquid phase, and the other one to calculating the composition of the liquid phases in equilibrium (rich in biodiesel and rich in glycerol). Both models considered the overall composition of the mixture as input. The obtained results showed that, with the first model was able to correctly classify the phases of the mixture with 98% accuracy, and the second one predicted the equilibrium compositions with a 4,98% error (lower than the one obtained using the UNIQUAC method).