Neural Networks Modeling of Dearomatization of Distillate Cuts with Furfural to Produce Lubricants

An important industrial stage in the solvent route production of API Group I lubricants is the dearomatizaton, a liquid-liquid extraction process. However, dearomatization feed is a complex mixture of components, a disadvantage for thermodynamic modeling approaches. As an alternative, we propose the use of artificial neural networks (ANN) to correlate compositions and crucial properties (e.g., specific gravity, viscosity index and refractive index) of raffinate and extract phases in liquid-liquid extraction. In order to improve the model reliability, data obtained in a pilot plant for dearomatization of three distillate cuts of Arab Light oil using furfural solvent were employed to fit and validate ANN models, evaluating different configurations. The dataset consists of six solvent/oil ratios at six different temperatures, resulting in a number of thirty-six experiments for each distillate cut. The ANN models fit suitably all investigated properties, for both fit and validate data sets, indicating the great potential of this approach as a predicting tool for reducing experimental efforts in characterization assays and also in predicting behaviour in industrial units.