Speaker: José Aniceto - CICECO - Aveiro Institute of Materials
Title of the talk: Machine Learning models for the prediction of diffusivities in different solvents
Abstract: The diffusion coefficient is extremely important for the design, simulation and scale-up of separations and chemical reactions. Experimental diffusivity data is still scarce for many compounds and thus accurate models are needed to provide reliable estimations. Currently, the Wilke-Chang equation is the most well-known and most used equation to calculate solute diffusivities in a solvent. Several other models have been proposed (e.g., hydrodynamic equations, phenomenological theories, empirical equations, molecular dynamics) with various degrees of success and applicability.In our group, we have developed predictive machine learning models to estimate diffusivities of solutes in different solvents, namely for supercritical carbon dioxide and water, two solvents of particular interest, as well as more general models aimed at polar and nonpolar solvents. The best results have been consistently provided by Gradient Boosting algorithms which showed, for the case of supercritical carbon dioxide, an average absolute relative deviation of 2.58 % for the test set. By comparison, the Wilke-Chang model showed worse performance with deviations of 12.41 %. The best conventional model (DHB) resulted in deviations of 4.27 %, although this model is correlative, thus it requires previous experimental data to fit the system parameters, which is not always available.José Aniceto, is a Junior Researcher at CICECO-Aveiro Institute of Materials. He has a M.Sc. degree in Chemical Engineering from the Faculty of Engineering of the University of Porto and a Ph.D. from University of Aveiro. His research activities focus on chromatographic purification processes and supercritical fluid extraction of bioactive compounds. Recently he has become interested in the application of machine learning in the field of chemical engineering.
Orcid: 0000-0001-5206-0142