Separation of organic molecules using porous polymeric membranes: Model development using advanced hybrid CFD and artificial intelligence

We proposed a new computational methodology for computational modeling of mass transfer in membranes with application for molecular separation. The computational method centered on hybrid computational fluid dynamics (CFD) and machine learning (ML) in which the data of CFD are used for training machine learning model. The case study is a polymeric membrane for removing a solute from a solution, and the mass transfer equations are solved using CFD. Therefore, the data set has two inputs: r and z which are the coordinates. Also, the single output is C which indicates the solute concentration, and the quantity of data points is more than 8,000. The data analysis and modeling in this study were carried out utilizing the Adaboost (Adaptive Boosting) model. Essentially, this model requires a base model to be built, and in this case, we have chosen three models to build this model, namely Decision Trees (DT), Theil-Sen Regressions (TS), and Support Vector Regressions (SVR), and compared their results. The coefficient of determination (R²) of boosted DT, TS, and SVR are 0.9978, 0.9171, and 0.9908 respectively. Also, in terms of RMSE their error values were 3.96E + 01, 3.03E + 02, and 7.99E + 01. Finally, Boosted Decision Tree model is selected as the most accurate technique for the membrane simulation. The MAE metric for this model is 1.01E + 01 and its MAPE error rate is 9.07E-03.