Separation of organic molecules from water by design of membrane using mass transfer model analysis and computational machine learning

This work investigates the utilization of ensemble machine learning methods in forecasting the distribution of chemical concentrations in membrane separation system for removal of an impurity from water. Mass transfer was evaluated using CFD and machine learning performed numerical simulations. A membrane contactor was employed for the separation and mass transfer analysis for the removal of organic molecules from water. The process is simulated via computational fluid dynamics and machine learning. Utilizing a dataset of over 25,000 data points with r(m) and z(m) as inputs, four tree-based learning algorithms were employed: Decision Tree (DT), Extremely Randomized Trees (ET), Random Forest (RF), and Histogram-based Gradient Boosting Regression (HBGB). Hyper-parameter optimization was conducted using Successive Halving, a method aimed at efficiently allocating computational resources to optimize model performance. The ET model emerged as the top performer, with R-2 of 0.99674. The ET model exhibited a RMSE of 37.0212 mol/m(3) and a MAE of 19.6784 mol/m(3). The results emphasize the capability of ensemble machine learning techniques to accurately estimate solute concentration profiles in membrane engineering applications.