Computational simulation and investigations on the molecular separation of Ibuprofen from aqueous solution using polymeric membranes

Abstract

We have developed computational modeling based on machine learning and computational fluid dynamics for purification and isolation of Ibuprofen from aqueous solutions. The considered separation and purification system is a membrane-based process which is used to selectively remove Ibuprofen from liquid phase. The computational fluid dynamics (CFD) method was performed to obtain the concentration of the drug in the membrane system, and then the concentration values were used as inputs to a number of machine learning models to build the hybrid model of process. Indeed, we dealt with a dataset of 8K data points generated from a CFD simulation. As core models, Multilayer Perceptron (MLP), Lasso, and Support Vector Regression (SVR) were used. To improve efficiency, a technique known as bagging has been added to these models. We optimized the models to find optimal hyper-parameters. With R 2 metric, all three models have scores above 0.995 so all models have acceptable performances. When we consider the MAE metric, the lowest error is related to the BAGGING+MLP model with a value of 5.805 × 10 1 , and the BAGGING+LASSO and BAGGING+SVR models have an error rate of 1.401 × 10 2 and 1.055 × 10 2 , respectively. Finally BAGGING+MLP can be introduced as the most accurate model. • Machine learning modeling of molecular separation via membranes. • Optimization of machine learning using Bagging technique. • All three optimized machine learning models had scores above 0.995. • The method of MLP indicated to perform better than other models.