Computational simulation of mass transfer in membranes using hybrid machine learning models and computational fluid dynamics

Abstract

In this work, a modeling approach has been developed for describing component separation by selective membrane systems. The developed modeling strategy in this research is based on computational fluid dynamics (CFD) integrated with machine learning (ML) approach to reduce the computational costs and evaluate the possibility of ML model for integration to CFD models. The considered case study is a membrane-based molecular separation for removal of species from water. CFD was performed to solve mass transfer equations, and the data in form of concentration profiles have been used for ML computations. The data set in this research has about two thousand data vectors, each of which contains two inputs r and z and one output C which is the species concentration in the feed channel of membrane. Three models, Multi-Layer Perceptron (MLP), Support Vector Machines with RBF kernel (RBF-SVM), and Decision Tree (DT) were chosen for modeling since they were anticipated to perform well on the provided data by adjusting their hyper-parameters. After optimizing the models, their results were carefully checked with different criteria. All three models showed acceptable results with R2 score criteria higher than 0.9. MAPE decreased to 3.57 × 10−3 and improved to 3.6 with the RMSE criterion. We can introduce the best model in this research as the MLP model for description of the mass transfer in the membrane.