Arash Khosravi

In recent years, the application of membrane technologies has increased for both gas separation and carbon capture and Artificial Intelligence (AI) can play a crucial role in reducing the costs and removing the implementation related hurdles of these technologies. In this study, in order to address the limitation of experimental research and accelerate the pace of identifying new and effective membranes for being used in the gas industry, a Machine Learning (ML) technique has been developed based on polymers' physical and chemical properties. In particular, the Random Forest (RF) algorithm is used to predict membrane performance in terms of permeability and selectivity for CO2/CH4 separation. Then, the Shapley Additive Explanations (SHAP) method was used to interpret the results. In addition, in order to introduce polymers to ML model, fingerprinting and molecular descriptor approaches were used simultaneously. The results proved that the Topological Polar Surface Area (TPSA) is one of the most influential parameters on membrane performance. Furthermore, findings revealed that polar groups in polymer backbone structure have a negative effect on permeability, while they are positively correlated with selectivity. Another outcome of the present study was about the negative effect of aromatic rings on membranes permeability.