| Popis: |
Rationale: Membrane-based gas separation technologies are advancing in leaps and bounds [2]. They are used in various gas separation processes such as carbon capture, air separation, hydrogen recovery, and many more [3-5]. The performance of membranes in the gas separation domain is typically described by permeability and selectivity coefficient [6]. Although both permeability and selectivity should be high for an optimal polymeric membrane, there is a trade-off relationship between those two factors [7]. Therefore, tackling this limitation has motivated researchers in order to explore different classes of polymeric membranes, such as thin film nanocomposites (TFN) and mixed matrix membranes (MMMs) [6]. MMMs can be synthesized by incorporating porous nano-fillers into polymeric matrix [8]. In contrast, TFN membranes are synthesized by forming a polyamide (PA) layer on top of the polymer support layer through the interfacial polymerization (PI) technique. By decreasing the TFNs thickness and using nanoparticles (NPs) as fillers, researchers have succeeded in increasing the membrane permeability and maintaining or increasing its selectivity for gas separations [9]. Traditionally, experimental research used to play a crucial role in developing innovative polymeric materials. However, the synthesis of a large number of possible membranes is expensive and time-consuming. Therefore, many theoretical methods and models have been developed to speed up the pace of materials discovery [10]. These models are based on many assumptions and are ineffective in describing the features of complex materials and systems. So, to address these problems, machine learning (ML) and Deep learning (DL) algorithms have been developed. ML is one of the most valuable methods that has recently entered the membrane separation toolbox. This collection of statistical methods aids researchers in analyzing large datasets and speeding up the discovery of new polymeric materials and membranes with high performance [2]. However, it seems that many studies focused on regression-based techniques, and the outcomes of this review can shed light on the current status of Artificial Intelligence (AI) in this field and aid researchers in finding innovative methods for improving the quality of research in the discovery of next-generation membranes for gas separation. This review paper will discuss various applications of machine learning in membrane-based gas separation. Firstly, an introduction to ML will be provided, including its principles, databases, and featurization techniques which are used in the membrane separation realm. Then, pre-processing methods, different ML algorithms, the potential of transfer learning for being used in membrane separation, and hyperparameter tuning methods will be explained. Finally, the main challenges and limitations of machine learning in this domain will be concluded. |
