Insights into Synthesis and Optimization Features of Reverse Osmosis Membrane Using Machine Learning

Reverse osmosis membranes have been predominantly made from aromatic polyamide composite thin-films, although significant research efforts have been dedicated to discovering new materials and synthesis technologies to enhance the water-salt selectivity of membranes in the past decades. The lack of significant breakthroughs is partly attributed to the limited comprehensive understanding of the relationships between membrane features and their performance. Insights into the intrinsic features of reverse osmosis (RO) membranes based on metadata were obtained using explainable artificial intelligence to understand the relationships and unify the research efforts. The features related to the chemistry, membrane structure, modification methods, and membrane performance of RO membranes were derived from the dataset of more than 1000 RO membranes. Seven machine learning (ML) models were constructed to evaluate the membrane performances, and their applicability for the tasks was assessed using the metadata. The contribution of the features to RO performance was analyzed, and the ranking of their importance was revealed. This work holds promise for metadata analysis, evaluating the RO membrane against the state of the art and developing an inverse design strategy for the discovery of high-performance RO membranes.