(197be) Machine Learning-Enabled Modification of Polyamide Reverse Osmosis Membrane

Polymeric membrane design is challenging, and finding new polymer is time-consuming due to the complex interplay of various factors that influence the membrane's performance. Among the various membrane separation techniques, reverse osmosis (RO) is the most common architecture globally, offering excellent performance and simple operation for water desalination. Polyamide (PA)-based polymeric membranes are the current RO membranes in the market where their performance is typically evaluated by water permeability and salt rejection. There is a tradeoff relationship between permeability and water/salt selectivity and the goal of any modification is to push the boundaries of permeability/selectivity tradeoff. The present upper bound correlation defines a trade-off that empirically showcases the current state of the art in separation performance. Surface modifications (e.g., surface coating) is one of the most effective strategy to improve the performance of the PA-RO membranes.

Machine learning (ML) models as an effective and cost-saving tool can assist traditional trial-and-error processes to accelerate the discovery of polymers with anisotropic diffusional properties used in membrane fabrication. Though more number of studies have been performed on using ML to predict transport performance of polymeric membrane and accelerate the discovery of high-performance materials used in membranes, modifications of the PA layer of surface RO membrane has not been well explored with ML. To the best of our knowledge, there are still no ML model to modify polyamide layer with a reported ability to increase water permeability of RO membranes. To address this challenge, we explored the potential of ML to enhance water permeability of polyamide RO membrane and pushed the permeability/selectivity tradeoff across the boundary.

In this study, we trained an explainable screening ML model on a dataset of different polymers used as the active layer in RO membranes. By applying SHapley Additive exPlanations (SHAP) analysis to our model, we identified moieties with positive and negative contributions toward water permeability. We leveraged the results from the SHAP analyses to modify the PA layer of the commercial RO membrane by sharping the intensity of functional groups' peaks with positive effect and suppressing those with negative effect, resulting in a substantial increase in water permeability. The membranes were characterized using FTIR, XPS, SEM analyzer and tested for water permeance, and NaCl rejection using a dead-end stirred cell. Compared to the commercial membrane, the modified membrane improved the water permeance from 1.7 ml/30min to 3.4 ml/30min.

Our results demonstrate the potential of ML to replace the traditional trial-and-error to modify polyamide layer of PA-RO membranes for any particular separation application, paving the way for the development of more efficient and sustainable RO membranes for water treatment and purification applications. The proposed approach can be extended to other fields of materials science and engineering, where the identification of novel materials and/or modification of molecular structure with tailored properties is essential.