Accelerating Water Desalination Screening of 2D Membranes via Machine Learning Surrogates

High Resolution Image Download MS PowerPoint Slide To mitigate the significant computational costs and time associated with molecular dynamics (MD) simulations, there is a growing utilization of Artificial Intelligence (AI) to enhance the efficiency of water treatment and seawater desalination processes have been conducted. In this study, we have developed a surrogate machine learning (ML) model by training on MD simulation results. The purpose of this model is to predict the desalination performance of 2D membranes that have been documented in the literature. The desalination performance of 2D membranes is determined by two targets: water flux and ion rejection rates. We have identified features that are correlated to those targets. The features of the system include the physics of the pores and membranes, applied pressure, partial charges on the atoms, pore geometry, and mechanical properties of the membranes are investigated. To augment the size of the data set, we have applied multiple well-known 2D nanomaterials with outstanding desalination performance. We compare different ML models and use evaluation metrics to validate the results and verify the prediction results by applying different random values. The findings of this study underscore the efficacy of machine learning algorithms, particularly ensemble methods, as robust tools to predict membrane water flux and ion rejection in reverse osmosis processes. These algorithms show a notable strength in accurately forecasting membrane flux and ion rejection, offering valuable insights into utilizing water desalination calculations in time and energy.