Artificial Intelligence Enhanced Water Desalination

 Water scarcity is currently affecting the lives of billions of people all around the world, and the situation is worsening. Reverse osmosis (RO) water desalination is used as a predominant industrial solution of water scarcity. RO water desalination is to apply pressure on saline water toward a permeable membrane, while the membrane can filter out unwanted ions and allow fresh water to pass through. Traditional polymeric membranes used in the RO process are generally energy-inefficient because of the low water permeability. In recent years, 2D materials such as graphene with artificially-created nanopores have been widely researched as more efficient substitutions for traditional membranes. There are two factors that cast influence on the desalination performance of 2D materials: the material and the nanopore geometry. In this work, various 2D materials are compared for their performances in RO desalination using molecular dynamics (MD) simulations. Physical reasons behind the superior performances of materials such as \ce{MoS_2}, MXene, and metal-organic frameworks are unveiled. Harnessing the power of artificial intelligence, a deep reinforcement learning (DRL) model is trained to rapidly optimize graphene nanopore geometry by balancing the water permeation/ion rejection trade-off. DRL optimized nanopore geometries inspire the creation of more uniquely-shaped nanopores with much higher water permeation than regular circular nanopore without compromising the ion rejection capability. Lastly, neural network is used predict the ion concentration profile under nanoconfinement. It serves as a faster and accurate surrogate for molecular dynamics simulation to deepen our understanding about the electrical double layer in nanoconfined space, which is a critical physical phenomenon associated with RO desalination.