(329c) Estimation of the Structure of Confined Water between Hybrid Materials Using Convolutional Neural Networks (CNN)

Two-dimensional (2D) materials such as graphene, boron nitride, and molybdenum disulfide find applications in a wide range of fields, including catalysis, sensing, energy storage, biology, and electronics. However, it is important to note that these materials are also highly dependent on factors such as surrounding temperature, pressure, and humidity. Our recent study investigated the structure of confined water between stacked 2D sheets of our novel layered hybrid materials using conventional analysis of molecular dynamics (MD) simulations trajectories. The principal factors like hydrophilicity and slit width between two 2D sheets significantly affect the structure of water confined. In this study, the spatial and chemical characteristics of sheets were integrated with deep learning models, i.e. Convolutional Neural Networks (CNN), to determine the precise structure of water around sheets. This study showcases the proof of concept of AI/Deep learning methods to model the structure of confined water around the hydrophilic/hydrophobic hybrid sheets. Specifically, the CNN model was trained on the three-dimensional structure of the hybrid material system in water to estimate the atomic density profile of oxygens in water and the infrared (IR) spectrum of water molecules. A 3D grid structure of water was allowed to convolute and extract the physical and chemical features that significantly affected the ordering of water around these hydrophilic/hydrophobic sheets. Four convolutional layers were fully connected, followed by five dense neural layers used to accurately train the model. Moreover, this surrogate model was used to accurately predict the structure of water around the sheets with their altered slit spacings and for the sheets replaced with their other counterparts.