Determination of clay-water contact angle via molecular dynamics and deep-learning enhanced methods

Molecular dynamics modeling is a useful tool to study the interface physics of the clay-water system at the nanoscale. A key parameter to characterize the interface physics of unsaturated soils is the contact angle between clay and water. In this paper, we propose a deep-learning enhanced numerical method to accurately determine the clay-water contact angle at elevated temperatures from molecular dynamics simulations. In this approach the water droplet on clay is treated as a 3D point cloud. The spatial location of individual water molecular is derived from the coordinate of the molecule’s center-of-mass. At each point, the covariance matrix method is used to fit a tangent plane to a surface spanned by its k -nearest neighbors. The clay-water contact angle is calculated from the normal of the tangent plane. The covariance matrix method is enhanced by a deep-learning algorithm through the randomized Hough transform. Through the transform, each normal is mapped to a Hough accumulator with many bins. The final normal is obtained from the most voted bin via the deep-learning algorithm. Numerical results have demonstrated that the deep-learning enhanced algorithm is robust in characterizing variations of the contact angle between clay and water at elevated temperatures.