Prediction of the adsorption properties of liquid at solid surfaces with molecular scale surface roughness via encoding-decoding convolutional neural networks

GY Li and YT Guo and T Mabuchi and D Surblys and T Ohara and T Tokumasu, JOURNAL OF MOLECULAR LIQUIDS, 349, 118489 (2022).

DOI: 10.1016/j.molliq.2022.118489

Molecular dynamics (MD) simulation can effectively analyze the transport properties of liquid at the solid surface with different nanoscale roughness, while high computational costs are required. Herein, a deep learning encoding-decoding convolutional neural network is proposed to predict the adsorption density distribution of atomic and organic liquids under different molecular scale surface roughness. Compared with the previous deep learning studies focusing on simple macro adsorption parameters, our deep learning method realizes the prediction and visualization of micro scale adsorption behavior with very high accuracy. The data-driven deep learning algorithm replaces the MD extensive sampling and simplifies the operation process, which improves the computational efficiency of a single model 36000-fold. This study proves the good coupling between MD and deep learning method, which is helpful for designing surface geometry to obtain desirable interfacial transport properties of molecular liquid and complementing the nanoscale model system enabling the interactive visualization. (C) 2022 Elsevier B.V. All rights reserved.

Return to Publications page