Modeling and scale-bridging using machine learning: nanoconfinement effects in porous media

Area of Science:

• Computational physics
• Materials science
• Chemical engineering

Background:

• Simulating fluid behavior in nanoporous media (e.g., shale, <50 nm pores) is crucial for applications like hydrocarbon production.
• Fluid properties like density, viscosity, and adsorption are significantly altered by confinement effects at the nanoscale.
• Existing pore-scale Lattice Boltzmann Methods (LBM) require corrections for confinement, while Molecular Dynamics (MD) is computationally prohibitive for large-scale corrections.

Purpose of the Study:

• To develop a computationally efficient Machine Learning (ML) surrogate model to bridge the scale gap between MD and LBM.
• To accurately capture nanoscale adsorption effects in nanoporous media across a wide parameter space.
• To enable accurate upscaling of fluid behavior for large-scale simulations.

Main Methods:

• Developed a Machine Learning (ML) surrogate model trained on a limited set of Molecular Dynamics (MD) calculations.
• The ML model computes upscaled adsorption parameters, considering variations in fluid density, temperature, and pore width.
• Validated the ML model's ability to represent confinement effects.

Main Results:

• The ML surrogate model achieved a speedup of 7 orders of magnitude compared to brute-force MD simulations.
• The model effectively captures adsorption effects across a broad range of physical parameters.
• Successfully bridged the gap between MD and LBM scales with significantly reduced computational cost.

Conclusions:

• The developed ML workflow provides a computationally feasible method for incorporating nanoscale confinement effects into larger-scale simulations.
• This scale-bridging approach is agnostic to the specific physical system and can be generalized to other applications.
• The ML model significantly enhances the efficiency of simulating fluid dynamics in nanoporous materials.