Modeling rarefied gas-solid surface interactions for Couette flow with different wall temperatures using an...

Area of Science:

• Fluid Dynamics
• Statistical Mechanics
• Machine Learning

Background:

• Rarefied gas flows exhibit boundary phenomena like velocity slip and temperature jump.
• Existing boundary models struggle with complex gas-solid interactions and local nonequilibrium states.
• Accurate modeling of gas-surface interactions is crucial for understanding complex flow conditions.

Purpose of the Study:

• To develop a novel statistical gas-solid surface scattering model using machine learning.
• To investigate the performance of the Gaussian Mixture (GM) model against established models.
• To assess the GM model's capability in simulating rarefied gas flows with complex boundary conditions.

Main Methods:

• Employed unsupervised machine learning (Gaussian Mixture model) to analyze molecular dynamics (MD) simulation data.
• Developed a statistical scattering model based on MD-derived collisional data.
• Simulated Couette flow for Argon (Ar) and Helium (He) gases between gold walls.

Main Results:

• The GM model demonstrated excellent agreement with MD simulation data for postcollisional velocity distributions.
• GM model significantly outperformed the Cercignani-Lampis-Lord (CLL) scattering kernel in predicting accommodation coefficients.
• For Helium, the GM model's energy accommodation coefficient closely matched MD results, unlike the CLL model.

Conclusions:

• The Gaussian Mixture model provides a highly accurate statistical representation of gas-surface scattering.
• The GM model shows significant potential for developing generalized boundary conditions in complex, nonequilibrium rarefied gas flows.
• This machine learning approach offers a superior alternative to traditional scattering kernels for detailed gas-surface interaction modeling.