.png){kind=icon}
.png){kind=icon}
Return
Physics-informed machine learning for wetting hydrodynamics
.png){kind=icon}
PRE
AI
DOI：10.1016/j.jcp.2025.114596.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This study presents a data-driven approach for modeling wetting hydrodynamics phenomena in the form of droplet transport on chemically heterogeneous surfaces, utilizing prior physical knowledge. The training datasets are generated using direct numerical simulations of the twophase Stokes equations, applicable in the low Reynolds number limit. The resulting data-driven models are based on the Fourier neural operator and are trained to correct the time derivative of the contact line as predicted by a low-order asymptotic approximation, an approach that was proven to be accurate and generalizable in earlier studies that focused on wetting hydrodynamics in the long-wave regime. More specifically, two data-driven models are trained to augment the low-order contact line velocity prediction: (i) a model focusing on correcting the purely translational velocity component of the droplet, and (ii) a model focusing on the higher-order corrections to the contact line velocity. The corrected contact line velocity is then used to advance the solution in time, using standard time-integration schemes. The resulting physics-informed machine learning workflow is proven to accurately capture the contact line dynamics on a range of different substrate heterogeneity profiles and can provide a reliable and efficient alternative to costly direct numerical simulations in optimization tasks, where a large number of simulations are needed.
Keywords:
Wetting hydrodynamics
Capillary driven phenomena
Chemical heterogeneities
Data-driven modeling
Fourier neural operator
Journal
IF:
3.8
Papers:
1.5W
Citations:
7.4W