CoNFiLD-inlet: Synthetic turbulence inflow using generative latent diffusion models with neural fields.

Area of Science:

• Fluid dynamics
• Computational science
• Artificial intelligence

Background:

• Eddy-resolving turbulence simulations need accurate stochastic inflow conditions.
• Existing methods like recycling or synthetic generators have limitations in realism and computational cost.
• Deep learning (DL) offers potential but often struggles with robustness and error accumulation.

Purpose of the Study:

• To develop a novel deep learning-based inflow turbulence generator.
• To address the limitations of current methods in producing realistic and robust synthetic turbulence.
• To create a versatile tool for various Reynolds numbers without retraining.

Main Methods:

• Integration of diffusion models with a conditional neural field.
• Parametrization of inflow conditions using Reynolds numbers.
• Validation using a priori and a posteriori tests in direct numerical simulation and large-eddy simulation.

Main Results:

• CoNFiLD-inlet successfully generates realistic, stochastic inflow turbulence.
• The method demonstrates effective generalization across a wide range of Reynolds numbers (10^3 to 10^4).
• High fidelity, robustness, and scalability were confirmed through comprehensive validation.

Conclusions:

• CoNFiLD-inlet provides an efficient and versatile solution for synthetic inflow turbulence.
• The novel approach overcomes common challenges in turbulence simulation setup.
• This method advances the capability of deep learning in computational fluid dynamics.