Abstract:Simulating spatiotemporal turbulence with high fidelity remains a cornerstone challenge in computational fluid dynamics (CFD) due to its intricate multiscale nature and prohibitive computational demands. Traditional approaches typically employ closure models, which attempt to represent small-scale features in an unresolved manner. However, these methods often sacrifice accuracy and lose high-frequency/wavenumber information, especially in scenarios involving complex flow physics. In this paper, we introduce an innovative neural differentiable modeling framework designed to enhance the predictability and efficiency of spatiotemporal turbulence simulations. Our approach features differentiable hybrid modeling techniques that seamlessly integrate deep neural networks with numerical PDE solvers within a differentiable programming framework, synergizing deep learning with physics-based CFD modeling. Specifically, a hybrid differentiable neural solver is constructed on a coarser grid to capture large-scale turbulent phenomena, followed by the application of a Bayesian conditional diffusion model that generates small-scale turbulence conditioned on large-scale flow predictions. Two innovative hybrid architecture designs are studied, and their performance is evaluated through comparative analysis against conventional large eddy simulation techniques with physics-based subgrid-scale closures and purely data-driven neural solvers. The findings underscore the potential of the neural differentiable modeling framework to significantly enhance the accuracy and computational efficiency of turbulence simulations. This study not only demonstrates the efficacy of merging deep learning with physics-based numerical solvers but also sets a new precedent for advanced CFD modeling techniques, highlighting the transformative impact of differentiable programming in scientific computing.

What problem does this paper attempt to address?

This paper mainly discusses the challenge of simulating spatiotemporal turbulence with high fidelity in computational fluid dynamics (CFD). Due to the complex multiscale nature of turbulence and high computational demands, traditional methods often use closure models to represent small-scale features in an unresolved manner. However, these methods often sacrifice accuracy and lose high-frequency/wavenumber information, especially in scenarios involving complex flow physics. The paper proposes an innovative neural network-based differentiable modeling framework aimed at improving the predictability and efficiency of spatiotemporal turbulence simulation. This framework combines deep neural networks with numerical partial differential equation solvers, achieving the synergy of deep learning and physics-based CFD modeling through different programming frameworks. Specifically, a mixed differentiable neural solver is constructed on a coarse grid to capture large-scale turbulent phenomena, and a Bayesian conditional diffusion model is employed to generate small-scale turbulence predictions based on the large-scale flow. Two innovative hybrid architecture designs are investigated and their performance is evaluated through comparison and analysis with traditional large-eddy simulation techniques, including physics-based subgrid-scale closures and purely data-driven neural solvers. The results show that the neural differentiable modeling framework has the potential to significantly improve the accuracy and computational efficiency of turbulence simulations, demonstrating the effectiveness of integrating deep learning with physics-based numerical solvers and setting new standards for advanced CFD modeling techniques. The paper also emphasizes the transformative impact of differentiable programming in scientific computing and discusses the challenges faced by existing data-driven deep learning models in dealing with complex turbulent flows, such as long-term prediction stability, reliability, and dependence on a large amount of annotated data. Two hybrid architecture designs are proposed, one using neural networks as corrective terms for closure models and the other parameterizing the interpolation and discretization schemes of numerical solvers as trainable neural networks. Overall, the paper attempts to address how to leverage deep learning and innovative hybrid model designs to improve the accurate and efficient simulation of complex spatiotemporal turbulence, overcoming limitations of traditional methods.

Neural Differentiable Modeling with Diffusion-Based Super-resolution for Two-Dimensional Spatiotemporal Turbulence

End-to-end differentiable learning of turbulence models from indirect observations

Bayesian Conditional Diffusion Models for Versatile Spatiotemporal Turbulence Generation

Generalizable data-driven turbulence closure modeling on unstructured grids with differentiable physics

Differentiable hybrid neural modeling for fluid-structure interaction

Learning Incompressible Fluid Dynamics from Scratch -- Towards Fast, Differentiable Fluid Models that Generalize

Conditional Neural Field Latent Diffusion Model for Generating Spatiotemporal Turbulence

Reconstructing Turbulent Flows Using Physics-Aware Spatio-Temporal Dynamics and Test-Time Refinement

A Data-driven Approach for Turbulence Modeling

An interpretable framework of data-driven turbulence modeling using deep neural networks

Neural Dynamical Operator: Continuous Spatial-Temporal Model with Gradient-Based and Derivative-Free Optimization Methods

Deep Structured Neural Networks for Turbulence Closure Modelling

Integrating Neural Operators with Diffusion Models Improves Spectral Representation in Turbulence Modeling

Physics-enhanced Neural Operator for Simulating Turbulent Transport

Benchmarking Autoregressive Conditional Diffusion Models for Turbulent Flow Simulation

Deep learning to develop zero-equation based turbulence model for CFD simulations of the built environment

Data-driven Reynolds-averaged Turbulence Modeling with Generalizable Non-Linear Correction and Uncertainty Quantification Using Bayesian Deep Learning

Teaching the Incompressible Navier-Stokes Equations to Fast Neural Surrogate Models in 3D

Neural Ideal Large Eddy Simulation: Modeling Turbulence with Neural Stochastic Differential Equations