Abstract:We integrate neural operators with diffusion models to address the spectral limitations of neural operators in surrogate modeling of turbulent flows. While neural operators offer computational efficiency, they exhibit deficiencies in capturing high-frequency flow dynamics, resulting in overly smooth approximations. To overcome this, we condition diffusion models on neural operators to enhance the resolution of turbulent structures. Our approach is validated for different neural operators on diverse datasets, including a high Reynolds number jet flow simulation and experimental Schlieren velocimetry. The proposed method significantly improves the alignment of predicted energy spectra with true distributions compared to neural operators alone. Additionally, proper orthogonal decomposition analysis demonstrates enhanced spectral fidelity in space-time. This work establishes a new paradigm for combining generative models with neural operators to advance surrogate modeling of turbulent systems, and it can be used in other scientific applications that involve microstructure and high-frequency content. See our project page: <a class="link-external link-http" href="http://vivekoommen.github.io/NO_DM" rel="external noopener nofollow">this http URL</a>
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the spectral limitations in surrogate modeling of turbulence using Neural Operators. Specifically, although Neural Operators perform well in terms of computational efficiency, they are deficient in capturing high - frequency flow dynamics, resulting in overly smooth prediction results. To overcome this problem, the authors propose a new method, that is, enhancing the resolution of Neural Operators through Conditional Diffusion Models, thereby improving the representation of turbulent structures.
### Main Problems
1. **Spectral Limitations**: When dealing with turbulence, especially under high - Reynolds - number conditions, Neural Operators are unable to capture high - frequency flow characteristics well, resulting in overly smooth prediction results.
2. **Spectral Bias**: The spectral bias of neural networks causes them to tend to learn low - frequency components while ignoring high - frequency components. This is particularly important in turbulence modeling because the energy spectrum of the turbulence system still has a non - trivial energy level in the high - frequency part.
### Solutions
1. **Combining Neural Operators and Diffusion Models**: By conditioning the diffusion model on the output of the neural operator, the ability to capture high - frequency flow characteristics can be enhanced, thereby improving the resolution and accuracy of the prediction results.
2. **Verification Method**: The authors verified the effectiveness of this method on different datasets, including high - Reynolds - number jet simulations and experimental Schlieren velocity - field data. The results show that, compared with using Neural Operators alone, the method of combining diffusion models significantly improves the consistency between the predicted energy spectrum and the real distribution, and the proper orthogonal decomposition analysis in space - time also shows better spectral fidelity.
### Specific Cases
1. **Kolmogorov Flow**: Through the conditional diffusion model, the predicted vorticity field eliminates Gibbs oscillations, restores finer features, and is consistent with the real power spectrum.
2. **Buoyancy - Driven Transport**: The conditional diffusion model shows better spectral similarity when predicting the concentration field, reducing the over - smoothing phenomenon in Neural Operator predictions.
3. **Airfoil Turbulent Wake**: In the turbulent wake of the NACA0012 airfoil, the conditional diffusion model can better restore high - frequency components and is more consistent with the real pattern.
4. **Three - Dimensional Turbulent Jet**: In the three - dimensional turbulent jet, the conditional diffusion model restores high - frequency components from the output of the Neural Operator, especially in the later - time - state prediction.
5. **Helium Turbulent Jet in Air**: In the experimental data, the prediction results of the conditional diffusion model are more consistent with the real data in both the spectrum and the energy - decay spectrum.
### Conclusion
This research establishes a new paradigm. By combining generative models and Neural Operators, the surrogate - modeling ability for turbulence systems is significantly improved. This method is not only effective in turbulence modeling but can also be applied to other scientific applications involving micro - structures and high - frequency content.