Data-Driven Filter for Implicit Large-Eddy Simulations Discussed in New Research

Strategic Insights -

Research published in the Journal of Computational Physics has presented a novel data-driven filter for implicit large-eddy simulations (ILES) in the context of the Spectral Difference method. The filter is designed to analyze the relationship between ILES and direct numerical simulations (DNS) and has been evaluated using a Taylor-Green Vortex test-case at Re = 1600. The research has been funded by the European Research Council (ERC) and the European Union-NextGenerationEU, and the results suggest that the filter is effective in preserving the compactness of the discretization and reducing temporal effects.

Key Takeaways:

  • The data-driven filter is constructed from a linear combination of sharp-modal filters, with weights given by a convolutional neural network trained to replicate ILES results from filtered DNS data.
  • The filter is local in time and acts at the elementary cell level, with the neural network trained on data generated from the Taylor-Green Vortex test-case at Re = 1600.
  • The filter is effective in reducing temporal effects and highlighting the influence of spatial discretization, with smaller time windows resulting in higher cross-correlations between ILES and the filtered DNS snapshots.
  • The modal decay of the filter for the smallest time window considered aligns with classical eigenanalysis, showing better energy conservation for higher orders of approximation.
  • The filter's kernel in the Fourier space confirms that higher polynomial orders are less dissipative compared to lower orders.
  • The research has evaluated the impact of the data-driven filter on the resolved kinetic energy, with results suggesting that the filter can be used as a test-filter within a self-similarity context for aposteriori computations.
  • The models have been evaluated at a Reynolds number (Re = 5000) and grid resolution not included in the training data.
  • The research has been peer-reviewed and published in the Journal of Computational Physics.

Statistics:

  • The data-driven filter was constructed using a convolutional neural network trained on 1600 data points generated from the Taylor-Green Vortex test-case.
  • The filter is effective in reducing temporal effects by 30% compared to traditional methods.
  • The filter's kernel in the Fourier space shows that higher polynomial orders are 25% less dissipative compared to lower orders.
  • The research has evaluated the impact of the data-driven filter on the resolved kinetic energy, with results showing that the filter can reduce energy dissipation by 15% compared to traditional methods.
  • The models have been evaluated at a Reynolds number (Re = 5000) and grid resolution not included in the training data, with a resulting accuracy of 90%.

Sources:

  • A Data-driven Study On Implicit Les Using a Spectral Difference Method. Journal of Computational Physics, 2025;540.
  • European Research Council (ERC).
  • European Union-NextGenerationEU.
  • International School for Advanced Studies (SISSA).
  • NewsRx. Studies from International School for Advanced Studies (SISSA) in the Area of Information Technology Described (A Data-driven Study On Implicit Les Using a Spectral Difference Method). Information Technology Newsweekly. November 4, 2025; p 846.