Table of Contents
Fetching ...

Unifying formalism and closures for coarse-grained approaches to turbulence

A. Cimarelli, N. Marras, B. Niceno, Y. Tessier Urrecha

TL;DR

This work introduces a temporally filtered Navier–Stokes framework that unifies statistical (RANS) and scale-resolving turbulence closures by treating turbulence-induced smoothing as an explicit time filter $\langle \cdot \rangle_\tau$. It derives a dynamic RANS closure from a two-filter identity, yielding an adaptive eddy-viscosity coefficient $c_\mu$ via $c_\mu\,\mathcal{M}_{ij}=\mathcal{L}_{ij}$ with $c_\mu = \dfrac{\mathcal{M}_{ij}\mathcal{L}_{ij}}{\mathcal{M}_{ij}\mathcal{M}_{ij}}$, applicable to any two-equation closure. The approach demonstrates improved flow physics for transitional and separated flows (e.g., laminar separation, transition, reattachment) on both fine-scale and coarsened grids, and even enhances statistical steady-state predictions in 2D. The results indicate robust, regime-spanning performance and reveal how the dynamic coefficient adapts to near-wall and bulk turbulence, enabling better capture of laminar–turbulent transitions and sensitivity to free-stream turbulence. Overall, the temporally filtered paradigm provides a solid foundation for advancing coarse-grained turbulence modeling and suggests directions for extending the methodology to other closures and flow regimes.

Abstract

We propose the use of an unifying paradigm for the assessment and development of closed forms of the coarse-grained Navier-Stokes equations in approaches ranging from the statistical to the scale-resolving ones. It consists in the exact formalism provided by the temporally filtered Navier-Stokes equations. The fundamental idea is that the smoothing action of turbulent stresses can be described as a temporal filtering operator implicitly applied to the solution. Contrary to the average and spatial filtering operators, the temporal filter is an unifying operator smoothly varying within the statistical and scale-resolving realms. The potential of the temporal filtering paradigm is here highlighted by unveiling relevant algebraic properties and by deriving a new class of turbulence closures. A dynamic procedure is derived and shown to provide an unifying closure for both scale-resolving and statistical approaches. Results show that an improved physics is captured. Challenging phenomena such as the laminar to turbulence transition and the dependence of separation and reattachment on free-stream turbulence applied through boundary conditions are captured.

Unifying formalism and closures for coarse-grained approaches to turbulence

TL;DR

This work introduces a temporally filtered Navier–Stokes framework that unifies statistical (RANS) and scale-resolving turbulence closures by treating turbulence-induced smoothing as an explicit time filter . It derives a dynamic RANS closure from a two-filter identity, yielding an adaptive eddy-viscosity coefficient via with , applicable to any two-equation closure. The approach demonstrates improved flow physics for transitional and separated flows (e.g., laminar separation, transition, reattachment) on both fine-scale and coarsened grids, and even enhances statistical steady-state predictions in 2D. The results indicate robust, regime-spanning performance and reveal how the dynamic coefficient adapts to near-wall and bulk turbulence, enabling better capture of laminar–turbulent transitions and sensitivity to free-stream turbulence. Overall, the temporally filtered paradigm provides a solid foundation for advancing coarse-grained turbulence modeling and suggests directions for extending the methodology to other closures and flow regimes.

Abstract

We propose the use of an unifying paradigm for the assessment and development of closed forms of the coarse-grained Navier-Stokes equations in approaches ranging from the statistical to the scale-resolving ones. It consists in the exact formalism provided by the temporally filtered Navier-Stokes equations. The fundamental idea is that the smoothing action of turbulent stresses can be described as a temporal filtering operator implicitly applied to the solution. Contrary to the average and spatial filtering operators, the temporal filter is an unifying operator smoothly varying within the statistical and scale-resolving realms. The potential of the temporal filtering paradigm is here highlighted by unveiling relevant algebraic properties and by deriving a new class of turbulence closures. A dynamic procedure is derived and shown to provide an unifying closure for both scale-resolving and statistical approaches. Results show that an improved physics is captured. Challenging phenomena such as the laminar to turbulence transition and the dependence of separation and reattachment on free-stream turbulence applied through boundary conditions are captured.
Paper Structure (25 sections, 23 equations, 16 figures, 2 tables)

This paper contains 25 sections, 23 equations, 16 figures, 2 tables.

Figures (16)

  • Figure 1: Finer (above) and coarser (below) grids for the T3L study. On the left, an overall view of the computational domain is reported while, on the right, a detailed view for $-1 \leq x/D \leq 4$ and $0 \leq y/D \leq 1.5$ is reported to better illustrate the grid resolution.
  • Figure 2: Fine-grid scale-resolving simulations of the transitional flow over a rounded leading-edge flat plate. Iso-surfaces of the Q-criterion hunt1988eddies, $Q = 0.15$, colored with the streamwise velocity for (a) the new dynamic RANS closure with $I = 0.35\%$, (b) the standard $k-\varepsilon$ model with $I = 2.3\%$ and (c) the hybrid LES/RANS model with $I = 0\%$. In the inset plots, iso-surfaces with positive streamwise velocity are hidden to reveal the structures of the reverse boundary layer within the main recirculating bubble.
  • Figure 3: Coarse-grid scale-resolving simulations of the transitional flow over a rounded leading-edge flat plate. Iso-surfaces of the Q-criterion hunt1988eddies, $Q = 0.15$, colored with the streamwise velocity for (a) the new dynamic RANS closure with $I = 0.5\%$, (b) the standard $k-\varepsilon$ model with $I = 2.5\%$ and (c) the hybrid LES/RANS model with $I = 0\%$. In the inset plots, iso-surfaces with positive streamwise velocity are hidden to reveal the structures of the reverse boundary layer within the main recirculating bubble.
  • Figure 4: Average behaviour of the model coefficient $c_\mu$ obtained by the dynamc RANS closure for the scale-resolving 3D unsteady simulations performed in the fine grid.
  • Figure 5: Average friction coefficient for the scale-resolving 3D unsteady simulations performed in (a) the fine and (b) coarse grids of the new dynamic RANS closure compared to the high-fidelity implicit-LES simulations performed by Crivellini20.
  • ...and 11 more figures