Skip to content

ENH: EBRSM: new elliptic-blending Reynolds-stress turbulence model

Kutalmış Berçin requested to merge feature-elliptic-blending-reynolds-stress-model into develop

Acknowledgement

OpenCFD would like to acknowledge and thank Prof. Rémi Manceau, Dr. Michael Karl Stoellinger, and Dr. Ardalan Javadi for their contributions, elaborate suggestions and help, and critical recommendations. Highly appreciated.

Aim

Implement and evaluate the elliptic-blending Reynolds-stress turbulence models proposed by Manceau (2015) - Appendix C.

Methodology

  • Plane channel flow at ReTau=180, 395, 590 (Moser et al., 1991) and =4179 (Lozano-Duran & Jimenez, 2014)
  • NASA Turbulence Modelling Resource on various physics:
    • 2DCC: 2D Convex curvature boundary layer
    • 2DML: 2D Mixing layer
    • 2DB: 2D Bump-in-channel
    • 2DZP: 2D Zero pressure gradient flat plate
    • 2DANW: 2D Airfoil near-wake

Results

Plane channel flow, ReTau=180

Plane channel flow, ReTau=4179

2DCC: 2D Convex curvature boundary layer

2DML: 2D Mixing Layer

Meta-data

EP#1805

  • linux64ClangDPInt32Opt (clang11)
  • linux64GccDPInt32Opt
  • linux64GccSPDPInt64Debug
  • Alltest: No new error

Discussion

  • In general, EBRSM yields better predictions for R and turbulence quantities in comparison to kOmegaSST model.
    • But not always: For example, for 2DML: 2D Mixing Layer, the R predictions of EBRSM is worse than those of kOmegaSST. The reason seems to be that the EBRSM could not reach the target convergence levels for this specific case.
  • Predictions for U are similar.
  • In general, less stable than kOmegaSST.
    • For example, 2DZP: 2D Zero pressure gradient flat plate and 2DANW: 2D Airfoil near-wake cases are unstable.
  • Initial values/initialisations seem to be important to EBRSM in terms of numerical stability and fidelity.
    • The two-step automatic initialisation method proposed by (Manceau (n.d.)) should be preferred over precursor simulations.
  • Low-quality mesh cases are challenging, especially meshes with high-aspect ratios are prone to instabilities.
  • Max relaxation factor for R=0.4. Avoid R~O(0.05) at all costs.
  • Realizability conditions are not automatically satisfied.
  • Multiphase and compressible cases are not explored by the academia.

Future work

  • Test scope should be extended further for:​
    • Multiphase-flow cases​
    • Compressible flow cases
    • Dynamic-mesh cases​
    • Overset meshes​
    • Mesh (un)refinements​
    • Collated-data format​
    • Hybrid and single precisions​
  • New methods should be developed to stabilise EBRSM for low-quality meshes.
Edited by Kutalmış Berçin

Merge request reports