More Cells = Better Results? Biggest CFD Myth
CFD Toolbox
0:00 / 0:00
More Cells = Better Results? Biggest CFD Myth
111 просмотров · 7 дн. назад
CFD Toolbox
128 подписчиков
111 просмотров · 7 дн. назад
More cells does not mean better results - and the reasons are more interesting than the slogan.
This video builds the honest error budget of a CFD solution. Truncation error falls as you refine; that is the part everyone knows. But error terms can also cancel on a specific mesh, handing you an artificially accurate answer that gets WORSE when you refine and the cancellation breaks - non-monotonic convergence, drawn here from a computed dataset rather than asserted. And modelling error - turbulence model, boundary conditions, geometry simplifications - sets a floor that no amount of cells lowers.
Then the method that separates engineering data from unverified claims: the three-grid Grid Convergence Index, worked end to end on screen with every number derived - refinement ratio 1.5, observed order computed from the three solutions, Richardson extrapolation, GCI on the fine grid, and the asymptotic-range checks that the bare GCI number cannot give you. Plus the caption the method deserves: GCI bounds discretisation error. It is verification, not validation - it cannot see a wrong turbulence model, and a shared mesh-quality bias can converge cleanly to a wrong value.
Also: the real cost of uniform refinement in 3D (8x cells, 16x with the explicit time step, 256x for two levels); the y+ trap, with the wall-function band drawn from the actual log-law; why compensating for first-order schemes with more cells is a bill you cannot pay (first-order smearing shrinks only linearly with cell size - ten times less smearing costs a thousand times the cells in 3D, while second order is a scheme switch, not a mesh bill); and targeted, adjoint-guided refinement - cells where the flow demands them.
Software-agnostic. The numerics, not the buttons.
Timestamps:
0:00 The 50-million-cell disaster
0:37 The two enemies: truncation vs round-off
1:26 Error cancellation: right answer, wrong reason
2:02 The camera resolution analogy
2:40 The 8x cost trap in 3D
3:21 What mesh quality actually means
4:01 The y+ trap
4:43 First-order schemes: the silent killer
5:23 The three-grid method: GCI
6:11 Targeted refinement
6:58 The checklist
7:50 Think like a surgeon
Key References:
Roache, P.J., Verification and Validation in Computational Science and Engineering
ASME V&V 20 (Standard for Verification and Validation in CFD and Heat Transfer)
Ferziger & Peric, Computational Methods for Fluid Dynamics (error decomposition)
Eca & Hoekstra, J. Comput. Phys. 262 (2014) (discretisation uncertainty estimation)
Fidkowski & Darmofal, AIAA J. 49 (2011) (output-based / adjoint mesh adaptation review)
ANSYS Fluent and OpenFOAM user guides (quality metrics, wall treatment)
#CFD #ComputationalFluidDynamics #Meshing
---
Disclaimer: This video is for educational and informational purposes only and does not constitute professional engineering advice. Always consult a qualified professional for project-specific requirements. Codes and standards vary by jurisdiction.