- **Main CFD Forum**
(*http://www.cfd-online.com/Forums/main/*)

- - **mesh size and length scale
**
(*http://www.cfd-online.com/Forums/main/4102-mesh-size-length-scale.html*)

mesh size and length scale
Hello, All:
Does any one know any article that discusses the mesh size in CFD simulation and the length scale of turbluence. What is the minimum mesh size in terms of the local length scale? Thanks. Min-Hua Wang |

Re: mesh size and length scale
What type of flow are you considering? Laminar or Turbulent? Homogeneous or not?
What is your simulation tool ? DNS or LES ? What is the capacity of computer you can use? In my case, I'm doing LES of turbulent flow around square cylinder. Grid scale is determined based on Kolmogorov scale. I'm using Cartesian grid without adaptation and Cray SV1 (8 CPU). When Re=22000, I use about 100000 cells. I don't use explicitly parallel scheme. About 3 week, I can get reasonble output data. With this data, I can get time averaged results having reasonable accuracy. Now I try 8 times bigger cell number case to see fully 3-dimensional phenomena. With current numerical scheme and SV1 computer, it will take more than 6 months to get time-averaged data. Actually it is not my thesis topic but future research topic. So I just waiting the results. Actually there are several kinds of scales. Such as Taylor micro scale, Kolmogorov length scale, etc. But not easy to understand the physical meaning. Even though you have the information of the smallest scale of your flow,It is also difficult to discretize your computational domain accurately. It is one of the biggest topic in CFD. (Yes. if your computational domain is very simple. You can discretize the domain easily.) One of the reasonable approach is just try and see the results. And modify your resolution until you get reasonable results. |

Re: mesh size and length scale
There is an article by Moin and Mahesh in Annual Review of Fluid Mechanics 1998 on Direct Numerical Simulation. They discuss the grid requirements. The mesh spacing is also dictated by the numerical scheme that you are using. For spectral representations, I think most DNS studies use something like (Delta x) = 1.5*eta.
chidu... |

All times are GMT -4. The time now is 10:41. |