Gird independency study
 

Hi all,

I am simulating a case with RNG k-eps model together with Enhanced wall treatment, i've checked and sure all the values of y plus are less then 1,after the solution is converged, i don't know should i generate finer mesh to recalculate it again? can anybody answer me why should we do grid independency study? if it's so, should i refine mesh to which extent, e.g. add more nodes close to the wall or other parts of the computational domain?, what is the ratio between the old mesh and the new finer mesh?

Thanks for any advice!

The easiest way to do is perform simulations on N*N*N 2N*2N*2N .. upto three to four levels of grids. Just ensure that y+<=1 for the cases. Check the paper by Davison and Peng on k-omega model. They performed grid independence study.

Hello,

The previous poster has already given excellent advice with regards to method, i.e. doubling the number of nodes in each direction; however, I would like to address the why part of your question.

Error is present in your solution as a result of the space between your nodes. As such the results you produce can change as a function of the mesh used. What grid independence shows that the errors from the grid spacing are small enough that they are no longer significant and that your solution has stopped changing. This applies through your entire domain.

For instance you may find that as you change your mesh the y+ value of your first interior node will also change. In my own opinion, solutions that are not grid independent are similar to solutions that have not converged in that they are not fully solved.

I hope I have been helpful. Have a good day.

CAR

Thanks very much for both of your reply!Now it's become clear!

Hi harishg, could you tell me the detailed name of the paper you suggesting me to read?

Hi,

I just want to tell you that from a mathematical point of view obtaining a grid convergence has no meaning (for turbulent solutions). Let split the problem in two cases depending on the flow behavior:

For DNS: it is possible to perform a such grid convergence and this check is anyway demanded by each reviewer when you want you paper get published... I agree with your previous posting!

For Turbulence: the situation is somehow different, i.e., it appears an interaction between your discretization error(mesh size and time stepping) and the MODDELING ERROR (you know that each turbulence model introduces its own modelling error). Of course, in a unsteady simulation these errors do interact and sometimes even that you double your number of mesh points you will get wrong results... Therefore, today in order to validate numerical results it is the demand not to perform a grid convergence but to compare them with experimental data. Nevertheless, today one of the main directions in the research area is to quantify the interaction between these two errors.

I hope that you have now a global overview.
Ioan