Hi everyone,

I'm trying to carry out a grid independence study for a simple cylindrical pipe of 150mm length. I'm using an inlet velocity (z velocity) of 1.5 m/s and the k epsilon turbulence model. The mesh is a hexahedral cooper mesh. I find that with some 450,000 cells (a very fine grid for a pipe of radius 25mm and length 150mm), the result of pressure is only just getting to be grid independent. I simply don't understand why so many cells are required! Am I doing something fundamentally wrong for such a fine grid to be necessary? Although I must say, for my case the error obtained at about 155,000 cells is acceptable. So I'm going to be using that grid size.

The reason I'm baffled is that, with a complex pipe geometry (a lobed twisted pipe) of same length I get the order of accuracy I require at about 55,000 cells.

Please help me understand this....

If you use k-eps model with wall functions, do you respect the condition on y+ with such a fine grid?

Hi

ap

Hi,

I wanted to try to get the result without adapting the grid at the walls. The plan was to apply wall adaption at a later stage (because I'll be using this pipe connected to more complex pipe geometry later) since at the moment I'm only looking for rough results to see when I get near enough to a grid independent result.

I used standard wall function and didn't do anything with the y+. D'you think I should adapt the grid at the walls to get y+ in the region of 30-60?

I guess this might get me results close enough to the grid independent case with fewer cells?

thanks in advance, chanchala

Before attempting to obtain a grid independent solution, be sure to respect all conditions required by the model you use.

If y+ is not between 30 and 60, the wall function is not valid and this may influence the solution and its stability.

From the data I know on your case and your grid (which seems really fine), I'd say y+ is lower than 30, isn't it?

If so, start with a coarse grid, adapt it to satisfy the y+ condition, and then work on the grid-independency of the solution always checking the y+ condition.

Hi

ap

Hi

I was wondering that Is it a converged solution?, I mean all the residual are flat and values no longer changed.

Have you checked the Grid Quality?

Skewness(in Gambit) should be lees than 0.65 for hexahedral cell and 0.8 for tetrahedral cell)

Bambo

Hi,

Bambo- yes, I get a converged solution and the skewness is less than 0.6.

ap- I can't coarsen the grid in fluent using y+ adaption. It will only allow me to refine. How can I get around this? Should I coarsen the grid at the wall using another way (other than y+), if so what do you suggest?

Also, I thought the finer the grid at the walls the better the solution. Please explain to me why I should therefore attempt to coarsen the grid to get y+ atleast 30?

Thanks for your suggestions....

You told you're using k-esp turbulence model. If you adopt wall functions for the wall treatment (default option in FLUENT), you have to use a grid which respect the condition 30 < y+ < 60.

Wall functions are based on empirical considerations, and are not valid for values of y+ outside of this range. If you want to use a finer grid, just enable the Enhanced Wall Treatment, which requires y+ close to 1.

If y+ adaption doesn't work, just create a coarser grid in GAMBIT and start with it, but pay attention not to refine the wall region (see the manual and grid considerations on y+ for turbulence models).

Hi

ap

Hi ap,

Thanks for all your help. I'm going to have a look through the Fluent manual again regarding y+ adaption and wall functions and afterwards I'm sure I'll have a few questions for you, so please make sure you check here again in a day or two.

For the moment, I managed to get to grid independence for the simple cylindrical pipe case by using a coarser grid and refining using y+ adaption as you said.

However I have another complex pipe geometry where it's a lobed pipe to start with and the cross-section changes gradually to a circular pipe. Now in this case (using a standard wall function and k epsilon again) I can't seem to achieve grid independence unless I use a million odd cells (its the same size as the simple pipe). I did y + adaption initially but then as the grid got finer and still not grid independent, I can't because when the grid is that fine the y+ value for most of the cells at the wall tends to be lower than 30 (and fluent doesn't coarsen the grid using y+)! So I guess the solution is unstable as you suggested. I'll think about your suggestion of using enhanced wall function here.

Also, I haven't tried other turbulence models for this particular geometry so it may be that k epsilon is not sufficient.

Use non-equlibrium wall functions, which are sensitized to pressure gradient to obtain the grid-independent solution.

Read Wilcox, "Turbulence modeling for CFD" should be the title, if I'm not wrong.

Hi

ap

Hi Of course you can't coarse the grid in Fluent and u shouldn't coarse at walls either. Yes u are right the finer the grid at the walls the better the solution.

What is percentage error you are getting.

I think you have to mesh the pipe carefully. Don't just jump into cooper method. First you can mesh the surface then the volume of the pipe. Also try the second order solution for comparison.

You can read the "Guide for the Verification and Validation of Computational Fluid Dynamics Simulation", AIAA (1998).

Bambo

Hi Standard wall function only applicable between 30 < y+ < 150. So if your model y+ is too small (y+<30 then the Standard wall function is no longer applicable).

You can use option 3in k-e model, "Enhanced Wall Treatment", for new wall treatment (if the near-wall mesh is fine enough to be able to resolve the laminar sublayer y+<5). Otherwise, You could control number cells in wall by defining boundary layer.

Bambo

Standard wall functions are appliable only if y+ is between 30 and 60.

Enhanced wall functions requires y+ approximately 1.

Hi

ap

Fluent is able to coarsen the grid, you just have to choose the right option in the Adapt menu.

The principle "The finer the grid, the better the solution" is true only if the model is valid on such a fine grid and if you haven't reached the grid independent solution.

Hi

ap

Hi As far as I know you can only coarse the adapted mesh only. You CANNOT coarse the original mesh at all in Fluent.

Bambo