Hi,

How to obtain a grid-independent solution in Fluent? OR how to ensure that solution obtained is independent of grid/mesh size?

Thanx.

Sharad

Hi,

You have to adapt your mesh e.g. by gradient. Then start the iteration and get a new solution. If the values in your flow field differ only by 0-3 % the solution should by mesh independent.

Armin

Dear Armin,

Thanx for reply. Please elaborate what do you mean by "You have to adapt your mesh e.g. by gradient." Please explain in detail.

Sharad Dugad

I mean you should refine your mesh. For example by gradient or iso value etc. It depends on your problem what is suggestive.

Armin

Basically to ensure a grid-independent solution, you must run each Fluent simulation with two different grids. These grids must be significantly different from one another. That means you cannot just alter one or two cells. If you run Fluent on two completely different grids and you get the same answer, then you are grid-independent. If you get two significantly different answers, then which do you trust? Answer: neither can be trusted since the result depended on the grid (and you must create a 3rd, 4th, 5th,… grid until you find some that give you the same result).

Here are great methods to generate two completely different grids: 1) Use different cell shapes. If you used rectangles in one grid, try triangles with the second grid. 2) Use a lot more cells in one grid than the other. If one mesh used 10,000 cells, have the other mesh use 20,000 cells.

Unfortunately many people don't take the time or effort to do either of those methods. Armin Gips mentioned a method that is quicker and easier, but doesn't guarantee that you are grid-independent. Simulation errors most likely occur where there is a large gradient (where there is a large change in a variable over a short distance in space). Fluent has a menu that will add extra cells where there are large gradients (go to the adapt menu and select gradient). Then you must have a Fluent solution, then choose what variable Fluent use when it looks for gradients, then you must tell Fluent what to consider a large gradient, then you must tell Fluent to Adapt. Fluent will then double the cells anywhere where the variable gradient is large. Finally, you must iterate again. Unfortunately, there is no guarantee that you altered the grid in the right spot - or you may have just altered only one or two cells. However, this is better than doing nothing.

Be realistic when you check for grid independence. Suppose you are interested in a temperature at a point. If that temperature goes from 300K to 300.1 K with different grids then you should realize that 0.1K is not a significant change. If changing the grid altered the temperature from 300K to 302K, then this may or may not be significant (this depends on your required solution accuracy). If changing the grid altered the temperature from 300K to 1000K, then you most likely have really bad grids and need to create a good grid before you can even hope to trust the Fluent results.

I forgot to mention one thing.

A grid independent solution is a must for any simulation. However, it doesn't guarantee that you can trust your simulation results. If you used a bad model, then you will get grid-independent bad results. Only if you use a good model and you get grid-independent results can you trust your simulations.

Dear Scott,

Thanx a lot for your elaborate answers. I would appreciate if you please elaborate, in a similar manner, what Armin Gips meant to say. I did not understand what he mentioned. Could you please explain how to go about his "short-cut" way?

Thanking you again.

Sharad Dugad