how to do a grid independent study
 

Hello everyone

I want to do a grid independent study on my model, the model is only 5mm width, 2D
I add inflation layer on the wall and set the element size in the middle.
what kind of critiria is for the grid independent? After a coarse mesh, decrease the inflation layer and element size by the same ratio?
And for my pc , it can't stand the cell size smaller than 0.08mm. Can I finish the grid independent study on this model?

I have read all the threads about grid independent in this forums, and I also asked my instructors, he has no idea, so please tell me if you know , thank you in advanced.

Systematically decrease the element size by half and quantify the error, such as the maximum value of the streamfunction. For example, suppose there are two regions in the flow, namely, the bulk of the flow and the boundary layer. Start with the default mesh. In the bulk of the flow, decrease the element size and quantify the error. Repeat until the error is within tolerance. In the boundary layer, repeat the same procedure. That is, keep decreasing the mesh until the solution is independent of the grid.

Good luck!

how to know that the solution is independent of the grid?

Untill your result comes to the value within an acceptable error compared to the last one.

Hi
I have simulated flow over solids.I have applied periodic boundary conditions in GAMBIT.I have run the case in FLUENT and the case got converged.After I change the grid size to finest and most coarser, the values far away from the converged case ,I see the same results .Why is this happening?

What tolerance are you setting for your convergance criteria.
Increase it a little bit, say 10e-5 and then see your results.

Grid Independence Study
 

In ANSYS CFX i am performing thermal analysis of motor. I have done meshing for the motor in HYPERMESH tool and exported file to ANSYS CFX. I have selected the element length size of 5 mm and i got around 3.5 million elements.After initializing domain interfaces, boundary conditions i have run the simulation and got the results. In order to verify the results, again in HYPERMESH for the same meshed model i just increased the number of elements by re-meshing the model. Again in CFX after simulation i am getting the same results. The result won't vary what i got from the previous meshed model. Does this mean i have reached mesh independence study and got the expected results?
Please give me your feedback, I am new to CFD. Since i am from the electrical background i don't have much idea about the mesh independence study.

Hello,

I am new to fluent. I have the following mesh file. What command do I use to do a mesh independent study? Can anyone please tell me?

There is no command that you can use. You will have to manually refine the grid and see if the parameter you intend to measure varies.

this is a very good article about a methodology using Richardson extrapolation
https://www.grc.nasa.gov/www/wind/va.../spatconv.html
also you can read the book from p roache 'verification and validation in computational science...'

I recommend to go back to the mesher and generate a new grid. But if that is not possible then use the adapt feature. Mark whatever region (or all regions) to adapt. The default level of refinement is 2.

Grid Independence Study
 

So what's the deal for 3D meshes with polyhedral elements? J.D Anderson in his book mentions simple parameter comparisons.

For example,

Mesh with 31 elements, a=0.534
Mesh with 62 elements, a=0.533
Exact analytical solution, a=0.528

But in my case I don't have the luxury of an analytical solution. Except for the knowing the exact mass-flow-rate.

Thanks guys

For real problems you can assume that there never is any analytical solution in CFD. That is because if you had the analytical solution, then you wouldn't be doing CFD to figure out what he solution is.

But that is why you do something like a richardson extrapolation to estimate what the solution is. It doesn't have to be a richardson extrapolation, it could be something better.

[QUOTE= It doesn't have to be a richardson extrapolation, it could be something better.[/QUOTE]

when you say something better what do you mean? Do you know of a better procedure?

Any type of sequence accelerator is appropriate. Richardson extrapolation is just one simple method that gives you O(n+1) from an O(n) method. It is attractive because often it is easier to do the O(n) method more times than to do the O(n+1). But Richardson extrapolation is only a linear sequence transformation and it uses only information from say the n and n-1 terms to predict the n+1 term.

The Aitken's delta-squared method for example is a non-linear sequence transformation, again using only O(n) information but can give you much more information because it uses information from the n, n-1, and n-2 terms in the sequence. Thus is has more long range order.

The idea is not that you must use Richardson extrapolation always, it is simply a good place to start. There are entire family of methods that are available. We are just scratching the surface with Richardson & Aitken's methods.

The obvious analogy is you don't always use 1st order upwind or 1st order Euler discretization. Eventually you move on to higher order and more accurate methods when it becomes appropriate to do so.

Yes, your simulation results are independent of the grid size used, provided you used the same settings for both cases