CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   FLUENT (http://www.cfd-online.com/Forums/fluent/)
-   -   Grid Independence Study (http://www.cfd-online.com/Forums/fluent/106724-grid-independence-study.html)

Vidit Sharma September 6, 2012 09:18

Grid Independence Study
 
1 Attachment(s)
hello,
I am doing a simulation of film cooling and need to complete its grid independence study. I have completed four grids and the then plotted its effectiveness versus X/D. U can see the plots in the attached image.
So, can somebody tell me I need to make more grids or i can select the grid with minimum number of cell???

thank you.

RodriguezFatz September 6, 2012 09:31

What means "effectivenes" ?
You need to plot the differences between the cases, not the cases.
Richardson Extrapolation can help to estimate the error.

Vidit Sharma September 6, 2012 09:36

Quote:

Originally Posted by RodriguezFatz (Post 380562)
What means "effectivenes" ?
You need to plot the differences between the cases, not the cases.
Richardson Extrapolation can help to estimate the error.

thank you sir for the reply..

So, u r saying that I hav to take the differences of the results and then plot them.? Like the difference between the results of Grid2 and Grid4 is about the order of 1*10^-3, then its acceptable to take the grid with minimum cells ?

RodriguezFatz September 6, 2012 09:41

No. You need to see the differences between your solutions to be able to judge about the grids. I can't see anything in your picture.

What is the difference in your four grids? 20, 24, 28... is this the number of gridpoints?

Vidit Sharma September 6, 2012 09:44

Grid 1 is of 20 Lakhs
Grid 2 is of 24 lakhs
Grid 3 is of 28 lakhs
Grid 4 is of 37 Lakhs
Is it fine to take these grids or i need more finer grids?

flotus1 September 6, 2012 09:50

Plot the efficiency at specific X/D ratios over the cell size (not the number of cells). Then you can judge grid dependency of your case.

BTW: what is lakh?

RodriguezFatz September 6, 2012 09:53

I didn't know that "lakh" is an english word, up to now. My dictionary says, it is a "hundred thousand" in India, is that right?

For grid refinement study, You need to refine your original grid in each dimension by some factor. If you can, use a factor of 2 in each dimension. That means your 2nd grid has 2*2*2 = 8 times the grid cells than your 1st grid. Your 3rd grid will have 64 times the grid points than your first grid.
Than have a look at your values of interest.
Is that possible to you?

Vidit Sharma September 6, 2012 10:01

Sorry, grids are having 2, 2.4, 2.8, and 3.7 million cells.
Yeah Its possible to do refine the grids with the factor u r suggesting but it will take more time to converge. If now u r able to under stand about no. of cells in the grids which i am using. What else I need to do?

RodriguezFatz September 6, 2012 10:03

Longer version:
In grid refinement, You need to know, if your coarsest grid can give you a sensefull solution at all. For it, you need to refine your grid at least two times and see, if your solutions keep converging into the same direction, i.e. keep growing or decreasing during refinement. If your 2nd grid gives a larger solution than your 1st grid and your 3rd grid gives a smaller solution than your 1st grid - you are fucked up. Then, your coarsest grid is garbage for sure and you have to refine again. If your solutions keeps moving into the same direction - it can but does not have to be safe.

Then, you have to look at how the difference between (first - second) and the (second - third) solution develops. If the difference gets much smaller in the (second - third) case, you are safe. Otherwise you are not close to convergence.

I recommend to read about Richardson extraploation for CFD.

Vidit Sharma September 6, 2012 10:07

Thank U sir ..
Now I get it ... Surely, I will read Richardson extraploation for CFD.
:)

RodriguezFatz September 6, 2012 10:10

If you can, get a copy of "Computational Methods for Fluid Dynamics" by Ferziger and Peric. It's explained perfectly there.

Vidit Sharma September 6, 2012 14:10

Quote:

Originally Posted by flotus1 (Post 380567)
Plot the efficiency at specific X/D ratios over the cell size (not the number of cells). Then you can judge grid dependency of your case.

BTW: what is lakh?

sory, grids are having 2, 2.4, 2.8 and 3.7 million cells...

flotus1 September 7, 2012 02:50

again what learned:D


All times are GMT -4. The time now is 08:46.