CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)
-   -   Finite element method - number of elements (http://www.cfd-online.com/Forums/main/11206-finite-element-method-number-elements.html)

poornima April 11, 2006 02:01

Finite element method - number of elements
 
Hello,

I am using 2 dimensional finite element method to determine the temperature on a cylindrical wall in which fluid expands from inlet to outlet.

My solution is dependent on the number of grids (6-20). I understand that the accuracy of the solution depends on the number of grids, of course after a certain number the accuracy will not change.

But when i change the grids by 2 or 3, the solution is absurd, % error is more than 1000 sometimes.

How can I make my solution less sensitive on the number of grids?

Regards Poornima

Angen April 11, 2006 12:19

Re: Finite element method - number of elements
 
Your solution should be grid independent. So either the number of grid is to low for your problem or you have a bug in your program.

Angen


poornima April 12, 2006 07:35

Re: Finite element method - number of elements
 
I agree that the solution should be grid independent. but i am able to find solution for 6X6 grid and 10x10 grid but not for 10X6 grid, what could be the problem in this case. i guess i need to do scaling so that the equations are not a strong functions of number of grids. but i am struggling with how to design the scaling factors. i need some help in this regard.

Poornima

Anindya April 12, 2006 12:08

Re: Finite element method - number of elements
 
Your Grid of 10x10 or 10x6 is really too small for any cfd problem, even in 2D. Please increase your grid to say 100x60 or 100x100 an then compare the solutions.

Márcio April 13, 2006 08:25

Re: Finite element method - number of elements
 
Despite the very coarse grids, it seems the problem is not on the grid dimensions, but on its shape. You only get acceptable results in square grids, right? Maybe you should check your mesh generation and element matrices calculation.

-Márcio

poornima April 14, 2006 01:42

Re: Finite element method - number of elements
 
The geometry in context is an annular cylinder and hence the shape of the grid is cylindrical. I am only trying to discetrize the equations and solve, why should changing the number of grids result in an absurd solution. Surprisingly the problem disappears when I change the dimensions i.e the thickness and the length. Is it that aspect ratio is playing a significant role in the solution. If so, please suggest me how to overcome this problem.

poornima April 14, 2006 07:23

Re: Finite element method - number of elements
 
I suspect that the problem is because of the coupling between deltar and deltaz. because i am able to get solution for 100*6, 6*100, but not for 100*10, I am able to get solution for 10*10, 6*6 , 6*10, but not for 10*6. The ratio of D/L is known as aspect ratio, can any one suggest how to take this into consideration inorder to obtain solution to the problem.

regards Poornima


All times are GMT -4. The time now is 10:27.