CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   FLUENT (https://www.cfd-online.com/Forums/fluent/)
-   -   Mesh size (https://www.cfd-online.com/Forums/fluent/75640-mesh-size.html)

 Mohsin May 1, 2010 11:17

Mesh size

Hello

I got converged result for a 0.2 million cells geometry. Then i changed the mesh size to 0.1 million cells and the result was not converginbg with all the conditions same just i changed the mesh size and i couldn;t get converged solution. Can anyboyd tell what is the problem. If every mesh size has different result then how can one get a validated result if experiments cant be done.

Waiting fro your valued comments on this. thanks
Mohsin

 DoHander May 1, 2010 18:04

What was your simulation about ? Have you used a turbulence model ? What are the conditions for your flow ? I need to know more about your problem in order to give you a suggestion.

Do

 Mohsin May 2, 2010 06:57

I have a verticle cylinder. Coal micron particles are coming with nitrogen gas and it gets out of the cylinder from the 3 outlets. My goal is to get a uniform flow at the outlet. I am using DPM model and KE turbulence model with Realizable model. Velocity inlet is used for teh inleta nd pressure outl;et is used for the outlets.

I used 1st order discretization scheme for that and the result was converged but it didn't give me that accurate result. then i used 2nd order discretization scheme to get more accurate result but it wasn;t converging although i reduced under relaxation factors stil i couldn;t get converged solution. Then I reduced the mesh size from 2 million to 1 million but then it was not even converging with 1st order then.

I really am stuck now. Trying something to get a more accurate solution as 2nd order doesn;t give me converged solution.

My convergence criteria is :

Continuity: 10^-5
X,y,z,k,epsilon=10^-4

Thanks for your time.
Mohsin

 DoHander May 2, 2010 08:53

I'm more of an Aerodynamic's guy, so I can't help you with the physics of your problem. What I know is that if you want to obtain a grid independent solution you start with a coarse mesh on which you've obtained a converged solution (say your 2 mil mesh) and you REFINE this mesh, you actually increase the number of points in your mesh say to 3 or 4 mil and check if you obtain similar results.

Typically a first order solution will converge easier then a second or higher order, but will not give you an accurate result.

What you can try first is to use your initial mesh (2 mil one), start from scratch, calculate until your residual is about 10^-2 with first order and then switch to second order. If you are doing a steady calculation this will work. Do not start from scratch with a second order solution.

Do

 Mohsin May 2, 2010 09:01

Thanks for your time.

Yes You are rite I shud increase the mesh size from 0.25 million to 0.3 milllion and see the effect.

Also I applied the same technique of initially 1st order then second order so many times but cant get converged solution from second order.

I will let you know about my result from 3 million cells which i m doing just now. For 0.25 miilion case my worst element was at 0.77 skewness level. But now i cant reduced the skewness for 0.3 million cells its 0.85 now. Lets see what result it gives for that. Your valued comments will really be of help. thanks

Mohsin

 Chris D May 2, 2010 10:48

How do you know that the solution is not converged? Are you just basing that on the fact that the residuals have not decreased below a certain point, or is there something else going on in the solution? What else are you monitoring besides the residuals?

 Mohsin May 3, 2010 03:01

Actually I m counting the number of particles at the outlet and as the number of partuicles should be 33.33 percent each outlet becasue of symmetric condtions. I was getting 34, 35 and 31 percent particles at the three outlets for 1st order and lower size mesh but when i increased the mesh size i was getting 33.2, 33.9 and 33.0 i-e standard deviation reduced to 0.29.

My question is that how to get grid indepenednt siolution. For example in my case i increased the mesh size from 0.25 million to 3 million then i was getting better result. Should i do it for 4 million and check. If i m getting bad result that would mean that I shud carry on with 0.3 million cells. or if i m getting similar result then it shud be grid indepenednet. Otherwise if i m not getting similar result that would mean i shud increase the mesh size more etc. SHould i do it this way?

 Mohsin May 3, 2010 04:03

I increased the mesh size to .36 million from 0.3 million and the result is not converging even after double iterations (The residuals are considered as a criteria now). So that means I chekcd for

0.1 million cells (residuals not converging below a specified level)
0.25 million (residuals not converging)
0.3 Million (residuals converged and gave quite a good result even 2nd order converged and i got a result having a standard deviation of 0.29. 0 Standard deviation would be ideal as there may be calcuation errors).
0.36 million (the result is not converging based on rediuals but close to that for example at 10^-4 level of residuals I stoped the iterations and checked the Standard deviation which fell down to 0.14 which means its a very good result).

So by seeing this

I should say that as I increased the mesh size from 0.3 millionb to 0.36 million i should also reduce my eresidual level say for continuty from 10^-5 to 5*10^-5.
Or i should work with 0.3 million cells with standard deviation 0.29.

Please Ur valued comments are required. Thanks
Mohsin

 All times are GMT -4. The time now is 22:30.