CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   ANSYS Meshing & Geometry (http://www.cfd-online.com/Forums/ansys-meshing/)
-   -   [ICEM] 32G memory still not enough? (http://www.cfd-online.com/Forums/ansys-meshing/110304-32g-memory-still-not-enough.html)

foolboy007 December 7, 2012 12:35

32G memory still not enough?
 
The most powerful desktop I can access has 32GB memory.
but I still hit the limit when I try to mesh the geometry to get y+=1.
I also tried to use coarse mesh and split the prism to get y+=1. However, that gives me very poor quality mesh.

Could anybody give me some suggestions?

energy382 December 7, 2012 15:11

which kind of simulation, how many elements (total) etc.?

foolboy007 December 7, 2012 15:23

Quote:

Originally Posted by energy382 (Post 396329)
which kind of simulation, how many elements (total) etc.?

Fluent RSM model.
Rightnow, ICEM CFD shows 140M Cells. It might have more cells because it reached the memory limit and kept reading hard drive.

diamondx December 7, 2012 16:27

140M cells is too much... even for a 32 Go

foolboy007 December 7, 2012 18:14

Quote:

Originally Posted by diamondx (Post 396337)
140M cells is too much... even for a 32 Go

if I make a coarse mesh and keep y+=1,I have to have high aspect ratio cells.
Does that have any problem with RSM model?

energy382 December 7, 2012 19:20

Why do you use RSM?
This model is very complex and difficult to resolve numerical.

But anyway.....140 M cells?!!! That's definitely too much for 32GB

mjgraf December 7, 2012 19:46

not sure if the fluent solver is the same, but the CFX solver best practice guideline states

Aspect ratios away from the boundary layer should stay below 50.

For well resolved boundary layers
at high Re numbers, the near-wall aspect ratios can be of the order of 10^5-10^6
.

Then, of course, you need to use the double precision solver.

The only issue this causes is the quality reports that are run flag this as a poor mesh.

Quote:

Originally Posted by foolboy007 (Post 396297)
The most powerful desktop I can access has 32GB memory.
but I still hit the limit when I try to mesh the geometry to get y+=1.
I also tried to use coarse mesh and split the prism to get y+=1. However, that gives me very poor quality mesh.

Could anybody give me some suggestions?


foolboy007 December 8, 2012 00:58

Quote:

Originally Posted by energy382 (Post 396348)
Why do you use RSM?
This model is very complex and difficult to resolve numerical.

But anyway.....140 M cells?!!! That's definitely too much for 32GB

coz the near wall velocity significantly decreases at y+=1-4.
particle deposition rate would be high (wrong) if y+>3.

Far December 8, 2012 01:51

which RSM model you are using? It has been shown repeatedly that RSM model is not always better than the two equation models.

Which effect you want to model with the RSM model? How many no. of layers you are using in boundary layer and why? What about the steam-wise resolution?

It looks like to me a DNS mesh !!! BTW what is the Reynolds number you are trying to model?

foolboy007 December 9, 2012 00:36

Quote:

Originally Posted by Far (Post 396373)
which RSM model you are using? It has been shown repeatedly that RSM model is not always better than the two equation models.

Which effect you want to model with the RSM model? How many no. of layers you are using in boundary layer and why? What about the steam-wise resolution?

It looks like to me a DNS mesh !!! BTW what is the Reynolds number you are trying to model?

hi Far,

I used Fluent RSM with enhanced wall function
as I said the particle deposition rate would be high with k-e model because of the over prediction of velocity normal to the wall.

right now, I just arbitrarily choose 3 layers.
I am sorry I do not quite understand what is steam-wise resolution

Far December 9, 2012 00:42

Quote:

Originally Posted by foolboy007 (Post 396470)
hi Far,

I used Fluent RSM with enhanced wall function
as I said the particle deposition rate would be high with k-e model because of the over prediction of velocity normal to the wall.

right now, I just arbitrarily choose 3 layers.
I am sorry I do not quite understand what is steam-wise resolution

I am sorry, it is stream-wise mesh resolution i.e. along length of model.

Only three layers in boundary layer? What is the maximum aspect ratio?

Did you try the SST model with enhanced wall functions and curvature correction. You will almost get the same behaviour

foolboy007 December 10, 2012 17:25

Quote:

Originally Posted by Far (Post 396471)
I am sorry, it is stream-wise mesh resolution i.e. along length of model.

Only three layers in boundary layer? What is the maximum aspect ratio?

Did you try the SST model with enhanced wall functions and curvature correction. You will almost get the same behaviour

I cannot get the ICEM to finish meshing because of the limit.
With k-e with enhanced wall function and coarser mesh, the results were not so good.

I have not try SST. Do you mean the transitional SST?

Far December 10, 2012 23:59

I think Fluent has also implemented different versions of SST model like curvature correction. But CFX has full range of those models.

Just try SST model with well resolved boundary layer and somehow coarse mesh in the other regions.


All times are GMT -4. The time now is 23:00.