[ANSYS Meshing] To ask a question about O-grid generation

lnk · July 11, 2012, 11:40

Quote:

Originally Posted by flotus1

You have to ask yourself one question:

Is it important that the geometric feature (the sharp angle) is represented by the mesh?

Or wouldn't the flow field be almost exactly the same if the sharp angle is not represented by the mesh? In this case: change the geometry to allow better meshing.

If the mesh doesn't follow the geometry, what should I associate the edges at the connection part to? To the circle or the rectangle? I tried both of them but neither of them shows a good result. If I just leave them there without associating, it turns to the result of the picture above (at the previous post) which is same as associating to the rectangle. At this attachment, you can see what if I associate it to the circle. The mesh at some places are missing.

diamondx · July 11, 2012, 13:29

that's what i'm trying to fgure out ...

diamondx · July 12, 2012, 00:04

a little update...
Here is what i tried today:

Only one problem: i couldn't do this blocking after i changed the geometry by keeping only the tube and the small box in the middle, i couldn't do it whith the hole geometry. I don't know why...

BrolY · July 12, 2012, 03:53

This blocking is nice but doesn't change the fact that the elements at the conection between the circle and the rectangle are very bad. You can't avoid this, unless you change your geometry.

I think what flotus suggested was to redo your geometry by deleting the triangle parts ... because it might not be relevant for your calculation. And then, redo another blocking which would be simpler.
Flotus, correct me if I misunderstood you

flotus1 · July 12, 2012, 04:03

That is exactly what I meant.

If you give some Information about the flow you want to simulate, then maybe we can figure out how the geometry can be changed without altering the flow field significantly.

Far · July 12, 2012, 05:37

We can modify the blocking in the corner inside the box , but i think there is no option at the two extreme corners where lines have to be tangent to circle.

We can use options:

1) modify the geometry

2. apply tetra meshing in the box surrounding the circular pipe.

lnk · July 12, 2012, 06:24

Quote:

Originally Posted by flotus1

That is exactly what I meant.

If you give some Information about the flow you want to simulate, then maybe we can figure out how the geometry can be changed without altering the flow field significantly.

As in this picture, the flow is from the z direction in the tube. Some go y direction, some go -x direction, some still go z direction in the tube. The flow speed is quite low. The flow regime is basically laminar.

Best regards and many thanks,
lnk

diamondx · July 12, 2012, 10:37

Quote:

This blocking is nice but doesn't change the fact that the elements at the conection between the circle and the rectangle are very bad. You can't avoid this, unless you change your geometry.

you guys are right, i'm stubborn...

BrolY · July 12, 2012, 10:50

It's a quality to be stubborn when you have to deal with blocking

lnk · July 18, 2012, 15:41

Hi. Here is the problem we talked about a lot. I'm thinking about why don't we use domain interface to solve this problem? We don't have to make the nodes match at the connection part any more with the domain interface at CFX.

Since I'm fresh to this method, I'm wondering is the domain interface method accurate or not? By this method we can solve every geometry very easily. But life shouldn't be that easy. What's the drawback of the domain interface problem?

Best regards,
lnk

flotus1 · July 18, 2012, 16:30

If you can wait until tomorrow, I can show you a picture which illustrates the drawback of non-conformal interfaces.
The accuracy is poor and the interpolation slows down the solution process.
Before adding interfaces, i would rather choose tet or poly meshes.

Edit: Here are the results of a simple heat conduction case with nonconformal interfaces. There are three interfaces in X-direction. You can clearly see the discontinuities in the temperature distribution, even when the mesh size on both sides of the interface is similar.

lnk · July 19, 2012, 04:14

Quote:

Originally Posted by flotus1

If you can wait until tomorrow, I can show you a picture which illustrates the drawback of non-conformal interfaces.
The accuracy is poor and the interpolation slows down the solution process.
Before adding interfaces, i would rather choose tet or poly meshes.

Edit: Here are the results of a simple heat conduction case with nonconformal interfaces. There are three interfaces in X-direction. You can clearly see the discontinuities in the temperature distribution, even when the mesh size on both sides of the interface is similar.

Thank you very much for your answer. May I ask if the accuracy is always this bad, in what case should we still use it?

Best regards and many thanks,
lnk

lnk · July 26, 2012, 04:19

Hi

I'd like to test GGI to solve the problem. May I ask which button may I use to unmatch the mesh across the connection part?

Best,
lnk

flotus1 · July 26, 2012, 04:30

The two faces of the meshes you are trying to join via an interface have to contain non-conformal nodes.
If the meshes are identical on the two faces, fluent matches the nodes automatically and creates an internal interface. I don't know how to prevent this.

So the easiest way is to create meshes with non-matching nodes at the interface.

Concerning your first question: as always, it depends on what you are trying to simulate. Non-conformal interfaces are definitely a no-go in a LES for example, but for the steady-state calculation of a global parameter like pressure drop between inlet and outlet the interpolation might be no problem.
As you can see, the question cannot be answered universally, so checking the influence for a specific case is definitely a good idea.

July 11, 2012, 13:29		#22
diamondx Super Moderator Ghazlani M. Ali Join Date: May 2011 Location: Tokyo, Japan Posts: 1,385 Blog Entries: 23 Rep Power: 28	that's what i'm trying to fgure out ... __________________ Regards, New to ICEM CFD, try this document --> https://goo.gl/KAOIwm Ali

July 12, 2012, 00:04		#23
diamondx Super Moderator Ghazlani M. Ali Join Date: May 2011 Location: Tokyo, Japan Posts: 1,385 Blog Entries: 23 Rep Power: 28	a little update... Here is what i tried today: Only one problem: i couldn't do this blocking after i changed the geometry by keeping only the tube and the small box in the middle, i couldn't do it whith the hole geometry. I don't know why... __________________ Regards, New to ICEM CFD, try this document --> https://goo.gl/KAOIwm Ali

July 12, 2012, 04:03		#25
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,399 Rep Power: 46	That is exactly what I meant. If you give some Information about the flow you want to simulate, then maybe we can figure out how the geometry can be changed without altering the flow field significantly. BrolY likes this.

July 12, 2012, 10:50		#29
BrolY Senior Member AB Join Date: Sep 2009 Location: France Posts: 323 Rep Power: 21	It's a quality to be stubborn when you have to deal with blocking diamondx likes this.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Grid Generation	Raghunandan K.V.	Main CFD Forum	2	October 4, 2008 09:26
cartesian grid generation method	Abu Taleb	Main CFD Forum	7	April 14, 2001 09:49
Cartesian grid generation method	Abu Taleb	Main CFD Forum	0	April 8, 2001 12:15
Cartesian grid generation method	Abu Taleb	Main CFD Forum	0	April 8, 2001 12:03
grid generation help	zhe zhang	Main CFD Forum	2	November 12, 1999 22:48

July 12, 2012, 03:53		#24
BrolY Senior Member AB Join Date: Sep 2009 Location: France Posts: 323 Rep Power: 21	This blocking is nice but doesn't change the fact that the elements at the conection between the circle and the rectangle are very bad. You can't avoid this, unless you change your geometry. I think what flotus suggested was to redo your geometry by deleting the triangle parts ... because it might not be relevant for your calculation. And then, redo another blocking which would be simpler. Flotus, correct me if I misunderstood you

July 12, 2012, 05:37		#26
Far Super Moderator Sijal Join Date: Mar 2009 Location: Islamabad Posts: 4,553 Blog Entries: 6 Rep Power: 54	We can modify the blocking in the corner inside the box , but i think there is no option at the two extreme corners where lines have to be tangent to circle. We can use options: 1) modify the geometry 2. apply tetra meshing in the box surrounding the circular pipe.

July 18, 2012, 15:41		#30
lnk Senior Member lnk Join Date: Feb 2011 Location: Switzerland Posts: 118 Rep Power: 15	Hi. Here is the problem we talked about a lot. I'm thinking about why don't we use domain interface to solve this problem? We don't have to make the nodes match at the connection part any more with the domain interface at CFX. Since I'm fresh to this method, I'm wondering is the domain interface method accurate or not? By this method we can solve every geometry very easily. But life shouldn't be that easy. What's the drawback of the domain interface problem? Best regards, lnk

July 26, 2012, 04:19		#33
lnk Senior Member lnk Join Date: Feb 2011 Location: Switzerland Posts: 118 Rep Power: 15	Hi I'd like to test GGI to solve the problem. May I ask which button may I use to unmatch the mesh across the connection part? Best, lnk

July 26, 2012, 04:30		#34
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,399 Rep Power: 46	The two faces of the meshes you are trying to join via an interface have to contain non-conformal nodes. If the meshes are identical on the two faces, fluent matches the nodes automatically and creates an internal interface. I don't know how to prevent this. So the easiest way is to create meshes with non-matching nodes at the interface. Concerning your first question: as always, it depends on what you are trying to simulate. Non-conformal interfaces are definitely a no-go in a LES for example, but for the steady-state calculation of a global parameter like pressure drop between inlet and outlet the interpolation might be no problem. As you can see, the question cannot be answered universally, so checking the influence for a specific case is definitely a good idea.