# elliptical solver - grid lines going into the surface

 Register Blogs Members List Search Today's Posts Mark Forums Read

January 20, 2014, 04:29
elliptical solver - grid lines going into the surface
#1
New Member

Paul
Join Date: Apr 2011
Posts: 12
Rep Power: 13
Hi,

I am trying to create my 3D volume grids. I am creating the domains that make up the block, specifically I am using the elliptical solver to get orthogonal/smooth grid lines. However, at sharp corners, the grid lines curve and go into the surface. I have attached screen shots of the grid.

I have tried using Laplace as well as Thomas-Middlecoff functions for the interior control function. I have used both Multigrid and SSOR for the solution algorithm. I have tried both of the boundary control functions. The spacing off of the surface is 10x smaller than the smallest surface spacing. The angle control is orthogonal at the surface and "current grid" at the outer boundaries. The blending function for both the spacing and angle is "default exponential" but I have also tried linear blending function.

I can't seem to figure it out. Anything that will fix this problem?

thank you.

-Paul
Attached Images
 close-up.jpg (30.1 KB, 55 views) farview.jpg (97.0 KB, 56 views)

January 21, 2014, 09:54
#2
Senior Member

Chris Sideroff
Join Date: Mar 2009
Location: Ottawa, ON, CAN
Posts: 425
Rep Power: 20
There's no doubt the shape of this domain will make it difficult to smooth. If possible I would suggest an alternate topology (e.g. split the domain into 2). This is further exacerbated by the tight spacing normal to the wall.

It's difficult to make out what the rest of the domain shape looks like but here's a quick tip if you can't/don't want to change the topology.

I notice the grid line coming from the sharp (convex) corner goes straight up. Change the distribution of the points on the opposing edge of the domain so more point are to the right of the sharp corner. Doing this should make it easier for the grid line coming from the sharp corner split the angle more equally.

It's tough to explain this in words so I threw together a quick example.

In the first image, you'll notice the top connector has a lot more points to right of the sharp corner. I created a breakpoint on the top connector directly above the sharp corner. Then I changed the dimension of the sub-connectors such the right sub-connector had about 1.5-2x more points than the left one. That's step 1.

The next step which helps reduce the poor quality cells at the sharp corner is force the smoothing to propagate the orthogonality further into the interior. To do so, on that edge only change the default blending function constant to 24. See the second image. For my example, I found this particularly helpful for the tight near wall spacing.

The last step is to solve the domain. The tip here is to make sure you run a lot of iterations. Because the blending function constant is increased, I find it takes a lot more iterations for the smoothing to settle out. I ran it 100 iterations. See the last two images for the results.

I think this is about as good as you can do without modifying the topology.

Quote:
 Originally Posted by syoo Hi, I am trying to create my 3D volume grids. I am creating the domains that make up the block, specifically I am using the elliptical solver to get orthogonal/smooth grid lines. However, at sharp corners, the grid lines curve and go into the surface. I have attached screen shots of the grid. I have tried using Laplace as well as Thomas-Middlecoff functions for the interior control function. I have used both Multigrid and SSOR for the solution algorithm. I have tried both of the boundary control functions. The spacing off of the surface is 10x smaller than the smallest surface spacing. The angle control is orthogonal at the surface and "current grid" at the outer boundaries. The blending function for both the spacing and angle is "default exponential" but I have also tried linear blending function. I can't seem to figure it out. Anything that will fix this problem? thank you. -Paul
Attached Images
 domain-sharp-corner-01.jpg (76.9 KB, 43 views) domain-sharp-corner-02.jpg (50.3 KB, 40 views) domain-sharp-corner-03.jpg (72.2 KB, 48 views) domain-sharp-corner-04.jpg (41.7 KB, 40 views)

Last edited by cnsidero; January 22, 2014 at 10:39.

 January 23, 2014, 09:18 #3 Senior Member   Chris Sideroff Join Date: Mar 2009 Location: Ottawa, ON, CAN Posts: 425 Rep Power: 20 Paul were you able to resolve the problem?

 January 23, 2014, 13:42 #4 New Member   Paul Join Date: Apr 2011 Posts: 12 Rep Power: 13 Chris, thanks for your reply. I have been trying various things. The blending does improve the mesh quality quiet a bit, however, it doesn't always prevent the grid lines from going into the surface. Redistribution of the points on the top connector does help a lot. When I use the steps you have provided on a 2-d mesh, it works fine, but when I use them on a 3-d mesh I still have same problem on some parts of the grid. I am probably expecting too much from the software and trying to As I have tons of blocks to make, I am trying to be consistent in my grid making method. For the most part, I think I will have to divide up the grid and make changes to the topology. Still in works though. -Paul

January 23, 2014, 16:52
#5
Senior Member

Chris Sideroff
Join Date: Mar 2009
Location: Ottawa, ON, CAN
Posts: 425
Rep Power: 20
Paul,

Here's another approach that builds on the idea from my previous post.

I know changing topology can be a difficult thing, especially in 3D, because once the new topology propagates through the grid system it may not be compatible. You can, however, split the domain at the sharp corner without changing the topology. By splitting it, you introduce a new connector which will give you more control. I'll use my previous example to explain.

Drew a Bezier curve connector from the sharp corner to it's corresponding point on the opposite edge. Modify the slope control points so the angle around that the sharp corner is evenly split. You can even use the existing domain to help guide the shape of the new connector (and make it better). Now dimension the connector to match the other topological direction and set the end spacing and distribution at the sharp corner to match the other near wall distributions. See the first image.

Now instead of deleting the domain and rebuilding two new ones with the new connector, split the domain at the grid line at the sharp corner. Now, merge replace the old connector (1st) with the new connector (2nd). You'll find this feature in Grid > Merge ... Merge by Picking. This avoids having to rebuild all of your blocking.

Now smooth the domains. Use the exact same setting in the elliptic smoother I used previous except at the two edges lying on the new connector. Set their angle control function to be interpolate. See the second image. Again, you may need to run a 50-100 of iterations before it settles down.

This approach will give you more control at the corner and definitely remove the grid lines cutting through it since you're forcing it to be in the right location with a connector.

See the last image for my result. I checked the quality of this approach vs my first approach and it results in a much better quality. The worst minimum included angle for suggestion #1 was ~25 deg whereas here it ~34 deg.

Quote:
 Originally Posted by syoo Chris, thanks for your reply. I have been trying various things. The blending does improve the mesh quality quiet a bit, however, it doesn't always prevent the grid lines from going into the surface. Redistribution of the points on the top connector does help a lot. When I use the steps you have provided on a 2-d mesh, it works fine, but when I use them on a 3-d mesh I still have same problem on some parts of the grid. I am probably expecting too much from the software and trying to As I have tons of blocks to make, I am trying to be consistent in my grid making method. For the most part, I think I will have to divide up the grid and make changes to the topology. Still in works though. -Paul
Attached Images
 domain-sharp-corner-05.jpg (11.4 KB, 24 views) domain-sharp-corner-06.jpg (33.4 KB, 29 views) domain-sharp-corner-07.jpg (52.0 KB, 28 views)

Last edited by cnsidero; January 24, 2014 at 09:18.

 January 23, 2014, 18:05 #6 New Member   Paul Join Date: Apr 2011 Posts: 12 Rep Power: 13 that approach is awesome. I'm going to try that. thank you.