Hello,

I am using Gridgen to generate an unstructured grid for a ships propeller. I use that "On DB Entities" command in the domain menu. Everything works so far, but at one end of the wing where there is no connector, the tetrahedrons form some kind of "saw blade" unless I set avgDs to a very small value. I have created two screenshots to clarify what I mean:

http://img204.imageshack.us/my.php?image=gg12bc.png

http://img221.imageshack.us/my.php?image=gg28ce.png

Is there a way to refine the grid only on one side of the propeller?

Bye, Chris

Putting a connector along the leading edge will go a long way toward solving your problem.

In fact, I noticed something about your mesh topology that I suggest you change. At the wingtip, you have a connector that appears to start at the trailing edge, go around the top, around the leading edge, and then back down the bottom where it terminates at the trailing edge. In other words, the connector begins and ends at the same node. We recommend that you don't do that because it might confuse the topology engine for certain operations.

Since you're using the On DB Entities command, I assume that the topology you're getting comes directly from the database - the wing is a single surface combining both upper and lower surfaces together.

You have two choices for repairing this. First, you can delete the domain on the wing (don't delete the connectors). Then split the root and tip connectors at the leading edge. Next, create a connector of type DB constrained Line from root to tip. Then create 2 new domains using Assemble for the upper and lower wing surfaces.

There's still another option for handling this issue. Make two structured domains - one each for the upper and lower wing surfaces. Then create an unstructured domain by diagonalizing the structured quads. Use those diagonalized domains as faces for the unstructured block. The good thing about the structured domains is they result in a more regular faceting of the wing surface due to the grid's structure.

Hope this helps.

This problem occurs in areas of high curvature, such as around the leading edge. As John says the solution (for a fully unstructured surface mesh) is to introduce local clustering. In gridgen this is most easily introduced by adding additional connectors and froming the surface using multiple domains. Alternately a semi-structured approach (structured - followed by conversion to unstructured by adding the diagonals) can also be used, the main problem then is the number of surface cells required to avoid problems with cell aspect ratio and skewness.

Another suggestion would be to use curvature based clustering with either the Max Ang or Max Dev parameters. Max Ang (aka swing) is the maximum turning angle allowed between normals of adjacent triangles. Max Dev (aka sag) is the maximum chordal deviation allowed between a triangle's center and its underlying database.

Hi,

thanks for your help John and HelpfulSoul and sorry for my delayed response. I think I will try the method with the structured domains next, as the Max-Ang/Max-Dev-parameters only improved the triangulation locally at the wingtips, whereas inserting a connector at the leading edge caused the domains not to be bound to the database anymore. BTW: The wingtip is not described by a single connector, but by two connectors where there is one very small connector at the thinner side.

Thanks, Chris

There's no reason why inserting a connector at the leading edge should cause the domains to lose database adherence. In fact, you should probably have a connector at the leading edge even if you use a structured domain.

It depends on how that connector was drawn. If you drew it at a 2 Pt Curve, or a DB constrained Line, it should be automatically attached to the database and, therefore, the domains should be automatically attached to the database. If you drew the leading edge connector as a Line, you need to do two other things: 1) project that connector onto the database and 2) using the unstructured solver, tell the two domains that they're supposed to be constrained to the database, project them, and do some further mesh refinement to smooth things out.

The behavior you see with Max Ang and Max Dev is correct - it will cluster to regions of high curvature and use only the Average Spacing value on the flat portions of the wing.

The other thing you need to decide is whether the problem on the leading edge is the faceting or the irregularity of the faceting. If the latter, going to structured grids is your best bet.