Tips for Handling Complex Geometries, Very Large Grids
 

This thread is aimed at providing a forum for questions, answers, and comments that arise from watching our latest Let's Talk Meshing video. The link for the video will be posted below in a separate post on Wednesday, August 6th at 10AM CDT.

This Let's Talk Meshing video steps you through the meshing process for creating an unstructured grid around a moderately complex geometry. We pay particular attention to solid modeling, uses for scripting, local grid refinement, and creating orthogonal cells in the boundary layers. The geometry was provided by NASA and Gulfstream as a test case for participants in the BANC-III aeroacoustics workshop held in conjunction with the 2014 AIAA Aviation conference in Atlanta, Georgia, in June.

Aeroacoustic simulations require a well resolved flow field, from which pressure fluctuations are used to calculate frequency spectra. Such simulations are most often performed with hybrid RANS/LES or DDES, requiring very smooth transitions in cell volumes and highly resolved surface grids. We will show you how to create such a grid in Pointwise and will demonstrate best practices for handling complex CAD geometries and obtaining a mesh suitable for this type of application.

It's nearly 10AM CDT. Here's the link to the Let's Talk Meshing video: http://youtu.be/2mqwGFvaxMM. Enjoy!

To paraphrase a question we received via email: Is it possible to use a script to study a series of geometries with slight changes between configurations (such as the shape or position of the torque arm)?

The answer is yes, though it takes a little bit of setting up. If the changes being made to the geometry are parametric, then it may be possible to simply import the grid onto the altered geometry. If the changes are more drastic, then I would recommend creating several copies of the part that varies in independent files (one for each configuration), and dropping each component into the larger assembly by Appending the file via a script. That is, you would open the main file without the component that changes, then Open and Append the file that contains only that component (which can contain a surface grid as well). You can then grab the domains, assemble the blocks, set up your viscous BL settings and initialize.

We should also mention, if it is simply a matter of the positioning of the torque arm, using the Edit, Translate tools on both the database and grid entities allows you to reposition the grid without needing to redefine the block.

Does memory usage scale linearly with cell count?

Essentially, yes, although it depends a bit on the complexity of your surface grid and other settings (such as the number of T-Rex layers and iterations for the isotropic tetrahedra generator). From our tests without multiple iterations, we found initialization requires around 130 MB of RAM per 1M cells. (Your results may vary!)

John and Travis thanks putting together for such a helpful discussion on a very practical model. There were many subtle yet powerful suggestions you mentioned. I often find it's the small tips are what makes the difference between an OK mesh and a good mesh so it's nice to have a resource for these. In particular, the tips about joining quilts and fault tolerant meshing to eliminate sliver cells due to tangent regions are very useful.

I don't have any questions but I have a one thing to add if I might. During the solid modeling portion, John introduced a method for trimming coincident models (the torque arm with the nub) so that a single model could be created. These situations of course are quite common with CAD assemblies. The technique John used was to create an short extruded surface normal to the edge of the quilt (with a script) so that there was a surface available to enable trimming with the adjacent, coincident surface.

I have an alternate method that I use for achieving the same effect. I create duplicates of the curves defining the boundary of the quilt, select the copied curves, project them onto the coincident surface, and trim the coincident surface by curves. For a simple B-spline surface, to create the boundary curves you simple drawing DB constrained curves using the UV parameters: (0,0)-(1,0), (0,1)-(1,1), (0,0)-(0,1), and (1,0)-(1,1). For trimmed surfaces, you need to find to edge curves, which are usually hidden, and copy/paste them without a transformation. Clearly this can be quite tedious for complex geometries so I wrote a script that helps automate this. Anyway, just thought it might be a helpful alternative for assembling multiple models.

Thanks again and I look forward to the next meshing video.

Hello Chris,

Thanks for watching. I like your suggestion for the other approach to trimming operations; trimming by curves seems to be a more robust operation in general, but creating the curves manually can be a bit tedious. Great tip for another script.

John