[davidkingnj]

Hi guys!

I'm having an problem with SHM (clearly). I noted the problem when I had first meshed the geometry but was not convinced that it would lead to problems with the simulation. Now that the simulation is complete I am a bit more dismayed.

So, a quick overview just for an idea and sense of scale I am in charge of meshing an airfield to analyze structure impacts on the velocity field at the runways. Large mesh intended to be run on a thousand cores. I ran a rough simulation while I was away on vacation and figured I would make my workstation earn the "work" part of it's name. The simulation reinforced my concerns about the 'odd' meshing.

I've uploaded the four sample images to my Dropbox. Although I attached images, it just seemed easier to share higher res images this way; I can also just add more images. I'll reference them as I go along.

First in the image prepended with "0_", I noticed after running SHM that the area surrounding the buildings seemed odd. Not that it is particularly easy to get a good feel for cell boundaries in a 2D slice of a 3D mesh but something just didn't feel right... .

So, no biggie, perhaps it just looked funny. I'm not an expert at using SHM and figured that I was going to be away for ten days so just let it run on the computer and see what it looked like when I returned.

After running the simulation and visualizing the data I became somewhat concerned. Overall, when zoomed far out, it looks nice which is great, but I have some concerns about the accuracy due to local issues that seem directly related to the mesh. I noticed a "blockiness" to the velocity field results that I feel I should not have been seeing (which you can see in the image prepended with "1_"). I've provided a closer image of a particular area to highlight this issue (image prepended with a "2_"), and also an image of the velocity field overlaid with the surface boundaries on that particular slice (image prepended with a "3_").

I'm a bit of a loss here and haven't found much regarding this topic on the Interwebs.

I guess ideally I am hoping that someone with some experience dealing with this issue will have some advice on cleaning up the mesh. I have computation time on Vesta at Argonne and cannot waste those hours trying to resolve something which is only going to result in useless simulations (which is what I consider this). Is there any way to avoid this?