I am attempting to model an airfoil. It seems to me that a wind tunnel simulator should excell at that, if nothing else. I am using STL files as input and no matter how the model is constructed or what size it is the Falcon math is leaking "air" through the model, especially near the thinner trailing edge. Any scale, any resolution, any velocity, any result mode does not change it. Contructing the model with very high resolution, over 1000 faces or very low, just 42 makes no difference. Making it with quads or tris changes nothing. Putting inner layers or a shell makes no difference. No matter what I try the math is not respecting the faces. This happens all over the model but is worst at the leading edge and the trailing edge.
The result is that the calculated airflow is entirely wrong. I am modeling a standard Clark-Y airfoil at zero degrees angle of attack. The result should be total flow adherance to the boundary layer with no separation. It doesn't make any difference what Reynolds number is used (assuming the software calculates the Reynolds # based on scale).
Unless this is fixed the program is useless. I have attached images showing the problem.
Here is another good example. I left off the end of the model to make it visible.
edit: The question is: What sort of collision detection are you using? One of the most efficient is variation on marching ray but an inappropriate choice of parameters will cause this sort of problem.
I have found that cranking up the matrix resolution beyond the interface setting helps but it doesn't make it go away entirely. Also, then Falcon crashes constantly and of course it also becomes very slow. Slow is to be expected even though I have a fair amount of processor power. CFD can bring any computer to its knees. The crashing is another matter. It is very likely related to the crashes I have seen already at permitted values. Since it is related to the amount of memory used I suspect a logic race condition.
The issue of not finding the surface looks like it is a sampling problem. Nyquist sampling theorem applies in cases like this. In order to reliably discover the surface the minimum sample distance must be less than half the shortest voxel distance. Nyquist is commonly quoted as saying the sample frequency must be at least twice the frequency of the sample but that is not quite correct. The sample frequency must exceed twice the frequency of the sample or else it can still alias at twice the sample frequency exactly.
As a technology preview, sometimes we collect the data, store it away, and investigate more thoroughly if the technology moves beyond the preview stage. That's one of the advantages of tech previews, they're not supported products yet. They may never become supported products. But it's always good to unearth these issues sooner rather than later.
Log into access your profile, ask and answer questions, share ideas and more. Haven't signed up yet? Register
Start with some of our most frequented solutions to get help installing your software.