you use SALOME just to make 2D surfaces - prepare STL - then you throw it into snappy or cfMesh

Oh yeah i forgot to reply, I figured it out

you can do tetra meshes there easily but everything else is very complicated

I just made a mesh for my step files after grouping them and exported that as STL

although I could only export the groups individually and not as one STL all noted as solids and had to recombine later

that wouldve been nice

Yeah, yeah, that's exactly it. I created two faceZones that divide the boundary layers into sections 1) per layer 2) per surface patch (images 2 and 3), and use those to control the scotch decomposition. Having per-layer control (like in the first picture) would allow me to use more cores, but, since I have many angles of the airfoil, I prefer to run multiple cases at once on fewer cores, instead. That means having the entire layer be on a single core is not a big deal, and I get better results. https://cdn.discordapp.com/attachments/400697259819270145/1432407861761413230/image.png?ex=6900f147&is=68ff9fc7&hm=63ecf101f131de8c3fbe399a5622a33f71a6f3fd80a4887bdc5feb97216e08db& https://cdn.discordapp.com/attachments/400697259819270145/1432407862252011540/image.png?ex=6900f147&is=68ff9fc7&hm=b7b01267cee747c75505d39d53c6f3bc6dcfc6dd56948a178173e5603d7ccfe8& https://cdn.discordapp.com/attachments/400697259819270145/1432407862637756416/image.png?ex=6900f147&is=68ff9fc7&hm=bf89ca2776e598c949b5b56157ac88f00af87999c44223fc1df4760008c71737&

It turns-out that the "better results" I was getting with 30 cores was just error cancelation

This is so unsettling that you'd need to resort to this kind of decomposition.. There must be something very wrong in the parallelization of the solver if your setup is OK?

It's not that I have to resort to it, I'm just trying to minimize the chances of unnecessary errors. Just switching to the PCG solver allowed me to get good results with 20 cores without special decomposition treatment, but I just like to do things properly as I can, where I can, since the results I've been getting are... not great.

Either way, you are welcome to check the case for mistakes/issues, if something doesn't sound right. I keep consistently getting higher lift numbers than I should, and cannot for the life of me figure out how to fix it, so if anyone found a mistake, that'd actually be pretty nice. https://cdn.discordapp.com/attachments/400697259819270145/1432410667264114861/2d_airfoil_rhoSimpleFoam.zip?ex=6900f3e3&is=68ffa263&hm=2e1543e75aa347d7b76ef9b5c1fc0f7891dafe209231bf4abdc3f8eefe420393&

I'll give it a shot and let you know, thanks. I do run renumberMesh right before initiation, number of cells is 81k, where 25k of that are boundary layers.

I might have to do something simular for my decomposition as I observe weird values in processors patches on the surface, (this might actually be a coding mistake somewhere in custom solver though... so not sure) but I might try that too now

OK, so I reduced the tolerance down to 1e-8, which didn't seem to help, so I also turned down relTol from 0.01 to 0.001, and now I get the same result when running the case with any number of cores and just plain scotch decomp. So, thanks! Never knew when and how to change those values. It's nice to finally use those properly for once.

so you mean to say... it was just a non-converged case? I'm not really used to suggesting anything useful

This is where my terminology fails me, but yeah, seems the pressure equation wasn't reaching sufficient convergence each iteration? I mean the initial residual for p did go down to 1e-5 in each case, enough to trigger stopping the simulation, but obviously something was still undercooked underneath

relTol controls how much the solver lowers residuals relative to the beginning of the iteration. So with higher relTol it will need more iterations to get to desired tolerance but the result should be the same

Now the question is, how can I know that my linear solver convergence criteria are too loose in the future, short of trying to run (sometimes very large) cases with both, many and few cores and looking for differences?

Ah, that's interesting, then. Maybe the lower relTol caused an overshoot?

residuals aren't the only thing and might often not tell you anything at all, get yourself a function object with quantity of interest and finish the simulation when it stops changing (even more so with unsteady flows that never converge)

probably not - it's there because ... SIMPLE algorithm (and stability thingamabobs)

Oh no, I already do that. It just converged to different solution depending on the number of cores used.

huh

beats me

that's a 2D case?

yessir

Well then, nobody knows why the solution worked, but it did, so.. a good enough outcome for me

What solver or type of solver should I use to simulate a fairing on the arms of a drone to smooth out the airflow around the arm? Here is a rough outline of my criteria: - 3-5" propeller - Tip speed of up to mach 0.6-0.8 - Air speed of up to 70-100mph I'm currently constructing an fpv freestyle drone. The propellers are placed on top of the arm, so the high speed air exiting the propellers causes a high pressure zone on the arm, greatly increasing drag. The easiest solution is to put the propellers below in a pusher configuration, but based on testing I found of this configuration it experiences a high amount of drag at high speeds, causing a decreased top speed. My idea is to put an aerodynamic fairing on top of the arm to hopefully reduce the high pressure region and increased drag. To aid with the design, I want to use CFD to figure out the effectiveness of this idea and the optimal design of the fairing. It appears I will need to use a compressible flow simulation due to being within transonic speeds. What I'm unsure of is whether or not the tip speed is high enough to warrant needing a specializing solver for high speed airflow, or if a more generic compressible flow solver will cut it. I'm also curious to how hard producing an accurate simulation will be, as it appears what I'm trying to achieve is far from basic unfortunately. I'm also unsure of what computing power I should expect to need. I've been wanting to build a homelab for a while but a highschooler budget doesn't exactly like that idea too much. I've only ran a few example cases in OpenFOAM, so I'm trying to figure out what I need to do to build up to achieving such a simulation. I'm also interested in learning OpenFOAM for general use, as my main hobbies (rocketry and fpv) would greatly benefit from it. Thank you all for any help!

Using kOmegaSST the solution should be accurate with y+ greater than 30 if there is no flow seperation occuring and clean flow over the surface correct?