Thanks for sharing! Our cars have such a different shape (and it will be even more different from the next race, since I will switch to high nose layout) that we could not copy each other because the flow around the cars is completely different.
I'd find it nice if we could get .ini file switches to turn off streams.
I checked the mesh log: it looks that the number of elements is about the same (in my case 6x10^6, it requires about 8-10 gb of RAM, plus about 4gb used by the system and other applications). I can't explain what happens to your simulations.
I am not sure, but maybe you can turn off the streamlines erasing the "include... streams" command in controlDict2 (but for sure it is not the best way to do this and probably streamlines are not the cause of your problem).
I'll do a final check by running the same geometry in two versions. May be it's something I did.
If your hx monitoring surfaces are very detailed (triangles smaller than 10mm), you might see that the process simplefoam will work on them a bit longer, but I mean 10-20 minutes, not hours.
Thanks for sharing your streamlines, very clean geometry (especially compared to mine).
Last edited by CAEdevice on 10 Sep 2017, 08:16, edited 1 time in total.
I'm re-running one sim with 2017.7 and another couple very similar ones with 2017.9 at an AWS VM. I'm logging the time of every command on runCase.sh. Let's see what comes out.
Hi, I have send out an email about the next race. It contains the link to submit the cars. If someone has not received it, please let me know, you might simply not be on the list for some reason.
Regarding the speed of the simulations:
This year they are slower than last year. This is due to the increased accuracy compared to last year. There is not much that can be done about that. There is no free lunch. The only time a drop in speed could have been witnessed during software changes this year was when switching to versions 2017.5 and 2017.7. 2017.5 introduced a new OpenFOAM version. This was mainly done because of convenience when working on Windows. From what I saw it is a bit more stable so it should perform better as well. 2017.7 introduced better meshes. This should have increased the solving speed as well. The monitoring surfaces will not really affect speed at all. I have run tests on that. I was worried that to fine resolutions would hurt speed as they are checked every iteration. There was not much to see there, so no worries here. They will slightly affect post-processing time as more streamlines need more time. But again from very coarse to very fine we are talking about two minutes. Also do not worry about the resolution of the engine intake and exhaust outlet. This does not change anything. The monitoring surfaces also do not affect the mesh at all.
So why are the simulations from rsja slower? I hope the general computer configuration stayed the same. If yes, one of my first suspicions is the design. This car was always tessellated very poorly, best rookie yes, performance is great, CAD is like some people last year. This poor CAD files can create bad cells which simply slow down the solver. I know myself that when tuning the car further and further, the designs start you get dirtier and dirtier and you start going for more and more small crazy details. They will all slow down the process. And the other part is, the car is getting better and better. With increased aero performance the air is going through increased crazy changes of what is happening to it. This results in larger gradients which are hard for the CFD solver. So while you are making your car better your are giving the CFD solver more work!
There are two things you should check. I hope you still have some log files from earlier designs left on your hard disks. Check the checkmesh.log files. See if there are more complaints as the season progresses. If yes, you have to work on your stl quality. Also check if there are now simply more cells.
Then check the solver log files. Check in the first few iterations how many pressure solving iterations are performed. If you had few starting into the season when you say CFD was still fast and now it is solving many more, then this is either due to worse meshes or the air working harder because of your better designs.
I am very happy with the solver setup, 2nd order model improved the simulation of vortices: I have a couple of vortices that run below the floor with significant effects. During 2016 it was impossible to use this kind of effect.
The solution time is around 6hours on a machine with 6 or 8 cores and as much RAM as required (64gb available).
We could slightly reduce the 1st order iterations (1000>>500÷800) but it would not change very much about total time.
Considering the behaviour of my car in the 2nd race, I would like to propose to use the "long" template for all the first three car in the next race (maybe 3000-3500 iterations), averaging the data using the last 100 iterations.
I have ran a different cooling setup - using the same meshing workflow, 90% exactly the same meshes - and simulation times dropped drastically. So I'll go with the flow complexity here.
LVDH will also notice that my crazy graphs are gone, everything is way smoother for sepang, even on the same cooling setup.
Now regarding 2017.9, I'll come with the exact numbers later, but while a given model took 5h15' in 2017.7 it took 6h08' in 2017.9. This done with a c4x4 AWS VM, which is more than twice faster than my laptop.
I've rolled back to 2017.7 and would still like to be able to disable per part calculations and streams.
Per part calculations work fine when processed on linux & OF4.1, but won't make it to the report if processed in BLueCFD.
Well, now there is a problem.
Must have developed a hole in my car somewhere while making changes. The slices for the pressure are showing the inside of the car blue, strangely enough, though the slices showing the air speed are still showing the car as white.
Hopefully, i can find it quickly.
How are you all going people making improvements??
How are you all going people making improvements??
I don't have a reference beacuse I will use a new car for efficiency races (with high nose layout), but I don't think I will get the impressive efficiency values I read in this thread.
