Rotating Blades Simulation - CDi values

458 views
Skip to first unread message

Gustavo Jorge Resende

unread,
Jul 6, 2021, 3:37:19 PM7/6/21
to OpenVSP
Hi everyone,

I'm studying the behavior of wingtip-mounted propellers and using the X-57 as a reference model.

From the VSP Hangar I've downloaded the X-57 Complete and Simplified models. Then, I've created a third model, which has only the clean wing (from the simplified model) and the Cruise Propellers (from the complete model). For convenience in the simulations, I scaled the model so it would be in FT units (I tried to make it in meters, but for some reason, the propellers would not go further than 3.55 LengthUnits, it crashed every time I tried to go beyond that).

In a first simulation, I used the 'Auto Time Step' flag for 5revs and 2250 RPM for each prop. It calculated 120 time steps, with 0.00111 sec each. This simulation predicted an induced drag of 0.01112.

Then, in a second simulation, I used 900 time steps with 0.000148 sec each. This would give a resolution of 2 degs per time step, 5 revs at 2250 RPM. And for this simulation, the predicted induced drag is 0.03447.

Caught my attention such a difference in CDi just for changing the time steps. Furthermore, according to the literature, when comparing to the case without the propellers, the first CDi (0.01112) seems more reasonable.

I was wondering if I made some mistakes or with there are some cautions to make when changing the time step. Any hints or guidance would be appreciated.

-------------------
Some parematers of my simulations:

Wing tesselation: 
Num_W = 121 | Num U = 100

Cruise-Prop tesselation:
Num_W = 65 | Num_U = 81

VSAPAERO settings:

Sref = 66.666660 
Cref = 2.108185 
Bref = 31.622775 
X_cg = 0.000000 
Y_cg = 0.000000 
Z_cg = 0.000000 
Mach = 0.000000 
AoA = 0.000000 
Beta = 0.000000 
Vinf = 253.171500 
Rho = 0.001869 
ReCref = 2791100.000000 
ClMax = -1.000000 
MaxTurningAngle = -1.000000 
Symmetry = NO 
FarDist = -1.000000 
NumWakeNodes = 64 
WakeIters = 6 
NumberOfControlGroups = 0 
TimeStep = 0.000148 
NumberOfTimeSteps = 900 
Preconditioner = Matrix 
Karman-Tsien Correction = N 
Stability Type = 0 
Num Unsteady Groups = 3 
Num Unsteady Props = 2 


Brandon Litherland

unread,
Jul 7, 2021, 9:27:59 AM7/7/21
to OpenVSP
Hi Gustavo,

I won't pretend to know _exactly_ what's going on with the changing time step in the unsteady rotor cases.  I've seen similar behaviors in the current VSPAERO by trying to capture smaller time steps.  The first thing that I would check is how far into the far field the wakes are travelling.  Look at your solutions in the VSP Viewer.  For a decent analysis, you want your prop wakes and wing wakes to go roughly a semispan distance aft.  VSPAERO is relatively cheap to run so it doesn't hurt to let it crank for a while.  Another option is to run from Steady State which sets up your wing wakes (and treats propeller blades as fixed) for the first N Wake Iters and then begins an unsteady run.  When performing wing/prop interaction I usually run this way.  Something else to consider is the length of the distance a 'particle' of air would travel in the time step you are using.  Check your time step vs freestream speed.  Your Vinf is 150 KTAS so for the Auto case, your time step accounts for about 0.25 ft of forward travel. So for the full run, the wing moves about 33 feet.  In your second case, each step accounts for 0.04 feet and the wing moves again about 33 feet.  So have you set up your wing/prop resolution to properly catch these steps?  In the first case, you'll get away with a rougher chord-wise resolution because the distance is about right.  For a half-inch step size, you'll need a higher resolution.  Of course, this is all predicated on my experience with higher fidelity solvers like OVERFLOW where the surface speed has an effect on the final solution.  My relationship with unsteady potential codes is less developed but I'm getting there.  

So if the Viewer shows some differences in how far the wakes have travelled, set FarDist to something like 100.

Gustavo Jorge Resende

unread,
Jul 7, 2021, 10:44:07 AM7/7/21
to ope...@googlegroups.com
Hi Brandon,

Thanks for your answer!

Indeed, the wake travels much more in the first simulation than the second one. Both images below show the wake at the last time step.

image.png 
image.png

When you ask about the setup for the wing/prop resolution for the second case, you meant their mesh resolution?
If that's the case, I did not change the mesh. Both simulations have the same mesh resolution.

I did another simulation changing the FarDist to 3x the wingspan value. But it was for the case with 'Auto time step' with 5 revs. Not much has changed in CDi values.
After you answer, I think that it was expected.

I'll run the case with refined time steps changing the FarDist value as you suggested.

Thank you again!



Atenciosamente,

Gustavo Jorge Resende
Engenheiro Aeronáutico
PEE XXVIII
Tel.: (34) 99267-7400


--
You received this message because you are subscribed to the Google Groups "OpenVSP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openvsp+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/openvsp/dc4d1544-1cb9-4e2f-b06f-7f70e477ac67n%40googlegroups.com.

Rob McDonald

unread,
Jul 7, 2021, 11:08:10 AM7/7/21
to ope...@googlegroups.com
I would try to only compare cases with similarly developed wakes.  Your first image is superior in this regard.

Overall, it looks like your spanwise and chordwise resolution is much higher than it needs to be for both the wing and props.  I would aggressively de-res, then work to get everything working as best as possible, and then do some refinement work to figure out what resolution is really needed.

Are you running in VLM mode or Panel mode?  I would primarily run this case in VLM mode.

VLM mode needs less chordwise clustering than Panel mode -- it is generally very efficient in chordwise panel resolution.  For the propeller, you can probably get away with 5-10 chordwise VLM panels.  For the wing, 10-20.

For the prop, 10-20 spanwise panels should be plenty -- moderately clustered at the tip.

For the wing, you will want pretty tight clustering in the region of the propeller.  I would be tempted to split the wing geometry such that you could separately control the paneling on the outboard section vs. the inboard section.  Look at your load distribution plots when evaluating whether your spanwise paneling resolution is good enough.

Rob


Message has been deleted

Gustavo Jorge Resende

unread,
Jul 7, 2021, 2:13:26 PM7/7/21
to ope...@googlegroups.com
Hi Rob,

I'm using just the VLM mode. And you are right, this mesh is too refined.
A while ago I did a mesh convergence study for this wing and compared the results with the ones available in one of the X-57 articles.

The image below shows the CL vs AoA for different meshes resolutions and, in purple are the values available in the article.

image.png
And here are the time it took for each simulation:
image.png

To be honest, I don't know why I kept the mesh with 100 kutta nodes. Must have forgotten to change it back to 50 at some point in my studies...

These results do not consider clustering towards the tips, though. It is something I have not tried yet, thanks for the advice.

Here is the loading distribution for the simulation with 'Auto time step' flag with 5 revs:

image.png

Near the wing tip the curve seems well behaved. But I'm not so sure about these little peacocks near the wing root.
What do you think?

I'll use your recommendations on the propeller panels for the next simulations. It should reduce some simulation time.

Thanks again!!!

Atenciosamente,

Gustavo Jorge Resende
Engenheiro Aeronáutico
PEE XXVIII
Tel.: (34) 99267-7400

Rob McDonald

unread,
Jul 7, 2021, 4:16:39 PM7/7/21
to OpenVSP
I am not sure what is causing the bumps in the load distribution.  For the most part, they seem to be self canceling.  I suspect you will see a lot less of that if you coarsen the mesh substantially.

You'll want to monitor the load distributions in an unsteady manner -- the tip load distribution will dance around, you need to make sure you have sufficient resolution for a full period of oscillation.

Rob

Pranav

unread,
Jan 25, 2023, 6:42:16 AM1/25/23
to OpenVSP
Dear Brandon,

I was also planning on using OpenVSP for my thesis related to propeller wing interaction and was just trying to to see if I could find something related to it on this forum. I found this discussion quite relevant to my topic and while I was reading it, I had a question. When you said, ' So have you set up your wing/prop resolution to properly catch these steps?' I did not quite clearly get what you meant here. How I perceived this statement is: The mesh cell length should be of a similar scale as that of how much the wake would travel downstream in one time step. I am not sure if I interpreted this correctly.

Could you please let me know if I understood it right?

Best regards,
Pranav Zinjarde

Brandon Litherland

unread,
Jan 25, 2023, 4:40:32 PM1/25/23
to OpenVSP
In the case above, you can tell from the trailing wakes that the model has high surface tessellation which is likely unnecessary for what is being examined.  Particularly with prop-wing interactions, you'll want to try and set higher resolution on the wing in regions of propeller-blown flow but not so excessive as to overburden your solver.  The latest versions of VSPAERO assume 18 degrees of rotation per step if automatically set in unsteady blades mode.  This can be altered, however.  Similar to achieving good results from a wing in isolation, there will eventually be diminishing returns on accuracy (and even where the solution will diverge) with increasing resolution on Propellers and time steps.  Start with something that runs successfully and then perform a grid/time step study to find where the breaking point is.  In the case of the simplified X-57 geometry above, I'd begin with the wing at NumW = 41 and NumU = 21 with appropriate clustering.  The prop should be similarly refined on the surface though you can go lower as the blades are quite slender.  Begin with auto time steps.  If you feel that the wakes are progressing too far per step, manually adjust them down to achieve the desired result.  If your time steps become too small (and this is where my familiarity with exactly how VSPAERO is treating adjacent vortex strengths breaks down somewhat), I suspect that the influence of each wake filament gets too close to the upstream vortex on the surface and they start to interact.  I could be completely off base, though.  Keep in mind that VSPAERO is a Potential Method solver and does not solve for boundary layer or other viscous effects.  I will not resolve detached eddies in the volumetric flow field and so the time step rules that typically apply to high-fidelity Navier-Stokes solvers don't apply here.  

Sorry this isn't a great answer to your question but my recommendation is to start with the settings I provided and then experiment a bit to see how close you get to other data.

Reply all
Reply to author
Forward
0 new messages