Oswald too high

[Floris Mariën]

Hi all

In my simulations with VSPAero, the Oswald Efficiency is unrealistically high (sometimes even above 1). The simulations are mostly just only wings with e.g. span = 20 and chord = 2.

Does someone of you know how I can get this right?

Thanks in advance!

Floris

[C P]

For VLM refine your wing chordwise and spanwise, paying attention to cluster panels toward tips.

Panel method works also with coarser panneling

Make a comparison between the two, before saying you have the " right" value.

Corrado

[Rob McDonald]

Please post an example file if you're still having trouble.  Not much to go on here.

Rob

--
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/b7c3137c-8d1a-4f7b-b086-cb9bf63aec03n%40googlegroups.com.

[Floris Mariën]

As attachement you can find the .csv-file from the simulation done with VLM at:

M = 0.2

Alpha = 10

10 Iterations

64 Wake nodes.

The model has the default span and slices chordwise (33) and spanwise (6), but is adapted to a rectangular wing with chord = 2.0 (file= RefWing.vsp3). 

After refining and clustering the panels better, the result got better (Trial1.vsp3; Trial1.csv). Still, the Cdi is almost twice the value gained by textbook calculations. Does someone know how this is possible?

Thanks for the help! Really appreciate it!

Floris

Op woensdag 2 juni 2021 om 07:57:26 UTC+2 schreef Rob McDonald:

[C P]

Here my thought, waiting for Rob's answer:

1-  your first question was addressed to solve the problem of Oswald factor greater than 1. I setup a quick wing model, span=20 and chord= 2 as you mentioned as the mostly simulated. With the suggestion above i never had  e>1;

2- in your second question you refer to induce drag as the problem. If we are talking about the same wing, I cannot increase (double) CDi if Oswald factor is increasing as you mentioned in first point. Then it would be useful to fix "who is ( or seems to be) the problem" .  For example: if you made your own hand calculation, which "e" and "CDi" values did you obtain ? Did you cross check those values taking AR, pi, CL and look at the result you have putting all in a formula ?

3- RefWing and Trial1 models attached totally differ. The first has span 5.2, chord=1 AR=5.2 while the second span=18, chord=2 AR=9. What is it for, to refine the second in order to make a comparison with the first ? This doesn't help you in understanding where the problem is. If the reference wing has span=20, chord = 2, then all efforts should rely on such a reference model so that you can develop the feeling of the relation between what you change in the model and what you have from analysis. Then you could take those results and compare them with hand calc or exp. data or whatever information you have. 

Corrado 

[Floris Mariën]

Thanks for the answer!

These are my results where I was talking about. 

I did the hand calculation as follows:

I calculated the lift curve slope according to DATCOM (1978):

Then simply multiplying with the AoA and I got the Cl.

As the outcome shows, this is very close to the simulation, so that's fine. Also for different geometries, Mach numbers,... this calculation is very close to the outcome of the simulations.

Then, I found the theoretical oswald factor according to  Niță and Scholz (2012), by first finding the value for Hoerner's curve f(λ):

With this oswald factor and lift coefficient, I calculated the Cdi as follows:

I think, the simulation provides an overestimation of the induced drag (73.26% higher then the hand calculation) and as a consequence the oswald efficiency is unrealistically high. This is all going out that the textbook calculation is correct. 

The theory I used can be found at:

https://www.fzt.haw-hamburg.de/pers/Scholz/arbeiten/TextBudziak.pdf

https://www.fzt.haw-hamburg.de/pers/Scholz/HOOU/AircraftDesign_7_WingDesign.pdf

To answer on your third point, my bad! I uploaded the wrong file. As attachement the right one.

Thanks again!

[Rob McDonald]

As Corrado mentioned, you need to work on the resolution of your model -- in particular, 6 spanwise panels is very inadequate.

Are you using VLM or Panel method -- for this, VLM should work well.

For VLM, you don't need much chordwise clustering -- I would set it to 1.0 LE and TE for starters.

For a straight wing, you don't need clustering along the center line -- but some clustering at the tip is not a bad idea.

I would start with the paneling set to:

Chordwise - 33 panels (default) clustering 1.0 1.0

Spanwise - 16, clustering 1.0 root 0.5 tip.

Try that and see how it does.

Rob

--

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/375539c2-6260-4339-937e-053606c3a621n%40googlegroups.com.

[Floris Mariën]

Hi again

Thanks for the tip!

Here some of my results with adapted clustering (done with Panel Method):

The clustering seems to have very little influence on the simulations. The oswald factor is more realistic now, but the Cdi is still way to high compared to the textbook calculation. Do you guys know how I can get this closer? Or is it just a consequence of the low fidelity of the textbook formulas?

Op woensdag 2 juni 2021 om 18:12:37 UTC+2 schreef Rob McDonald:

[C P]

Did you make a comparison between VLM and Panel method  for the same wing , but changing panels refinement for each method?

Did you make a comparison with other formulas for evaluating "e" and CL_alfa ( Datcom, Torenbeck, Roskam,Raymer,..) ?  CL is squared in CDi formula, so small changes, deriving from different formulas , will impact CDi in non trivial way. Beside, if also "e" changes, overall CDi value will change again. Empirical fomulas are not perfect but a comparison among them will help in  centering a useful value. 

Search for exp. data of simple geometries to tune you model and to fix which one of the many formulas gives the most satisfactoy result for your needs.

Corrado

[Floris Mariën]

I now did the same simulations with VLM. Due to a slightly higher outcome for Cl, the oswald is very close to the textbook calculation (which is based on Datcom). But still, the Cdi is very far off.

I'll look for experimental data now. 

If you have any other tips, please let me know.

Thanks for the help!

Op vr 4 jun. 2021 om 11:24 schreef C P <corp...@gmail.com>:

You received this message because you are subscribed to a topic in the Google Groups "OpenVSP" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/openvsp/jnmeIvECqZs/unsubscribe.
To unsubscribe from this group and all its topics, send an email to openvsp+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/openvsp/cf3c2005-13d2-4ea3-a123-e3ce066305dcn%40googlegroups.com.

[Rob McDonald]

Comparing percentages when the numbers are very close to zero can be very misleading.  If your Oswald is right - then your error in CL is just as much the problem as your error in CDi.  But, the percentage errors are vastly different.

I would suggest you do a bit more of a grid refinement study - and keep an eye on CL's progression as well.

You might also try a comparison with higher and lower AR -- the equation from DATCOM will be better for high AR than for low.  9 should be high enough for a relatively decent match, but it is interesting to see how these things diverge with AR.

Rob

To view this discussion on the web visit https://groups.google.com/d/msgid/openvsp/CAPpE2cCQeNTYvWWDP%2B%3DVj6kiJdvLPjqJYfqyfWbOV%2B-UaXAs2g%40mail.gmail.com.