ACHIEVING THE BEST REFLEXED AIRFOILS FOR FLYING WING USE IN THE SMALL PLANE CATEGORIES

Homebuilt Aircraft & Kit Plane Forum

Help Support Homebuilt Aircraft & Kit Plane Forum:

WINGITIS

Well-Known Member
Joined
Jun 24, 2020
Messages
378
Location
Wellington, New Zealand
I've never seen him on HBA, you will need to contact him or Selig directly.
Hi WM

I wanted to do the CFD comparison test first....

As it indicates, at this stage, that the MH algorithm for CL/AOA is closer to the CFD than the XFLR5/XFOIL combination then we need to do more tests first to substantiate what is going on.

Cheers
Kevin
 

Norman

Well-Known Member
HBA Supporter
Joined
Nov 28, 2003
Messages
3,114
Location
Grand Junction, Colorado
2: We need to find a NACA report with wind tunnel data for various thicknesses of a common airfoil, such as 23012, 23015, 23017, 23019 to make another comparison.

DOES ANYONE ALREADY HAVE SUCH A REPORT? The airfoil type is irrelevant for this test.
NACA-ARC-L5C05 by Abbott and Von Doenhoff, figures 39, 40, and 41. You can buy it as a 700 page book titled "Theory Of Wing Sections" for about US$14.00 or download it as a huge PDF. Those figures are 11 pages. Here's the first part:
 

Attachments

Norman

Well-Known Member
HBA Supporter
Joined
Nov 28, 2003
Messages
3,114
Location
Grand Junction, Colorado
Here is a control set of various thickness airfoils based on the N60R reflexed airfoil.

The original is 12% and I have just changed the thickness in 2% steps to 20% keeping the original camber line.

It shows that just changing the thickness on an already optimized airfoil will not result in a better airfoil in XFLR5.

The NACA WIND TUNNEL MAX CL result for the 12% is 1.4 at 18 Degrees AOA at the 3,100,000 RN, no speed specified.

Next I will put them all through JAVAFOIL and then through the CFD.

View attachment 102728
I ran those airfoils through XFLR5 and got some weird results. The curves show more variation in their shapes than they should. I suspected that you introduced some statistical noise by thickening them (actually it's most likely a magnification of the noise that was already in the original coordinates (most coordinates before 1935 were noisy to some extent)) So I smoothed them with 2 passes through the Hanning filter and ran them again. Apparently we can attach .ZIP files so here's my project file with the originals de-rotated and re-paneled and the smoothed versions. The pressure distributions are still a bit lumpy but I didn't want to change them too much. The N60 probably isn't the best choice for doing this kind of comparison because the original coordinates may have been lifted from the wind tunnel model with a CMM rather than being generated by an equation like all of the NACA sections from the 4 digit sections onward. (very little noise in the NACA coords) I've done this kind of comparison a few times before and am looking forward to seeing how a clean set of airfoils works out for you. I talked about the results I've seen in this post and and attached a graph.
 

Attachments

Last edited:

WINGITIS

Well-Known Member
Joined
Jun 24, 2020
Messages
378
Location
Wellington, New Zealand
NACA-ARC-L5C05 by Abbott and Von Doenhoff, figures 39, 40, and 41. You can buy it as a 700 page book titled "Theory Of Wing Sections" for about US$14.00 or download it as a huge PDF. Those figures are 11 pages. Here's the first part:
Thanks Norman

I already have a 550 Page version, where in there are the comparisons, am I looking for the 23000 series or something else?

Cheers
 

WINGITIS

Well-Known Member
Joined
Jun 24, 2020
Messages
378
Location
Wellington, New Zealand
I ran those airfoils through XFLR5 and got some weird results. The curves show more variation in their shapes than they should. I suspected that you introduced some statistical noise by thickening them (actually it's most likely a magnification of the noise that was already in the original coordinates (most coordinates before 1935 were noise to some extent)) So I smoothed them with 2 passes through the Hanning filter and ran them again. Apparently we can attach .ZIP files so here's my project file with the originals de-rotated and re-paneled and the smoothed versions. The pressure distributions are still a bit lumpy but I didn't want to change them too much. The N60 probably isn't the best choice for doing this kind of comparison because the original coordinates may have been lifted from the wind tunnel model with a CMM rather than being generated by an equation like all of the NACA sections from the 4 digit sections onward. (very little noise in the NACA coords) I've done this kind of comparison a few times before and am looking forward to seeing how a clean set of airfoils works out for you. I talked the results I've seen in this post and and attached a graph.
Which ones did you run through XFLR5, the closed TE ones do not work properly in XFLR5, I did the XFLR5 ones with the open TE examples.

The closed TE ones do work fine in JAVAFOIL, they also have to be closed TE in the CFD.

The ZIP file did not work "ERROR READING THE FILE SAVED"

The 12% one was already Derotated, That is what the DR means in the file names, which has normal results in XFLR5 then the thickness was adjusted, which then has introduced those bumps.

That is another of those Forced versus Non Forced transition bugs/issues within XFLR5 I believe.

Proven by the fact that JAVAFOIL does not show them!

The original 12% N60R was generated by a formula.

I chose the N60R because it is reflexed and I had the wind tunnel data for it.

THE ACTUAL SMALL DETAILS DO NOT MATTER IN THIS CASE THOUGH..

I THINK THE MAIN ISSUE IS THAT THE CL/THICKNESS TREND WITHIN XFLR5/XFOIL APPEARS TO BE WRONG.

That is what we need to sort out.

Someone else on here must have access to another CFD tool for a validation check!!??

So far there are three issues that appear to be wrong with XFLR5/XFOIL

1: Forced Transition variances are far to high and create inconsistent results, airfoils can be tuned for either option presenting a 35% CL variance which does not match any other real data.
2: The CL variance with Thickness trends oppositely from CFD and Javafoil results.
3: The Drag Bucket analysis at high AOA appears to be incorrect.

They all affect the results in a large enough way that one may as well just choose any reflexed airfoil....probably the M6 or a simple Horten and run with that!

I am keen to get it right because my SDN algorithm can make the best airfoil for best results in any analysis tool at any settings.....if the tool is wrong by a BIG %, I need to move to a more correct one, in order to make the BEST REFLEXED AIRFOIL for actual flight.

Kevin
 

Norman

Well-Known Member
HBA Supporter
Joined
Nov 28, 2003
Messages
3,114
Location
Grand Junction, Colorado
I already have a 550 Page version, where in there are the comparisons, am I looking for the 23000 series or something else?
Sorry about that, I keep forgetting that paper pages are diferent from PDF "pages". The PDF file is 700 pages not the book. All of your N60R files that I downloaded from here had open trailing edges. The minimal smoothing I did made the analysis run more smoothly with fewer failures to converge. Smoothing also seems to increase the error your concerned about ie CLmax goes up and CD goes down. Decreasing NCrit to 6 or increasing Mach increases the predicted drag a bit and increasing Mach also takes a tiny bight out of CLmax. However neither changes the XFoil results enough to bring it into agreement with your CFD results because panel codes always have errors near CLmax.
 
Last edited:

Retiree

Well-Known Member
Joined
Nov 29, 2010
Messages
104
Location
USA
I am keen to get it right because my SDN algorithm can make the best airfoil for best results in any analysis tool at any settings.....if the tool is wrong by a BIG %, I need to move to a more correct one, in order to make the BEST REFLEXED AIRFOIL for actual flight.

Kevin
Hi Kevin,
What parameters does your SDN algorithm generate for the reflexed airfoils. Are the parameters just numeric that have meaning only to the algorithm. Or are the parameters physical quantities that designers can relate to like lift coefficient, airfoil thickness, amount of camber, ...? If the parameters are physical, can you share them with us?
Doug
 

WINGITIS

Well-Known Member
Joined
Jun 24, 2020
Messages
378
Location
Wellington, New Zealand
Hi Retiree

It does not generate parameters perse.

You give it parameters against the polar outcomes you want and attach weightings to them.

It then "OVER TIME" pops out an airfoil very accurately with 499 points/panels. NO SPECIFIC AIRFOIL DESIGN KNOWLEDGE IS REQUIRED!

No understanding of such things as Stratford distributions or any other components one then has to stuff into Inverse Design etc etc.

As there as so few airfoil designers around and they are mostly inaccessible and secretive, then this tool was clearly required, so I and with input from some other professionals from some adjacent specialities built it.

The airfoil it generates is then tested with an airfoil ANALYSIS tool, can be anything...

Then "OVER A BIT MORE TIME" it generates the best airfoil that meets the parameters that the ANALYSIS tool confirms is the best match WITH THE HIGHEST PERFORMANCE ACROSS ALL THE POLARS specified.

If the tool is only say 90% as good as NATURES physical parameters/limits you get a 90% best possible airfoil.

Hence I need to know which is the most accurate ANALYSIS tool.

If we were only talking about a 2-5% difference between tools then who would care, that is about the same limit as the building tolerances, BUT the difference is clearly more than that, up to 35% within XFRL5 alone by varying only 1 setting(Forced Transitions)

If one is going to all the trouble and effort to design an aircraft for maximum performance then there is no point doing all that work only to get an average or less than average result when its built...

It is clear many people have probably built aircraft with their FAVOURITE airfoil and would never even know that it could be less than average, that is a scary concept and not one I wanted to find myself in!

For flying wings it is EVEN MORE IMPORTANT to get the best airfoil so one does not lose to much CL when one increases Reflex to keep the CM positive.

Of course one has to know enough to be able to specify the input parameters you require to achieve your design requirements, it cannot do that part for you.

Cheers
Kevin
 
Last edited:

WINGITIS

Well-Known Member
Joined
Jun 24, 2020
Messages
378
Location
Wellington, New Zealand
Sorry about that, I keep forgetting that paper pages are diferent from PDF "pages". The PDF file is 700 pages not the book. All of your N60R files that I downloaded from here had open trailing edges. The minimal smoothing I did made the analysis run more smoothly with fewer failures to converge. Smoothing also seems to increase the error your concerned about ie CLmax goes up and CD goes down. Decreasing NCrit to 6 or increasing Mach increases the predicted drag a bit and increasing Mach also takes a tiny bight out of CLmax. However neither changes the XFoil results enough to bring it into agreement with your CFD results because panel codes always have errors near CLmax.
Hi Norman

The closed TE ones are in POST #160.

From the MH site Javafoil also claims to use the Panel method as part of the calculations, but still varies a tremendous amount from XFLR5.

In terms of the CFD, WE need someone else to run the two (Closed TE) airfoil files from POST #160 through an alternate CFD tool to see if it mimics what mine does.

Until then we cannot be sure which tool is the most representative of REALITY.

Cheers
Kevin
 

Norman

Well-Known Member
HBA Supporter
Joined
Nov 28, 2003
Messages
3,114
Location
Grand Junction, Colorado
The airfoil it generates is then tested with an airfoil ANALYSIS tool, can be anything...

Then "OVER A BIT MORE TIME" it generates the best airfoil that meets the parameters that the ANALYSIS tool confirms is the best match WITH THE HIGHEST PERFORMANCE ACROSS ALL THE POLARS specified.

If the tool is only say 90% as good as NATURES physical parameters/limits you get a 90% best possible airfoil.

Hence I need to know which is the most accurate ANALYSIS tool.

If we were only talking about a 2-5% difference between tools then who would care, that is about the same limit as the building tolerances, BUT the difference is clearly more than that, up to 35% within XFRL5 alone by varying only 1 setting(Forced Transitions)

If one is going to all the trouble and effort to design an aircraft for maximum performance then there is no point doing all that work only to get an average or less than average result when its built...
2D Panel codes will always have errors and they will be different for different airfoil sections so one test case isn't showing that one is more accurate than the other. It's just showing where the errors are for each implementation of the panel method for that particular test case. Although I haven't seen anything like the 35% you mentioned. If I didn't know better I'd say you were comparing 2D results to 3D which do look that far off. Also comparing them to CFD isn't exactly without some ambiguity either because not all CFD is created equal. If you consider these errors too large for your needs you should give up on panel codes and stick to CFD.
 

WINGITIS

Well-Known Member
Joined
Jun 24, 2020
Messages
378
Location
Wellington, New Zealand
Hi Norman

The 35% difference is between the RGREAT and MAGMAGEXP airfoils....both 2D, both designed within the XFLR5 tool analysis capabilities.

RGREAT IS OPTIMISED WITH NO TRANSITIONS AND MAGMAGEXP WITH "1" in the XFLR5 transitions.

THE CL VARIANCE IS 35%...

The results have been previously posted here a few times.

THEY ARE TWO EXTREME EXAMPLES OF WHAT CAN BE ACHIEVED WITH XFLR5.

They both have the same input parameters into the SDN, the SDN created the best airfoil using the limits set by the transition variable in XFLR5 in both cases.

I have run the same SDN parameters in a limited manner into Javafoil, the results are much tighter no matter which options are chosen, for instance the transitions makes little difference, no matter which one of the 9 transition modes is used, never mind the values.

The main difference is when the STALL MODE is changed between "CALCFOIL" and "EPPLER" but only on extreme airfoils.

I could keep running this and it would hone down to a more likely REAL candidate "OVER TIME"

However WE need to have someone do a further validation check in another 2D CFD tool before it is worth doing any more new airfoil creations.

I agree if another CFD tool matches the one I use pretty closely across say 3 different reflexed airfoils I will move to that exclusively but I dont think we are quite there yet!

The mere fact that JAVAFOIL and XFLR5/XFOIL are so far apart and yet both use panel methods is disturbing.......

Just taking the variance of CL with Thickness trend alone being reversed between the two would indicate they are virtually useless, or one of them is at least....!
 

Retiree

Well-Known Member
Joined
Nov 29, 2010
Messages
104
Location
USA
Hi Retiree

It does not generate parameters perse.

You give it parameters against the polar outcomes you want and attach weightings to them.
Hi Kevin,
What kind of parameters do you give SDN to get the reflex airfoils?
I also wanted to ask about your CFD runs. How many points do you try to get into the boundary layer and how many points do you have in your computational domain?
Thanks,
Doug
 

WINGITIS

Well-Known Member
Joined
Jun 24, 2020
Messages
378
Location
Wellington, New Zealand
Hi Doug

If you specify "POSITIVE CM" on the inputs for all AOA, assuming your talking about medium to high AOA, you will always get a reflexed airfoil.

This can be in a reflexed form "camber line" similar to a Cubic or Quintec Polynomial, but the SDN AI never quite generates a true formula, that is one of its advantages, it creates entirely unique and new designs of airfoils!

The 2D CFD I use has a (480,000) cell resolution, its professional edition runs at a tunnel resolution of (1 million cells) for more accurate results.

How it deals with the near and far fields of the flow I do not know.

If someone else can confirm its results with another CFD package, as discussed previously, then I will move up to the 2D professional version.

There is also the 3D version with a Cartesian mesh consisted of 31 million fluid and solid cells, equally divided into the near field and far fields of the flow.

Cheers
Kevin
 

Retiree

Well-Known Member
Joined
Nov 29, 2010
Messages
104
Location
USA
Hi Kevin,
Can you discuss any conclusions from your reflexed airfoil investigation? What airfoil characteristics did you decide were most interesting, thickness, pitching moment, lift coefficient range?
Doug
 

WINGITIS

Well-Known Member
Joined
Jun 24, 2020
Messages
378
Location
Wellington, New Zealand
Hi Kevin,
Can you discuss any conclusions from your reflexed airfoil investigation? What airfoil characteristics did you decide were most interesting, thickness, pitching moment, lift coefficient range?
Doug
Hi Doug

Yes here are my general observations:

1: A thickness of between 12% and 14% is Optimal in XFLR5 depending on what your forced transition setting is set to, at 0 its 12%, at 1 its around 14%

2: Varying thickness has a reverse trend between XFLR5 and JAVAFOIL(See posts) this has not yet been fully resolved but the CFD tests compare more favourably with JAVAFOIL in that respect.

3: A POSITIVE CM always comes with a lower C/L, different reflexed designs just hone in on the optimum point where CM is 0 or Positive and C/L is the MAX it can be.

4: For 3: above, the ideal is a flat line at 0 CM for all angles of AOA that the wing needs to operate at.

5: The power factor is a good reflection of the success of the airfoil and is directly proportional to the bell curve of the lift/drag graph. A big fat bell curve with the highest value possible being ideal.

6: The airfoil can also be tailored to have the center of the lift/drag bell curve at a set AOA, so one can optimize for say a "cruise" airfoil or an "aerobatic" one.

7: Certain Horten(Formula settings) and the Munk M6 airfoil are a pretty good reflexed airfoil to start analysis with.

8: Profili gives similar results to XFLR5.

9: SOMEONE ELSE needs to do some additional CFD comparisons of the examples presented(ABOVE) to establish that JAVAFOIL should probably be used in place of XFLR5 for airfoil analysis.

10: At some point I will add a NACA 2412 "Cessna 150" airfoil modified and a Liebeck LD-104E modified to be reflex airfoils as extra reference airfoils.

In summary, in my view, there is not YET enough valid information presented here to be able to confirm a relevant airfoil recommendation for any given purpose because the analysis tools have not been proven to be accurate enough.

For a single given requirement maybe one or other tool can show a valid output of a certain airfoil, but across a range of different airfoils and requirements the tests are not conclusive enough to either justify the time to build a wing using the PROJECTED ideal one or secondly take a risk flying with it.

I am happy to continue research here over time if others continue to put effort in and post it here.

Otherwise the thread will just, or could just, turn into a place to post various Reflexed airfoils that people have or come across.

Which whilst not as productive is still interesting for everyone, including myself.

Cheers
Kevin
 

WINGITIS

Well-Known Member
Joined
Jun 24, 2020
Messages
378
Location
Wellington, New Zealand
Hi Folks

It is well into the NEW YEAR, so has anyone got any new, rare or just strange reflexed airfoils for "US ALL" to have a look at?

Santa it seems did not bring any!

Cheers
Kevin
 
Top