Oh you are talking about a 6DOF driving simulator! I'm not.

Don't forget Mach number. Can't forget compressibility considerations in your attempt at Mach 1 on that dry lake bed.

All I know is MSFS seems to run slower, have less things working (wake turbulence, helicopter ground cushion effect) and generally feels less fluid than X-Plane when it comes to simulating aircraft - and that's all I'm interested in from my flight simulator.  If CFD is really "better" than BET, it appears to have some significant drawbacks when it comes to implementation in a consumer flight simulation product.

 

 

1 hour ago, blingthinger said:

Oh you are talking about a 6DOF driving simulator! I'm not.

Don't forget Mach number. Can't forget compressibility considerations in your attempt at Mach 1 on that dry lake bed.

neither CFD nor BET give you values for what happens to an object moving in a fluid, they calculate what happens to the fluid and the pressures the fluid exerts on the objects surface.

The "flight model" is what you do with the output from CFD/BET.

Edited December 13, 20241 yr by mSparks

1 hour ago, mSparks said:

neither CFD nor BET give you values for what happens to an object moving in a fluid, they calculate what happens to the fluid and the pressures the fluid exerts on the objects surface.

Time-accurate/unsteady CFD simulations can be made to engage dynamic remeshing that would permit the object to move in a fluid. 

BET is not bi-directional and does not calculate what happens to the fluid (see SAS443's college textbook quote ). Austin's special BET sauce does track wing wakes/tip vorticies as well as adjusts induced AoA based on control surface movement.

 

1 hour ago, mSparks said:

The "flight model" is what you do with the output from CFD/BET.

The 6DOF bit is a key element there, agreed. flight model = force calculations + equations of motion. 

46 minutes ago, blingthinger said:

and does not calculate what happens to the fluid (see SAS443's college textbook quote ).

I mean in terms of

https://en.wikipedia.org/wiki/Reynolds_number

The "hard" part you are trying to solve with CFD/BET etc is where Laminar Flow transitions to turbulent flow - the calculations in either are pretty straight forward, but where it occurs in the flow is very non linear. NNs are pretty good at non linear cause and effect.

22 minutes ago, mSparks said:

The "hard" part you are trying to solve with CFD/BET etc is where Laminar Flow transitions to turbulent flow

No, BET doesn't go there. You can bundle turbulent transition behavior in the airfoil coefficient tables though. Depends entirely on what physics solver was used. XFOIL (easily the most common airfoil solver among XP devs) has a reasonable boundary layer transition model but isn't good for predicting stall. CFD is a mixed bag. RANS CFD does not deliver that without explicit transition models and those are problematic to say the least. Nor is it a turbulent flow prediction in the sense that everyone thinks of. LES can get there. DNS absolutely will but that's not an option in this context.

Edited December 13, 20241 yr by blingthinger

If I model a wing, run said wing through a CFD toolbox to generate lift and drag polars and then use said polars to feed a BET flight model, I've not only combined the advatages of both methods, but also gave both the MSFS and X-Plane communities a collective anyeurism because I've committed the ultimate heresy by combining the methods of both flight simulators. If that isn't an achievement, I don't know what is.

Edited December 13, 20241 yr by Bjoern

4 hours ago, Bjoern said:

If I model a wing, run said wing through a CFD toolbox to generate lift and drag polars and then use said polars to feed a BET flight model, I've not only combined the advatages of both methods, but also gave both the MSFS and X-Plane communities a collective anyeurism because I've committed the ultimate heresy by combining the methods of both flight simulators. If that isn't an achievement, I don't know what is.

DO IT!

I haven't yet wasted a day or ten setting up and playing with pytorch, but after watching half baked video series like

I know imma gonna.

6 hours ago, blingthinger said:

No, BET doesn't go there. You can bundle turbulent transition behavior in the airfoil coefficient tables though. Depends entirely on what physics solver was used. XFOIL (easily the most common airfoil solver among XP devs) has a reasonable boundary layer transition model but isn't good for predicting stall. CFD is a mixed bag. RANS CFD does not deliver that without explicit transition models and those are problematic to say the least. Nor is it a turbulent flow prediction in the sense that everyone thinks of. LES can get there. DNS absolutely will but that's not an option in this context.

 

My (admittedly fairly simple, I looked at this stuff enough to decide I am VERY happy to just leave what LR do best to LR) understanding is that turbulent flow has significantly lower velocity that laminar flow, and the accuracy - or rather any inaccuracies - are predominantly centred around capturing when that happens, BET simplifies it to "elements" that are either turbulent or laminar flow, "full" CFD has a large number of points that are either turbulent or laminar, after that the calculations are well characterised.

BET absolutely depends on the velocity of the flow over the element.

Edited December 14, 20241 yr by mSparks

13 hours ago, mSparks said:

turbulent flow has significantly lower velocity that laminar flow

You're in the ballpark. The velocity everywhere on the airfoil surface is always 0. For a given freestream velocity what does change is the slope of the velocity gradient in the boundary layer right at the wall. This yields very different skin friction values. Laminar flow generates lower skin friction. The "holy grail" is to retain laminar flow over all the airfoil surface right to the trailing edge. At typical operating Reynolds numbers this is not possible but is why glider pilots will sometimes polish the wings to eliminate any surface contamination (insect cemetery) that might trigger turbulence.

Laminar flow is also more susceptible to separation/stall but that's a different chapter (see golf ball dimples).

Every airfoil will have at least some laminar flow at the leading edge. Exactly where the transition point occurs varies on angle of attack, Reynolds number, etc. This video is a CFD simulation of a NACA 4412 at Re=400k.

https://www.gauss-centre.eu/fileadmin/research_projects/2020/CSE/GCS_Aflo_video2_1080p.mp4

Study summary.

This is a relatively low Reynolds number: on par with a glider at landing. They used 430 million points (9600 cpu cores to run!) in the 3D point cloud surrounding the airfoil surface in this LES  (they also claim to have resolved all turbulence vortices which technically makes it a DNS...not sure why they chose to turn on the LES model). Higher Reynolds number (or speed in this case) will generate progressively smaller turbulence vortices which would demand even more dense 3D point cloud to fully resolve.

The airfoil suction surface starts to transition at around 15% chord. Fully-developed turbulent flow is probably around 30-40% chord. Bonus in that video is 15 sec mark when they start rippling the pressure side surface to reduce drag. Kinda cool.

I pulled up a 4412 in Xfoil for chuckles and see the following skin friction trend at AoA=5 deg:

Turbulent transition on the suction surface (blue) starts at ~50% chord. Increasing to AoA=10 deg pushes that up to 10% chord. I can also push it forward by changing the freestream turbulence. There's plenty of knobs to turn. I don't know what AoA that CFD was run at and overall, no clue which is more accurate for this particular case (CFD vs Xfoil) because there's no experiment data to compare to. 

The predicted (either CFD or Xfoil) overall lift/drag/moment coefficient at that AoA gets fed into the BET airfoil polars. That's the only way BET knows about any transition. BET doesn't track a velocity over the airfoil surface. It does know what the velocity is at the leading edge, but it's not tracking or simulating turbulence in any way. Though as I mentioned before, Austin's BET does track wing-tip vortex trajectory which could be considered turbulence in the surrounding air.

11 hours ago, blingthinger said:

BET doesn't track a velocity over the airfoil surface.

Isnt this what airfoil maker is for though?

Edited December 15, 20241 yr by mSparks

7 hours ago, mSparks said:

Isnt this what airfoil maker is for though?

Not quite. You can set the general airfoil shape which XP then uses for thickness information and drawing the wing visually in sim if you aren't making your own 3D model, but it never digs into that level of detail.

Airfoil maker lets you set the total-airfoil-average force coefficient across the full range of angles of attack (0 to 359 deg) for as many Reynolds numbers as you want. It essentially builds three 2D surfaces describing the airfoil performance. One each for lift, drag, pitch moment. Actually you'd want Mach too but Austin does that separately with the Prandtl-Glauert correction.

That Xfoil simulation I ran above would be but one point on that 2D surface: angle of attack=5 deg and Re#=400k. You have to run Xfoil in polar-dump mode to tell it to solve for each angle of attack and then hand-fit the resulting curve in airfoil maker at that specific Reynolds number. Then rinse/repeat for other Reynolds numbers.

So you're capturing the overall average effect of turbulence transition just not down to the airfoil-surface level of detail. Because of the fact that you can bake this info into the airfoil performance curves, XP is the only consumer sim that lets you simulate this level of aerodynamic detail. CFD requires many millions of grid points to simulate turbulence transition. That does not run in real-time. 

@mSparks Reading through your post, you and I have the same general thought process on this it seems. You've made many great points which I agree with.

Guys!!! 

Puleeeze, do me a favour and use your expertise, which leaves me in total sensation of analphabetism, to help LR/Austin finally fixing the bug that is causing BeT to create a right rolling moment on twins with CW rotating props, so intense that it hides prop effects and the expected left turning tendências... 

Don't know if it also affects other multiprops?

14 minutes ago, jcomm said:

Puleeeze, do me a favour and use your expertise, which leaves me in total sensation of analphabetism, to help LR/Austin finally fixing the bug that is causing BeT to create a right rolling moment on twins with CW rotating props, so intense that it hides prop effects and the expected left turning tendências...

Only if you stop the constant whining about it.

52 minutes ago, Bjoern said:

Only if you stop the constant whining about it.

Ok!!! 😃