MSFS flight model

[blingthinger 649]

5 hours ago, Krakin said:

Did PMDG ever say their hiccups were related to the FM? All I remember them complaining about was stuff related to avionics.

Thats....a really good point. I brain-farted thinking the FM was the only concern, but it was an entirely new system mostly everywhere. Dunno. But I have also been stunned at the list of items that were kicked back to legacy status. Lots of stuff that would be very very useful still.

[blingthinger 649]

5 hours ago, jrw4 said:

computation costs might be that are associated with evermore detailed real-time computations

 

15 hours ago, Christopher Low said:

How many parts does a 3D model of a plane need to be broken down into before the end result gets anywhere near the real thing? I mean, a thousand surfaces sounds impressive, but that's still a rather blocky representation of a real world smoothly curved airframe.

If looking purely at the CFD side, this is a complicated problem. It depends on the airframe and the operating condition. On the coarse end for say a 737, you might be looking at 100,000 surface elements. It gets tricky because you need smooth airfoils. Faceted wing surfaces are obviously going to perform incorrectly, especially at stall. There are tricks to play with grouping cells and elements near areas of high curvature, but that only gets you so far. Also you still have to fill the air volume around the airframe, so looking at 100's of thousands of cells. Asobo's currently at 8000 total.

The gridding method Asobo is using is not well suited for complex geometries + low grid count...this is another reason why I think they are modeling flat plates even though they don't explicitly say so.

Plus you have to morph the grid cells as the control surfaces move: more equations and time involved there. Landing gear and fowler flaps, oh my...

For a single instantaneous snapshot in time (assuming mesh morphing is complete), with the latest compute GPUs ($5000), you might be looking at 10 min to converge a solution. 30 FPS is 0.03 sec, so that's a 20,000x speedup needed for real-time CFD. Those are optimistic numbers and considering FM only. 

I'm not holding my breath for this happening in my lifetime. And even for that grid resolution there would still need to be fudge (or "artistic") factors applied to the final movement.

For now, we get to watch these pretty CFD videos that would have taken a month to simulate on a thousand CPU cores to get 0.03 sec of flight time. ~5 billion cells in the most refined grid. https://twitter.com/drchriscombs/status/1441047753690333184

 

 

[NZ255 309]

1 hour ago, blingthinger said:

Thats....a really good point. I brain-farted thinking the FM was the only concern

When asked will PMDG take advantage the new CFD FM, soft body etc. the dev stated “Our flight models are custom already. We are not relying to the default MSFS.”
https://forum.pmdg.com/forum/main-forum/general-discussion-news-and-announcements/172449-use-of-new-msfs-flight-model-features

[blingthinger 649]

4 minutes ago, NZ255 said:

When asked will PMDG take advantage the new CFD FM, soft body etc. the dev stated “Our flight models are custom already. We are not relying to the default MSFS.”
https://forum.pmdg.com/forum/main-forum/general-discussion-news-and-announcements/172449-use-of-new-msfs-flight-model-features

That's really got me thinking now. The capability does exist to just move the aircraft around wherever you want, but I would think you'd lose ground contact checking at the least. I wonder if that just means they are injecting forces components along the way? Or is it a full on replacement? The latter would be a huge undertaking.

[NZ255 309]

11 minutes ago, blingthinger said:

That's really got me thinking now. The capability does exist to just move the aircraft around wherever you want, but I would think you'd lose ground contact checking at the least. I wonder if that just means they are injecting forces components along the way? Or is it a full on replacement? The latter would be a huge undertaking.

I don’t know. Talk around the water cooler is that majestic q400 (FSX and p3d) does the setting the X,Y, Z  velocity etc via simconnect, but I’ve not seen devs theorise that’s how PMDG does it. https://www.fsdeveloper.com/forum/threads/independant-fde-i-want-to-know.435978/#post-730438 

In the majestic after exiting slew mode it takes a few second to start moving again. That could be a tell of doing it that way but I’m not sure. I’ve not seen any other addon misbehave like that

But yes it is interesting, we assume we’re talking about the sim but actually a dev can overwrite everything about flight model. And if it’s bad people will blame the sim, not the dev. 

Edited September 14, 2022 by NZ255
Changed my answer

[blingthinger 649]

3 minutes ago, NZ255 said:

I assume it’s the later. I believe fully moving the plane (setting the X,Y, Z etc) is what majestic did. Because after exiting slew mode it took a second to start moving again. 
But yes it is interesting, we assume we’re talking about the sim but actually a dev can overwrite everything about flight model. And if it’s bad people will blame the sim, not the dev. 

Interesting. A full replacement would surprise me to be honest. Partly because it's obvious there are places they can improve and that flies in the face of their reputation for quality (which is very very good). Take the damped roll-yaw coupling behavior that flightdeck2sim keeps bringing up, for example. If they are overriding everything, that should be a relatively quick fix given their history in P3D. In other words, I would expect them to have figured that one out by now.

[NZ255 309]

Just now, blingthinger said:

Interesting. A full replacement would surprise me to be honest. 

Please note I changed my answer a bit after re-reading. This is all speculation from my part, with some more speculative sources cited 🙂 

[flying_carpet 657]

Talking of hundreds or thousands of surfaces for (real time) calculation - not only - in a flight sim is marketing gibberish. A very easy understandable example is the approximation of a circle with polygons: https://demonstrations.wolfram.com/ApproximatingPiWithInscribedPolygons/

[mrueedi 702]

13 hours ago, blingthinger said:

The capability does exist to just move the aircraft around wherever you want, but I would think you'd lose ground contact checking at the least.

This would be a good starting point, if you want to reinvent the wheel:

GitHub - tml1024/flying-brick: Silly sample aircraft for MSFS built from scratch

Something to note, is that Fenix did not choose to an external flight model for their A320, which even is an FBW aircraft. Only PMDG.

16 hours ago, TrafficPilot said:

Landing too fast is one of the most common errors with students and low hours PPL holders - probably because we've drummed into their brains how dangerous (and easy) it is to stall an aircraft in the circuit.

Regarding the floating during landing this imho explains probably most of the issue you guys have! Check your final approach speed. As hard evidence is harder to counter than words, I made a little excercise. First I checked the final approach speed of the real Cubcrafter XCub ("stabilized final approach speed, 45 knots"), then fired up the sim and landed with exactly this speed. The result? No floating, the stick pulled fully at touchdown, can anybody tell me whats the issue with that landing?: 

 

[tweekz 2,165]

18 hours ago, blingthinger said:

The gridding method Asobo is using is not well suited for complex geometries + low grid count...this is another reason why I think they are modeling flat plates even though they don't explicitly say so.

Plus you have to morph the grid cells as the control surfaces move: more equations and time involved there. Landing gear and fowler flaps, oh my...

As you mention this, it is interesting that Laminar Research solved some gear effects by a fake momentum, which one of the devs of Flight Factor complained about a few years ago in another forum. Here's what he said:

Quote

A huge fake pitching momentum proportional to CG shift and aircraft weight is added continiously all the time. Instead of computing total reaction forces from gears on ground, he simply add's this fake momentum to produce different gears loading with different CG shift. Everything is ok while aircraft standing on ground, but this momentum is not removed even in flight. It's added to momentum generated by lifting surfaces with respect to CG position, and each XP aircraft should compensate it continiously by elevators\stabilizer producing unnesesary drag. More over, it's not added in sim/flightmodel/forces/ daterefs, making intergration of external flight model very intresting... more over, all XP aircrafts can rotate only around some fixed point (default CG), but not around it's actual CG as they should (and it was implemented even in FS2004, 14 years! ago). Do you know what was an answer from Austin? "Yes, this is a fake, but it's to difficult to make things properly, so we do not want to do anything". No, i find a way how to overcome this, as alot of other XP fakes, but you can continue to trust.

 

Edited September 14, 2022 by tweekz

[Alpine Scenery 1,609]

The ground effect issue has been around for ages on the default planes. Showing one good landing on one particular plane doesn't prove anything. There is a sweet spot and you can still land any plane, but the ones affected have about 1/5th of the real window of how a real plane would act (according to a pilot in the official forums). Hence, the way he described it was, in a real plane you might feel the ground differently or the plane start to try to lift off the ground slightly when coming in just a tiny bit off, but in MSFS it shoots you back into the air. Hence, a decent sloped curve is not there in the descent, as it goes immediately from being on the ground to having way too much lift. His words not mine. My perception is the same, compared to every other sim, there is too much ground effect. 

As noted, many 3rd party planes do not suffer the issue, but some do. The problem is prolific on certain planes in the game. The excuse that we shouldn't be flying default planes, well this isn't so hard to fix. If Carenado and Milviz can make the plane land properly by adjusting a few values, so should Asobo.

Edited September 14, 2022 by Alpine Scenery

[Christopher Low 5,332]

4 hours ago, mrueedi said:

This would be a good starting point, if you want to reinvent the wheel:

GitHub - tml1024/flying-brick: Silly sample aircraft for MSFS built from scratch

Something to note, is that Fenix did not choose to an external flight model for their A320, which even is an FBW aircraft. Only PMDG.

Regarding the floating during landing this imho explains probably most of the issue you guys have! Check your final approach speed. As hard evidence is harder to counter than words, I made a little exercise. First I checked the final approach speed of the real Cubcrafter XCub ("stabilized final approach speed, 45 knots"), then fired up the sim and landed with exactly this speed. The result? No floating, the stick pulled fully at touchdown, can anybody tell me whats the issue with that landing?

So you think that guys who have been flying in other simulators for years without problems are suddenly struggling to understand how to maintain a stabilised approach??

Edited September 14, 2022 by Christopher Low

[mrueedi 702]

1 hour ago, Christopher Low said:

So you think that guys who have been flying in other simulators for years without problems are suddenly struggling to understand how to maintain a stabilised approach??

@TrafficPilot and I are saying, that if you float, it could be that you approach too fast. My demo shows, that if I don't approach too fast, the floating is not there. Try it for yourself and report here!

And ... I think, that if it would be easier to land in the sim, it would not do justice to reality. Because landing tail draggers requires training and is not so easy.

And ... it could also be, that the guys who flew for years in other sims indeed got familiar with planes, which make landing too easy. There is the video of the real world 172 pilot, who specifically said "thats pretty close to the real thing landing, it floats like a Cessna".

[Alpine Scenery 1,609]

It does not just happen on Tail Draggers, people have tested this on several planes, they are not right...
There are probably 50 threads about this already. I am not sure about that particular plane, as I do not fly it, but take the DA-62, add any amount of wind angle to the wind over 5mph, and you get inconsistent landings. A DA-62 in RL is barely affected by a 6mph wind. On certain runways, it will either "hold you over the runway" or shoot you into the air, or it will angle you into a weird hopping pattern, it's a 2-part problem. A DA-62 can handle a 25 knot crosswind (almost 30mph) in real life with the right technique. The second part of the issue is that many many airports have incorrect slopes and bumpy mesh on the runways. Part of the reason for the issue, is because any slight angle of the runway contributes to the wind calculation turbulence error. Luckily most overly bumpy or mesh errored runways are generally fixed with any payware airport.

A lot of people will just think, oh wow I had a bad landing that time, umm no....

Edited September 15, 2022 by Alpine Scenery

[blingthinger 649]

6 hours ago, tweekz said:

As you mention this, it is interesting that Laminar Research solved some gear effects by a fake momentum

I was actually talking about generating a CFD grid around the landing gear: very complicated...but the landing gear/momentum discussion is good dirt!