8 hours ago, SAS443 said:

Any word on the actual CFD model quality/resolution? I mean a fine vs coarse model is night & day regarding computational complexity. It's not like we have an abundance of CPU overhead as it is. 

edit: or is the heavy lifting done on Azure  side?😁

See https://www.avsim.com/forums/topic/626199-is-anybody-else-excited-about-the-new-20-km-cfd-in-su11/?do=findComment&comment=4870207 ... the set of Navier Stokes equations used for CFD calculations in MSFS are custom and likely a subset of the full set (https://en.wikipedia.org/wiki/Navier–Stokes_equations) , and the fineness/coarseness of this processing can be controlled via flight model config and/or dev mode, with CPU utilization going up exponentially as the fineness increases (no processing in the azure cloud here, it's all local)

As always for aircrafts, CFD is just another technology in the MSFS aerodynamics engine/toolbox that aircraft developers can utilize to implement their aircraft's flight models. At the end of the day, it's the individual aircraft's flight model that makes or breaks a bird.. i.e. a well implemented aircraft FM based on legacy table-lookup methods can fly better and more realistically than a poorly implemented aircraft FM based on the latest CFD and other tech. MS/Asobo are using the default C172 (glass cockpit version) to provide an example of aircraft CFD implementation in its FM, and it's not bad at all, and provides a competent flying experience. There are also various 3rd party aircrafts that use CFD like the Sting S4, etc.. this thread on the official forums keeps a good tallying list: https://forums.flightsimulator.com/t/new-propeller-cfd-soft-body-simulation-airplane-list/504719 

Also good to keep in mind that even if CFD is used in the FM there will always have to be tweaks/tuning of parameters/scalars/etc in the flight model ... CFD calculations are based on the shape and geometry of the aircraft, and currently the MSFS SDK doesn't allow for very fine grained definitions of the aircraft shape/geometry. In theory, if such fine definitions are possible, then CFD calculations run for that shape should be enough to calculate the actual airflow on it and its flying characteristics.

As for this new expanded CFD technology coming in SU11, MS/Asobo are using the same CFD tech to calculate atmospheric and environmental airflow.. and when it comes to geometry & shape, those are already well defined (i.e. terrain shapes, scenery objects, etc) so the resulting airflow simulation around them in the sim is quite realistic.. i.e. to the point where even a 1 foot tall sand berm on a beach has airflow impact in SU11: https://forums.flightsimulator.com/t/testing-the-new-airflow-simulation/549454/5?u=lwtxb

 

Edited November 6, 20223 yr by lwt1971

2 hours ago, lwt1971 said:

As for this new expanded CFD technology coming in SU11, MS/Asobo are using the same CFD tech to calculate atmospheric and environmental airflow.. and when it comes to geometry & shape, those are already well defined (i.e. terrain shapes, scenery objects, etc) so the resulting airflow simulation around them in the sim is quite realistic.. i.e. to the point where even a 1 foot tall sand berm on a beach has airflow impact in SU11: https://forums.flightsimulator.com/t/testing-the-new-airflow-simulation/549454/5?u=lwtxb

 

Thanks for the answer @lwt1971!  @SAS443, as a partial answer to your question about the resolution for the 20 KM CFD, it appears the 20 KM CFD's resolution is 1 foot in the circumstance above, if that picture of the sand berm above is accurate (and there were no config modifications). I don't know if it's 1 foot for the entire 20 KM CFD though - from my software developer instincts, to keep the performance up and not tank the FPS, I doubt the resolution is 1 foot for the areas that are almost 20 KM away, it's probably only 1 foot for the closest areas to the plane (this is just my educated guess).

From the other pictures I saw of the immediate CFD airflow lines around the plane, the resolution appears to be finer than 1 foot though.

6 hours ago, abrams_tank said:

I don't know if it's 1 foot for the entire 20 KM CFD though - from my software developer instincts, to keep the performance up and not tank the FPS, I doubt the resolution is 1 foot for the areas that are almost 20 KM away, it's probably only 1 foot for the closest areas to the plane (this is just my educated guess).

I agree with your hypothesis. There's no way they are using 1ft resolution everywhere. In fact, not likely anywhere at all. There are just as many flow lines passing right through that ground bump, as there are seemingly going over it. It's possible that that bump just happens to be at an interface between two CFD cells that are following the terrain contour. There's a ton of interpolation going on there too. All of their fancy flow visualization streamlines rely heavily on interpolation around the airframe solution, which is much much more coarse than their flow lines suggest. Just like XP has been showing for many many years.

I'm guessing they are using a Chimera/overset style mesh for the weather. You start with a coarse base mesh for the big distances (20km) and overlay a more refined mesh near the aircraft, much like this 2D CFD mesh around a wing section w/ flaps extended (blue on top of green):

This example is the type of resolution fully-resolved CFD runs with. And that's just 2D. Currently Asobo's flight model mesh is placing no more than a couple cells across the chord of the wings.

So in the weather model, the tight mesh around the aircraft (eg blue above) moves around with the aircraft itself and interpolates from the green mesh to get the general feel for a region. And there won't be much of a performance hit. You only have to run this CFD when the winds change, which is every few...minutes probably? Every half hr? Whenever the weather updates significantly enough.

Any bumpiness of turbulence is going to be nothing more than a few random sinusoid oscillations added on top of the base wind velocity. That's not simulated by the CFD.

Makes me wonder what XP is doing. @Janov recently made a youtube video in XP12 demonstrating terrain-induced updrafts across a canyon: up on one side, down on the other. Much like what Asobo is generating w/ the CFD here. It's possible that LR was the 'first' on this one, but they've been tight-lipped on this capability. 

 

 

Asobo certainly has the more advanced model, given it accounts for trees and buildings. So even though CFD on the fixed-wing FM is laughable, using it for terrain-induced wind variation (and possibly other weather effects) is pretty cool.

I'm certainly not excited...but I will admit to some jealousy!

@blingthinger - Excellent post ! I believe your hypothesis is probably very close to what ASOBO designed.

I got the impression from one of the videos recently published by ASOBO that probably some sort of convergence between air masses could be modeled, which would be great for the simulation of soaring flight.

Their base for convective / thermal lift is also very interesting, although up to SU10 at least it looks a bit incomplete and in some aspects even buggy, but I guess it's WIP and they will probably release an updated version with SU11-40th. At least their base idea is rather interesting and I've seen it explored in only one of the soaring sims I used in the past - SilentWings.

Finally, rotaries are going to be my first test when I get my hands in SU11. I have positive feelings about it, but so far just feelings...

 

Edited November 7, 20223 yr by jcomm

37 minutes ago, jcomm said:

I got the impression from one of the videos recently published by ASOBO that probably some sort of convergence between air masses could be modeled

That's a challenging one to model. Vertical mesh resolution becomes a big deal for that one, and they simply can't afford the mesh cells. You need to go up quite a few 10kft to capture the vertical updraft motion.

Plus, 20 km radius is wayyyy too small for that kind of simulation. Probably better to stick w/ Meteoblue's non-CFD model there...but then you run into all those quirks of interpolation on the weather model volume.

It's a tough one for sure!

56 minutes ago, blingthinger said:

That's a challenging one to model. Vertical mesh resolution becomes a big deal for that one, and they simply can't afford the mesh cells. You need to go up quite a few 10kft to capture the vertical updraft motion.

Plus, 20 km radius is wayyyy too small for that kind of simulation. Probably better to stick w/ Meteoblue's non-CFD model there...but then you run into all those quirks of interpolation on the weather model volume.

It's a tough one for sure!

Indeed, but for some of sea breeze effects, for instance, 20km vertical would suffice.

Some of my collest glider flights irl have been flown in sea breeze convergence, and several miles can be flown with barely 1m/s keeping you at a more or less constant alt. You just have to be careful with teh return 🙂

Then, unless we run some customized code over the GRIB data most of the vertical movements captured by the models aren't really the ideal indication of actual activity. We do it around here with ECMWF model, but then customized calculations have to be performed for smaller areas, eventually coupling it with ARPEGE  data too...

OFC I don't think ASOBO's approach will be this detailled, but I'm curious!

Edited November 7, 20223 yr by jcomm

2 hours ago, jcomm said:

Their base for convective / thermal lift is also very interesting, although up to SU10 at least it looks a bit incomplete and in some aspects even buggy, but I guess it's WIP and they will probably release an updated version with SU11-40th. At least their base idea is rather interesting and I've seen it explored in only one of the soaring sims I used in the past - SilentWings.

SU11 is completely reworking thermals and convective airflow, it's not just an update on what's available now. Thermals are now simulated based on a great many more factors than they were <= SU10 (previous posts on this thread have more details).

In terms of the cocoon radius (20km) and resolution in terms of physical length the various CFD driven airflow operates on, the best info we have now are on the official forum threads and testers' findings, and partly also in the SDK docs (hoping they'll add more details to the SDK docs about how this expanded CFD works in simulating atmospheric airflow). The proximity airflow simulation has the highest resolution of about a foot, as also mentioned by Seb in an earlier Q&A. The larger atmospheric airflow simulation in the 20km cocoon obviously will have a coarser resolution. Underlining my own below.

The sim calculates the solar radiation in watts (factoring in clouds). It checks the albedo (colors) in the satellite imagery and uses them to determine how much the surface should be heated, which can result in rising air. Rising air is also created directly below clouds. The rising air is an input into the larger atmospheric flow CFD simulation, along with the wind, resulting in air that is moving in various directions, including being puilled downward. Everything in the scene affects the wind in the proximity airflow simulation: Terrain, buildings, trees, etc.

Updraft simulation

The updraft simulation simulates rising air - the results of which are fed into the atmospheric airflow simulation. The updraft simulation particles are created close to the ground based on the solar radiation simulation and clouds. Their velocity is based on the solar radiation simulation and wind speed, and their initial angle is based on the terrain. These particles are affected by the wind, but not the terrain or the scenery (after they are created). They rise on an angle relative to the terrain and are blown by the wind. The higher the wind, the less altitude the particle will gain before dying. Right now (SU11 Beta) there is a minimum wind speed of about 3 knots to get proper thermals.

Atmospheric airflow simulation

The atmospheric airflow simulation covers a large area around your plane and is affected by the DEM, live data, and updraft simulation. The upper-altitude of this simulation is the boundary layer. This simulation is always done a small distance above ground level on a single slice of the atmosphere (a single voxel in height) and does not increase in altitude with your plane. Objects and trees do not affect this simulation. Since this simulation has long-lived particles, it can result in interesting behavior, such as tumbling air (rotor turbulence).

Proximity Airflow Simulation

The proximity airflow simulation covers a much smaller distance and the particles have a lower lifespan, but it updates much more frequently and is affected by just about every object in the scene ( including trees and buildings), as well as live data, at a resolution of about a foot. This simulation is done on a single slice of the atmosphere (one voxel in height) at the altitude of your aircraft. The atmospheric airflow simulation is an input for the proximity airflow simulation until the boundary layer, after which it is just the wind (live data) that affects the proximity airflow simulation.

Updraft simulation → Atmospheric airflow simulation → Proximity airflow simulation → Aircraft
 

As for the local aircraft CFD capabilities available for use in flight models since SU9, would be great if someone like @simbol chimed in with their thoughts given the experience in developing that aircraft's FM using CFD... did it prove useful in improving the flight model for the Sting S4? Could a similar flight model for the S4 have been achieved without CFD? I remember you had some interesting times trying to get more info during the SU9 days on the developer forums 🙂

Given the capabilities of the core MSFS aerodynamics engine pre-SU9, obviously aircraft devs like Fenix (A320), Milviz (C310), and PMDG (737) have developed their flight models quite nicely without aircraft CFD, and like I said before it all comes down to the fine tuning of the flight models in capable hands at the end of the day. But I also wouldn't dismiss local aircraft CFD as useless, since it's obviously made the default C172 better than before, and is being used in great aircraft like the Sting S4.
 

Edited November 7, 20223 yr by lwt1971

1 hour ago, lwt1971 said:

SU11 is completely reworking thermals and convective airflow, it's not just an update on what's available now. Thermals are now simulated based on a great many more factors than they were <= SU10 (previous posts on this thread have more details).

In terms of the cocoon radius (20km) and resolution in terms of physical length the various CFD driven airflow operates on, the best info we have now are on the official forum threads and testers' findings, and partly also in the SDK docs (hoping they'll add more details to the SDK docs about how this expanded CFD works in simulating atmospheric airflow). The proximity airflow simulation has the highest resolution of about a foot, as also mentioned by Seb in an earlier Q&A. The larger atmospheric airflow simulation in the 20km cocoon obviously will have a coarser resolution. Underlining my own below.

The sim calculates the solar radiation in watts (factoring in clouds). It checks the albedo (colors) in the satellite imagery and uses them to determine how much the surface should be heated, which can result in rising air. Rising air is also created directly below clouds. The rising air is an input into the larger atmospheric flow CFD simulation, along with the wind, resulting in air that is moving in various directions, including being puilled downward. Everything in the scene affects the wind in the proximity airflow simulation: Terrain, buildings, trees, etc.

Updraft simulation

The updraft simulation simulates rising air - the results of which are fed into the atmospheric airflow simulation. The updraft simulation particles are created close to the ground based on the solar radiation simulation and clouds. Their velocity is based on the solar radiation simulation and wind speed, and their initial angle is based on the terrain. These particles are affected by the wind, but not the terrain or the scenery (after they are created). They rise on an angle relative to the terrain and are blown by the wind. The higher the wind, the less altitude the particle will gain before dying. Right now (SU11 Beta) there is a minimum wind speed of about 3 knots to get proper thermals.

Atmospheric airflow simulation

The atmospheric airflow simulation covers a large area around your plane and is affected by the DEM, live data, and updraft simulation. The upper-altitude of this simulation is the boundary layer. This simulation is always done a small distance above ground level on a single slice of the atmosphere (a single voxel in height) and does not increase in altitude with your plane. Objects and trees do not affect this simulation. Since this simulation has long-lived particles, it can result in interesting behavior, such as tumbling air (rotor turbulence).

Proximity Airflow Simulation

The proximity airflow simulation covers a much smaller distance and the particles have a lower lifespan, but it updates much more frequently and is affected by just about every object in the scene ( including trees and buildings), as well as live data, at a resolution of about a foot. This simulation is done on a single slice of the atmosphere (one voxel in height) at the altitude of your aircraft. The atmospheric airflow simulation is an input for the proximity airflow simulation until the boundary layer, after which it is just the wind (live data) that affects the proximity airflow simulation.

Updraft simulation → Atmospheric airflow simulation → Proximity airflow simulation → Aircraft
 

Very good writeup!  This gives more details of how the 20 KM CFD works. I really like the direction that Asobo is heading. It's amazing how much the flight model in MSFS has advanced over the last 2 years.

2 hours ago, lwt1971 said:

but it updates much more frequently and is affected by just about every object in the scene ( including trees and buildings), as well as live data, at a resolution of about a foot. This simulation is done on a single slice of the atmosphere (one voxel in height) at the altitude of your aircraft. The atmospheric airflow simulation is an input for the proximity airflow simulation until the boundary layer, after which it is just the wind (live data) that affects the proximity airflow simulation.

This is very interesting. So, they're trading cells from the vertical dimension to increase resolution in the horizontal directions. That would open the door for 1ft res, though the tradeoff is that it's more difficult to simulate the up/down drafts. Merging masses of air will essentially be two huge rotating vortices spinning against each other. Can't get that vortical motion in CFD w/out the vertical mesh cells. And sure enough, the updrafts are calculated prior to the CFD run. They are fed in as boundary conditions. 

I'm betting that the proximity simulation isn't running at every flight model update. Probably ever few seconds at the fastest. Still won't be much performance penalty which is, again, pretty cool.

 

2 hours ago, lwt1971 said:

Could a similar flight model for the S4 have been achieved without CFD? I remember you had some interesting times trying to get more info during the SU9 days on the developer forums 🙂
....
But I also wouldn't dismiss local aircraft CFD as useless, since it's obviously made the default C172 better than before, and is being used in great aircraft like the Sting S4.
 

Yes, a similar flight model could be created w/ out CFD. It's well known that PMDG is ignoring the CFD because it currently only influences prop wash. Or, in other words, it's no savior. I mean, they ignore Asobo's model in general, so... not very advanced, I guess.

The CFD implementation is a bandaid to make up for a lack of sufficient FSX table correlations. Granted, such tables couldn't even be bought from the airframers, so there really is no other choice.

You can thank XP's flight model for that competition/inspiration. 

The marketing department wants everyone to see CFD as advanced flight modeling because it helps sell product. Engineers who use it all the time laugh and pity the 3rd party dev who has to do the clay molding from there. It's more art than engineering. You have to apply nonsensical, unrealistic values to the model to get it to work. Hence the "interesting times" you just referred to.

Edited November 7, 20223 yr by blingthinger

27 minutes ago, blingthinger said:

Yes, a similar flight model could be created w/ out CFD. It's well known that PMDG is ignoring the CFD because it currently only influences prop wash. Or, in other words, it's no savior. I mean, they ignore Asobo's model in general, so... not very advanced, I guess.

Well it obviously is the case that great FMs can be developed without local-aircraft-CFD (the aircrafts I cited like the Fenix A320, Milviz C310, PMDG 737 are good examples)... I am just interested to hear from aircraft developers like simbol (of FSReborn) on their experience in using CFD for their birds. If it was useless or mostly marketing gimmicks then devs wouldn't have employed it... I'm also not hung up on how it's billed as "advanced" or not by whomever. IMO, for local-aircraft-CFD to really shine MSFS needs to gives aircraft developers ways to define the shape/geometry of aircraft in much more detailed/comprehensive ways. Short of that, CFD is always going to need to be complemented by tweaking/tuning in the FM of various parameters/scalars/etc. But with with a combination of such tweaks *and* CFD, does a FM in MSFS have more advantages vs not using CFD? That's what I'd like to learn more about.

Local-aircraft-CFD aside, the core MSFS aerodynamics engine is quite capable and IMO no lesser or greater compared to XP's.. regardless of how great/good the core platform of a sim is, an aircraft's flight model is always going to live and die by the fine tuning/tweaks that the aircraft developers do.

That said coming back to the topic of this thread, i.e. this new atmospheric airflow modelling in SU11 driven by CFD, is quite novel.. and I'd say falls squarely in the "advanced" category, especially compared to other sims.

 

Edited November 7, 20223 yr by lwt1971

2 minutes ago, lwt1971 said:

this new atmospheric airflow modelling in SU11 driven by CFD, is quite novel.. and I'd say falls squarely in the "advanced" category

I agree.

 

3 minutes ago, lwt1971 said:

I'm also not hung up on how it's billed as "advanced" or not by whomever

Ah, but you do. And the marketing department cares too. Tell you what, when NASA starts requesting job experience with Asobo's flight model (CFD or otherwise), or airframers use it as a prototyping tool, you can start calling it "advanced".

 

3 minutes ago, lwt1971 said:

IMO, for local aircraft CFD to really shine MSFS needs to gives aircraft developers ways to define the shape/geometry of aircraft in much more detailed/comprehensive ways.

Yup. And that's not coming any time soon. The entire flight model hinges on the flat plate geometries described in the SDK. They don't even trust it beyond prop wash, and that's sketchy at best. Lots of fudge factors. Not "advanced".

16 minutes ago, lwt1971 said:

That said coming back to the topic of this thread, i.e. this new atmospheric airflow modelling in SU11 driven by CFD, is quite novel.. and I'd say falls squarely in the "advanced" category, especially compared to other sims.

Well said! I definitely agree that is falls in the "advanced" category compared to other sims!

25 minutes ago, lwt1971 said:

this new atmospheric airflow modelling in SU11 driven by CFD

Also, it's technically not driven by CFD. It's driven by Meteoblue's weather model. The CFD is nothing without the boundary conditions extracted from the global model. Though it's true that the forecasts they are using are most likely CFD-based (e.g. if it came from NOAA), that would mean competitor sims are using CFD too. 

Augmented/enhanced by CFD is more accurate.

28 minutes ago, blingthinger said:

Ah, but you do. And the marketing department cares too. Tell you what, when NASA starts requesting job experience with Asobo's flight model (CFD or otherwise), or airframers use it as a prototyping tool, you can start calling it "advanced".

Well you seem to have insights into the MS/Asobo marketing department that I don't. To be honest, I could care less about XP and the usual stuff XP fans parrot about NASA and whatever else 🙂 ... this is the MSFS forum, and we're discussing atmospheric airflow CFD so don't want to devolve this into a sim vs sim back-and-forth. If some in this MSFS forum feel that the great technologies coming out of Asobo in the aerodynamics space are "advanced" so be it, no need to feel insecure about how XP compares to it.
 

Edited November 7, 20223 yr by lwt1971