the game uses drag coefficients, so yes they could use that data
2 Likes
The OP is suggesting that the flight models should be developed/trained using the virtual wind tunneling rather than manually adjusted.
Presently, if the wiki is correct - the current methodology is very, very, vaguely listed here:
https://wiki.warthunder.com/Flight_model_creation
- Finding and studying documentation, determining reliable data.
- Creating an internal “aircraft passport”.
- Calculating required parameters using known data.
- Introducing the values into the flight model.
- Thoroughly testing the model – all the values from all the tests and those indicated in the aircraft passport must coincide.
- If data differs from reference data by over 1%, further refinement work is performed on the model.
From my own professional experience - this methodology seems analogous to creating a QSAR model using purely experimental data.
Basically, you want to create a simple Matrix of Observed Parameters * Vector of coefficients = Vector of Observed Outcomes for each desired physics quantity.
From what I can assume based on the linked page, the matrix and vector of outcomes are populated purely using manuals, blueprints and the like and the coefficients are manually generated and adjusted based on the observed outcomes differing from what the books say.
What OP suggests instead is…
Take the matrix of parameters and observed outcomes and instead generate the coefficients using CFD modelling/“Wind Tunnels” rather than manual tuning.
Phrased differently, OP wishes for a more accurate initial guess for the flight models to be made.
Frustratingly, I cannot find any details beyond this wiki page or people complaining about stuff so it might well be Gaijin already employs some form of simulation/modelling given the “calculating required parameters” entry but it has limitations and whatnot.
As such, I find it difficult to talk anything about this subject without feeling like I’m guessing 90% of what I’m saying or phrased less civilized: everything we’re discussing here is pulling things out of our arse.
Our Best bet is probably looking at how FlyOut generates its flight models “on the fly” (heh) as it’s a game that allows players to design planes and it computes flight models based on that.
Edit:
X-plane is often hailed as having some of the best flight models from what I could see within flight simmer community.
This is what they describe as their process of automatically generating flight models which works even for user designed planes.
5 Likes
I don’t know much about aerodynamics, but i think it may be a better way to build flight models compared with the way of searching information and source manually.
Just like the J10A this time, I believe the developer has collected some information about it. But how much information collected, and how accurate it is, remains unknown for players.
Then they made a jets far from our impression. On the one hand, there isn’t much vaild information due to the confidentiality. On the other hand, we doubt the validity of some information they used (like the interview).
From this aspect, collecting data from wind tunnel test is definitely a good way to convince all of us if it is feasible.
2 Likes
Anyway it would be closer to IRL than now.
It use Gaijin iteration of it. But not actual principle.
You know, come to think, it would be nice for Gaijin if they also showed how they create their flight models, which sources were used for modeling and such (provided, ofc, if said sources can be publicly displayed and free to access)
Ofc, you need to check origin data to be sure your model is correct. But IRL planes fly so exactly because of it’s shape and laws of nature. Not because someone worte how they should flying.
So if you have correct model, you can put any shape form into it and it will fly same as in IRL.
One caveat I would add is:
Modelling things accurately within a narrow regime is much easier than having a generalized model that covers a wide range of behaviours. This is to say that a flight model that can handle subsonic flight excellently and in a performant, real-time manner might struggle with dealing with transonic/supersonic/near-ypersonic regimes and vice versa. This is especially true if your model requires parameter fitting rather than an ab initio calculation.
Another caveat is that modern planes are often designed with computer assistance with “quirky” design principles that may require a higher resolution/finer integration grid than what we might be able to obtain using rougher approximations in reproductions. Now, this might impact RCS more than flight characteristics, but even then.
Speaking of flight characteristics, modern fighters rely heavily on their fly-by-wire systems. I’m confident that this alone can cause significant performance disparities as I doubt our Instructor, amazing as it is within a gaming context, can manage all the engines and control surfaces as well as a millitary fight computer.
And control surfaces and engine performance themselves will cause differences that geometry alone won’t recreate. Thrust vectoring is a significant contributor to how modern jets control and fly (to the point that some proposed “Next Generation” jets even do away with vertical stabilizers for better RCS and instead use thrust vectoring for yaw control).
This all is not to say we shouldn’t try and implement more sophisticated parameter fitting for the flight models, but I do not believe methods used for creating simulators for general aviation and commercial flight can work to the same effectiveness as for high performance jets.
Original data is necessary for sure.
Yet, I think the biggest problem is feasibility.
One problem with your suggestion, those Virtual Wind Tunnels (VWT for later) are done by aerospace engineers, to analyse the data, and always carries some sort of errors,…and always required to verify those data, through test flight on prototypes
GAIJIN do not own any Aerospace Engineers able to oversee and validate those results.
Also, VWT are giving different results depending on which calculation software is used, because of how those softwares made calculation about Fluid mechanics (especially non-stable ones)
Those VWT might help having FlightModels more accurate overall, but they will not fix every errors, and requires People able to oversee those results.
(And i’m saying this as an Engineer, so i do know what i’m talking about)
And second tweak of your suggestion, is how you make it sounds so great, when it will just be costly, unsure and definetly out of reach of Gaijin Company right now.
(They’re not alike DCS, and stay an arcade game overall)
IMHO Gaijin more financially reliable than ED anyway. ED uses VWT for at least missiles if they cant get legal data about it, they said it in an interview few years ago.
agreed, gaijin should modify its errors.
1 Like
Good idea indeed.But details need more research.Hope that gaijin will do this.
Yes but they’re far more precise than Gaijin,… and don’t pick only the ED part and ignoring other things.
Gaijin do not own the skills and abilities in current staff to do that.
Give me like half a year and I might be able to do this in a rudimentary fashion (studying for it currently)
I do know the weakness of such system,…
I’m not smart enough to understand the science behind it, but if I guess it would result in more realistic performance relative to the engine running the game environment?
I wonder how expensive doing what FAR does would be computationally in the context of Gaijin’s pvp and damage models and all that.
The source code: Ferram-Aerospace-Research/FerramAerospaceResearch/FARAeroComponents at master · dkavolis/Ferram-Aerospace-Research · GitHub
In 2013, this was posted. It has been improved and iterated on since, but I cannot find a plain text explanation more recent than that:
Here’s a non-exhaustive list of things modeled in FAR right now; most of this is using approximations rather than exact solutions since the latter would be painful for the processor:
Wings
Effects of aspect ratio, taper ratio, wing sweep, camber (not intrinsic to any piece; must camber wing with multiple wing pieces placed at angles), and flaps accounted for.
Subsonic flow based off Prandtl lifting line theory, with stall above a given angle of attack.
Supersonic flow based off a combination of linear approximations for wings and shock-expansion theory for airfoils (using 10% thick diamond airfoils).
Transonic flow approximated as a smooth blending of the above.
Drag is a combination of the induced and wave drag factors predicted by the above, along with a simple approximation to account for transonic wave drag and a constant base friction drag.
Bodies
Effects of taper ratio and fineness ratio are considered. Sudden changes in cross-sectional area produce more pressure drag.
All flow regimes are approximated using simple potential and viscous flow calculations.
Extra drag added for unused attach nodes, which indicate a very un-aerodynamic surface exposed to the airflow; nodes facing the flow are assumed to have a stagnation pressure coefficient across the entire surface while nodes facing away from the flow are assumed to have half of the minimum pressure coefficient in that flow.
Most of the above stuff is based off of the 1965 Stability DATCOM that I managed to get my hands on. Not all of it is completely exact, but it does seem to provide a decent flight model.
3 Likes
I’ve used FAR back in 2013-2014 when played KSP. It wont requiers NASA PC. We’re now in 2024, i’m sure Gaijin servers handle it easy. And even if not - they can move most expensive calculations to the client-side.
But within the context i should say - if someone made a 3rd party mod for a game back in 2013 with such capabilities and it actually worked as intended, i cant see any reasonable obstacles why Gaijin cant do the same for War Thunder today.