Question to devs regarding performance accuracy

[SJ_Butcher]

With the new yak coming I read that plane will be the most accurate in performance terms which is 1% of margin error, do you plan in the future to standardize all current planes in the game?

 

Regards

[Nocke]

They can't just decide to want a given margin, and then establish it that easily. The specs, like climb rates, speed at given altitudes, roll rate and so on are not introduced into the model as fixed values, they are the outcome of tweaking other internal simulation parameters. As the model will never exactly match reality, this typically means if you force one spec very narrowly to a desired values, others will inevitably go away from their target values.

[TP_Silk]

Even in the real aircraft of the day there will have been much more than a 1% variance in performance. I don't really see the point in modelling to one particular aircraft that was tested on a given day or standardised average performance figures. Getting it close enough is more than adequate for my needs. If it's slightly better or worse than some spec I can imagine that I have been assigned an aircraft that is better/worse than the average, just as will have happened to many pilots back in the day.

[Lensman]

I hope that someday in the future a development team will make a digital simulation of my Ford Focus car and get back the 15mpg that I've been missing since new according to the manufacturer's specification ...  

[-TBC-AeroAce]

I don't understand why in this day and age why there is a need to rely so heavily on period data. Don't get me wrong I really respect what the Devs currently do but it surely would not cost much these days to do a CFD study for each aircraft.  Half of the work is already done as they have the aircraft geometry. Doing CFD may even come up with some revelations. 

[A_radek]

 

27 minutes ago, AeroAce said:

Doing CFD may even come up with some revelations. 

 

I think it could, even with these in game models. I'd find tests like these a great read.

 

Though we would still be arguing over quality of surface fit&finish, mass, power and so. 

 

In the end, as you can't do detailed cfd's in realtime on our pc's or pre-calculate every possible situation - it would still have to boil down to a simplified and very fast to calculate model. One presenting deviations from whatever the cfd tests came up with.

 

Was as it not mentioned they had calculated fuselage side area incorrectly for some aircraft and that this was the root of the old "roll with rudder like no tomorrow" problem? That sounded very interesting.

[-TBC-AeroAce]

2 hours ago, a_radek said:

 

 

I think it could, even with these in game models. I'd find tests like these a great read.

 

Though we would still be arguing over quality of surface fit&finish, mass, power and so. 

 

In the end, as you can't do detailed cfd's in realtime on our pc's or pre-calculate every possible situation - it would still have to boil down to a simplified and very fast to calculate model. One presenting deviations from whatever the cfd tests came up with.

 

Was as it not mentioned they had calculated fuselage side area incorrectly for some aircraft and that this was the root of the old "roll with rudder like no tomorrow" problem? That sounded very interesting.

 

Of course, one cant do CFD on the fly. You use the CFD data to create better aerodynamic derivatives to be used in the linearised dynamics equations. These derivatives (i.e how so roll effect lift or pitch effect roll.....) are what is hard to estimate for these models and are the black art some people talk about when they say you cannot change one thing without effecting them all. The point is the better the CFD the better these derivatives can be calculated. There is no reason that all the aerodynamic derivatives could not be found to a very high degree of accuracy using CFD instead of having to hit and hope and get it as close as possible using old data.

 

Yes you are also correct that the mass distribution and centre of gravity cannot be found without actually knowing it or having an extremely detailed model with all the mass information for each component but in practice, these are only 3 or 4 numbers that are not very dynamic and only being a function of fuel burn making them very simple to estimate if all the aerodynamic information is good.

Edited March 5, 2018 by AeroAce

[A_radek]

 

Though again I wonder if there has been any simpler cfd tests done already. The models we have, even with those airfoils consisting of ~20 straight lines, could with interpolation still give some rough estimates usefull for something. And I'm sure that devblog I can't find mentioned calculating fuselage shape for each aircraft. So unless I'm dreaming that up it would imply they already went there.

[Nocke]

but even if you had a valid cfd model for each plane you still had to boil that down to some simpler model which can be run in real time in the game. Thats where I suppose the difficulties lie. This simpler model will never reproduce exactly all you can ( perhaps) get out of your cfd model. And please don't underestimate the necessary effort to create a really reliable CFD model.

[-TBC-AeroAce]

Ok so I have made many real time flight simulations that will run on a normal PC but never with custom graphics, only a basic representation of what the plane is doing.

 

The real time simulations are done using what is known as the linear equations of motion. Flight is actually very non-linear so you cannot just solve the base equations you need to break them up into ranges where the flight characteristics appear to be linear. For example, drag vs speed is a big U shape curve, so to solve for drag in real time you need to break the curve up into straight line sections that represent the real curve. This means that using programming logic any curve can be represented simply by saying for this speed range use this linear value for the next speed range use the next linear value .....

 

Now the linear equations of motion require you to know a number of these relationships as mentioned above and they are called flight derivatives. They basically relate a change in one thing to changes in everything else it may effect in a simple (mathematical!!!) way.

 

Now what I suspect the DEVs are doing is using the historical data to extrapolate these derivatives to be used in the linear equations. But using this method will always leave you without the full picture because there is never enough data to get a good number for all the derivatives.

 

My argument is that with CFD you could set up a way of gaining all of the derivatives needed by the linear equations of motion without having to guesstimate a load of them. This data would then produce a much more unbiased game as the math will be doing the talking.

 

Just so I'm clear the accuracy of the game come from the accuracy of the numbers you select for the derivatives. If the derivative data is very accurate there is no reason that the game cannot achieve an extremely high accuracy. You don't need to run a live CFD to get it correct you just need to run it once get the data and manage it correctly and then feed that into a "corner cutting model" that will gain very accurate results. This is pretty much a very simplified way of how planes have been designed for 30 years ish, with great results.

 

GARBAGE IN GARBAGE OUT!

 

There is the above way or the panel method way as used by Laminar Studios or whatever they are called uses.

 

 

Edited March 5, 2018 by AeroAce

[Nocke]

so lets stay just with your drag vs. speed curve, as an example. to be able to really describe all you need to reproduce the details beeing asked for here you need that as a function of angle of attack, elevator, aileron, rudder position. cowling, radiator settings. canopy open? gear out? flaps? meredith? Are we side slipping? and probably more. It would take a master thesis for each plane actually to do all that correctly in a cfd model.

[Nocke]

well meredith no .. please forget that

[Nocke]

Actually what I would like to say is just this: I can't imagine that it should be possible, with the resources available to the developers, to establish a CFD model for each plane, capable of making predictions exact enough such that you would not need historical data anymore, as suggested somewhere above. I am doing CFD for my living, however not planes, so perhaps my impression is wrong. Somebody around here who could seriously estimate the required manpower?

[-TBC-AeroAce]

27 minutes ago, 216th_Nocke said:

so lets stay just with your drag vs. speed curve, as an example. to be able to really describe all you need to reproduce the details beeing asked for here you need that as a function of angle of attack, elevator, aileron, rudder position. cowling, radiator settings. canopy open? gear out? flaps? meredith? Are we side slipping? and probably more. It would take a master thesis for each plane actually to do all that correctly in a cfd model.

 

Yes and no.

 

TO be honest it would cost about a DEVs salary for at max a week to get all the numbers needed for a plane. These days it does not cost that much to get this kinda work done. So it is a question of worth. I like the idea cause science and math settles it not all kinds of bias we currently have.

 

CFD now works in the "shut up and calculate" kinda way reminiscent of the early days of quantum physics. You don't need a thesis and now auto meshing works so well all you need to do is tell the program what you want to look at (more or less).

Get with the time's bro

Edited March 6, 2018 by AeroAce

[Nocke]

Well ... I disagree. I get dizzy just starting to think what all this would require. You are certainly aware e.g. of the discussions here about roll rates at high speeds? To get this from your model, you would have to couple it with a mechanical model of your plane to get the deformation of the wing. And I bet, once you really start trying to resolve the discussions going on here about wrong flight models there are more things coming in.

And excuse me, but it is really hard to believe for me that it should take a week to get the numbers, then one could fill all into an existing model, press a button, and you have all the data needed? Have you done that?

[Nocke]

22 minutes ago, AeroAce said:

CFD now works in the "shut up and calculate" kinda way reminiscent of the early days of quantum physics. You don't need a thesis and now auto meshing works so well all you need to do is tell the program what you want to look at (more or less).

Get with the time's bro

Thats what the CFD software producers want you to believe.

[-TBC-AeroAce]

5 minutes ago, 216th_Nocke said:

Thats what the CFD software producers want you to believe.

 

Well if you didnt know the Navi-Stokes equations, that CFD is based on has no known solution and has a big prize for anyone that can close it. That is why I said "shut up and calculate" BECAUSE IT WORKS IN THE REAL WORLD!!!

 

Ok all im really trying to say here, in the long run, is that it would be so much better if we could read in a geometry file and get the results based on science and math instead of using just historical data that is known for being a bit shit!

 

Some basic CFD + Historical records could really make the difference!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Edited March 6, 2018 by AeroAce

[Nocke]

It does NOT wirk for wgat is needed here. Another example: To really make a difference, your model should be able to gelp to decide the discussion about how fast the La5Fn should really be. As far as I understand that comes down to the differences between a soecially prepared prototype and a serial production plane. That means your modek would need to oredictuvely describe the effects if surface finishing, gap closures abd alike stuff. That is simply not possible with your ine click auto mesh model. B.t.w. the closure problem is not about the Navier-Stokes equations, its about turbulence modelling.

[-TBC-AeroAce]

26 minutes ago, 216th_Nocke said:

It does NOT wirk for wgat is needed here. Another example: To really make a difference, your model should be able to gelp to decide the discussion about how fast the La5Fn should really be. As far as I understand that comes down to the differences between a soecially prepared prototype and a serial production plane. That means your modek would need to oredictuvely describe the effects if surface finishing, gap closures abd alike stuff. That is simply not possible with your ine click auto mesh model. B.t.w. the closure problem is not about the Navier-Stokes equations, its about turbulence modelling.

 

NOOOO GEOMETRY IS KEY !!!!!! IF U HAVE THE PRODUCTION GEOM THEN U CAN WORK IT OUT!!!!! (Given power and moments of inertia)

 

Yes it would not work for small production differences. But that is not the point! It would be to fully understand an average set of the derivatives for each plane!(don't nitpick)

 

And yes I know it is about turbulence but tell me a real world without it! You would need it for the prop wash at least! (was trying to keep it simple for people that are not as informed as u)

 

 

As complicated as this subject is, keeping a base level of production standard would initially be key. If you wanted to model production defect all you would have to do is carefully add random numbers multipliers to the base model.

 

You are not seeing what I am trying to say which is that CFD is a tool that is being underutilised by all sim DEVs as a whole.

 

PS just search "CFD out sourcing"

Edited March 6, 2018 by AeroAce

[Panthera]

I should note here that whilst CFD programs are great and all, they still cannot replace actual real life flight testing which pretty much always brings to light things about a design the computers didn't predict.

 

So whilst CFDs of every plane ingame would be neat, it wouldn't solve every single accuracy issue we have ingame atm.

[1_Robert_]

I hope you all realize that with each post you guys make in this thread, simpler folk like me feel dumber by the post. 

[ShamrockOneFive]

1 hour ago, 1_Robert_ said:

I hope you all realize that with each post you guys make in this thread, simpler folk like me feel dumber by the post. 

 

They are just specialized in this area. I read with great fascination but I don't think I could contribute meaningfully either

 

I am interested in the philosophy of the debate. Do the devs go for 1% accuracy based on published data as tested from real aircraft or should they (in the future anyways) use an extremely detailed analysis of the aircraft's shape and composition to create numbers to base the modeling off of. I suspect that in the future that may be how things go... Especially if, as you guys say, the tech to do that is becoming more available/cheaper to do.

[unreasonable]

While I am all in favour of as many checks as possible, I do not see that substituting CFD modeling for historic tests data as proposed actually solves the OP's issue.

 

Suppose the modeling is done as AeroAce suggests for some specific aircraft, producing a value for some factor of 100.  The historic tests suggest a value of 96.

Which do you model in the FM? The CFD value may not have the issues of a historic test - but perhaps it is simply missing some variable that is present in the original physical aircraft. The community reaction is bad enough as it is when actual test data is used if the results do not seem to correspond with anecdotal accounts: imagine the uproar if the historic tests were overthrown in favour of a necessarily incomplete theoretical model.

 

I can see this only being useful if the CFD modeling were sophisticated enough that it could explain the difference between different test results in terms of the the difference in the conditions of the tests. Given the multitude of undocumented variables in a real test, I doubt that this is achievable. 

 

Even after you have decided whether to model the FM value to aim at getting 100 or 94, what we have seen is that it is not trivial for the the FM to deliver that value while at the same time maintaining the values for other variables. So I see this whole argument as a red herring.

 

Going back to the OP's point - I am sure the devs would like to get closer to a 1% margin, and if they can gradually do that that is fine, but what we currently have is well within the natural variation of actual production aeroplanes.  This really is not at all important for the health or future of the sim. 

[blitze]

6 hours ago, ShamrockOneFive said:

 

They are just specialized in this area. I read with great fascination but I don't think I could contribute meaningfully either

 

I am interested in the philosophy of the debate. Do the devs go for 1% accuracy based on published data as tested from real aircraft or should they (in the future anyways) use an extremely detailed analysis of the aircraft's shape and composition to create numbers to base the modeling off of. I suspect that in the future that may be how things go... Especially if, as you guys say, the tech to do that is becoming more available/cheaper to do.

 

Maybe a combination of the two?  I see them complimenting each other not either or.  Also modelling in game would have to be relatively simple to an extent given the metric of the game itself and the hardware range it is being played on. 

 

Anyway, even if we get an ideal model for a given aircraft design, we are talking 1940's era production here with the added bonus of War time duress on the industry.  I would think a 95% to 97% range would be good for our purposes.  Still, given Germany's industry duress, they did manage to produce equipment to quite a high standard but alas, they in their over engineering and tolerances perfection failed to account for variations in operating environment.  Then there was infield maintenance.

 

So, do we aim for 100% ideal representation of modelling a given aircraft and then scale the performance back 2% to 5%^to give more real world performance??

 

How would CFD deal with atmospherics as well taking into account air densities at altitude and under a variety of seasons / weather??

 

Really, aside from being there and then flying the actual machines in combat, everything is a relatively good estimate be it from historical test data or modern computational analysis.

[Holtzauge]

Someone once said that to a man with a hammer all problem look like nails and this also seems to be true for CFD:  Just because CFD produces pretty pictures like Panthera’s excellent cow avatar you should not be fooled into believing that it splits the Gordian knot because doing a full blown numerical Navier-Stokes CFD on problems such as a complete fighter with rotating prop at high angles of attack would require humongous amounts of computing power and time. Maybe the NSA could do it but I doubt it.

 

That being said of course CFD can be put to good use: For example, I think CFD would be a great tool to model a fillet between the wing and fuselage on a jet airliner to get it just right to minimize the drag. Here you have mostly attached flow (that is what you are aiming for anyway) and you have constrained conditions (small variations in angle of attack, Reynolds number and Mach) since you are optimizing for cruise conditions.

 

Another good example of how CFD can be used productively is David Lednicer’s CFD analysis of the Spitfire using VSAERO which showed that the armoured glass was set at a too steep angle which lead to flow separation. But Lednicer’s analysis was a high speed analysis, i.e. low angle of attack with largely attached flow, not a turn rate analysis close to stall with a lot of vorticity and separated flows and doing a full blown CFD solving the viscous flow according to Navier-Stokes for a fighter close to stall (as in a turn rate analysis for example) with a rotating propeller model and surface imperfections etc.  is a wholly different kettle of fish. Even if you do it just for a single angle of attack at one particular Mach number it will take an enormous amount of computing power and time and how can you ever hope to do it for the whole flight envelope for even one aircraft?

 

Finally, I can’t for the life of me understand why anyone would think a CFD model would be better than flight test data? Calibrated and corrected flight test data is the truth: CFD is just a model, and show me the CFD model that models the effects of the propeller slip stream, gun ports, steps, leakage from door wells, engine cowls, aeroelastic twisting etc. I will take calibrated flight test data averaged over a number of representative specimens over CFD any day of the week…...

Edited March 6, 2018 by Holtzauge

[A_radek]

That's depressive. Some of us like the thought of using cfd's and you people are ruining it. But ok, let's 3D print and head for the wind tunnel instead Yay!

[Eicio]

You just can't be perfectly accurate because even the vehicles in the real life aren't accurate to themselves, that is to say that there are many more parameters that should be considered like the materials quality: one same vehicle may not be assemble in the same materials and of course that changed the way they behave, the assembling could be different if you are building it in a peaceful plant or in a dark tunel under fire, the vehicles especially in the second world war were not copied pasted and you planes sure look quite the same but a welding may have been done a little differently and so on...

 

You can't achieve a perfect recreation of a vehicle in a simulation, all you can do is create a more likely representation of it and that is fine because it means you can always try to improve it.

[Holtzauge]

1 hour ago, a_radek said:

That's depressive. Some of us like the thought of using cfd's and you people are ruining it. But ok, let's 3D print and head for the wind tunnel instead Yay!

 

Well this will get you even more depressed: When your nifty 3D printer has done it's job and you head over to your windtunnel you will get the wrong results there as well since you will not be running it at the same Reynolds number as the IRL plane. That's unless you have one of those really BIG 3D-printers that can print you a full size copy of course. Then when you run it in your windtunnel how big is it? If it's infinitely big you will be OK but if it's finite then you will need to apply windtunnel wall and blockage corrections to your results. So to get true results you end up needing a fullsize model running in an infinite windtunnel which sounds a bit like a testflight don't it.......

Edited March 6, 2018 by Holtzauge

[Pupo]

Holtzauge pretty much nailed it. CFD can be used to answer very specific questions. Do you want to know the lift generated by a spitfire wing at one specific angle of attack? Not a problem. Do you want to know the drag caused by the radiator at one specific speed? Sure, it can be done. Do you want to know how much thrust the propeller is generating at one specific speed, and AOA? Will do.

 

Do you want to do all of these at the same time? Forget about it. Each of the above mentioned cases require completely different simulation setups, there is no "1 method to rule them all" in CFD. 

 

This does not mean CFD is useless. When designing a new part, you can easily identify a problematic part, say the radiator is producing too much drag. You then measure the radiator drag on a wind tunnel,  probably using a scaled model. You then use the collected data to calibrate your CFD model. Then,  use the calibrated CFD model to identify the origin of the problem and test new possible designs. Finally, you produce a prototype of the new radiator and put it to test in the wind tunnel. And, if it's working as it should, you start producing this new radiator and replacing the old design.

 

Calibrating an entire lumped parameter model (which is what IL-2 is) based on CFD simulations would be insanely expensive, and somewhat futile. You'd still need to calibrate your CFD simulations using test flight data. So, you might as well just use the flight data to calibrate IL-2's FM's and call it a day. As for the expensive, a few weeks ago I ran this simulation of a motorbike riding on a road and it took my 12 Core machine 24 hours to complete (0.2 seconds of real time). And the results were pretty, but borderline useless since the mesh used was not refined enough to correctly capture most of the aerodynamic effects you are trying to model.

 

Anyway, If anyone has a suggestion on something they would like to see modelled using CFD, you can propose. If it's not too complicated I will give it a go and post results. Just for fun.

 

Cheers

Edited March 6, 2018 by Pupo

[A_radek]

I don't think anyone here thought building a full on virtual replica and testing it in a simulation of reality was a time saver or something we could rely on for it all. Correct me if I'm wrong.  I do very simplified fluid dynamic simulations at work with lot's of processing power at hand, so can well imagine the magnitude of a project that was to simulate an aircraft in every detail. And that's without considering building the required and dynamic geometry.

However I have seen mentioned here a simple cfd could reveal something. So here's a thought to shoot down:

Say we have no valid test data for a given aircrafts roll rate, only very positive pilot anecdotes. So let's run a simulation with the left wing half in the hopes of getting an idea of how much force a 20 degree aileron deflection would need at x speed. To keep it simple let's only consider aileron shape/hinge type and leave out flex, surface and mass. Pretend the engine is off. Perhaps we even know the gear ratio between stick and aileron deflection. Would a cfd simulation like this, even when excluding some variables not give us a ballpark figure? 

(Holz I know nothing of wind tunnel testing. Had no idea you needed to take walls In to consideration which makes sense. Though is there no way to reliably recalculate and compensate for the small scale?)

 

Edited March 7, 2018 by a_radek
I think Pupo answered my post.

[A_radek]

7 hours ago, Pupo said:

Anyway, If anyone has a suggestion on something they would like to see modelled using CFD, you can propose. If it's not too complicated I will give it a go and post results. Just for fun.

What format mesh can your software take Pupo? I'd be happy to contribute with the geometry for a community test.

[Nocke]

3 hours ago, a_radek said:

Say we have no valid test data for a given aircrafts roll rate, only very positive pilot anecdotes. So let's run a simulation with the left wing half in the hopes of getting an idea of how much force a 20 degree aileron deflection would need at x speed. To keep it simple let's only consider aileron shape/hinge type and leave out flex, surface and mass. Pretend the engine is off. Perhaps we even know the gear ratio between stick and aileron deflection. Would a cfd simulation like this, even when excluding some variables not give us a ballpark figure?

 

but ... exactly this is a good example for why the sort of simplified modelling you can do quickly might not provide you with something usefull at all. As far as I know elastic bending of parts IS of key importance to get the roll rate correct. I would not be surprised if in this simplified model you are proposing all planes of comparable size and weight would roll just identically. And what would it help you to know the force required for a given aileron deflection? You want the resulting roll rate.

[A_radek]

Nocke. I omitted to write you would need to calculate for both, but the cfd could reveal how much force the aileron/wing at an X angle is producing towards the roll. And you might very well be correct that calculating roll rate from the latter means nothing if not considering stiffness of the wing/aileron.

I'm not so sure they would all roll equally though? Perhaps if we average roll rate throughout the speed envelope for every aircraft and then compare it all starts to even out. But there are after all different types of ailerons/wing shapes, not only different size/force lever ailerons. So if we lack proper real life test data, don't we get something better to base our guess on than just a guess from space?
 

[SJ_Butcher]

Lol this topic turned into a research simulation of fluids and aerodynamic... My question was focused into the existent data, with the new yak setting new parameters of accuracy, is there  any plan to carry the old planes into the new standard?

[Holtzauge]

@a_radek: About the windtunnel corrections: I have myself only done this in lab exercises while at university but I have later on socialized with the guys doing windtunnel tests on the JAS39 Gripen and discussed the problems with them. Not only do you have Reynolds number (Re) effects but many models have strips added to the to trigger the boundary layer to go turbulent and if you go to higher windtunnel speeds to get a higher Re you end up with Mach problems. NACA had a very nice variable density tunnel that fixed a lot of issues but of course the small models can never capture gun ports, cowl and wheel bay leakage, steps etc. so as usual, no free lunch and you are left with a bunch of corrections and assumptions about the results.

 

Then when it comes to the roll modeling in CFD I think Nocke and Pupo pretty much covered that problems but let me add that while the idea to do only half and mirror the results may be good in theory this may only be done for symmetrical problems and the thing with roll is that it will not be symmetrical: First of all the change in the flow field will be different because of the airfoil camber and then there is in many cases a differential in the deflection of the aileron and then you have in many cases frise type ailerons etc. so you need to model the whole thing. Then you are faced with determining how much deflection say 50 lb will get you and I would say the flow field around a frise type aileron which relies on separated flow is a tall order to model.

 

@216^th_Nocke: Appreciate your input on CFD modeling: Nice to hear the opinion of someone who does it for a living. Would you mind telling us what it is you model in CFD? Sounds like an interesting job. I do simulations as well but these are not CFD but involve radio propagation in urban and suburban environments etc. which is also challenging in a way since you in an urban environment have multiple radio paths between a base station and a mobile and these vary very quickly due to reflections and diffraction etc. and just like in CFD you can’t model everything down to the Al window sills, concrete reinforcements and paneling of every building……

 

@Pupo: Thanks for the info on the bike simulations and your input on how the modeling was done. Was an interesting read. One thing I think could be interesting to see is a 2D model of a schematic P-51 type diffusor/exit and see how this works if the boundary layer is ingested versus with a boundary layer separator like they had IRL. Probably tricky to model the boundary layer diverter in 2D but maybe this could simply be modeled in CFD as a diffusor with a flat plate in front of it to generate a boundary layer versus one where the flow is created just in front of the diffusor to simulate a boundary layer separator? OTOH I guess the details of modeling a bounday layer with a transition from laminar to turbulent at some point in CDF is still tricky business?

Edited March 7, 2018 by Holtzauge

[Nocke]

@Holtzauge I am generally working on simulation of crystal growth processes.

This involves, e.g. for the Czochralski process for the production of silicon crystals, be it for photovoltaics or electronics, the calculation of melt flow, heat and species transfer in hot, rotating metallic melts of up to several hundreds of kg.

Interestingly, velocities here are of course much lower than in aerodynamics. However, due to the high densities, we actually are dealing with Reynolds numbers in absolutely the same range as air passing over a wing at ww2 prop driven planes.

If I have learned something over the last 25 years doing this, its this: Its ALWAY more complicated and involved than you thought initially, and I bet its the same with the planes.

Its usually easy to set up a first model, as has been proposed above. You use some default settings, go ahead, get some results. Now comes the hard part - evaluate these results.

Usually you will find you are not that bad, not completely off, makes sense perhaps - but it does not really tell you anything new you did not know beforehand from experience, or from simpler engineering estimates. Once you really want details you did not basically know already, the effort explodes. You have to fine tune your boundary layers, the transition to the turbulence model. Tons of other things costing a lot of work and time.

Last but not least you will first spend a lot of time just calculating bullshit because of errors you have introduced at some point without noticing.

As has been said (by you?), DNS (direct numerical simulation, i.e. full resolution of the problem without turbulence modelling), is just not possible, except for selected test cases which you then let run for 1 or 2 months on some supercomputer, if you have access. So you need to fine tune and calibrate your model. Once that is done, as has been said as well, then, you can probably start to try to identify a smaller subproblem which you can tackle using your calibrated model to get some new insights.

 

It would of course be different if you could find someone with at least several years of experience in modelling the flow around WWII prop planes, and by chance owning his own computer cluster for the calculations. Then, I guess, it could be done. For all the planes we have here, I would estimate perhaps not much more than 5 to 10 years should be enough  

 

... but I could be totally off with that estimate. Have been and still am way of, in both directions, in my estimates of the time it will take to achieve something with the simulations ...

 

[Holtzauge]

How absolutely fascinating Nocke: Metallic flows that move at Re in the same range as airplanes! Who ever would have guessed that.....Enjoyed reading about that and the description of what you do. Sounds like pretty advanced stuff. Regarding the tuning and interpretation of results from simulations that you write about I can only agree based on my own 30 years of experience working in engineering: Things are usually more complicated than they seem and the more one sees the more one agrees with the old adage that "a little knowledge is a dangerous thing" and even though it sounds a bit cliche it's perfectly true that the more you learn the more you realize how much there is still left to learn which was not exactly the kind of attitude I displayed fresh out of university.......