Well, this is a very interesting project. 

All my testing so far has been reading datarefs, including  the aircrafts mass, wing dimension, flap settings etc, and injecting the data into the Dryden filters then changing the forces on the aircraft.

I am now entering a second round of testing. XP12 obviously knows the aircraft data and is injecting the turbulence.  This time I am analysing what turbulence XP12 is injecting into the sim, putting that data through the Dryden filters and then adding/or removing control forces as required to better fit a Dryden response.  The script is learning each aircraft and noting what additional/less effects are required to fit the Dryden filters.  This may be a better way to go.

I am a real pilot, I'd love to try this out. Flies maybe 40-50hrs per year in Cessnas and Diamond Stars.

Carry on your work @MrBitstFlyer, the majority of us roots for you and wishes you nothing but good fortunes.

Yes it continues to be quite impressive to watch and see what these LLMs are capable of with appropriate prompts and interactions. 

Development is going well.  I have now included MIL-F-8785 gust gradients. into the calculations.  Essentially this is what the script does.

• Analyze the turbulence XP12 is injecting into the sim.
• Analyze the turbulence with The Dryden model and MIL-F-8785 gust gradients.
• Inject the forces required to shape the turbulence to match the model specifications.

Feeling very good - the turbulence is +-5% to the model specifications..  I am nearing an initial release as a LUA script only.

Edited May 12May 12 by MrBitstFlyer

2 hours ago, MrBitstFlyer said:

shape the turbulence to match the model specifications

What variables is it using to make this match? E.g. local_y? local_vy? Normal force?

 

2 hours ago, MrBitstFlyer said:

model specifications

Please do elaborate on what you've interpreted these specifications to be.

1 hour ago, blingthinger said:

What variables is it using to make this match? E.g. local_y? local_vy? Normal force?

 

Please do elaborate on what you've interpreted these specifications to be.

The script is not checking aircraft height, verticle speed and trying to force those to match Dryden - this is what I tried to do in my first attempt and realised this wouldn't work. The script is generating its own Dryden based gust values internally which are logged in a CSV.  The gust values go plus and negative, so the script is measuring the size of the movement (gust). A force/moment is added to XP12 effects to improve how that turbulence feels.  An 'observer' constantly evaluates how the aircraft responds so it can increase or decrease its added turbulence effect to what XP12 is producing itself.

By specification I mean expected behaviour of the generated gust signals.  I am mainly checking Dryden based gust signals but also using MIL-F-8785 gust gradients as a small additional rotational cue for roll/pitch/yaw disturbance.  It is blended into XP12, rather than replacing XP12's own turbulence model.

23 minutes ago, MrBitstFlyer said:

expected behaviour

Behavior of which variables? Which physical datarefs are being compared to Dryden's spectrum curve fits?

Per MIL-STD-1797, the model's outputs are the amplitudes of velocity oscillation of a passing turbulent eddy in one of the 3 Cartesian directions: vx (or y or z), at a given frequency in the modeled spectrum for that component. I'm surprised to see you using terms like "filter" or "transfer function" or iterative training of an acft model, given that they are neither found nor implied in the report.

 

23 minutes ago, MrBitstFlyer said:

rather than replacing XP12's own turbulence model

Appears that you ran into the same brick wall I did on this one. I couldn't find a way to turn it off either.

 

23 minutes ago, MrBitstFlyer said:

how the aircraft responds

Same question as above. How what responds? Which datarefs is it monitoring?

Edited May 12May 12 by blingthinger

33 minutes ago, blingthinger said:

Appears that you ran into the same brick wall I did on this one. I couldn't find a way to turn it off either.

That is why I realised the way to go is add/remove effects to XP12's own turbulence.

 

35 minutes ago, blingthinger said:

Same question as above. How what responds? Which datarefs is it monitoring?

Normal force, Roll/pitch/yaw.

 

36 minutes ago, blingthinger said:

Behavior of which variables?

wind speed and direction, live/cockpit wind behaviour, used as a gust proxy, XP12 turbulence value where available, airspeed, altitude / AGL, aircraft mass and inertia, roll, pitch and yaw rates, bank/pitch/heading-type state, whether the aircraft is on the ground.

58 minutes ago, MrBitstFlyer said:

wind speed and direction, live/cockpit wind behaviour, used as a gust proxy, XP12 turbulence value where available, airspeed, altitude / AGL, aircraft mass and inertia, roll, pitch and yaw rates, bank/pitch/heading-type state, whether the aircraft is on the ground.

So everything except turbulence eddy velocity delta, which isn't an option because it isn't provided by the sim. That's comparing apples to oranges. The amplitudes will be all manner of mismatched and I couldn't consider that by itself an implementation of Dryden's model. So where is Dryden's model still entering the picture?

What values of length scale and intensity are used? Is it using a single value for each? If yes, that would certainly be a pity given that the report offers full up atmospheric distributions for both values.

 

1 hour ago, MrBitstFlyer said:

add/remove effects to XP12's own turbulence

This must be where it's using a transfer function somehow. Had to do my own LLM search on this bit as I've never needed to actively alter a real-time signal IRL.

13 hours ago, blingthinger said:

What values of length scale and intensity are used? Is it using a single value for each?

Yes, single at the moment for development to a stable state - which I'm at now.  Currently 

L, sigma but going to 

Lu, Lv, Lw
sigma_u, sigma_v, sigma_w

8 hours ago, MrBitstFlyer said:

but going to 

Lu, Lv, Lw
sigma_u, sigma_v, sigma_w

Very nice.

Silence on the other question though...there's some scaling or morphing of the spectra curves happening somewhere. Question is if it's a linear shift or a scale factor and if it's uniform to the full spectrum. The reason I asked about L and sigma is because that's what's setting the spectrum shape at a given altitude. Depending on how it's being scaled or shifted, you might still be able to claim that it's using the Dryden model. Your silence is worrying there. It might be starting with Dryden sure, but if it's then morphing the model beyond recognition simply because you ran an iteration on a plane and say 'this doesn't feel right', and it wrings the curves out of shape, it's suddenly no longer Dryden. If you're maintaining the spectrum slopes and therefore energy dissipation into the higher wavenumbers/frequencies, your claim holds water. Otherwise, I currently don't see how you can claim that it's using Dryden.

5 minutes ago, blingthinger said:

Your silence is worrying there. It might be starting with Dryden sure, but if it's then morphing the model beyond recognition simply because you ran an iteration on a plane and say 'this doesn't feel right', and it wrings the curves out of shape, it's suddenly no longer Dryden. If you're maintaining the spectrum slopes and therefore energy dissipation into the higher wavenumbers/frequencies, your claim holds water. Otherwise, I currently don't see how you can claim that it's using Dryden.

Left home at 5am and got home at 8pm - not that worrying😁

To be more accurate as far as Dryden is concerned - Dryden shaped gust core with XP12 specific response shaping.  Currently the gust channel sigma is accurate between 0.9 to 8.3%.  Testing continues.

Maybe this isn't currently close enough in your book, but I have been very aware of working the script to the numbers, rather than it feels right (even though it is feeling great).  I'll get it as close as possible, but I would imaging Laminar would need to put this in the base sim to be 100% accurate to Dryden.

13 minutes ago, MrBitstFlyer said:

Dryden shaped gust core

And what happens after that? If that statement was all there is to it, there would be no need for iterating on feel for each acft. Pending clearer descriptions, you're saying that 8.3% of apples are oranges. Not a matter of "good enough". It's simply not using Dryden. Maybe starting there to some degree but the model itself appears to get smeared away based on what you're describing. It's a good starting point to "feel" better, but "feel" isn't "apply" as you've been claiming.

LR wouldn't need to implement it internally for manual weather mode. I mean...that's exactly why the MIL document is bringing it up: to set a standard for flight simulators. The algorithm would have to calculate more accurate forces using the turbulent velocities and some basic geometry definition (fuselage areas, force coefficients, much like XP already calculates wing forces) but it's a complete model as is.

If sigma is the primary knob it's turning to tweak Dryden's equations, it's going to remain pretty close to what IRL data sometimes looks like. Spectrum plots in Fig 7 look pretty darn similar to the polynomial equations that Dryden came up with.

https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2022JD037491