For the past couple of weeks I have been researching turbulence in XP12.  We know XP12 has a good flight model, but I wanted to see if that could be enhanced by applying the Dryden Turbulence model MIL-STD-1797A — "Flying Qualities of Piloted Aircraft".  This adds effects and texture to the turbulence in XP12.  I haven't flown for many years, so I have a question - in a banked turn, can wind induced yaw actually stop the turn?

Any comments on this would be welcomed.

Claude didn't give provide a page number in MIL-STD-1797?

https://www.abbottaerospace.com/downloads/mil-hdbk-1797-department-of-defense-interface-standard-flying-qualities-of-piloted-aircraft/

Appendix A, p.804. Looks like a nice complete set of data to implement the model. How many frequency modes is claude using from those spectra regressions? Also, did it describe what it means by "effects and texture"?

By "wind induced yaw", do you mean a cross wind? In other words, can a cross wind (be it steady or transient gust) null out the forces of a coordinated turn and hold the plane in a banked attitude with no heading change? Depends on whether or not the arriving wind mass (either front or turbulent eddy) induces an opposite yaw moment. If you have a perfectly ideal/straight wave front arriving at the nose first, you'd end up with a brief period of opposite yaw moment. If the air-mass velocity delta holds constant everywhere as it envelopes the fuselage, it would balance out and the turn would eventually continue. If it's a series of turbulent eddies, you'd get the rock-and-roll of..well...turbulence.

Edited May 7May 7 by blingthinger

One thought about that model: it's only providing/simulating the 3 velocity components of a turbulent structure acting directly on a point in space (e.g. fsx table model acting on acft geometric center). Unless you create your own approximation (or send claude hunting for another model of) pitch, roll, and yaw moment, you'll be missing a fair amount of realism. In other words, it'll be very much "on rails" and will miss most of the dynamic sensation of turbulence.

Though that'd really mostly be a problem for low flight speeds. An airliner moving along at Mach 0.8 with a lot of momentum is enough to penetrate the turbulent eddies quickly enough that the angular moments really don't have enough time to make their presence known. Most of the time.

To the OP, yes, if a strong column of rising or sinking air hits the wingtip of a glider, it can surely "unbank"....

Turbulence such as that experienced in rotor clouds can not only "stop a coordinated turn" but actually flip your aircraft (glider and glider tug) upside down, as I've seen happening while flying in mountain ...

Turbulence can really be tricky.

In terms of flight simulation, X-Plane 12 can make different surfaces of the aircraft, like wings and tail surfaces, be hit asymmetrically by rising or sinking air. Supposedly this should be goos enough to represent the effect of "feeling the thermal", or the most powerful rising or sinking air as the tips of the wings of our gliders cross those areas ...

Edited May 7May 7 by jcomm

Didn't I spot something in the library related to that recently...?

Ah, yes.

https://forums.x-plane.org/files/file/99862-xpt_turbulence/

6 hours ago, blingthinger said:

Appendix A, p.804

And to clarify, that's the pdf page count. That MIL document is very interesting. It even includes a discussion of what to simulate when considering the turbulent wake of an aircraft carrier (p. 812).

Also, I hadn't seen table LIV before writing:

Quote

it's only providing/simulating the 3 velocity components

It does have the rotational moments. Very cool.

Edited May 7May 7 by blingthinger

10 hours ago, blingthinger said:

By "wind induced yaw", do you mean a cross wind? In other words, ....

Yes to everything you suggested.  Claude is just a tool and can produce quite silly results at times, and as I haven't flown for many years input from you lucky pilots is a valuable input to iron out flaws in the model data/XP12 transition.  Thanks for the input.  Certainly the Dryden model does introduce a cyclic element to a turn which felt excessive to my past memories of flying.

4 hours ago, Bjoern said:

Didn't I spot something in the library related to that recently...?

Ah, yes.

Yes, XPT is a recent release at the org and I have tried it, but it doesn't give me the feeling I remember from flying, especially low down on approach over undulating terrain.  Obviously my script will use the same turbulence models (I am using some elements from the more sophisticated Von Karmon model), but the trick is how that data is used within the XP12 environment.

I have flown hundreds of circuits while testing at flat, undulating, ridge top, coastal airports in XP12 using default XP12 turbulence and comparing that with what my script produces.  I am very pleased with the results so far. When released I can tune it depending on input from pilots of different aircraft types/environments.

 

Edited May 7May 7 by MrBitstFlyer

1 hour ago, MrBitstFlyer said:

you lucky pilots

I'm not that lucky. Maybe some day. But I am one helping make their dreams and careers a reality. Turbulence modeling is up one of my alleys, so to speak.

My beef with LLMs is when there's no attribution to the source of the blood/sweat/tears. I mentioned claude because I remembered you talking about it in the haze thread and your description also seemed a bit LLMish. I wanted to ask questions but kind of knew I wouldn't really be speaking to you.

I've toyed with this off and on a few times but never found a complete-enough model with atmospheric spectra (plus scaling factor distributions at altitude!) for all 3 axes. I also never found a way to disable turbulence in real weather mode or easily funnel the real weather turbulence intensity data to a plugin.

 

1 hour ago, MrBitstFlyer said:

Certainly the Dryden model does introduce a cyclic element to a turn which felt excessive

The report says to use von Karman (but then table values are for Dryden, weird). This sounds like a bit like it's mixing in more low frequency content than you'd expect. If I'm understanding the model correctly, I reiterate an earlier question: how many frequency modes is the lua applying to generate the final signal? Or is it applying the continuous spectrum somehow? You're not using real weather mode to test are you?

46 minutes ago, blingthinger said:

I wanted to ask questions but kind of knew I wouldn't really be speaking to you.

Eh?  I was clear in the haze thread I was using Claude, so it was clear I was the source of the ideas.  Claude isn't going to create a haze script or a turbulence one either, without a great deal of human input.  Without Claude I wouldn't be able to create anything, even though I have ideas and motivation and a lot of love for XP12.

For those that don't know much about AI, Claude AI is like having an expert in coding with little knowledge at all how I want XP12 to feel.  All my time is spent shaping, correcting and giving guidance of what I'm after and Claude does the drudgery of typing the code and error checking (no insult intended, I loved coding when I was doing it!).  I acknowledge your expertise in turbulence modelling and make no claim at all in that field.  What I do know is how XP12 feels to me and Claude AI helps me get those ideas into the sim. 

This may be of interest to some - two examples of an interaction with Claude

My input - feeling pretty good. Could you check the relationship between yaw and altitude plus the effect on roll. On approach the yaw amount feels good - floaty, not too sharp. as I climb (not exceeding 2000ft) yaw seems to have too much influence, with even a 45 degree bank seemingly not turning the aircraft for a while - maybe this is excessive yaw?

Claude's  response - Good observation — let me trace the altitude dependencies for yaw through the code. Found the problem. There are three compounding issues at altitude:

1. gust_t has zero altitude fading — wind speed gust yaw is equally strong at 2000ft as on the ground
2. bm increases with airspeed — climbing faster = stronger yaw moment coefficient
3. w_t at altitude has a much longer time constant — at 500ft, L≈75m (2.3s gusts). At 2000ft, L≈213m (6.9s gusts). So yaw moments persist in one direction for 5+ seconds, which is what's fighting your turns

The net result is yaw that's actually stronger and more sustained at 2000ft than on approach. Physically wrong — real surface-layer gustiness drops off rapidly above the boundary layer.

Notice how Claude didn't just know there was a bug in the yaw formula?  Claude AI  does this error checking and bug fixing much quicker, but needs human input - saves a lot of time for us non coders 😁

Another example

My input - should we use that second order dryden filter. Apart from the oscilation the turbulence is feeling really good now.

Claude response -  Good call, the second-order filter is actually the correct Dryden implementation per MIL-STD-1797A. The lateral and vertical channels are specified as second-order transfer functions:

H(s) = σ√(L/πV) × (1 + √3·L·s/V) / (1 + L·s/V)²

One could ask why Claude didn't just use that second order filter in the first place!

I am testing mostly in manual so I can set known conditions, but then also test each version in live weather - it has to be tuned to what XP12 is doing itself, and that was another interesting challenge.  If it feels good to me, maybe others will too.

Edited May 7May 7 by MrBitstFlyer

1 hour ago, MrBitstFlyer said:

Claude's  response - Good observation ..

<snip>

Claude response -  Good call, ...

Oh yeah ... 😉. You always get responses like this from AI, they want to ingratiate so that you feel well and use them more often. 

 

1 hour ago, MrBitstFlyer said:

One could ask why Claude didn't just use that second order filter in the first place!

If you ask Claude several times, sooner or later, it’ll end up contradicting itself 😄.

True story: I can't talk about Claude because I don't know it (in detail), but just today a colleague told me an example of his experience with AI (unfortunately I don't remember which one). He asked a question and the response felt strange to him. So he asked again, but the AI insisted that the response is correct. So, the colleague asked again with some more details. Still, AI responded in the same way as before. So the colleague, who actually knew the correct answer asked again like "Are you sure? Please think again." Finally, the AI said: "Oh ... I'm so sorry ... actually you are right and I was wrong." Other colleagues, who also have experience with AI confirmed that they (partly) have made the same experience and myself, too. So ... answers may be correct, BUT ... they can be plain wrong as well.
Disclaimer: I'm not saying that Claude is completely and always wrong but ... Conclusion: AI CAN be very helpful, but it also can lead you into the wrong direction if you believe everything you read.

The next steps on the internet will be that we have more and more AI generated content and one AI will use the input from other AI's and the result will be ... pure puzzle-solving based on probabilities.

Edited May 7May 7 by uwespeed

16 minutes ago, uwespeed said:

If you ask Claude several times, sooner or later, it’ll end up contradicting itself 😄.

Totally agree there!  Sometimes the answers are completely odd. As long as you understand this it is an excellent tool to speed up development, but there is plenty to laugh about along the way!

59 minutes ago, MrBitstFlyer said:

so it was clear I was the source of the ideas

Not at all. The LLM is doing the vast majority of the "thinking" both here and in the other thread. You're just feeding it your feelings on the topics. You're only clarifying your usage here in this thread because I brought it up. If I remember correctly, it was the same situation in the other thread (queried by someone else). You also don't mention it in your file post over on .org. Curious... 

1 hour ago, MrBitstFlyer said:

I acknowledge your expertise in turbulence modelling and make no claim at all in that field

I was merely probing for an intelligent conversation about the topic. Wondering if maybe I was wrong about your usage of a LLM. All good.

 

26 minutes ago, uwespeed said:

"Oh ... I'm so sorry ... actually you are right and I was wrong."

First thing I do if I ask any of them any questions is turn off the fluffy conversational responses. I'm continually stunned that people actually flock to it and respond to it. That side of the tech has turned into a sad commentary on humanity, though I suppose the wound has to be acknowledged before it can get proper treatment...

I'm going to bow out of this thread. I am excited to find a tool that I can use to create scripts for my enjoyment, scripts that I otherwise couldn't create.  I'll release at the org for others to try out and use to see if it increases their use of our wonderful XP12.  I am not interested in being psychoanalysed. Why do I need to justify my use of AI - maybe I should also say I use a word processor instead of a typewriter?

I posed the question for some useful input from pilots to improved what I - sorry, what Claude AI can produce all by itself, just needing my feelings along the way.

2 hours ago, uwespeed said:

Oh yeah ... 😉. You always get responses like this from AI, they want to ingratiate so that you feel well and use them more often. 

 

If you ask Claude several times, sooner or later, it’ll end up contradicting itself 😄.

True story: I can't talk about Claude because I don't know it (in detail), but just today a colleague told me an example of his experience with AI (unfortunately I don't remember which one). He asked a question and the response felt strange to him. So he asked again, but the AI insisted that the response is correct. So, the colleague asked again with some more details. Still, AI responded in the same way as before. So the colleague, who actually knew the correct answer asked again like "Are you sure? Please think again." Finally, the AI said: "Oh ... I'm so sorry ... actually you are right and I was wrong." Other colleagues, who also have experience with AI confirmed that they (partly) have made the same experience and myself, too. So ... answers may be correct, BUT ... they can be plain wrong as well.
Disclaimer: I'm not saying that Claude is completely and always wrong but ... Conclusion: AI CAN be very helpful, but it also can lead you into the wrong direction if you believe everything you read.

The next steps on the internet will be that we have more and more AI generated content and one AI will use the input from other AI's and the result will be ... pure puzzle-solving based on probabilities.

Depending on the activity, I’m taking to asking several AI’s the same set of prompts. 

The results are frequently different.

And yeah, there always seems to be more to be had by forcing a deeper dig. 

1 hour ago, MrBitstFlyer said:

I am not interested in being psychoanalysed

And you aren't. You're the one who suddenly hyped it up with wild fonts in a wall-of-text post and deflections of being "clear". All you had to say was "yes I'm using it again". That's it.

For my part, knowing that it's the driving thought process behind your work simply means that I won't bother trying to engage in a technical sense any further because I might as well just ask chatgpt and I'll end up with statistical parity with the answers. That's it.