Plane: stall detection

[magicrub]

Stall detection is needed. Must work with, and without, pitot tube.

related to stall recovery issue #2669

[magicrub]

There are many indirect ways to declare our state to be "stalled". My favorite candidate is roll and roll_demand diverge but for this feature to be robust we need to define what a stall is.

[marcmerlin]

Since I happen to be an airplane pilot, you are correct that you can get one wing that stalls first and that will give you a clue that you stalled as the wing dips.
However, you can have a coordinated stall where no wing dips and just the nose stalls.
IMO, it's easier and more reliable to notice that the nose just dropped a bunch without any control input asking for it to do so.
(actually, "dropped" is a misnomer, you can get an accelerated stall in a 60 degree bank angle where the nose buffets and basically stalls but not towards the ground. You have compute the direction vector for your current attitude and see that the nose slipped from there).
Your other stall is a stall/spin where indeed one wing drops and depending on the plane, you can end up in a spin.
https://www.youtube.com/watch?v=F6d4VKdc34A (watch 10 seconds after start)

[magicrub]

Thanks for the insight Marc. I think when we are actively in a stall the aircraft is basically not responding to controls so pitch, roll and yaw will all diverge - not just roll or pitch. The question is what will diverge first and by how much and is that a reliable source for a trigger or do other things need to be considered? We need as much of a generic solution as possible.

If we can't get something totally generic, we could look at specific airframes via params such as if the are elevons configured. If we go that route, we would need a table of what aircraft shows what signs of a stall. That may get complicated but would hopefully be more reliable.

In all cases I think it's safe to say our commanded altitude and desired altitude diverge.

[marcmerlin]

Agreed, going with commanded altitude vs actual altitude is a pretty safe way to detect almost all stalls (except accelerated buffet stalls in bank turns, but those are likely pretty rare in the list of stalls people see, and not something I'd be worried about anyway).

Either way you have a +1000 from me on stall detection and recovery :)

[magicrub]

Does that mean you're volunteering to flight test? :)
On Oct 12, 2015 3:44 PM, "Marc MERLIN" notifications@github.com wrote:

Agreed, going with commanded altitude vs actual altitude is a pretty safe
way to detect almost all stalls (except accelerated buffet stalls in bank
turns, but those are likely pretty rare in the list of stalls people see,
and not something I'd be worried about anyway).

Either way you have a +1000 from me on stall detection and recovery :)

—
Reply to this email directly or view it on GitHub
#2933 (comment)
.

[marcmerlin]

On Mon, Oct 12, 2015 at 03:51:07PM -0700, Tom Pittenger wrote:

Does that mean you're volunteering to flight test? :)

While I'd be sad to crash my $1000+ long range FPV glider, I can probably
try this at altitude if the code isn't going to do reboot the pixhawk or do
something too extreme I can't manually recover from :)

Incidently I just tried stall prevention in AC3.4 this morning in training
mode. I'm not really sure it worked, I was able to pull up the nose a lot
and get the aircraft to stall. That said, it may just need more tuning.

Marc

"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
Microsoft is to operating systems ....
.... what McDonalds is to gourmet cooking
Home page: http://marc.merlins.org/

[iskess]

I disagree that accelerated stalls are rare, I think they are very common and the most dangerous, especially the turn from base to final approach close to the ground.
Every unexpected stall I experienced on an RC plane was related to an aggressive roll. That downward aileron stalls the wing you are wanting to lift, and often leads to a spin. I believe divergence in roll will be the most important indicator to the most dangerous type of stall. The ailerons should immediately go neutral and the rudder and elevator need to be used to recover.
I volunteer for flight testing. Sounds like fun!

[magicrub]

Ha! Thanks for the enthusiasm Iam. Glad to hear you're on board for
testing. You're always up for a challenge :)

Let's try and keep the stall recovery chat to the stall recovery issue.
It's easy to get side tracked on that topic because I'm sure it will get
complicated for various aircraft types.
On Oct 12, 2015 4:21 PM, "Iam Bouret" notifications@github.com wrote:

I disagree that accelerated stalls are rare, I think they are very common
and the most dangerous, especially the turn from base to final approach
close to the ground.
Every unexpected stall I experienced on an RC plane was related to an
aggressive roll. That downward aileron stalls the wing you are wanting to
lift, and often leads to a spin. I believe divergence in roll will be the
most important indicator to the most dangerous type of stall. The ailerons
should immediately go neutral and the rudder and elevator need to be used
to recover.
I volunteer for flight testing. Sounds like fun!

—
Reply to this email directly or view it on GitHub
#2933 (comment)
.

[magicrub]

Ok guys, I may be doing some stall testing next week on my skywalker x8 flying wing. They're notoriously hard to recover from so recovery is a secondary goal. Lets start posting some logs of stalls of different aircraft and situations so we can get a model of what a stall looks like data-wise. Here's a couple things to look for in the logs:

• abs(altitude_demand - altitude) is large
• abs(roll/pitch/yaw - roll/pitch/yaw_demand) is large
• rather large unexpected sink rate.
• airspeed_demand - airspeed spikes up then all of the above happens.

Anything else we can look for?

[iskess]

Here is a nice example of a tip stall during and autolanding in a 2013 Skywalker 1880 a few years back. It includes a video for the full experience.
http://diydrones.com/forum/topics/autolanding-failure-tip-stall-on-turn-to-final-causing

[iskess]

Here is a stall/spin/crash during the autotuning of a Multistar Twinstar II. I believe it was my correspondence with Tridge and Paul over this crash that lead to the STAB_PITCH_DOWN parameter. I didn't realize that FBWA would fight the plane's natural nose down tendency when the throttle is cut, and allow the airspeed to bleed off. Combine that with the steep bank angles required for auto tune, and it was a recipe for a spin.

2014-04-29 15-11-20.zip