yaw torque issue in large vectored tailsitter

tridge

tridge

Member
I'm having trouble getting the yaw torque right for a large dual motor vectored tailsitter.
On a smaller vectored tailsitter we're not having any problems. The two motors are set to opposite directions (one clockwise, one counter-clockwise) and it flies really well.
The larger vectored tailsitter behaves quite differently. The way it flies looks like both motors are turning in the same direction, but they are set to opposite directions and I've confirmed the directions by starting the motors in the vehicle editor and seeing which way they spin.
You can see the problem in this video:
https://www.youtube.com/watch?v=HO4XeMSih1A
starting at around 1:50. The model is hovering in no wind, and should have a small total yaw torque, with the motors pointing straight up. Instead the motors are at a large angle in order to hold the yaw.
For comparison, here is our smaller aircraft with the same setup:
https://www.youtube.com/watch?v=W2OIl7i3EaI
that one behaves as expected, with the motors only needing to tilt when you command some yaw rate.
Can anyone suggest why the larger aircraft behaves like this?
All the model files are here:
https://github.com/ArduPilot/SITL_Models/tree/master/RealFlight/Tridge/CUAV_TVBS
 
I reviewed both the CUAV_TVBS model and the Caipirinha.

I don't see any obvious flaw in the TVBS model from a theoretical standpoint, but...

I wonder if the problem might lie in the use of symmetric components in the TVBS. The Caipirinha uses two separate movable pods as motor mounts, while the TVBS is using a symmetric motor mount component. This should work OK, but it's possible that you've exposed a bug in the tracking of symmetric prop effects.

I'd recommend trying a setup like the Caipirinha, with individual physics for the left and right mounts and their children. If that works, then we know there is an issue with symmetry and V-pitch propellers (at least) and I can file a bug report.
 
Jeremy Sebens said:
I reviewed both the CUAV_TVBS model and the Caipirinha.
Click to expand...
thanks!
I wonder if the problem might lie in the use of symmetric components in the TVBS. The Caipirinha uses two separate movable pods as motor mounts, while the TVBS is using a symmetric motor mount component. This should work OK, but it's possible that you've exposed a bug in the tracking of symmetric prop effects.
Click to expand...
I've actually tried it both ways with the same result.
It also doesn't seem to matter what propeller I use, so it doesn't seem related to using the vpitch props.
What I think I need to do is slowly morph the Caipirinha to be the CUAV_TVBS, testing at each stage to see how yaw torque is handled and hopefully find the key difference that way.
Cheers, Tridge
 
tridge said:
I'm having trouble getting the yaw torque right for a large dual motor vectored tailsitter.
On a smaller vectored tailsitter we're not having any problems. The two motors are set to opposite directions (one clockwise, one counter-clockwise) and it flies really well.
The larger vectored tailsitter behaves quite differently. The way it flies looks like both motors are turning in the same direction, but they are set to opposite directions and I've confirmed the directions by starting the motors in the vehicle editor and seeing which way they spin.
You can see the problem in this video:
https://www.youtube.com/watch?v=HO4XeMSih1A
starting at around 1:50. The model is hovering in no wind, and should have a small total yaw torque, with the motors pointing straight up. Instead the motors are at a large angle in order to hold the yaw.
For comparison, here is our smaller aircraft with the same setup:
https://www.youtube.com/watch?v=W2OIl7i3EaI
that one behaves as expected, with the motors only needing to tilt when you command some yaw rate.
Can anyone suggest why the larger aircraft behaves like this?
All the model files are here:
https://github.com/ArduPilot/SITL_Models/tree/master/RealFlight/Tridge/CUAV_TVBS
Click to expand...

Check to see if both of your motors are running similar RPM's, pitch, and throttle settings. A High pitch prop running at a low rpm will require different torque than low pitch High rpm. I have a feeling that your auto pilot is lugging one motor with high pitch to balance the thrust of the high rpm low pitch

I am running manual controls with fixed v-pitch props and I can not recreate the problem in your video
 
Paul Riseborough has had a look at this now as well, and discovered that the key factor is the "Prop Wash Factor" of the motors. I had that set to a higher number on the CUAV_TVBS than I had it on the Caipirinha. When I lower it on the CUAV_TVBS the "yaw torque" issue disappears.
Jeremy, can you explain a bit about what the Prop Wash Factor does? The docs in the vehicle physics editor help text don't give any hints as to why this would produce a yawing effect like we observe.
Paul had some other air flow modelling issues he noticed while experimenting which he will post separately.
 
flexible2g said:
If you wish to try these things, post it here, as beta.
Click to expand...
I should explain that this plane isn't intended to be flown manually in RealFlight. It is setup for flying with the new LINK feature, with ArduPilot controlling it.
Dual-rates and expo just break the linearity of the aircraft for the autopilots control.
Regarding the colour scheme, I know it isn't great (especially the mirror-like effect on the tail). I'm a beginner with 3ds max and am still learning how to control things like that.
Cheers, Tridge
 
legoman said:
Check to see if both of your motors are running similar RPM's, pitch, and throttle settings.
Click to expand...
yes, they are all the same on both motors.
See the comment I just posted about the prop wash factor being the key difference.
Cheers, Tridge
 
tridge said:
Paul Riseborough has had a look at this now as well, and discovered that the key factor is the "Prop Wash Factor" of the motors. I had that set to a higher number on the CUAV_TVBS than I had it on the Caipirinha. When I lower it on the CUAV_TVBS the "yaw torque" issue disappears.
Jeremy, can you explain a bit about what the Prop Wash Factor does? The docs in the vehicle physics editor help text don't give any hints as to why this would produce a yawing effect like we observe.
Paul had some other air flow modelling issues he noticed while experimenting which he will post separately.
Click to expand...

tridge said:
yes, they are all the same on both motors.
See the comment I just posted about the prop wash factor being the key difference.
Cheers, Tridge
Click to expand...

The prop wash Factor increases or decreases the airflow coming off the propeller. with the airflow blowing up or down on the wing it negated most of the roll athourity of the thrust vectoring. I did notice the roll rate not increaseing with thrust vectoring with the manual controls on both TVBS and the caipirinha. It was my second guess
 
tridge said:
I've actually tried it both ways with the same result.
It also doesn't seem to matter what propeller I use, so it doesn't seem related to using the vpitch props.
What I think I need to do is slowly morph the Caipirinha to be the CUAV_TVBS, testing at each stage to see how yaw torque is handled and hopefully find the key difference that way.
Click to expand...

Makes sense. I'll be interested to hear what you find out!

<edit> looks like I was slow on the uptake and results were already reported! I'll dig in below.
 
Last edited:
flexible2g said:
You must have a force to control yaw. Newtons' law.
Click to expand...
It does work fine, there are several people flying real vehicles of this type. Roll comes from differential thrust. Pitch comes from pitching the motor tilt. Yaw comes from differential tilt in the motors. Thrust from the motors.
Here is a video of the same type of aircraft (just smaller) in RF8:
https://www.youtube.com/watch?v=W2OIl7i3EaI&t=320s
Here is the same aircraft, but the real aircraft this time:
https://www.youtube.com/watch?v=nIXRRIeIZe0&list=LLJzpx7cFPmzIW1RbGpZesYA&index=26&t=71s
 
legoman said:
The prop wash Factor increases or decreases the airflow coming off the propeller.
Click to expand...
I think it is a physics bug in RealFlight. This type of vehicle does fly very well with a lot of yaw authority in practice.
Here is a nice video showing just how much yaw authority this sort of vehicle has in a real vehicle:
https://www.youtube.com/watch?v=koOqenMpvck&t=84s
We can just set the prop wash factor to zero and we can fly it in RealFlight, but I suspect that is just hiding the issue. Hopefully Jeremy can work out whats going on.
 
flexible2g said:
This screen shot shows the problem. The motors are not set up for movement in RF.
Click to expand...
That is the Cat Tailsitter, which is a non-vectored aircraft. I assumed you were talking about the CUAV_TVBS model (which is what I started this topic to discuss). The CUAV_TVBS is vectored.
The Cat tailsitter is not supposed to tilt its motors. The moveable pod at 90 degrees is needed to make the motors appear with the right orientation, but there is no servo attached.
This is based on a real aircraft, and it flies well. It also flies well in RealFlight. Yaw for this aircraft comes not from tilting the motors, but from the huge elevons.
Note that the "manual" CatTailsitter that I uploaded to the swap pages is a variant of the one that is normally used with ArduPilot control.
Here it is under ArduPilot control in RealFlight8:
https://www.youtube.com/watch?v=yOp-3Uy1DLE
Sorry for the confusion.
 
Last edited:
tridge said:
Jeremy, can you explain a bit about what the Prop Wash Factor does? The docs in the vehicle physics editor help text don't give any hints as to why this would produce a yawing effect like we observe.
Click to expand...

The prop wash factor sets the strength of downstream effects of the prop, both axial and helical. In this case, that would include wash effects on the fuselage booms and the fins.

I'd be curious to know if the Caipirinha misbehaves if it's prop wash factor is increased. Prop wash interacts with the fins/rudders, which would be "blown" in the CUAV, but not in the Caipirinha.
 
Last edited:
Jeremy Sebens said:
I'd be curious to know if the Caipirinha misbehaves if it's prop wash factor is increased. prop wash interacts with the fins/rudders, which would be "blown" in the CUAV, but not in the Caipirinha.
Click to expand...
yes, it happens the same on the Caipirinha. If you set prop wash factor to 50 its fine. If you set it to 60 it gets a noticeable yaw problem.
Does anything different happen in the maths with values above 50?
 
btw, while testing this I also happened to notice that the "NavGuides" current reading is only reading for 1 motor in the Caipirinha. It also doesn't seem consistent with what is sent over the LINK protocol.
 
tridge said:
yes, it happens the same on the Caipirinha. If you set prop wash factor to 50 its fine. If you set it to 60 it gets a noticeable yaw problem.
Does anything different happen in the maths with values above 50?
Click to expand...

OK, I'm adding a case to our issue-tracking system.

To my knowledge, there is no discontinuity at 50%. I'll verify as soon as I can.
 
tridge said:
btw, while testing this I also happened to notice that the "NavGuides" current reading is only reading for 1 motor in the Caipirinha. It also doesn't seem consistent with what is sent over the LINK protocol.
Click to expand...

This is a known issue with our NavGuides. The RPM and energy consumption figures only correspond to one motor. I'll have to look into what value is reported in the FlightAxis Link state vector.
 
flexible2g said:
I deleted the rudder. Added symmetrical rudders, at 45%, centered under C/L of motors thrust. Increased power to 140%. Flew ok, but I noticed that when slowly adding for lift off, it moved to the right. You had the props turning backwards, that realty messed up the prop wash big time.
Click to expand...
Do you realize that the max files and FBX for the model are available for you to modify? I know that isn't the norm on these forums, but for the ArduPilot models we're making them available and you are very welcome to modify them so you can match the graphics to the changes in the physics models.
The files are here:
https://github.com/ArduPilot/SITL_Models/tree/master/RealFlight/Tridge/TailSitters/Cat
The changes you've made result in quite a different aircraft. The orginal design is a model of a real aircraft, and was built so we could test and improve ArduPilot control of that aircraft.
I think the reason you wanted those changes is that RealFlight doesn't have a builtin tailsitter controller. If the multicopter controller in RealFlight could be adapted to support tailsitters then I think you'd find this model very easy to fly without any physics changes.
Cheers, Tridge
 
flexible2g said:
The key word is vectored! The RF model you posed is not vectored!
Click to expand...
Indeed, the Cat is not vectored. Non-vectored tailsitters fly fine. They aren't as agile as vectored, but they are much more common. Please do a google search for dual motor tailsitters, or see the links I gave in the download for the Cat, or lookup the wingtra, which is a commercial aircraft of this type. Thousands of people fly them, and lots and lots of people have done PhDs and masters on how they fly.
Flying them without a flight controller is hard but not impossible. Flying them with a flight controller is easy.
 
You must log in or register to reply here.
Back
Top