Ref to altitude drop in alt hold Issue 1066 (now closed)

[David Pawlak]

Randy mentioned the altitude drop at the end of a horizontal run in Alt Hold (before closing issue #1066).

"... we beleive it's caused by an aerodynamic effect on the baro......  so far we've been asking people to solve this physical problem by physically changing their vehicle. It can apparantly be fixed by ensuring even airflow over and under the vehicle"

This problem is usually soved in "real aircraft" by using a static port, a small flat disk with a tiny hole flush with the side of the aircraft. You need one on each side, to cancel effects of sideways flight.

I was thinking of taking this on as a one shot device that could be added to just about any configuration. Not without it's challenges though, as eventually you have to interface an airtight tube from the static ports to the input holes of the sensor, and in such a way as untrained persons don't damage their sensors.

But I figure it's worth testing it out.

[Randy Mackay]

David,

 

     Yes, that question comes up a fair bit with people who have just started with ArduCopter.

 

     I don’t know if other flight controllers suffer from the same problem or if they’ve introduced a work around in software.  Someone suggested that we should reduce the impact of the baro for a couple of seconds in the situations that the problem occurs (when flattening out after moving at high speed).  I also don’t think anyone has tried the EKF to see if it does any better.

 

     Experimenting with a hardware solution sounds good to me.  The cause is still just a theory as well I think.  I’ve never seen anyone prove the cause and/or demonstrate a physical solution.

 

     On the pixhawk (or future flight controller) I suppose that two barometers might be possible in case that helps in reducing tubing.

 

-Randy

--
You received this message because you are subscribed to the Google Groups "drones-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to drones-discus...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Paul Riseborough]

It's definitely an aero effect on my quad because its measurement error (not control error) and varies with  airspeed and direction. If I put it into circle mode on a windy day, the cyclic variation in height around the circle is noticeable, yet according to the logs, the baro alt is constant. 

To measure static pressure without any of the dynamic pressure affecting the measurement it must be measured in undisturbed flow, where the air has not been made to change velocity (eg along the side of a slender pitot tube well in the front of an aircraft).

At 10 m/s airspeed, the air  has a dynamic pressure of just over 60 Pascals.  If the airflow relative to the  body is forced to accelerate by 15% as it flows around the body of the copter, the static pressure will drop by around 32% of the dynamic pressure which is  20Pa or 2 metres. this means the baro alt will read 2m high causing the copter to lose height in high speed runs. This error increases with the square of the airspeed.

There is also a significant amount of baro fluctuation downwind of large structures on windy days.

[Andrew Chapman]

Although wind would be an unknown variable, is it possible to model the effect based on speed and tilt angles and at least partially compensate for it, or is it vehicle dependent?

AC.

[David Pawlak]

It shouldn't take much to whip up a statick port jig. I've already got some conceptual designs which use a bit of silicone tubing, and some bits of tubing and plastic.

With any luck, next week I can do some tests. It may not be perfect, but it should be enough to confirm.

From there we can see how much else may need to be done.

[Paul Riseborough]

It will be vehicle dependant. 

[Paul Riseborough]

David, a high mounted  flat disc wit  a centre pressure tap in the top and bottom going to a Y joiner, and then to the static pressure sensor would be a start. 

The smaller the thickness to diameter ratio the better. If the disc is kept thin and is high mounted in undisturbed flow, the average of the top and bottom pressure in the middle of the disc should be close static across a reasonable angle of attack range. When the disc is tilted relative to the airflow, air will flow from the top (high pressure tap) to the bottom (low pressure tap). 

It will be important that the flow resistance from each pressure tap to the y joiner is initially the same . By making the length of tubing from the top and bottom pressure taps to the Y jointer different lengths (different flow resistances) the sensor can be calibrated so the static pressure change with tilt angle is minimised. For example if the static pressure drops when the copter is tilted, the length of tubing to the bottom pressure tap can be increased and vice-versa.

[David Pawlak]

OK, I'll get the connection to the sensor set up first, with an interface that will let me connect and disconnect various methods. It would be nice to see that end of run sink eliminated.

[Julien Dubois]

Yeah,

I've spent some time on the subject, made this model compensation but, as Paul said, it's model dependant, I have the alt_drop issue with a copter but not with another.
The right solution is a cased frame (eg: Iris, DJI phantom, Walkera X350, ...) to avoid the baro disturbance.
For my "open frame" copter that suffers from this issue, I've got great results compensating baro with a model + modulating TC_Z factor to give less weight to baro in the cases I know it's disturbed.
Modulating only TC_Z works as well for any frame, but we just have to ignore baro a bit more.

Anyways, I had great results because my alt_drop issue was due to baro. Maybe, others issue could come from accelerometer... and giving less weight to baro => more weight to accelerometer so that would just be worse.

Julien

[Paul Riseborough]

Even a cased frame is no guarantee as depending on where on the casing it vents to the outside flow will effect whether the pressure inside the casing increases or decreases with airspeed. On my Skywalker X-5 plane, the baro height rises 4m when it flies at 14 m/s airspeed despite the Pixhawk being mounted inside because the airflow accelerates and drops pressure as it flows around the fuselage.

In theory its possible to position vents on the casing  to minimise the pressure change inside, but this requires a fair bit of trial and error and some understanding of the pressure distribution around the casing. With my  X-5 I'm thinking of opening up a hole in a high pressure position (low on the nose) to offset the low pressure it currently sees. I would imagine that on a copter that can fly in any body relative direction, coming up with suitable vent locations could be tricky.

[Julien Dubois]

Yeah, of course and even 4m is not so bad as the dynamic pressure @14m/s is equivalent to about 10m.

What about a GPS based compensation (filtered offset) with a weight proportionnal to speed.
And as well modulating TC_Z to lower baro effect during braking phases at high velocity.

[David Pawlak]

Yeah, in small private aircraft the alternate (fallback) is cabin. It's close, but you open a vent or "window" and the altitude goes to.......

But it's close in an emergency.

In our case the drop is about a meter, not too much, but the other day I was flying fast at 2 m with my 3DR X8 in alt hold and came into a heavy curve pitching back and things started to unravel pretty quickly. Added some collective ( LOL) and just managed to come out of it.

[mlloyd]

With EKF, we have an estimate of both the ground and wind velocities, from which we can easily infer the air velocity. We know the barometer error is proportional to the square of the air speed. Could we just apply a correction to the air pressure?

i.e.

air_velocity_NE = aircraft_velocity_NE + wind_velocity_NE  (from EKF)

dynamic_pressure = barometer reading

static_pressure_estimate = dynamic_pressure + constant * |air_velocity_NE| ^ 2

The constant would be a tunable parameter but I'd guess a single well-chosen default value would work well for most aircraft.

There must be a software solution to this; the NAZA suffers from alt drop during forward flight much less than ArduCopter does. My NAZA used to lose 2-3 feet at most even at top speed. My ArduCopter (AC 3.2-rc1, Pixhawk, same frame and motors as before) loses 6-10 feet depending on the speed, and it even loses 3-4 feet of altitude when a gust of wind comes along, which NAZA never used to do. They either must correct like above, or have a super low-noise accelerometer (their accelerometer is encased inside a metal shield suspended in a foam harness), or both.

[Paul Riseborough]

You don't get a wind velocity estimate off the EKF for copter - without airspeed measurement, or the ability to make an assumption about direction of flight relative to body, it is unobservable by the current design.

 In theory airspeed could be estimated given an assumption about drag on the airframe. An airspeed/windspeed estimator for multi-rotors sounds like it could be a worthwhile project.

BTW, with your naza installation, was the naza located in the same position as the pixhawk? These errors are very sensitive to position of the sensor.

[mlloyd]

Alas, yes, I see, we'd have to make assumptions about drag on the airframe. Air speed is not directly related to the attitude required to stay put in say loiter mode; it depends on how draggy the craft is, and various other factors too I'd think (mass of airframe and actual thrust force, etc.) On the other hand, we might be able to estimate those factors too, given that we find out during flight what bank angle is needed to maintain a certain velocity of forward flight.

I wonder whether a correction based just on ground velocity might still be good enough to be useful in most situations, given that typical wind velocities are small compared to maneuvering velocities. It seems pretty easy to try - I'll have a go sometime over the next couple of weeks. I'll probably modify the EKF code to correct the incoming barometric pressure based on just the current estimate of VN and VE.

My NAZA was located in exactly the same position as the pixhawk, with one small difference. The NAZA was attached directly to the carbon fiber plate using very thin foam. The Pixhawk is in one of the 3D printed vibration mounts, the main difference being that air can flow both above and below the pixhawk, below it between the upper and lower parts of the vibration mount. This probably changes the aerodynamics quite a bit and might be contributing to the alt drop I'm seeing. Perhaps I'll try putting some foam in between the upper and lower parts to prevent air flow.

[David Pawlak]

I'll be testing with a Pixhawk. My first shot at a physical solution to this is to seal the small compartment that encloses the alt sensor, and drill a small hole in the Pixhawk case to allow access by a static tube. The static tube will run a short distance to a "T" connector, which vents on one side above the frame and on the other below. I will place a small uncompressed piece of "sponge rubber" in the tube leading from the "T" to the sensor to serve as a small filter which should also facilitate equalization from high to low pressure in the "T".

Testing Friday if all goes well. 

[David Pawlak]

First test didn't go too well. It's about the same.

But I probably have a leak in the static system. Everything up to the Pixhawk is fine, but I am being extra careful around the sensor. Tried using plastecine to seal around the sensor box, for short term tests, because I wasn't sure what silicone solvent (acetic acid) and its fumes could do to circuits and the sensor itself. I'll try to find a chance to look more deeply into this. I seated everything and had a look at the seal marks. they looked good, but plastecine may not be the best, even for the short term.

Graphs show the altitude deviations directly related to pitch angle. I did notice that there was absolutely no loss of altitude in any other direction, roll and rear.

I don't remember though if that was a pre-test condition, or if that has gotten better.

[Philip Rowse]

putting it at the top and bottom will create a flow through the tube due to the lean on the copter, it really will be a challenge finding the "right" spot to put the static sensors around a copter.  think of it this way... for your copter to fly, it creates an area of high pressure below it, when you are heading forward at an increased rate, you have a high pressure area above it as well, so by having your ports top and bottom, you are still creating an overall higher pressure as you tilt forward, and as such the copter (should think it has dropped down, and pull up...)

thoughts?

Philip Rowse

Electronics Engineering Dept

3DRobotics

Ballarat

Australia

[Robert Lefebvre]

I thought the higher velocity flow below the props means it has a reduced pressure?  Would that then mean that the higher pressure is in fact on the top?  I always get confused by that part...

[David Pawlak]

For the reasons you mention, this is not really a viable solution. Perhaps finding a common sweet spot would be difficult to impossible.

I mounted mine fairly far out front. The static holes (1.5mm diameter) are on a plastic plate about 7mm thick, A hole drilled from top to bottom.

A drilled channel also of 1.5mm  entering from the side to reduce at least forward back ram pressure effects connects a silicone tube to the static holes. Theres as small piece of sponge placed at the entrance to the silicone tube. The idea, to facilitad flow through from top to bottom and perhaps reduce any higher frequency "noise".

Being further forward. it should be somewhat lesser effected by differencial pressure of the motors, But there does seem to be the effect that you mention,

I don't seem to see altitude changes with velocity, but they are definitly present with pitch change.

As I mentioned as well. ZERO drop with side to side or reverse movement. I had about 35 deg lean angle and fair good velocity, and the altitude held impeccibly. The same with reverse.

I´m sending the log, if it's not too big!

[David Pawlak]

Looking at the logs again, it's the delta pitch that has the most effect. I see one spot between 27,000 and 28,000 where the pressure seems to change for no reason though. I don't have it in higher forward velocities long enough to really see what effect pure velocity has, but the real change (drop) in pressure comes as the pitch goes from negative to positive in forward flight.

Now that I have the staic attachment it should be pretty easy to change the configuration of the actual ports.

[Philip Rowse]

Hi Rob & David

    Here is a basic diagram that covers that off for the prop wash, its a bit hard to know what the real effect of the pressure under the copter would be, as if the props are far enough away from the body, then the effect on the underside of the body would come more from the effect of forward flight than the propwash.

    At low speed, we are only dealing with static thrust, so the pressure would build up under the copter. However as the forward speed increases, the body of the copter would be affected more by free stream air than the propwash, so the bottom port may be at a negative pressure, and the top positive.  However, that is assuming a perfect model, in reality there will be turbulent flow everywhere, so predicting where the best place for these ports should be is very difficult, even with a wind tunnel, as the one thing that a copter does well is fly on any angle, in any direction......

    The solution here, is to let the EKF do its job, and filter out some error in altitude when you have a trusted sensor on board like a Pixsy GPS etc.

    in the meantime, if you have more than just two ports, going to a center chamber filled with an open cell foam, then tap off that to your sensor, you may be able to average the static better. think of a pingpong ball on a pole, with lots of small holes, filled with foam, with a tube coming from the center down to your sensor.  similar to this:

 

So if you had this sticking up above your copter, it wouldn't matter which direction you are traveling, you would get a stable Static reading on your baro.

thats all taken from a sensor I am working on that will give Omnidirectional wind vector and speed information, as well as omni directional static.  the intention on a device like this, is that is would be individually calibrated, and verified in a wind tunnel / vacuum chamber. it would have its own controller, and just feed out calibrated wind speed and direction, as well as temperature.  The reality of this device, is that it would actually not plumb the pressure to a remote sensor, rather the sensors would be built into the sphere.

Unfortunately, this wont be a cheap device, the 20 x MS5611 sensors (or similar) will cost around $200, plus processor and the sphere....... 

Phil

Philip Rowse

Electronics Engineering Dept

3DRobotics

Ballarat

Australia

[Paul Riseborough]

Rob,

Bernoulli's theorem is based on conservation of energy and does not apply when energy is added to (eg by a propeller) or subtracted (eg turbulence behind a bluff body) to the flow

You can apply Bernoulli's theorem to the flow before the prop and you can apply it to flow after the prop, but not to flow through the prop.

The static pressure of the air reduces as it accelerates towards the prop dis upstream of the prop. There is then a step increase in static pressure going through the prop disc, and then that static pressure then drops again again as the streamlines contract after the prop and the flow accelerates. There is a step change in pressure through the prop, but not a step change in velocity.

Eventually the flow after the prop mixes with the surrounding air, the stream lines start to diverge and the flow slows down.

Paul

[Paul Riseborough]

As I have suggested before, a small thin flat disc mounted high above the body (similar to the location enjoyed by the GPS and compass on some multi-rotors) with a centre pressure tap top and bottom would do a reasonable job. It would act like a low aspect ratio lifting surface in forward flight, and wind tunnel tests for low aspect ratio flat plates shown the upper and lower surface to have roughly equal and opposite pressure changes due to angle of attack bu the time you get away from the leading edge region and back to the 50% chord point

The problem with a sphere or cylinder is that the average pressure over the entire surface is actually less than static. This  is due to flow separation and the fact that pressure is not fully recovered on the back side. To obtain static pressure from the pressure distribution you and have to apply a curve fit with corrections for Reynolds number - you can't just take an average.

[David Pawlak]

Thanks Paul, my first attempt focused more on getting a static port attachment working, and was a little pressured for time to fit the space that I had available to work on this. I do hope to get more time to tune this up, although I'm sure as you say, the most practical and consistent solution would be EKF. It's more of a curiosity exercise than anything.

I'm using an 3DR X8 octo. The static port is mounted pretty much forward, but is still between the front props. However in forward flight (I imagined) that the prop influence would be reduced somewhat. What's interesting, is that in sideways flight where there is likely to be more influence from propwash, pressure changes due to body angle and velocity were minimal.

I suppose on the nose, the airstream is still in leading edge effect, there is flow disturbance behind both the top and bottom ports... in sideways flight there would be none.

Does anyone know/remember if they see the drop effect in sideways or rear flight? Because right now I have zero and the graphs show little disturbance in those regimes.

I guess I have to mount a thinner disc (10cm dia?) up on a stick maybe 15 cm about midships. (And then, what will the influence of the stick be?? sheesh)

[Robert Lefebvre]

Ah, now that makes more sense to me, thanks! 

[Paul Riseborough]

I was playing around with some ideas using two of the thin SS tubing from inside the 3DR pitot static tubes. These would run side by side in a slot in some thin wood or pastic, potted in with epoxy. One tube would have a small hole drilled in it on the lower side, the other tube would have the hole drilled from the upper side. The the resulting disc would be less than 2mm thick, so could be made only 50mm in diameter. The disc would be supported from the back where the tubes come out (think flight deck on SS Enterprise). Provided it was kept thin, the  influence of the support would be minimal for forwards and sideways flight.

I'm on the waiting list for a couple of the new laser sensors which in conjunction with the EKF mods I am testing, should improve alt hold over flat(ish) terrain. But pursuing a accurate baro reading is still worthwhile, particularly for higher speed airframes.

If i only wanted to do high speed flight in the forward direction, I'd be temped to try connecting to the static line from a pitot-static tube assembly mounted forward between the two front props, and angle the prop upwards so it is aligned with the flow in forward flight. Cross winds would cause some errors with this setup.

I need to finish my work on integrating optical flow into the EKF before starting a new project - not enough hours in the day.

[David Pawlak]

I'd have tpments!o see where I could get some thin tube like that. I'm one of those poor suckers who can't receive shipments abroad.

But there has to be something.

[David Pawlak]

I tested an unmodified multi today (prop change) but thought I'd look at a "standard reaction" to the "alt hold issue" again.

As already posted, my 3DR X8 with modified static system  in "Alt Hold", "Hybrid", and I assume Loiter, registered a similar altitude drop at the end of a forward flight run as a normal vehicle (rev 3.2). However this new static system showed NO sign of altitude drop latterally or in reverse. NONE. It was very stable.in all regimes.

My tests today, without modified static system showed failure in all directions. Significant drops at the end of every run, irrespective of the direction.

So I think we can conclude that the present theory is correct. The problem is a result of pressure effects created by body angle and resulting differencial pressure  where the autopilot is located.

I was impressed. The drops are significant. Even in the forward direction. without the mods, they are worse. 

[Julien Dubois]

What  is the static system you are talking about?

--
You received this message because you are subscribed to a topic in the Google Groups "drones-discuss" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/drones-discuss/TsuDlbWn_qc/unsubscribe.
To unsubscribe from this group and all its topics, send an email to drones-discus...@googlegroups.com.

[David Pawlak]

A little detail in previous posts to this thread..

Basically I set up a sealed static port connection to the Pixhawk altimeter sensor. Fortunately the altimeter sensor rests over a plastic "chamber" which holds a black piece of foam against the sensor. So I "carefully" sealed the chamber around the edges and openned a small hole in the Pixhawk case which alowed me to attach a small tube.

This tube runs to a flat plastic plate which I have mounted far forward on my 3DR X8. The plate has a vertical hole, venting top to bottom, from which a thin tube  conects the centerpoint of the vertical hole to the static pressure tube coming from the Pixhawk.

The idea was to equalize the top/bottom pressure to mitigate the effect of increased pressure on the top where the pixhawk normally measures static pressure.

Initial tests with this configuration showed that the octo still showed a drop at the end of a fast forward run in anu of the alt-hold modes. However it showwed Zero drop in lateral or reverse runs.

This last test with an octo of similar characteristics and without the modified static system, showed the drop in all directions. Since doing this second reference test, I noticed that the forward drop was less as well with the improved static system.

I still want to improve the dynamics of the  "static plate", similar to something Paul mentioned above.

[Peter Plischka]

Very interesting David.

 

Can you send some photos so we can imagine that better?

regards Peter

[David Pawlak]

I triefd to take several photos before to attach here, but the contrast (for my cell phone) between the white plastic part and the black base is too much to see the detail. But I've taken the best I can and over written some lines to show the detail.

The static tube exits the pixhawk in the part circled in red above. The tube enters from the side to try and reduce ram effect... although it doesn't seem to work very well for that purpose.

I drew the lines ( within the second red circle), to show how the internal holes are drilled. You may see a yellow piece of spnge in the static tube going to the Pixhawk. The ieda being to try to reduce turbulence noise and try to equalize the pressure from top to bottom.

The plastic piece is far from optimum. It should be thinner, and more streamlined. But I'm working on a new design. that optimizes the dynamica a lot more. The real solution I think is in software. This is not a job for "the average user"

[David Pawlak]

Ooops the file

[Julien Dubois]

Thanks David for details.

That's an interesting first design.
I think you could learn from Philip Rowse post and make a network of tubes with nozzles at strategic locations (simple eg: 2 of your static system, 1 on top and 1 at the bottom of your copter, the hole of the tube perpendicular to main speed)
At most it has tubes, the better will be the resulting averaged pressure.
All the tubes would be connected to a kind of integrator (pingpong ball is great idea) that will filter the pressure variations. This chamber will act as a pressure equalizer.
From this chamber, put 1 tube that goes to your sensor.

The volume of the chamber and the tube inner diameter would give the filter time constant that has to be tuned regarding the expected dynamics.

Negative effect: global delay of baro measure (150ms) would be increased.
Positive effect expected : averaged measure with a single baro sensor, insensitive to speed

Julien

[David Pawlak]

Just in case it's not clear in my description, the vertical hole I have in the white plastic piece goes from top to bottom. The black plate it's on, has a hole in it as well, which mates flush.

But far from ideal. Something like Phil's design would obviously be better, and serve a wider range of functionality.

My next test will be more like Paul describerd. A sort of" Star Trek Enterprise" shaped disk. I'm doing this more to verify the "drop" theory, as I can fit it between the three builds I have going at the moment.

Thanks for the input. It's clear it's not a simple matter, especially in a 3D multi.

[Julien Dubois]

Ok David, I had not fully understood, sorry. It's clear now.

So you have at the front of copter, 2 nuzzles (one at top and one at bottom).
Of course pressure around the copter depends on its shape but I think we can have a nice equalization if we sum pressures from 2 symmetrical points front the copter's center.

Here, you've averaged 2 points symmetrical to a line passing through this center, but not directly symmetrical to this center.

So, maybe, if you add the same system at the rear of the copter, collect both front and rear systems together and send the result to you sensor, that would give a better average.

Julien

[David Pawlak]

Sounds promising and doable. Thanks