The problem with airspeed sensors

Airspeed sensors have been a major concern for both manned, and unmanned aviation alike. Yet almost every fixed wing or VTOL vehicle is dependent on it. So what is wrong with these sensors and why are they used?

How an airspeed sensor should work

A standard airspeed sensors consists of a pitot tube, a temperature sensor and 2 pressure sensors. The pitot tube has 2 ports (holes), one at the tip (like an injection needle) which is called the dynamic port, and one at the side which is called the static port. Each of these ports are connected to a pressure sensor. When the tube is pointing forward on a moving vehicle, it can measure the difference between the pressure from dynamic and static port. When factoring in the temperature it can estimate the current speed through air.

So whats the problem?

Over the years many problems have been identified with the use of airspeed sensors in both manned and unmanned vehicles. There is not one major problem so here is a list in random order;

For both manned and unmanned aviation

  • Icing: The same ice you see forming on the wings of an airplane can accumulate inside the tubes. When this happens the reading get unreliable or inconsistent at best.
  • Water: It might sound scary, but a few drops of rain inside the tube can cause the sensor to fail. The pitot tube holes are therefor made as small as possible.
  • Sand & dirt: Due to the small size of the holes sand or dirt can get lodged inside the tube rendering it useless.
  • Calibration: An airspeed sensors requires calibration before every flight. This is needed to account for the difference in readings the 2 pressure sensors will give. No 2 sensors will read exactly the same. The calibration is a snapshot that can be different from one moment to the next.
  • Drift: As the sensors are in operation they tend to drift due to temperature changes, how they drift is not always the same so it is hard to compensate for this.
  • Angle of attack: As you might imagine the system only works correctly if the sensor is pointing directly into the wind, but the pitch angle of any aircraft, especially unmanned, can change based on speed, air pressure, temperature or drag.

Specifically for unmanned vehicles

  • It’s fragile: The sensor needs to measure undisturbed air to work, and therefor needs to stick out in front of the UAV. The pitot tube is hollow and fragile and can easily break during transport or landing.
  • It needs special calibration: Not everyone is a UAV specialist, and properly calibrating an airspeed sensor takes a specialist. The board computer cannot tell if the calibration was done correctly, it just needs to assume it was.
  • Temperature drift: A UAV heats up as it flies. This means there is a drift in temperature that does not correlate to the air temperature. The board computer will try to compensate for a temperature that is not the actual air temperature.
  • Cross wind: Where manned aircraft usually have multiple airspeed sensors, UAVs normally have only one. If it flies in a cross wind there will be extra pressure measured on the static port (on the side).

The effect of incorrect airspeed readings

If the airspeed data is missing most vehicles UAVs will not takeoff or enter some failsafe mode. But usually the data is not missing, it is simply incorrect. With incorrect readings it will either fly too fast and therefor very inefficient, or it will fly too slow and in many cases crash.

Another serious side effect of bad airspeed readings is landing. Fixed wing UAVs are generally loaded to their maximum capacity. That means they need to fly their absolute minimum speed to land without damage. If they land too fast they start tumbling and usually break their wings or worse. VTOL fixed wing UAVs do not have this problem, but still suffer from inefficient or dangerous flight behavior when the readings are incorrect.

Eliminating the danger

During the research & development of the DeltaQuad VTOL UAV it quickly became clear that the airspeed sensor was a major concern. We therefor embarked on achieving the impossible: completely eliminate the danger.

Having members on the PX4 core development team can be a big advantage when pioneering in unmanned aviation. Using this advantage we managed to completely eliminate the need for an airspeed sensor by using a variety of other sensors to estimate our airspeed accurately and consistently.

Achieving accurate and reliable estimation was not easy, and required very specific tuning and testing, but the end result turned out better then we hoped. Using our estimated readings we not only eliminated all the issues in relation to airspeed sensors, we had also increased the efficiency of our vehicle by 20%

So if you are considering the purchase of a fixed wing or VTOL UAV, and you see a probe sticking out in front that look suspiciously like an airspeed sensor, make sure you think twice.

Views: 988


Moderator
Comment by Graham Dyer on April 2, 2019 at 9:21am

Will, or is this method be available in PX4/Ardupilot or is this just an advert for DeltaQuad?


100KM
Comment by sander on April 2, 2019 at 9:33am

@Graham all code we produce on the primary avionics is merged upstream and available in PX4. I also wrote a tuning guide in the PX4 docs. It just takes quite some tuning to get it right.

Comment by Mike on April 2, 2019 at 10:45am

Thanks for your post. I have found it very interesting although I have noticed some points of which I would like to know your opinion if possible. I'm not a PX4 developer like you and your team, so I apologize in advance for my lack of knowledge in the field.

Other onboard sensors have similar or even worse problems than those mentioned (temperature dependency, drift or random walk calibration ...)

I'm missing some more information about what other sensors have used to complement the absence of the Air Data System. It gives the impression that they are relying heavily on the GNSS system (from my deep ignorance) . In that case, I would also like to know the consequences derived to get safe platform performance after a GNSS failure (e.g. Jamming ).

We should not forget a very important fact: the air is the fluid in which aircraft moves and is what produces the precious lift.
An ADS system can give us great information at all times, but even more useful during dead reckoning phases. To this day, a manned aviation is totally inconceivable without using these autonomous systems (they do not depend on any out of our system, such as, satellites constellations for example), despite the inconvenience caused by doing a pre-flight, ice, or dirt...

Apart from the body axes and NED axes, there are the aerodynamic axes, as well: angle of attack and sideslip angle. These angles are defined with respect to the incident wind and play an importnat role during flight. On the other hand, an adequate wind estimation will make our operation more efficient. Have you got a reliable wind and crab angle estimation without ADS?

From my point of view, as a non-expert, the key to achieving a safe operation would be the redundancy of different sources of information. (It is said: "Information is power")

I think that several observers (e.g. sensors) with different characteristics allow us to cross information and even estimate variables that we can not measure directly. In this way, isolated sensor failure could be addressed with more guarantees.

Instead of removing elements, in my opinion, work should be done to generate the necessary logic to detect that a sensor is operating incorrectly or outside its declared range of use and isolate it.

I still think that an ADS, is a cheap sensor, easy to maintain and can provide a great added value to the whole system.

On the other hand, and from my humble opinion and without intending to offend anyone, I do not like the mention of the plane crash at the beginning of the post. I think it gives a wrong message: It could be misinterpreted as a marketing strategy by mistake.

Thanks and good luck with your new platform.


100KM
Comment by sander on April 2, 2019 at 2:11pm

@Mike

Thank you for your kind feedback :)

Other onboard sensors have similar or even worse problems than those mentioned

You are correct that drift occurs on the other sensors too, however, the drift on those sensors, although significant to the primary functions of the vehicle, is not very significant to the airspeed estimation. Also, the data from these sensors are a combination of several different types of sensors (gyro, accel, etc) and most of these sensors are redundant. The remaining error in the readings is therefor very minimal, and the drift based on temperature is linear and can be compensated for.

Take for example the pitch angle (one of the factors taken into account). In an extreme condition the gyro could be off by 2 degrees, although this is significant for the sensor itself, it is a relatively small error when attempting to maintain cruise flight.

It gives the impression that they are relying heavily on the GNSS system

No, not at all. GNSS is not used for this calculation, that would indeed be a very bad thing to rely on for the reasons you state, and also because it does not tell you anything about wind conditions.

The sensors used to maintain stable flight are accelerometer, gyro, temperature sensor, voltage sensor and barometer.

These sensors are combined to provide data using Kalman filters to weed out any malfunctioning or noisy sensors. The data they then produce accurately is attitude, velocity, heading, altitude, air pressure and thrust. Based on this data an estimate of airspeed and even wind force and direction can be made. The software essentially controls the vehicle the same way a skilled pilot would when lacking airspeed data,

> the key to achieving a safe operation would be the redundancy of different sources

Indeed generally the more sensors you can compare the more accurate the result. For example, if one of the redundant gyros is reading a change in heading, but both compasses and the other gyro disagree, you can safely drop the trust level on that gyro. Unfortunately the only thing you would have to check the accuracy of the airspeed sensor would be the synthetic airspeed i just explained and no third arbiter. It is not for nothing that most passenger airplanes have at least 3 airspeed sensors.

> despite the inconvenience caused by doing a pre-flight, ice, or dirt

Unfortunately our field tests showed that even when performing these tasks correctly, more then half of our failures in autonomous flight were still related to incorrect airspeed data. This very fact led us to address fail-safes for the sensor. Our fail-safes proved so good we decided to replace the sensor all together.

> I do not like the mention of the plane crash at the beginning of the post. I think it gives a wrong message.

I agree, i started writing this article when reading this news (slightly late), and further down my blog I do underline that i think we did something well on our products. I will have the reference removed.

Comment by Mike on April 3, 2019 at 1:40am

Thanks for your reply, Sander.


You have mentioned a just released "tuning guide" within PX4 documentation. Could you provide a link or an easy way to find it? I am afraid I did not find it, but I need to confess that I do not master the PX4 documentation estucture and likely I have been looking for it in the wrong place.


100KM
Comment by sander on April 3, 2019 at 2:22am

Hi Mike,

I did not say it was "just released", i wrote one about 12 months ago.

http://docs.px4.io/en/config_vtol/vtol_without_airspeed_sensor.html

Comment by Mike on April 3, 2019 at 2:47am

It's true! :)

Thanks

Comment by Andreas Gazis on April 3, 2019 at 5:04am

How do you get thrust without using a current sensor?


100KM
Comment by sander on April 3, 2019 at 6:47am

@Andreas: voltage and known thrust curve of the motor

Comment by Andreas Gazis on April 3, 2019 at 7:56am

I see, but voltage will also drop as the battery discharges so you need to account for that. In turn, this also assumes you really know your battery's discharge curve and internal resistance, plus this will vary with temperature and age. Is it not a hell of a lot more reliable to just get a current sensor which will work across any battery without having to guesstimate and guesscalibrate through voltage? Or, even better, read motor rpm through the ESC? These sensors are more robust than pitot tubes and reasonably cheap and standard.

Comment

You need to be a member of DIY Drones to add comments!

Join DIY Drones

Groups

Season Two of the Trust Time Trial (T3) Contest 
A list of all T3 contests is here. The current round, the Vertical Horizontal one, is here

© 2019   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service