Tuesday, April 3, 2018

Airball Probe v4 is alive – and flying!

A new version of the Airball Sensor Board

In Oopsie – but it's alive! back in January, I described the process of producing the Airball Sensor Board v3 (and the "oopsie" in the short-lived v2), which made a number of improvements from Ihab's initial design (retroactively named v1).

While Ihab has been busy testing the v3 boards, I've now produced Airball Sensor Board v4, which made a handful of important improvements and made the boards a LOT better:

  1. Integrated all necessary components from the Sparkfun Fio v3 which we were carrying on the previous boards in a socket. This allows us to reduce the size of the overall board stackup quite a bit and allows us to fix a number of issues with the Fio, and gives us more flexibility in laying out the board as well.
  2. Relocated the pressure sensors and re-arranged them.
  3. Replaced the (very tiny) TI TMP102 outside air temperature sensor with a bigger and more manageable TI TMP275.
  4. Integrated all battery charging and voltage regulation on the board (instead of the Fio) and added a new battery monitor chip to provide accurate voltage and charge/discharge rate from the LiPo battery, which I wrote about recently in Validating air data probe battery life – with data.
This is what the new boards look like when populated:


It takes a number of parts to assemble one:


When assembled with the new mechanical parts, it looks pretty nice! I hope you'll agree:


The internal plumbing is quite clean with the new design:

Flying my DA40 with my new Airball probe!

I took my DA40 up for a short flight to validate the probe (and the probe mount) that I built, to test radio reception, and to see if I couldn't collect some data to add to our growing data set and provide some new insights through analysis.

Here's a close up mounted to the plane:


And a glamor shot of the first Airball-equipped DA40 (at Carson City Airport):


Onwards and upwards!

Friday, March 23, 2018

Validating air data probe battery life – with data

Hello, world!

I've been hard at work on several aspects of Airball's air data sensor probe as well as the in-cockpit display unit. I recently received circuit boards for version 4 of the sensor probe. This new board fixes a lot of things, and adds several new bits of functionality... which I'll cover in another post. One of the things it adds, though, is a battery monitor chip (the Maxim DS2782), so we can now monitor the probe's battery status in real time.

Since I am now able to easily get high resolution battery voltage and current, it seemed like a good time to do a battery "rundown" test, and collect useful data. I wanted to validate a few things:
  1. The total useful runtime from a full charge. I had some ideas about what it ought to be, but I wanted to confirm that the math is not too far off base.
  2. How the probe behaves as the battery voltage drops.
  3. Whether the probe will safely cut off at any point, or just keep running and potentially damage the battery.
  4. Whether the battery monitor chip is working and sane.
The test went well, overall! For this test, I was using a 1 Ah LiPo battery from SparkFun and a XBee Pro 60 mW 802.15.4 radio. Here is the data collected:



Using this combination of battery and radio yielded at least 12 hours of usable runtime on a full charge – quite acceptable! And, the battery discharge curve looks exactly like it should for a LiPo battery. So I must be doing something right...

A few things still require follow-up, of course:

  • The runtime was actually longer than expected (we expected ~7.5h, but ran for about double that). The battery discharge rates (in both mA and mW) we saw aligned reasonably well with our predictions, but the battery capacity seemed to be much higher. I am not sure whether that is due to the battery cell having a higher capacity than advertised (maybe?!), or some other problem with the test setup, but more validation is needed.
  • Neither the battery itself nor any other component cut off the module at a safe low voltage level. (The battery should have a low voltage cutoff at 2.75V, but despite allowing it to discharge to 2.6V, it never cut off...) I will add some additional protection mechanisms (both software and hardware) to avoid over-discharge of the cell.
More to come soon about these air data probe improvements, and some brand new design work on the display unit!

Tuesday, March 13, 2018

Yaw consistency in flight test

Here we will analyze the yaw consistency of the flight test in this directory:

https://github.com/airball-aero/airball-embedded/tree/master/data/2018-03-09-01

Recall that we observed the positions of the Airball yaw indication, with the inclinometer centered, at the following locations:


The display is 480 px wide, so half of it is 240 px. We know the pixel sizes of the decorations -- they are laid out on a 20-pixel grid. And we know that full scale beta was +/- 15 degrees. So we can approximate the yaw angles that our probe computed at these conditions as:

cruise: (2.75 * 20) / 240 * 15 = 3.4 degrees
climb: (2.5 * 20) / 240 * 15 = 3.1 degrees
slow flight: (1 * 20) / 240 * 15 = 1.3 degrees

The total variation in yaw between cruise and slow flight is therefore (3.4 - 1.3) = 2.1 degrees.

Assuming that we did indeed center the ball properly in all conditions, this means we should expect, at that mounting location, to read about 2 degrees of yaw error as a result of speed variation.

Recall that this is with a "bad" mounting location that is really pretty close to the prop slipstream -- in fact, closer to the slipstream than the factory Pitot tube itself:


So for the Nth time, the question we're trying to answer here is: Is it a good idea for Airball to measure the angle of yaw directly, or is the flow near the airplane too disturbed, and should we instead settle for something like lateral acceleration normalized by our best guess of the free stream dynamic pressure?

My conclusion from this data is that we don't yet know. :) We should:

1. Try another test with the probe mounted farther outboard -- perhaps mounted a little distance vertically below the intersection of the strut and the wing.

2. In flight, play with the relative sensitivity of the Airball yaw measurement compared to the inclinometer ball, since theory says that, with reduced dynamic pressure, the inclinometer should get "mushy". Verify that the yaw deviation seen at slow flight is real.

3. Acquire some data where we correlate the probe inputs with a signal from an accelerometer. This should eliminate the pilot technique issues of trying to center the inclinometer then read the Airball location.

This should give us a bit more information to either validate our direct yaw measurement, or nudge us to do it indirectly instead.

Second flight test

We conducted a second flight test, in a Cessna 172M. The data folder is here, with lab notes included. We will be adding some blog posts about the interpretation of the data soon.

https://github.com/airball-aero/airball-embedded/tree/master/data/2018-03-09-01

Friday, March 9, 2018

Open Source!

We've been busy with this and that since our test flight, but we haven't blogged much. Rest assured, more news is on the way. We're hard at work on new probe mechanical parts and electronics, a new display, and a more powerful wireless connection.

But we do have a HUGE milestone for you: Our code is now officially Open Source under the MIT License! Check it out here:

https://github.com/airball-aero/

We are not yet quite sure of the best way to release our mechanical designs -- these are scattered around in SolidWorks, OnShape and Autodesk Fusion 360. But we're working on that. Otherwise, everything else about Airball, including Eagle files for our circuit designs, is in the repo!

Sunday, February 11, 2018

Yaw bias due to asymmetric mounting

One of the things we noticed when we flew yesterday is that the ball was consistently deflected left (i.e. showing the airplane yawed to the right) throughout the flight.

The probe was mounted on the right hand side of the plane. We can imagine the airflow probably looks something like this:



It's not surprising, therefore, that the probe "sees" an airflow that tends to come from the left.

The problem, though, is whether this "yaw bias" is consistent across flight conditions. If it is consistent, then for a given probe mounting position, on a given airplane, all we need do is add an offset in the display software, and we're done. If it's not consistent, then we need to do something more fancy.

We do have fancy tricks we can pull out, where we "fake out" a good-enough yaw reading using an accelerometer and scaling to the current IAS, but we'd rather not play these games. The closer we are to a simple glider yaw string, the better.

Now a word about estimating angle of yaw. It turns out (we describe this in our original 2016 paper) that the ratio of pressure difference between left and right holes and the pressure difference between center hole and static is a nondimensional quantity that, according to potential flow theory, is a predictor of yaw. (The same is true for upper and lower holes, and AoA.) It also turns out that the relationship to angle is very nearly linear, with a scaling factor of about 11.5 degrees for each unit of ratio difference. With that, we can do a very rough computation of angle of yaw based on our pressures, and this is what we'll be using here.

Our intention is to get a general clue as to whether we have a project, given our current setup, or whether we need to seriously go back to the drawing board.

My conclusion at the moment is that we still have a project. Let me show you the data. First, here is the center hole dynamic pressure plotted at the top, and rough-calculated angle of yaw below, for the entire test:


What a mess, right? When we're sitting on the ground, the data is all over the place. Notice though that, halfway into the first flight, we do some maneuvers, including slowing down (this is when we tried stalls). Zooming into this maneuvering period, we have:


This looks more like data, and you can see the near-constant offset of the yaw signal. If I had to guess, I'd say it's about negative 2.5 degrees or so. The question is how this yaw signal varied with conditions. We can zoom into the area near the left of the plot where we are slowing down graduallyin preparation for the stall to see more:


There is a trend towards the right-yaw signal correcting to the left towards the low end of the speed region -- but is that real yaw that happened as we slowed down?

Finally, for this section of the data, we can plot center hole dynamic pressure versus yaw to see if there's a correlation:
Clearly there is some sort of correlation. It seems to go from a rough average of about -3.2 degrees at high speed to -2 degrees at low speed, a difference of 1.2 degrees. Again, I have no idea if this is a variation in actual yaw or the "fake" yaw due to our mounting.


In the big picture, we are building an operational instrument for pilots, not a flight test instrument. The gold standard for a flight-test-worthy airdata probe is a large boom that sticks out in front of the airplane. We can't tell every pilot to install one of those. So what's good enough for piloting?

While this question remains open, we certainly can do more data acquisition to study this issue. There are various avenues, at least a few of which are:

  • Trying out various installation locations on the same airplane;
  • Trying maneuvers with an experienced pilot remaining coordinated, and looking for correlations in that data; and
  • Acquiring data from an accelerometer and correlating with our yaw signal.
One more thing that's worth mentioning is that the standard inclinometer ball is already "wrong". For  a given angle of yaw, the ball is less sensitive at lower airspeeds (in fact its sensitivity should follow the square of the airspeed, so the difference is not trivial). Our yaw signal should be scaled properly at all airspeeds -- and it does not lose sensitivity at the low speeds when you need it most.

We hope to figure out what is the "best" yaw estimation method given convenient mounting on everyday airplanes. It might be our current aero-derived yaw, or it may be something else. Stay tuned!

Saturday, February 10, 2018

Successful first test flight

Well we did it! Thanks to my friend KC Budd who flew our hardware on his Cessna 172, N2720L, we have a successful first flight!

We flew KPAO KHAF, stopped for lunch, then returned KHAF KPAO. The air was pretty bumpy in places, over the hills and near the ground at KHAF. We did enough maneuvering to calibrate the fiducials on the Airball display, and observed how it performed in stalls.

Now the main goal of this test was to see if we have a project, or if we're barking up the wrong tree. I had no idea what the actual data would look like -- the AoA and yaw may be so bunched up on one end of the scale or the other that it would be useless. The airspeed feedback due to the ball size may be bogus. The whole concept may be bogus. Who knows until you've tried it, right? I have done some testing on the old Android app with my other friend's RV-9A, getting data from his Dynon avionics, but this was the first time with an actual probe -- and a homemade one at that. Would this be a complete dud and should we pursue something else, like quilting?

Well so KC's initial feedback is that, to a first approximation, we have a project. I concur, but then I'm obviously highly motivated. :) The AoA varied enough in all phases of flight to give useful feedback, and the airspeed feedback from the ball size helped situate us. Climbing at Vy with the AoA pegged on the Vy symbol was pretty cool. Cruise looked like cruise. Stalling looked like you'd expect.

There is still some human factors work to be done to determine if applying some nonlinear transform to the vertical scale a little bit would make it more information-dense, but for now, this thing looks like data rather than nonsense. May it continue to be so.

There are videos, which I'll link to as soon as they are uploaded from my phone.

This is our route (outbound and return). You can see the wiggling we did outbound, which was when we were testing stalls and calibrating the display:


This is what the data for the whole flight looks like. About 2/3 of the way through you can see a sudden vertical drop in temperature; this is when we switched the system off at KHAF and went to eat lunch, then came back, switched it on, and flew home:


I will be analyzing this data and posting updates as I go. One thing that was interesting is that there was a yaw "bias" in the instrument, which caused us to always be off-center. My assumption is that this was due to a combination of the flow field around the fuselage, and the mounting of the probe. It was therefore hard to try out the slip indication. One important question will be whether a constant offset can indeed correct for this bias, or if it's nontrivially related to airspeed, power setting and other factors.

Here is the back of my car as I got all my stuff together:


Eventually I got the probe mounted. You might recognize the mount from the renderings I posted yesterday. This mount worked really well and was very rigid. I'm going to make a couple of improvements to it then post the design here, and the 3D printed parts on Thingiverse, since it might be useful to some people in its own right:






We mounted the display on the windshield with a Ram suction mount:


And here we are flying (note the yaw bias):


Here is the probe flying over the ocean:


It also works over mountains:


And it can even work near the coastline, in between ocean and mountains:


which surely you must agree makes it a very versatile probe indeed! :)

More soon; stay tuned!