This installment is more on the esoteric side. If you are just wanting to build an InqWind, and don’t really care about technical details, you can easily pass this one by. If on the other hand you are designing your own, there might be some value in some of these results. The main goal of the experiment was to get a testbed to evaluate using the AS5600 sensor and start the software development. Using 3D printed bearings with ceramic BB’s was just a tangent to satisfy my curiosity of whether they could be printed successfully. Let’s just jump in with the test unit.
The Mule
As hinted in the last installment, the test mule uses a 3D printed bearing made using ceramic BB’s and sized to contain the AS5600 sensor. It uses an ESP8266 WeMos development board for convenience so I can use the pin headers for wiring and don’t have to mess with power circuitry. The eventual InqWind will not use this development board. Also for this mule, I’ve added a GPS module so I can get accurate speed data. Fortunately for testing InqWind (but unfortunate for my sailing habit) winds around here are usually non-existent. To get any data in a timely manner without resorting to some kind of wind tunnel, it’s far easier to move the mule than to move the air. Using GPS can accurately get my speed.
It is integrated onto the same ESP8266 development board so I can pull wind speed values triggered by updates of the GPS unit and also so I can view real-time graphs as I keep on truck’n. The Mule cart is a pick-up truck. Although not a rocket ship able to explore InqWind’s disintegration speeds, it is up to any speed I’m willing to do on the public Interstate. It is also convenient for having the laptop handy for viewing real-time data results as well as actually programming and bug-fixes. Although I was able to test at the low end of hurricane speeds, I’m not exactly sure how I’ll test at the high end. Although I have access to a car capable of 150+ mph speeds, I can’t imagine any race track will permit me to do a SpaceX grade RUD on it. Do sailors expect their wind instruments to survive a hurricane? Maybe InqWind can just be checked-off as a best effort design.
Crash and Burn
No… not the Mule Cart. Let’s just start off with the failings discovered during testing.
- The 3D Printed bearings are a dismal failure in this use case. The friction alone kept the gauge from starting up until at least 5 to 7 mph. I’m using mph only because I can correlate it to the truck’s speedometer. It doesn’t have a knots option. Running it through the speed spectrum it was easy to see the influence of the friction on the results. At higher speeds this was largely overcome, but at low speeds it was apparent as will be discussed below.
- Noise was an issue. Even over the truck aerodynamic wind noise, and diesel engine noise, I could hear the bearings through closed windows. If you are experimenting with your own design, I’d say don’t bother with printed bearings. If you’re still wanting to pursue it, I’d say you’d might want to experiment with Nylon races and find some way of polishing them.
- I’m still getting a handle on the AS5600 sensor. It is perfect for use as a wind direction sensor with its 0.1° resolution. As I am also trying to use it for the wind speed indicator (describe in detail below) I am running into some issues. The datasheet seems to mix up sampling rates with sampling frequency. The label in the chart says sampling rate, but uses units of time instead of frequency. It is unclear if the number (150) and label are correct and the sensor is capable of only 150 Hz or if the label should be sampling period and the number and the units together are correct, indicating a 150 μ-second period and thus indicate a 6.7 kHz sampling rate. There are also some graphs and settings to enable a Fast Filter ability. I believe this is what I will need to use and I’ll need to explore this area more. As the sensor comes out of the box, I can confirm it does not like rates over 1kHz. At 200 Hz, I’m getting reliable data.
- The magnet, as discussed in the previous installment, is marginal for strength. If you are designing your own wind gauge, you should look for a vendor with 6 mm diameter, diametrically polarized magnets. I am finding that the 4 mm version, my vendor supplies forces tolerance to be very tight. Top quality 3D printers can meet these requirements, but it does not permit much wiggle room. I would have liked to open up the tolerances to make it easier to build.
Successes
Now that we have the bad news out of the way, let’s get on with the results that will get into the final InqWind.
- The AS5600 sensor appears to be quite usable for determining wind speed. This will be discussed in the next section at length.
- GPS units have come a long way. I played with a few Bluetooth GPS units about twenty years ago and found they lost signals quite easily. I received 5 of these BN-220ZF units as a free promotion. Actually, I thought it was a joke. Frankly, my expectations were pretty low. What I found is they are very reliable and easy to use. I hooked them up to the WeMos ESP8266 and am getting 7 to 10 satellites… inside my house!
- ABS Plastic is plenty strong and even though I made the Mule on the light side, it easily handled 70+ mph winds with no signs of harm.
Why Use the AS5600 for Wind Speed Determination?
Many wind speed indicators use hall effect sensors to count when a magnet goes by. This gives a frictionless way of deriving speed. Some may only use one sensor in which case they have a resolution of only one per revolution. Some use multiple sensors and get a resolution of say four counts per revolution. These are fine for high wind speeds. But at low speeds, they have to wait for some long duration to see how many counts they get before doing the speed calculation. So in these cases, you either get a delayed response that might be seconds old or an average speed that might miss short puff’s by doing an averaging of the speed. Or in the worse case, they can’t even read these slow wind speeds.
I have no clue if this is a novel use of the AS5600 sensor as no marine grade anemometers advertise what they use and I didn’t do a tear down of any. By using an AS5600 sensor to read the position, we can get the equivalent of 4000 counts per revolution. We can do two samples… say milli-seconds apart and get a very accurate speed estimate. In this way we can get a speed that is milli-seconds old instead of seconds old. We can measure changes in speed of fractions of knots instead of getting an average over multiple seconds. By using this sensor in this way, we are able to read speeds of even sub-one-knots and know we’re getting accurate resolutions of say… 6.1 versus 6.0 knots. We can get accurate values of puffs as compared to the average speeds… even in this single-digit speed realm.
Besides… it reduces the electronic part count and the number of different sensors and different software coding. All win-win. In the final InqWind, we should be able to get it down to the micro-controller board (ESP-12F), two AS5600 boards and a DC-to-DC voltage converter board. We should not need any small, secondary components or even a circuit board.
Mind Numbing Details
Engineers like charts and graphs, so here are a few and how they have driven the design. This first chart is an overall comparison of speed. The Blue line represents the truck speed. There was no noticeable ambient wind, but a few cars did pass me by and were passing by in the opposite direction. These would obviously affect the wind speed readings, but not the GPS speed.
The next two on this chart (red and grey lines) are the values calculated from using a 25 milli-second time slice and a 250 milli-second time slice. The differences are minor and I will be studying the time slice in more detail once I get a better Mule and a more controlled test. The last line (yellow) shows a wind speed based on a single sensor (1 step / revolution). What we get from this graph? Immediately obvious is that we can use the AS5600 sensor this way. We also can see that the ABS plastic didn’t mind over 60 mph (100 km/h). We also see the friction of the 3D Printed bearings are hopelessly useless in the low speeds. These have to go.
This is a detail of a high speed portion of the chart. Although this is far higher than we need for typical marine wind needs, it does illustrate good correlation of the three wind calculations relative to the GPS reading.
It is yet to be determined if the variability is due to digital sensor noise, minute gusts caused by other cars or simply bearing friction variations of the 3D Printed bearings. In future tests, we will attempt to eliminate or at least mitigate these extraneous factors.
In this final chart, we look at the low speed aspects of the sensor.
This shows the wind speeds down in the single digits. Much below this speed and the sensor stops due to bearing friction. But we can get several important points from this chart. Note how the trends of the two speeds using the AS5600 sensor do follow each other. This indicates that the differences are probably not due to digital sensor noise and are actual changes in the speed of the rotation. It does not identify whether it is variation of bearing friction or minute wind gusts… say from turbulence. We can also see that the counter version loses all fine detail. Each data point is the summation of the counter over the last 1000 milli-seconds. As there are very few counts over these slow speed times, every count affects the calculation. Even losing one count during a time slice dropped the speed calculation from a pretty steady 6.8 mph down to 4.5 mph. In other words, in this low speed realm, the resolution is over 2 mph. Although a sailor may not need to know about all these micro-fluctuations, we might want to know that the average is 6 mph or 5.9 mph. The counter method can’t give us that fine resolution. It should be easy to recognize that the AS5600 sensor gave us at least 1000x the speed resolution.
Looking Ahead
In summary, InqWind will be using purchased ball bearings, and will be using two of these AS5600 sensors – one for the wind direction indicator and another one for the speed determination. We will be making it using ABS and thus not using the far more expensive Carbon Fiber filled Nylon or Polycarbonate plastics. InqWind will also use a design (albeit having tight tolerances) so I can use the 4 mm magnets that came with my AS5600 sensors. If you’re designing your own… look for the 6 mm ones and save yourself some grief.
Inq