When you think anemometer, you probably don’t think “load cell” — but (statistically speaking) you probably don’t live in Hurricane Country, which is hard on wind-speed-measuring-whirligigs. When [BLANCHARD Jordan] got tired of replacing professionally-made meteorological eggbeaters, he decided he needed something without moving parts. Whatever he came up with would probably qualify for the Component Abuse Challenge, but the choice of load cells of all things to measure wind speed? Yeah, that’s not what the manufacturer intended them for.
In retrospect, it’s actually a fairly obvious solution: take a plate of known area, and you’re going to get a specific force at a given air speed. The math isn’t hard, it’s just not how we normally see this particular measurement done. Of course, a single plate would have to be pivoted to face the wind for an accurate reading, which means moving parts– something specifically excluded from the design brief. [Jordan] instead uses a pair of load cells, mounted 90 degrees to one another, for his anemometer. One measures the force in a north-south axis, and the other east-west, allowing him to easily calculate both wind speed and direction. In theory, that is. Unfortunately, he vibe coded the math with ChatGPT, and it looks like it doesn’t track direction all that well. The vibe code runs on an ESP32 is responsible for polling data, tossing outliers, and zeroing out the load cells on the regular.
If you’re feeling forgiving towards abominable intelligence, the problem might not be code, but could potentially be related to the geometry of the wind-catchers. To catch the wind coming from any angle, instead of a flat plate, a series of angled circular vanes are used, as you can see from the image.
Given that arrangement is notably not symmetrical, that might be what throws off the direction reading. Still, the wind speed measurements are in very good agreement with known-good readings. The usual rotating bird perch doesn’t measure direction either, so this solid-state replacement should be just as good.
If you like the idea of hacking components to do something the designer never intended, the 2025 Component Abuse Challenge runs until November 11th — just don’t wait until the 11th hour, because entries close at 10 AM Pacific.
Instead of intersecting plates, why not just a sphere? Sure, it would be more aerodynamic than the plates, but symmetry would be preserved in all directions, and simple calibration would compensate for the reduced drag.
Reduced CONSISTENT drag with a sphere too.
Ping-pong ball on a short fiberglass rod?
That 3d printed garbage has so much texture difference side-to-side, that I’m surprised it isn’t just a random number generator.
How to tell us you’re <70 IQ without telling us you’re <70 IQ.
Well, he gave his processor vibe code, it give him a vibe answer “Yup, there’s wind, dude.”
I don’t see the problem
I would have expected to see a sphere as the part exposed to the wind.
Load cells is very clever. The wind catchers… yeah I think those are part of the problem.
Maybe grilles like they have on wind tunnels to normalize the air coming in?
I’m thinking octagon with grilles on all 8 sides to straighten incoming airflow then loadcells with simple plate wind catchers in the center for each side.
With the assumption that it’s a buffeting issue, I’d suggest trying a piece of dense mesh or something similar where the many small buffets cancel themselves out leaving a useful average load and direction
I’ll have to look in my meteorological instrumentation textbook to be absolutely sure, but the math for determining wind direction from 2 pressure-plate anemometers (offset by 90 degrees) should be fairly simple. Anyway, this is what they do at places like Table Mesa, Colorado, where the site destroys traditional cups and vanes regularly.
s/direction/vector
I used that name and a hispanic friend guffawed and said “so, table table then?” and I looked it up and it’s north table mountain and south table mountain, no mesa involved.
Boulder county up closer to Rocky Flats has measured windspeeds over 100mph multiple times in the last couple of years. That used to be a once a decade event, now it’s happening twice in a month sometimes. It sure tears up anemometers and wind generators.
Yea, us Colorado old-timers still use the old names for things
Are you sure they are using load cell sensors and not ultrasonic wind sensors? Ultrasonic sensors are common in meteorology now, and famously tough and low-maintenance compared to mechanicals.
Yeah I was wondering the same thing, and they to direction.
Could this be done with the Trackpoint/pointing stick/isometric joystick from an old IBM Thinkpad? If so, you’d get an X/Y direction signal from a single sensor. Of course, the Trackpoint may not be strong or sensitive enough for the job, but maybe it’s worth trying. And it’s a use that IBM did not intend.
that is not a single sensor.
What about two optical encoders from a ball mouse? One to get the direction, one to get the degree of deflection of a heavy/spring loaded plate that the wind has to lift.
Anemometers rotate because measuring rotation speed is easier, more linear, and more robust than measuring non-linear force directly. Cool project, though!
Force-balance anemometers do exist, essentially a load cell behind a small plate is used to measure precise turbulence or gust measurements in wind tunnels.
Poor resolution or sensitivity to direction could also be due to both sensors being on the same plane?
If one sensor is “downwind” it could be skewing the reading when the model is expecting each input to exist in a vacuum from the other, so sensors stacked vertically may be a better approach?
The theory is interesting though, and this might be what I take a crack at implementing now that I am collecting parts for my own weather station.
… and then there is winter ice …
One of my all-time favorite reads is “The Avalanche Hunters” by Monty Atwater; his solution to freezing wind wanes and anemometers were military surplus red lights (obviously LED lights weren’t invented yet, so incandescent light bulbs) glowing bright red day and night. The story goes that those were bright enough to be seen during cloudless nights, and, sure enough local municipalities got few calls about “bright red eyes/UFOs hovering in distance” (accompanying photo shows two red lights facing wind vane and anemometer, hence “two red eyes” : – ]).
The only thing of worth here is the IDEA of using load cells.
Everything else is a lesson in what not to do.
3d printing is the wrong tool. Don’t print flat plates. (Even ones you will assemble into 3d shapes.)
The flat plates>sphere-ish shape is wrong. Use an actual sphere for consistent drag without all those 3d printer texture variations IF AND ONLY IF you have a 2-axis load cell.
If you are using single-axis cells (the only ones I have seen…) you use an array of flat plates at different angles, each on their own load cell. Then combine with math to get direction.
” vibe coding” eeuugghh. Why even bother.
We are starting to see so many “look at what I made!!!” projects where the ‘maker’ does barely any work.
Using GenAI isn’t acceptable.
Not anywhere.
It’s all trained on stolen code.
And even if the code was ‘open’, it sill requires proper accreditation.
The fact that people aren’t carrying pitchforks and torches is baffling.
We flip out when a company obviously and brazenly breaks GPL, but we think it’s cute when a widdle hakaw bwoy uses a tool that is created entirely with GPL violations or outright theft of closed source?
No. You come down wit a social ton-of-bricks, so it is well known that it is never ever EVER okay.
What?!?
Little 8-year old Timmy has been sneaking out and joyriding in a stolen car?!?
That is the appropriate level of response…
Giving an 8 year old a smile and a high five for doing felonies isn’t the correct response.
Neih3r is a shrug just because everyone else is doing it.
…requires proper attribution.
F’king autocorrect…
Apart from the flaws mentioned in the comments above, I think this is actually a pretty cool project. And hey, if it is a hobby project, there is no reason it should need to be perfect. We can all still learn.
But what I don’t get is why it is related to the component abuse challenge. Using load cells to measure… loads. Seems pretty much what they are intended for.