@AliT

Ok so it’s done! There’s an accelerometer taped to a little USB fan (yes, tape I know) and it feels the fan vibrating. That gets fed into a model I trained, and the model tells you if the fan is running fine or if something’s wrong with it. The way I make it “broken” is I stick a bit of blu-tack on one of the blades and it throws the fan off balance and the whole thing shakes differently. The model catches it in about a second and flips to FAULTY. I also made the Arduino’s LED light up when that happens so there’s an actual physical alarm, which is makes it better. Arduino sends 200 readings a second down the USB cable, the model runs in the browser, takes about 0.3ms. The dashboard shows the wave, a live frequency spectrum, and how confident it is also one fun thing I found: the fan spins at 37.5 Hz, so about 2250 rpm. check the video at last image slide or check more of it in my github: https://alioramuss.github.io/stardance-motor-fault-detector/

Video of it working. The fan runs normally and it says HEALTHY, then I stick a weight on one of the blades and it flips to FAULT. Second clip is the same thing again. Then I mess about tapping the sensor to see what it does.
How it works: there’s an accelerometer taped to a USB fan, wired up to an Arduino. The Arduino reads it 200 times a second and sends the numbers to my laptop. It doesn’t do the thinking itself, it’s only got 2 KB of RAM so there’s no room.
I recorded a minute of the fan running fine and a minute with a weight on it, chopped those into 5 second bits, and trained a model in Edge Impulse. It looks at a second of data at a time and decides from the frequencies.
First go it got 64.7% and just called everything faulty. Turns out I hadn’t normalized the features. Some were tiny numbers, some were in the thousands, so it couldn’t learn properly. Fixed that and it went to 100%.
Now it runs in a browser tab and reads the Arduino live. Takes 0.3 ms to classify. There’s a graph that shows the fan’s spin as a spike at 37.5 Hz, which is about 2250 rpm. Also got an LED on the board that lights up when it finds a fault
Being honest, that 100% is a bit flattering. Both recordings came from one session so it’s only ever seen this fan on one afternoon. And it only knows two things, so when I tap it there’s no option for “that’s something else”. It has to say healthy or faulty. Next job is recording loads of separate sessions and adding a fan off class.
You can try it yourself, no hardware needed:
https://alioramuss.github.io/stardance-motor-fault-detector/
I hope you guys like it

Added a live FFT graph to the dashboard today. You can watch the fan’s spin show up as a spike at 37.5 Hz, which is about 2250 rpm. Kind of cool that I measured my fan’s speed with an accelerometer.
Also made the Arduino’s LED light up when it spots a fault, so it’s not just on the screen. And added a slider for how sensitive it is, because it kept flipping between healthy and faulty when it wasn’t sure

Got and soldered the MPU-6050 accelerometer and wiring was done, most of the two hours was setting it up and installing Node.js and had to resolve issues with an ARM64 and as it did not have a C++ compiler, so i am using PowerShell right now

Hello, as my first devlog i want to say that my parts have arrived except the sensor, I used Arduino UNO to first test out the LED and counts over serial and i have attached a tape on my fan to represent it as a fauly motor, as it would cause an increase in vibrations