Audio Recording Form2 + Machine Learning = Issue Detection?

Experimental
1

Hi everyone,

Two days ago I heard a loud banging noise just after starting a new print on my Form2. Luckily I immediately found out that I stupidly forgot to attach the wiper after changing the resin tray. I could abort and start over rapidly without any damage to the printer and its accessories. I think I was quite lucky as I did not leave the place right after starting the print.

What saved me here is the noise from the machine that gave me quite a hint that something was wrong.

I think some sort of sound monitoring could be useful either as a warning or diagnosis tool. I’m not a coder but as a proof of concept I used Teachable Machine, a machine learning tool from Google and trained it using a microphone near my Form2.

I made several classes (Background Noise, Wiper, Build Platform going UP, Build Platform going Down) and the preview seems to be working okay with little samples you can try it online if you have a microphone to put next to the printer.

teachableMachineForm2
teachableMachineForm21600×799 187 KB

I think with more training on specific problematic sounds, and a better method for recording sounds it could make a great inexpensive feature. I am thinking about some noise-related posts in the forum that were related to:

  • z-axis stepper motor
  • suction issue
  • resin tray not mounted well
  • collision wiper/failing parts

I am not keen on reproducing these issues on my printer to train the model but I would totally have a Raspberry Pi with a microphone recording during the print and send me notifications if something weird is recorded.

PS: Happy to share TM file with samples (30Mo)
PPS: I also tricked TM by sending sliced mp3 recording of the print via a virtual microphone

3 Likes
2

how would you record ONLY the sounds from the printer and not those of the surroundings?
(Privacy issues)
Great idea though! I’d very much like to have a self-monitoring printer

3

Hey hoollto,

If I understand well, once the algo is trained with this tool, it is possible to convert to TensorFlow for local use (FAQ/Can I use my model outside TM?) without worrying having a microphone connected to Google.

1 Like
4

I love this idea. I wish the use of secondary signals like this were explored more widely across products.

Another one I wish got more uptake is the use of a feedback loop during printing to tighten tolerances. e.g. There’s a MarkForged printer that measures your model with a laser ranger as it prints, and one could use that kind of input to continuously calibrate and correct for many types of aberration.

1 Like