Correlation in between volumetric flow and extruder power

Developers
1

correlation_volumetric_flow_power
correlation_volumetric_flow_power1920×975 146 KB

I currently log my print data using the Moonraker Klipper API, for a project, which should detect a clogged nozzle by training a machine learning algorithm. The repo can be found here. The correlation factor of the mean value is 0.38. This is the data of printing 51h with Pla(+) and Petg. The black lines indicate a new print. The next step will be systematically clog the nozzle (or stop extruding). If this project will work my model won’t be working for everyone, so if you find this project interesting you can start to log your own print data.

This solution won’t be perfect and I don’t even know if it can work reliably ( a sensor is definitely better). I actually rarely have a clog (maybe 2 in a year), but I found this project interesting and I want to learn more about machine learning.

I got this idea from @TreoWayne, at this closed post.

3 Likes
2

Neat! Here are my two ideas:

  1. Except for an AMS environment, which may contain motorized feeds, could you place a spool on rollers that track the rotation of the rollers? Then you could put in a flag if the rollers stopped turning. Sometimes filament runout won’t detect this issue because filament would be in the detector switch, just not moving. But this would detect filament breaks as well as clogs, yes?

  2. Pressure sensor (collar) in between the bowden tube and the heat break. It would sense the bowden tube being pushed back because of the filament being pressed into the clog, removing the backlash (slop) from the gap between filament and interior of the bowden tube. Now this will probably detect retractions but retractions can removed if the pressure (or lack there of between tube and heat sink) is less than XXX milliseconds.

3

Thanks for your input. I know in general sensors are more reliable to detect clogs/tangled filament. My new printer(RatRig vcore 4) will have an orbiter runout sensor, which also measures the pulling/pushing force. When the force is greater than 2kg for 1 second it detects an issue.

I’m a mech e student and interested in machine learning. The university is quite theoretical, so in my free time i try to apply as much as i can. I’ve heard that some industrial printers already use this method. If you use mainsail (with fluid probably too) you can also plot the heater power. When you have a clog the power usage decreases quite a bit so you can see that something went wrong. I only have a 20W heater so the power(%) difference is quite large. With more powerful heaters the difference would be smaller and not that visible.

1 Like