4 ms·

>Is it really? I've looked into this a good bit myself and it seems that many who have tried this have trouble keeping the diameter consistent. If you'll check

>Is it really? I've looked into this a good bit myself and it seems that many who have tried this have trouble keeping the diameter consistent. If you'll check my medium post, I've linked a video where I am extruding filament. Problem with those hobbyist machines is motor torque/power rating. And second, they don't use a special screw and barrel. The motor used in professional extruders used in real filament factories starts at 5 horsepower for the smallest machine. If you miss those two things, there is no way you'll get within 0.03mm Without any PID feedback loop, I can get within 0.03 by setting puller speed manually. >I suspect that if it really was trivial, then you'd think just about any commercial supplier could get it right There is no reason to believe that junk filament is coming from people with good machine design. That could be coming from hobbyist underpowered extruder machines like filaestruder and likes. >Maybe you are just very good at it; +-0.02 is pretty good even by commercial standards. The entire first page of Amazon best sellers are +-0.03 or +-0.05. At higher extrusion speed you'll get more output per extruder machine time but you'll have lower filament dimensional stability. So the trivial part is have a powerful motor, use a good screw+barrel and don't extrude too fast. That said I've offered my telegram in medium post and YouTube video. So anyone who gets stuck can reach out to me. I am not leaving them hanging on their own. Also, I am looking to make a microcontroller firmware for extruders which basically measures the filament then sets the puller roller speed based on thickness measured. It pulls filament faster if it's thicker than target and pulls filament slower if filament is coming out to be thinner than expected. This will help us reach 0.0020mm accuracy easily! The problem I am facing is: 1. I am not good with micro controller and control algorithms 2. Laser guage which measure the filament without distorting it is expensive and starts at 2K USD. So anyone who can help solve these technical challenges are welcome. But you might not need this much dimensionally accurate filament. Selling within 0.03mm is easy. Anything better not sure how much it will help FDM quality but yes microcontroller firmware approach will help get more output out of this machine as biggest problem with extruding too fast is that dimensional stability goes haywire.

AxiomaticSpace 6y ago
Instead of a laser gauge, you could try using a camera hooked up with some edge detection software and use that to estimate the width. It probably wouldn't be as accurate, but it would definitely be cheaper.
- nitrogen 6y ago
  You'd be looking at very subtle variations in the color of single pixels, so you might as well just use a single optical sensor and light source into a single high res adc, using intensity to measure thickness. At some point dust in the air will be a limiting factor.
  - londons_explore 6y ago
    You'd make the filament go slightly diagonally across the field of view of the camera, and then you'd have thousands of part-filled pixels. As long as your material is consistent (no bubbles, shiny spots, etc.), You should be able to build a very accurate sensor this way.
  - elcritch 6y ago
    For high accuracy, the brightness is really more the limiting factor, as most of the edge finding/binarization algorithms use a intensity factor for cutoff. I used camera measurement for small pieces of tissue samples in biomechanical measurement (~1mm x 5mm). We had a small window in our lab, and cloudy days would distort measurements. I solved it mostly by going to a SLDR camera from a handheld and a backdrop with a type of luminance chart. Much more accurate than calipers for soft squishy (and wet) samples. See [1] Figure 2E (pg 22), you can average out small variations pretty readily. 1: https://pdf.sciencedirectassets.com/274145/1-s2.0-S1751616116X00112/1-s2.0-S1751616116304477/Jaremy_J_Creechley_Creep_Fracture_2016.pdf https://pdf.sciencedirectassets.com/274145/1-s2.0-S175161611... Though your idea of a light source and adc could work well if you put it in a darkened enclosure. It’d probably require a bit more hardware design or tuning but the software / microcontroller side would b simpler. A RPi PiCam with a macro lens would work well too, but the reverse with less hardware tweaking and more software.
  - econcon 6y ago
    I found this video https://youtu.be/kpsN78mQ6YY https://youtu.be/kpsN78mQ6YY They use a laser pointer to measure width of a hair.
jgable 6y ago
I can help you with the microcontroller and control algorithm side. It’s what I do. If you’re a software developer you can do it yourself with a small amount of guidance from someone like me. Email is my username plus 79 at gmail if you want to set up a quick discussion.
bsirkia 6y ago
Hey @econcon, I think my friend and I will give this a try. We're starting to spec everything out, and probably want to start out with laser gauge to have high accuracy. Would love to ask a couple questions if you have time, my email is just my username at gmail.com

AxiomaticSpace 6y ago

Instead of a laser gauge, you could try using a camera hooked up with some edge detection software and use that to estimate the width. It probably wouldn't be as accurate, but it would definitely be cheaper.

nitrogen 6y ago
You'd be looking at very subtle variations in the color of single pixels, so you might as well just use a single optical sensor and light source into a single high res adc, using intensity to measure thickness. At some point dust in the air will be a limiting factor.
- londons_explore 6y ago
  You'd make the filament go slightly diagonally across the field of view of the camera, and then you'd have thousands of part-filled pixels. As long as your material is consistent (no bubbles, shiny spots, etc.), You should be able to build a very accurate sensor this way.
- elcritch 6y ago
  For high accuracy, the brightness is really more the limiting factor, as most of the edge finding/binarization algorithms use a intensity factor for cutoff. I used camera measurement for small pieces of tissue samples in biomechanical measurement (~1mm x 5mm). We had a small window in our lab, and cloudy days would distort measurements. I solved it mostly by going to a SLDR camera from a handheld and a backdrop with a type of luminance chart. Much more accurate than calipers for soft squishy (and wet) samples. See [1] Figure 2E (pg 22), you can average out small variations pretty readily. 1: https://pdf.sciencedirectassets.com/274145/1-s2.0-S1751616116X00112/1-s2.0-S1751616116304477/Jaremy_J_Creechley_Creep_Fracture_2016.pdf https://pdf.sciencedirectassets.com/274145/1-s2.0-S175161611... Though your idea of a light source and adc could work well if you put it in a darkened enclosure. It’d probably require a bit more hardware design or tuning but the software / microcontroller side would b simpler. A RPi PiCam with a macro lens would work well too, but the reverse with less hardware tweaking and more software.
- econcon 6y ago
  I found this video https://youtu.be/kpsN78mQ6YY https://youtu.be/kpsN78mQ6YY They use a laser pointer to measure width of a hair.

jgable 6y ago

I can help you with the microcontroller and control algorithm side. It’s what I do. If you’re a software developer you can do it yourself with a small amount of guidance from someone like me. Email is my username plus 79 at gmail if you want to set up a quick discussion.

bsirkia 6y ago

Hey @econcon, I think my friend and I will give this a try. We're starting to spec everything out, and probably want to start out with laser gauge to have high accuracy. Would love to ask a couple questions if you have time, my email is just my username at gmail.com