ImageJ Circularity Analysis and Feed Tests

I didn't have a GBX up and running for a good chunk of last week, so I did some more ImageJ analysis. This time I figured out how to collect data on circularity, which ImageJ defines as:

4π × [Area]/[Perimeter]2  

A value of 1.0 indicates a perfect circle. As the value approaches 0.0, it indicates an increasingly elongated shape. 

Telling ImageJ to calculate this value is actually really easy. I went to Analyze > Set Measurements, which pops open a window. I checked "Shape descriptors," which tells ImageJ to return circularity values when the particles are analyzed.

So what kind of results did I get from the data? I did particle analysis on a variety of regrind materials.

Excluding the John Marsh materials, all regrind materials were ground up with our in-house SHINI granulator. I also analyzed some InnoCircle PET pellets to compare with the regrind.

Sifting was accomplished first with a 3D printed sifter with 5 mm diameter holes, 2 mm deep. Then the material was sifted through the same window screen we use in our dehydrator. This removes both particles too large for the 5mm sift, and too small for the window screen.

One thing to note is that the values for some of the unsifted regrind are skewed from small dust particles, especially for PETG and the HID Global polycarbonate. These dust particles are removed during the sifting and drying process. This explains why area values actually increase from the unsifted to the sifted material.

The above data confirms that our granulator produces particles on average between 3 and 3.5 mm in length, and an average circularity around 0.5.


Feed Tests

The above data characterizes the physical properties of our various printing materials, but it doesn't reveal anything about how it actually prints. So I've designed a feed test, which consists of strapping our feeding assembly (hopper, feed tube, and hopper attachment) to a shelf. The tube was mounted vertically to a piece of wood. On the actual printer the tube bends, moves, and vibrates during a print, but I'm not sure how to best simulate that besides just putting it on the printer. So I used with a vertical configuration as a starting point.


We have 2 tube sizes: the larger has a 1.5" ID, and the smaller has a 1" ID. First I tested just the tubes, without the hopper attachment. To test them, I taped up the bottom end of the tube and loaded the entire length of the tube by pouring the test material into the hopper. With the tube filled, I unblocked the bottom end and observed whether all the material flowed out via gravity. I did 5 trials with each material to confirm each result.

And... every single material flowed through both tube sizes except for the Starbucks cups, which had about four times the average particle area (41.46 mm^2) than everything else I tested. If I had a more diverse range of regrind material of various particle size, I may be able to more accurately pinpoint the the minimum particle area necessary to flow through the tube. But I think the conclusion from these tests is that the feed tube is not the limiting factor here. Instead, it's the hopper attachment.

I repeated the feed tests for both tube sizes, now with the hopper attachments. Hopper attachment rev2 attaches to the 1.5" tube, and revs 3 and 4 attach to the 1" tube. Here are the values I've tested so far:

There are 3 possible results for each test:

  • Yes: the material flowed through completely via gravity
  • Tapping: the material would get stuck, but could be dislodged and continue flowing with some very unscientific light "tapping" from me on the feed tube.
  • No: No tapping helped the material get unstuck and flow.

Unscientific tapping aside, we can generally conclude that revs 3 and 4 of the hopper attachment behave similarly, and are an improvement on rev 2. This could be because of the different lofting in the Solidworks model, or may be because the material doesn't have as dramatic of a bottleneck in the hopper attachment since it's flowing from a smaller tube. Regardless, I think we can also conclude that I need to think of a more granular (pun intended) and accurate way to test feeding behavior, and identify a more accurate threshold for particle sizes that can feed through GBX.


Please sign in to leave a comment.