This would make a great infill setting. Although the demo is only in 2D, 3D packing would need more thought.
One approach could be to optimise a 3D packing of variable sized polyhedrons (eg octahedrons) into an STL shell before slicing so that each layer of the infill will be supported by the layer below.
http://youtu.be/SAha2pBq-R0
Nice! do you have a link to a description of the algorithm?
@Anders_Wallin It was written by @Damian_Axford , a fellow member of @Swindon_Hackspace . You can find more information here : http://axrap.blogspot.co.uk/2013/01/variable-density-circle-packing-to.html
@Ashley_Webster A stronger lighter part, it’s how human bones are structured.
@Mark_Rehorst Also, the video is not running real time, it is slowed down a lot for the purposes of seeing the packing procedure taking place.
@Ashley_Webster I believe there will be a time saving for a comparable strength part because the mean average cell size would be larger than traditional infills. It’s the distribution that changes the strength.
Circles are really rough on simple-minded G-code generators but friendly to acceleration-limited machines.
My experience is that circles are not as good as long straight lines, but better than things with corners. Because the problem isn’t how fast you move, it’s how fast you can change direction. But similarly I haven’t done any detailed testing. Real G-code knows about arcs, but whether common FDM slicing software and machine firmware can do those I don’t know. (And if nothing else, lots of tiny segments clog your comm channel.)
It would be more efficient if the algorithm started with larger circles and didn’t put tiny circles all along the perimeter. Also, it needs to be 3D to work. Just think about how the pattern would need to adapt to fill a pyramid to see what I mean. As has been pointed out before, FEA meshers know how to fill a solid with polyhedrons. We just need someone with the know-how to do it.
As mentioned in the original text, I think octohedrons would make a great 3D packing shape. When sliced, the resultant slice would be lots of variable length lines. i understand long straight lines are quick to print, but savings will occur because the cell structure would be larger where strength is not required. The changes in direction would be akin to the hexagonal infills that we see a lot of.
So I’m beginning to wonder whether, if we’re going to go to the trouble of figuring out octagonal packings and so forth, we shouldn’t think about what it would take to just go full FEA. For the most part, infill is just because we don’t really know the load paths through a piece(or don’t care enough to know). But for most practical (as opposed to decorative) prints we either do know the load paths or some minimal FEA could figure them out. Classical infill tends to act mostly as support for top layers, and there are likely more efficient ways of doing that.
@paul_wallich I think there is room for both. FEA packing based on load paths and a “standard” packing algorithm where the only user modifiable parameters are average cell size and maybe a cell shape option. In the 3D world the top surfaces would also have smaller cells underneath it for strength and support.
@Steve_Wood_Gyrobot There’s an interesting option in slic3r to only put infill where it will be needed to support top surfaces. Wasn’t really ready for prime time last time I tried it, but I think the idea of having something like that able to leave the interior volume mostly empty could be both useful and fast. I bet that a smoothed interior perimeter a few shells thick, with zigzag fill connecting to the outer perimeter, would do the trick in 2D, and then you’d need to migrate inward at the maximum allowable overhang angle to support bridging at the top. (I can imagine doing this for designs in openscad, but doing it to STL would be way beyond my competence.)
Speaking of Slic3r, I believe it’s the current release that has a 3D fill option using octahedrons. Which I keep forgetting to try.
The problem with using minimal FEA to figure out a load paths is that FEA is rarely as minimal as you would like.
@Dale_Dunn yes, 3D hexagon infill, its a great start, now to vary the size distribution.