Started out with some photogrammetry to capture the geometry of the C-LEG, which will hopefully allow me to 3D print a bracket that fits the contours of the the C-LEG precisely.
This first scan was enough to play around with, but ultimately the glossiness and the bright sunlight caused enough gaps and distortions that I had to do a photoshoot later that night using our CNC machine as a light box. The even lighting from the LED rope was just the trick.
The next step was selecting a portion of the C-LEG’s surface to extrude into a form fitting shell. Blender was used to create a mirror image of the scan, and MeshLab was used to align the two sides and fill in the holes so I had a reconstruction of the entire CLEG (Agisoft was only able to reconstruct one side of it — I could of went back and tried another photoshoot, but decided it would be faster to just duplicate the half that worked). In the video you can see the mesh of the whole C-LEG next to the original scan.
Blender and MeshLab were used back and forth here: Blender allowed me to select a portion of the mesh freehand and export as a separate STL. MeshLab allowed me to offset this surface using ‘Uniform Mesh Resampling’ and then construct a volume around the surface using Uniform Mesh Resampling with ‘Absolute Distance’ checked off. This created an excessive and messy edge, however, so I brought it into Blender to perform a boolean intersection, extruding the surface that I selected earlier outward to overlap with the portion of the new mesh that I wanted to keep. With that cut performed, I used MeshLab one last time to perform a ‘Surface Reconstruction: Poisson” to smooth the corners. To cut a slit in the back of the model I used Tinkercad, because it’s quicker to align and subtract a cube, knowing what I know.
And it actually clipped on the way I had hoped, wrapping around the edges — but there was a considerable gap. The inner diameter of the print was 60mm, while the C-LEG is 55mm wide, so I uploaded the STL to tinkercad at 91% of the original size to continue to prototype #2:
I used some cylinder, cube, and hexagon shapes to throw together clamps that I can add nuts and bolts to for this print, to see if I can really clamp down on the C-LEG enough to hang some weight off of it.
Ended up printing copies at 93% and 96% of original size. It is not a perfect fit, but once tightened down with bolts, holds on pretty well. This one cracked due to the nut turning against the plastic — the white ABS must have shrunk more than the grey ABS, which had holes big enough for the nuts to sink into without forcing it.
I wanted to design a 3D printable file that would allow me to clamp my acrylic fairings to the cLeg. Luckily Kay and Gabe created a pretty solid model of the cLeg itself so I could use its contours in my design. Again, I had to cut the model down manually in Blender because netfabb does an awful job of exporting STL (hadn’t yet investigated if this is an actual or a free-demo-imposed limitation of the software).
I also used Blender to multiply the scale by 1,000 before importing into Tinkercad.
I like Tinkercad because it makes it super easy to perform boolean functions on complex stuff like our leg here. If you’re a tinkercadder, try ungrouping the model so you can see how I made a cross section of the leg: Tinker This
And I went on to improvise a way to screw two parts together — this is very much a test fit, to see if the idea makes any sense, to see if screwing two plastic pieces on with pressure is enough to keep it from falling off.
I went back and forth about 3D printing holes for nuts and bolts, but decided I would drill it out of the finished plastic — “machining” is way more accurate than printing for this kind of thing.
OK, the original Kinect scan was not very dimensionally accurate…I guess I could have compared measurements of the model with measurements of the object, that’s what I get for trusting technology -.- I can still use the blocks of plastic for practice fitting, though. They don’t clamp around the c-Leg, but I tested screwing the pieces together and drilling into the plastic to insert a brass threaded insert, and with a few taps of a hammer that turned our really nice.
Of course, the chunky tinkercad design isn’t so fashionable, but it served its purpose to confirm that I need to press the shape of the leg more accurately (the front fairing doesn’t sit flush with the printed model at all) and that countersunk M5 screws will do a nice job of securing the fairing to the bracket, though I’ll have to play with melting a countersink so the head of the screw doesn’t stick out the back.
Opened the original scan in Blender, selected the portion of the scan we’re concerned with. Separated it from the rest of the mesh and exported as stl.
Used Netfabb to repair the mesh and cut into a top and bottom portion.
Found out netfabb decimates the mesh upon export, from 27000 triangles to 24000 triangles, and manages to make it pretty rough, so it looks like this in 123D Make.
Inspected the files before and after repairing in Netfabb…
Used MeshLab’s Filters → Remeshing → Surface Reconstruction to “Shrinkwrap” the model and fill in gaps and holes without losing the smoothness of the scan.
Used this guide to split the part in two using Blender’s newish “Bisect” function.
Played around with different options in 123D Make…
Drew a circle in Blender against the scanned cLeg and stool, confirmed that the units of the scan are in meters. (The circle is .355 units wide and in real life the stools are too.) But, since many programs interpret these STL units as mm, our objects are 1000 times too small. Let’s scale them up.
Settled on a slicing pattern in 123D Make. Used the Modify Form feature to hollow out the leg and picked out the Radial Slices construction technique. The arrangement has some collisions at the bottom, but I don’t want a closed model anyway, so will do some hand-tuning in Inkscape. Exported the model to Blender to check that the cLeg would sit inside it okay.
Using Inkscape to layout the the pieces onto less than 4 sheets of plywood… (nesting is apparantly a difficult problem for computers, and not one that many softwares implement, certainly not free ones. just have to eyeball it)
123D Make is sadistic and exports the objects as groups of thousands of lines instead of a single vector, so editing the exported PDF is really really slow. (It takes 20 seconds to select a text object and type in a new number. I counted.) Learned things:
On the left: the highlighted list of numbers 4,000 characters long describes a single path. On the right side, two coordinates of a single line segment are surrounded by markup. The file containing 2 objects takes 73 lines of text. The file exported by 123D Make takes up 6151 lines of text for the same object. Inkscape modifies this information in memory every time you move something.
Inkscape operates with SVG, which is plain text, ASCII, one character equals one byte. When a path is exported as 677 individual paths, each one of those line segments has 8 lines of markup surrounding it in the SVG file (between the g tags, with information for position and stroke and fill attributes). So what could be saved (and manipulated in memory) as 20 bytes if it was appended to the list of nodes in a single path instead takes 389 bytes to be kept in memory as a standalone object.
I found this forum post describing how to select all the separate objects, Path → Combine them into a single object, and in node edit mode you can select all nodes and click Join Selected Nodes and it simply merges overlapping nodes. Hallelujah!
So to downsize the files enough that they’re not a pain to work with:
Open the PDF. I learned that if you check the “import via Poppler” option Inkscape can detect the colors of the stroke (otherwise imported PDFs have “undefined” color unless you adjust it yourself). So check that box. CtrlA, CtrlU to ungroup everything. Click a blue line segment. Edit →Select Same →Stroke Color. Path →Combine. F2 (Edit Paths by Node). CtrlA. Wait for a minute. Join Selected Nodes (one of the buttons in the node edit toolbar, 3rd from the left, just after the delete node button. It might take a few minutes for Inkscape to chew on this one. It will probably become unresponsive. Let it do its thing. In my case this trimmed the file size by 97% and I could then translate and rotate and rearrange the objects without any lag.
Two hours later…have my minified cut file, fitting on 1.5 sheets instead of 4. I’ve trimmed the tops and bottoms off the vertical slices so the end result will be open.
20 minutes of lasering + 10 minutes of assembling: Hey I made a thing!
Met up with Shawna and learned the design constraints: how the leg has to move, what areas can’t be covered, etc. We picked out a pattern she wouldn’t mind having wrapped around her leg:
I took a few measurements of the wooden positives and drew the outer shape of the fairing in inkscape by hand (using auto-smooth of course!). A few booleans later I had my cut file, which took about 20 minutes apiece. To get the part ready to shape against the mold, I stuck it in our convection oven to bake at 300F for about 5 minutes (don’t worry, we keep one oven for circuit boards and plastics and a different oven for foods 🙂
So the process was: semi-soften the plastic in the oven, then roughly fold it over the wooden form, then stick the whole wooden form in the oven. If you let it soften all the way when it’s sitting on the metal it’ll stick to the metal. Better to have it roughly on the wood form when it reaches maximum droopiness. When it’s as soft as rubber, we take it out and push the plastic against the form, holding it in place as it cools.
I’m pretty happy with the second prototype. I think the next step is 3D printing some bracket system, and maybe taking more time to plan out how the plastic will fold against the form — this was pretty heavily eyeballed.
