Disney Research has been on a serious roll with its 3D printing innovations and 3D printing patents. Fromhigh-res 3D printing processes, to replicating reflective properties onto 3D printed surfaces, to 3D printed wall-climbing robots, it seems as though Disney is looking to redefine how movie merchandise is made using 3D printing technology. But their latest study shows that they are also keen to bring 3D printing principles to other industries, for they have developed a new compiler that lets knitting machines behave like 3D printers and easily produce customized objects.

You might be as surprised as us to find out that knitting machines can’t really do that right now – even though they’ve been around for decades. For while industrial knitting machines can certainly produce very detailed, seamless, 3D surfaces quickly and autonomously, they are almost impossible to program for small, customized batch production. With their study, entitled A Compiler for 3D Machine Knitting, Disney Research associates Jim McCann, Lea Albaugh, Vidya Narayanan, April Grow, Jen Mankoff and Jessica Hodgins seek to change that. Also involved was MIT’s Wojciech Matusik.



As the Disney team explains, manufacturing principles are changing thanks to digitally controlled 3D printing and CNC milling innovations. “We believe that 3D machine knitting should join 3D printing in the pantheon of end-user-accessible additive fabrication, and that getting it there will require new tools, algorithms, and data exchange formats,” they say. That, in a nutshell, is what their compiler brings to industrial knitting machines. Featuring a transfer planning algorithm that converts commands into basic knitting procedures, it enables users to complete variations in knitted objects much easier and quicker.

For until now, you could use CAD software to design new garments or stuffed animals, but there is no single ‘print button’ for industrial knitting machines. “The industry standard tools for machine knitting provide high-level templates for a few standard objects, but otherwise leave the user to manipulate needle-level control instructions in a way that fails to divorce machine-specific details from actual fabrication operations,” they say, adding that customizing designs is a bit like being forced to write G-code toolpaths by hand for each and every print. That means making thousands of low-level instruction edits just to change a single design, and as it result little is customized aside from color. The world of on-demand 3D printing is very far away from knitting.



In an attempt to change this, the Disney team have developed a compiler that enables users to simply input specific high-level shape primitives (rather than detailed stitch descriptions). “These primitives are knittable by construction – that is, they can be automatically transformed into stitch level instructions. With our primitives, users can create and edit designs at a high level, and easily change knitting order, needle location, shape, and scale,” they explain. “Also, because our input format provides scheduling (knitting order and location) as well as shaping control, users can see which needles on the machine will be in use at a given time and adjust their design to avoid any conflicts; without changing its shape.”



But this required much more than simply letting knitting machine read g-codes. For knitting machines essentially feature hundreds of needles (or hooks) that perform four basic functions through a ‘knitting assembly language’ that is described in detail in the paper. The algorithm therefore had to incorporate all of these steps into a single system, while at the same time dealing with some of the other quirks of knitting. For instance, the algorithm also has to resolve the boundaries of every item to prevent unravelling. “Any loop not pulled through another loop could unravel, causing the final item to fall apart. Thus the starting and ending loops of each primitive must be made stable by being pulled through or pulling through another loop,” they say.

But there is one design advantage, and that is that most knitted objects (from gloves, sweaters, stuffed toys, to socks) consist of little more than tubes and sheets, albeit with varying bends and so on. As a result the compiler only needs to read tube and sheets primitives and convert that into knitting instructions. “After reading input primitives, our compiler breaks them into horizontal slices (courses). Computing these needle lists from the supplied input parameters is straightforward rasterization. Each course has an abstract “knitting time” value, a list (in counterclockwise order) of needles where loops will be formed, and a parameter value for every needle (running from 0 to 1 counterclockwise around the cycle) that will be used in linking,” its designers explain.


So far, the compiler has shown to work very well, and Disney have already knitted numerous unique shapes and toys, of which the snake visible above is perhaps the coolest example. The Disney research team have already said that they are planning to incorporate surface texturing options as well, which would certainly enable knitting machines to claim their place alongside the making machines of the 21rst century. Perhaps 3D printing service companies can start knitting-on-demand services too?

Image credit & Article via: 3ders.org