Free-form Curved Furniture Without Molds / Part 1





Introduction: Free-form Curved Furniture Without Molds / Part 1

About: Architect by training, Phil is a designer who codes. He abuses CNCs and industrial robots while building fine furniture, mixing digital fabrication and craftsmanship. He likes thinking about energy use with ...

Ever heard of Zipshape? It's a technique developed by Schindler Salmeron which lets you to draw a free-form profile curve for (say) a piece of furniture, and then cut a series of teeth into two sheets of thick material like MDF. Those custom-cut teeth are very specifically shaped and matched to each other such that they can "zip" together, but only when the sheets are bent into the curve that you drew at the beginning. The result is amazing: free-form bent furniture parts that you can glue up without molds of any kind. Draw a curve, cut the teeth, slather glue, stick it in a vacuum bag, and let it set.

Here's the rub: each tooth needs to be cut at a very slightly different angle in order to "zip" together properly. Thought you could rig your shopbot to do this for you? Nope, sorry. This process needs a full-fledged 5-axis CNC, or a 5+ axis industrial robot arm, or some crazy experimentation with an open saw blade. Not your usual fabber's tools. Here's how Schindler Salmeron illustrate it:

This instructable is the first in a planned series, the ultimate goal of which will be to make zipshape accessible for all with a regular 2-1/2 axis CNC (Ie, using only profile and stepped fixed-depth cuts). I will be developing browser-based online tools to handle the geometry of it, and I'll be inventing a new form of zipshape which makes it possible to cut it using only straight orthogonal cuts. Follow my series! By the end, you'll be able to draw a curve, download a DXF file, cut it on your shopbot, and glue it up into a beautiful piece of furniture. No molds required! Here's a rough table of contents for the series:

Part 1: Introduction to zipshape, my own experience with zipshape, prior work that's been done in the area, the problem to be solved, the proposed solution, tests proving the concept.

Part 2: Refinement of the tooth geometry, a more in-depth look at the geometry and math that make zipshape work. The building blocks of the future web-app to bring the process out of Rhino and into Chrome.

Part 3: To the CNC! Prior tests will have been lasercut in 2 dimensions; In part 3 we'll be moving to the CNC and fixing any bugs that come up with the shift to more furniture-like parts.

Part 4: Design of a finished piece, and Introduction of the web-based tool. Since it will be built on fundamentally different platforms, further 2D tests will be done to verify that the geometry is true to the original method. A final piece of furniture will be proposed, but for all those following along, this is the point where you can design your own.

Part 5: Construction of the new & improved 2-1/2 axis zipshape furniture (type to be determined in part 4).

Step 1: My Experience With Zipshape

My main exposure to Zipshape, aside from its being sort of "in the water" of design school and journals, was through a project I participated in last year while in the master's of architecture program at MIT. Under the direction of Prof. Sheila Kennedy, we made a series of "SOFT Rockers" for MIT's 150th birthday celebration: human-powered solar tracking devices that connect social activity and electricity generation through a unique piece of public urban furniture. This Instructable-series is more about the fabrication process than the SOFT Rocker project itself, so if you want to read more about SOFT Rockers just google them or see links here at engadget and gizmag. Big photos and a more thorough description are also available on my website.

We replicated the zipshape techniques from basic geometric principles, creating a grasshopper definition for Rhino 3D which generated tooth-geometry from an arbitrary curve. The geometric problems turned out to be the least of our worries: actually fabricating these parts was the real mess and took us several months to figure out. Initial attempts to use a Kuka 5-axis robot arm with a beefy router bit turned out not to be precise enough for tight-fitting parts, and we had to move to a much larger and even more expensive setup with a 5-axis CNC gantry; I understand there are only 3 comparable setups available in the North-Eastern United States. We got the parts made, but now that I've graduated I couldn't do it again the same way.

The reason such complex equipment is needed is because each tooth needs to be cut at a unique angle. It implies a completely different type of equipment from the more typical shopbot that can do simpler straight-up-and-down cuts: profiling, pocketing, contouring, and such. My mission here is to fix that, and make a soft rocker more makeable.

Step 2: Prior Work & Credits

The work that my team did on the Soft Rockers at MIT played a critical role in the learning process of how to make this stuff work. The images associated with this step show a bunch of failed tests, without which the current project would not be possible.

Our experiments owe their genesis, obviously, to the original Zipshape creators. Additionally, there's a student I've not met named Victor Leung whose many posts in the grasshopper forumandon his own website have been very helpful for reference. Here's a very helpful sketch he's put together that illustrates the principles of the complex geometry at play in zipshape: 

Step 3: The Problem I'm Addressing

As I've mentioned in the lead-up, the issue I'm tackling here is the variably-beveled cuts required to fabricate zipshape teeth. With those angled cuts, you have essentially two options for fabrication. Either you're limited to basically 2-D parts (the main limitation that Victor Leung has been operating under in his awesome laser-cutterable tests), or, if you wanted to make furniture out of extruded versions of the geometry, you need to have access to some completely exceptional equipment. Without those angled cuts, you could make zipshape furniture in hundreds of thousands of fab shops, all over the world. Zipshape would become accessible to the maker community, and could become a more impactful and useful method as a result.

Step 4: The Proposed New Zipshape Geometry

My proposed solution is simple: Instead of angled cuts, make them stepped. Using the zipshape algorithms exactly as they are but transforming the cuts into steps of consistent height but variable width should make it possible to manufacture zipshape parts on a run-of-the-mill CNC, or, with an additive process of stacking, even a 2-axis machine like a laser cutter or waterjet.

Step 5: Proving the Concept / First (failed) Attempt

Because it's much easier to prove the concept on a lasercutter than a full CNC, the proof of concept is a 2D representation of the final method, which would be extruded and cut on a CNC. The critical pieces is that I'm cutting only orthogonal lines in the 2D proof of concept, which means they could be cut in principle into the thickness of a piece of MDF using profile or pocket cuts and a straight bit.

My first tests were in 1/4" plywood. I used the grasshopper definition I had already to output the zipshape geometry from a curve I made for testing. I tried a couple of backing thicknesses and tooth geometries, but quickly hit a snag. Although all of the stepped teeth could fit into place, the lack of angled cuts meant there was no way for the teeth to actually get into position. Any angle at all while pushing the two halves together meant a tooth would be too wide for its fitted gap, which meant that assembly went fine through straight-ish sections, but curved sections caused everything to break apart. This is a little hard to describe. I've circled the problem and annotated it in the images above.

Step 6: Concept: Proved!

To solve the issue of assembly for square-cut teeth, I modified the original idea by adding a small flexure joint into the center of each tooth. Essentially this means that each tooth has a gap in its own center, which means the stepped teeth can flex just enough to get into position, but not so much that the overall structure is compromised.

I tested the additional flexure joint on one side and on both sides of the zipshape curve. Both went together easily, and both held the designed curve perfectly. The subtraction of material does make the overall assembly feel a little less rigid, so the best solution appears to be adding flexure to just one side, not both. The gaps that are produced may be fillable, also. I will be continuing research next time to see if the benefits of flexure can be maintained with a more tightly-fitting geometry.

The newly proposed zipshape-shape has another significant advantage over prior zipshape patterns: because the teeth are stepped and not sloped, they don't slide out of place when they're next to each other! The photo examples here are assembled with NO GLUE -- the curve is held comfortably and the two pieces stay together with their own friction, even when I pick up the curved stick by one end and hold it up for a photo. The internal sliding forces are perpendicular to the tooth surfaces, so they don't want to come apart.

Step 7: Summary & Curve Fidelity

So for this time I stuck to my familiar tools (grasshopper, rhino, lasercutter) and verified that my idea is conceptually feasible. Out of curiosity, I wanted to take a look at the various tests together, and compare them to the original curve I drew in Rhino. I expected the curve to match pretty well, given prior experience with zipshape. The test revealed a pretty good, though not perfect fit. To get perfect curve fidelity, I'll need to tighten the geometry somewhat or account for a "springback" factor, both of which which would be subject to the particular materials and machines used in fabrication. I'll save that for the next installment.



    • Creative Misuse Contest

      Creative Misuse Contest
    • Water Contest

      Water Contest
    • Clocks Contest

      Clocks Contest

    40 Discussions

    Call me obtuse, but why not simply purchase layers of veneer instead and then form the layers into whatever shape you desire? I would expect full laminated sheets would be stronger, albeit slightly less flexible. However, the total time to cut these shapes with a CNC must be enormous!

    A large bending (steam?) pot/oven and a large peg table would work for most of these designs, except those curving into mobius strips....

    The cost of that CNC device (while quite lovely, very unique, and yes I want one!) could pay for lots of materials and a few large hot tubs and a deck to put them on.

    4 replies

    Full laminated sheets are definitely stronger -- that's not the benefit here. The big advantage here is not strength or cut time, but one-off production: because no molds need to be built for a specific design, designs can be prototyped at much lower cost and mass customization becomes possible. A wooden chair can be sized and shaped exactly for you, for example.

    The value of mass customization is debatable, but it's certainly interesting and I think has great potential to remake the world of manufacturing.

    Because those of us with 5 axis cnc machines need something interesting to make, and we want to make it easier.

    Thanks for the comment! I have used bent laminations for other projects; I actually published one in an instructable:

    Bent laminations are beautiful and, I suspect you're right, stronger than a method like this. But they require solid-surface molds which are expensive and time-consuming to make. A pegboard is a good solution for eliminating the mold, but then you can't do wider sheets and your bend radius is pretty limited.

    I'm interested in finding ways to employ CNC's for more than they are obviously useful for. I'm not imagining a paradigm of "what's the cheapest way to make a particular curved piece" so much as "what are some exciting ways we can push the limits of this new technology?" In the case of this project, I'm proposing that we replace some of the labor ordinarily involved in bent furniture parts with precision made available by the CNC. The most obvious benefits being mold-free construction and easy parametric variation: it's just as easy to make every chair seat fit an individual person as it is to make them all the same.

    In the design stage, I'll try to design something that would be hard to make by other means :)

    Even if you can't make a version of zipshape which is simple enough and cheap enough for rapid prototyping or rapid manufacturing, it's still a valuable concept...

    In particular, zipshape is probably cheaper and faster than the conventional method for creation of solid-surface molds for bent laminations! :)

    Not only is a bent lamination likely to be stronger, but if you make many pieces of furniture/whatever using a single mold, then there will be less wasted materials, per piece, on average.

    What a great instructable! What an excellent way to (No pun intended!) bend wood to your will! I need to build a Recumbent Bike Seat. The whole Plywood and Hinges 'Thing' holds no appeal, THIS changes everything. Now I need to build my own laser cutter! Thank you for this design!

    Great stuff - I wonder if all the breakages will erase themselves when you extrude the form into a 3rd dimension. You must be getting twisting forces in that tight point at the bottom of the tooth.

    Very cool ... thank you for sharing your research !

    This is amazing. I don't think I understood half of what you said (math is not one of my strengths) but I still find it fascinating

    Isn't there something you can copy from the human body to make this work - like a spine-type of form with a cable inside to tighten or loosen to get the form you want? Like ToolboxGuy I too may be obtuse but it won't stop me from following you to see where you go with this, I just think there must be an easier way - when your pieces fit together there is no play because they lock where they lock - but what if the 'locks' were somewhere else? On the sides in the form of wings that can be adjusted?

    I am probably the only person that has actually gotten a chance to play with a CNC for two weeks (at Haystack) and was left cold by what came out - and by what else I have seen. Perfection leaves my eyes no where to rest, I feel like the person who can see the emperor has no clothes. I just don't like what I am seeing being made - yet. It is too static, for now anyway.

    This 'ible is wonderfully simple to understand for such a complex project, thank you for sharing.

    This is brilliant. Pseaton, I salute you. Not only for your creative solution, but for your commitment to making a revolutionary fabrication method available to a wider audience, using more commonly available tools.

    As a builder without any computer-controlled tools, I wonder if Zipshape fabrication could be accomplished using more traditional methods...

    Given a well-tuned table-saw (likely capable of accuracy to ±0.2 mm), do you think it would be possible to slice the requisite 'steps' into a sheet of material? What tolerances are required for your current level of success? What tolerances are you planning to use for your next iteration in attempting to improve curve fidelity?

    An unrelated question, have you found that gluing your curves improves their fidelity? It appears to me that the addition of the flexure joint is what's allowing your shape to relax. Do you think that is accurate?

    You might consider filling the flexure joint with a a block (or, in the case of a wider panel, a spline) after fitting the halves together. It would make the entire construction tighter, and perhaps truer? If you used the right wood glue to fit the spline, it will swell the surrounding material (think mortise and tenon joinery) and create a tighter joint than could otherwise be achieved.

    I'd be most interested to hear your thoughts. I (and my circular saw) might just have to try for a bent-wood chair; no steam-box required!

    I'm trying to figure out a simpler way to get the same effect - hopefully using a 3DOF router table setup. I propose:

    You have two boards - I'm calling them the leader and the follower. The follower has V-bit grooves cut over the whole surface - each groove is the same depth and they are adjacenct so there are no flats.

    The leader is cut with a variety of V bits. The goal is to cut the board so that each face (hypotenuse) is the same length, but the angle is controlled by the bit. If the follower is cut with a 90° bit and the master is cut with a 100° bit the assembly will curve at a constant rate. If you use two bits (cut 5 rows at 90° and then 5 rows at 100°) you will have a flat that transitions to a curve. If you "dither" the bits (rows 1,2,4,7 cut with bit 1 and rows 3,5,6,8,9... cut with bit 2 you may get a more continuous transition.

    The trick here is that depth of the cut on the leader board needs to vary so that faces retain the same length. After cutting the teeth on the leader board, you want to flip it over and cut the backside so that the board is approximately uniformly thick.

    The cut plan will be more difficult to compute but I think this solution will get you most of the effect with a lot less expensive hardware.

    3 replies

    I think we're after the same thing (3-axis process)... one problem you might hit is that in cutting the leader board to different depths with different bits so as to maintain the length of the sides of each cut, you would also be fixing the *spacing* of the cuts at the same time (cut deeper, and that groove is wider). The reason the original zipshape process has its extremely slight variations in angle because of this problem -- the spacing of the teeth is what fixes the curve (not the tooth angles), and then the teeth must have different angles in order to fit together in the same space.

    That said, I'd be excited to be proved wrong by a test! Would you be willing/able to run a test like this? I'd be happy to help setup the geometry in any way that would be helpful.... PM me if it sounds like something you might be interested in trying out!

    Good point! I haven't spent quite enough time on the CNC to work with tool paths as complex as the ones you describe. Would there be ridges left over in one direction or the other, depending on the direction of the passes? Or would there be a sort of micro-waffle pattern if you went both ways?

    I'm curious to learn more about the "Keller cuts" you mention, but a moment of googling doesn't show much. Is there something I could read to learn more?

    Thanks for the insightful comment!

    Yes there are ridges but as you tighten up the tool path you can achieve as close as a 16 or 32 RMS using moldmaker's machining practices. Typically you have your tool travel move in the direction of the longest straight line or in the case of a round object-around the direction of the largest radius so in your case you would be moving perpendicular to the direction you want to bend.

    "Keller cutting" is a general term referring to the use of a ball nose or other tool with a large corner radius to achieve a complex plane. The reason you would use it in this case is to reduce the number of tools needed. I believe a newer Machinist's Handbook talks about Keller Cutting.

    Looks interesting! Will be following you.

    Any reason you chose 2 step heights? Why not more, it would represent the original curves more closely.

    Will the depth to width ratio of the deep flexture joint you added in each tooth in one side be a problem/slow down production on a 2.5 axis CNC bot?


    1 reply

    Good question! You're right at the edge of my understanding of the geometry here... I'll be more immersed in it next time and may be able to provide a better answer but at the moment I'm a little unsure myself whether more steps would necessarily reduce the gaps or not. They'd certainly be closer to the original tooth shape, but honestly I'm not 100% sure what I've done here by adding the flexure... I'll try to figure it out geometrically and include that in the next installment, which is more about the geometry anyways!

    As for the depth of the flexure, it may well slow down production a bit. As I tried it it's about 3/16" thick and 3/4" deep, though the material I'd be using wouldn't be quite that thick most likely. Still, it's probably 3 passes with a 1/8" flat end mill, and that's definitely time. Not as much time as making molds for a single piece, though!

    Can't wait for the next installment, truly awesome stuff you guys are doing.
    I made an account just so I could vote for this Instructable.

    1 reply

    Thank you -- that's awesome! I'll get installment two underway soon... it may take me a few weeks to pull it all together, but stick with me and I hope to make it worthwhile!