Designing A Self-replicating Milling Machine

For his senior design project at Swarthmore College, [Julian] decided to build a metalworking equivalent to the RepRap. [Julian]’s final project is a self-replicating milling machine, and hopefully giving some serious metalworking power to all the makers with CNC routers and RepRaps out there.

At first glance, [Julian]’s mill doesn’t look like something you would find in a machine shop. The machine is built around a tetrahedral machine tool frame, giving the machine an amazing amount of stiffness with the added bonus of a degree of self-alignment. The spindle and motor are off-the-shelf units, but the entire bed assembly is made by [Julian] himself.

Right now, [Julian] still considers his project a very early prototype; there’s still a bit of chatter issues he’s working out, and the cost of the finished machine – about $1200, not including many hours of fine tuning – means it isn’t as competitive as other options. Still, [Julian] made a mill from scratch, and that’s nothing to scoff at.

14 thoughts on “Designing A Self-replicating Milling Machine

  1. From some pricing estimates I have found about 1200 not including tooling to be the bare bones price to get sub .003″ tolerances in tool steel. That leaves a small work area, low maximum bed weight, and slow speed.

    Ah, well budget the money and you get budgeted results back.

      1. @Nic

        It’s a reference to an obnoxious piece of machinist trivia “A Lathe is the only tool that can make itself”.

        It’s utter bullshit. You would need a 3rd axis to hope to build a lathe bed using a lathe, unless you scraped one out from a raw casting.

        A milling machine with a 3 jaw chuck tool for the spindle is just as capable, if not more capable, of making itself.

        Of course, when used machinery is selling for less than it’s melt value in iron on craigslist, the entire premise is pretty stupid. Using a Lathe or Mill to build a Lathe or Mill is as profitable as machining your own fasteners.

  2. I love to see any homemade CNC machine. But (surprise!) I have a few issues:

    1) “Equivalent to the RepRap”. RepRap is additive, milling is subtractive; with all the differences in materials and capabilities that entails. Not a valid comparison, in my opinion.

    2) “Self-replication” is a term thrown around a lot. But in reality, it can only make a tiny fraction of its own parts. And this isn’t even an CNC mill. My $60 manual drill press can make a few of its own parts, can I call it “self-replicating” too? Or would I have to heavily qualify that statement, as Julian did? Seriously, if a term must be followed by significant qualification to be true, the term isn’t appropriate to begin with; drop it.

    3) The goal of this senior design project was to create “…a bench-scale, inexpensive milling machine which is cost- and performance-competitive with existing alternatives…” We know in the end that it didn’t work out so well. That’s ok, but for a senior project (or anything, really) the goal should be believed at the start to be at least *possible*. It seems that Julian realized quite early on that it wasn’t, demonstrated by his choice to cannibalize the spindle, motor, controller, and Z-axis from an existing commercial mill; acknowledging that it was the most economical option. Hopefully his professor overlooked this in light of his otherwise excellent project, report and presentation, but I know some that wouldn’t.

  3. Hi everyone,

    Thanks all for your comments and interest! I’d like to briefly respond to some of the (really good) points that you’ve raised:

    Mark: Thanks! I decided not to weld the frame a) because of the heat distortion that you mention, as well as the residual stresses that welding leaves, b) because in the interests of making the machine “self-replicating”, I didn’t want to require that the user have access to a welder, and c) because I don’t trust my welding skills enough not to screw it up.

    1) Absolutely an important distinction, especially considering the (much) lower feature complexity that the milling process allows relative to most additive processes. I usually describe the machine as a “subtractive manufacturing analogue to the RepRap” – it got lost in translation somehow, I guess.

    2) I agree that the term “self-replicating” is questionable, especially for my machine (and especially in its current state). The definition that I tried to design to is that used by the RepRap – a machine which can make all non-standard parts required for its construction (my wording, not theirs). Although a Bridgeport can make many of its non-standard component parts (i.e. not the leadscrews, bearings, etc.), it would be extremely difficult if not impossible to make large modules such as the base or the knee – the Bridgeport has neither the work volume nor the machining capacity to do this. My machine, on the other hand, *theoretically* can recreate almost all of its components, with a few notable exceptions (the entire Z-axis and spindle assembly – I’ll explain this more below – the linear bearings, a few frame components that need redesign to fit within the work volume of the machine). However, I definitely see your point about the inaccuracy of the “self-replicating” moniker – I’d be curious to hear others’ suggestions for better terms.

    Also – no argument, my machine’s DEFINITELY not self-replicating yet, in any sense of the word. But it’s trying :-)

    3) I should mention two other constraints that were constantly acting in the background: 1) the need to make a functioning machine that could be used by future students and justify the Department’s expenditure, and 2) the limited time I had to work on the project. It was primarily these two issues that lead me to purchase not only the Z-axis/spindle assembly (which I freely admit still tastes like cheating to me) but also the linear bearing system. In their current incarnation, these systems are expensive, specialized, and are at the top of my list for future design modifications. Hopefully, either I or other builders will be able to pursue this in the future – I know some people have worked on this over at MIT’s Machines That Make group ( In the meantime, though, the machine is complete, and sufficiently functional for future students to mess around with.

    Like you suggest, however, this project does raise serious questions about whether or not making a self-replicating, precision subtractive machine tool in a cost-effective manner is even feasible. We’ve been making this type of tool for a long time, and have a solid design that hasn’t really changed since the 1900’s. Unfortunately, this design relies on manufacturing highly specific parts at large scale (i.e. so that casting and grinding operations can be made cost-efficient), and with large machines. Although I’m not convinced that a machine like mine is impossible to design as far as self-rep, with the glut of high-quality tools being basically given away on Craigslist like onemachinist points out, it’s going to be really, really hard to do it in a cost-effective way.

    Anyway, again – thank you all for your comments! Please get in touch if you have further questions/want to discuss the machine further.

    1. Julian writes:
      “The definition that I tried to design to is that used by the RepRap – a machine which can make all non-standard parts required for its construction (my wording, not theirs).”

      I took this to an extreme in the 1997 self-reproducing four axis mill “Frankenmill”: the entire machine keys on *one* nonstandard part. It is a 4 x 6 x 3/8 inch cold rolled steel drill jig with 15 (3×5) holes that accept 1/16 inch pins. The spacing of the legs of the cross vise is 1 inch and the legs are 3/8 wide. The holes are 1 inch apart. The jig is used to drill 8 holes in the cross vise and 18 holes in the drill press base to mount the cross vise in many positions allowing expanded reach. The plate was part of a slide on the cross vise that let the four jaw chuck be used for circular milling with a vertical instead of horizontal axis. I used the collet indexer to mill 120 teeth in the chuck for indexing that way, so hole patterns of 120, 60, 40, 30, etc. were possible, notably 5 and 6 holes for geodesic dome hubs.

      All the other parts, tools, and tool bits are standard. The reproduction happens economically; I did a job on it–I replicated a Bakelite dental casting articulator frame ball rail, a condyle joint emulator, in 6061 aluminum, at a decent price, earning money towards replication, but the main point was that with enough work and time, a working machine could be sold for the cost of the pairs of parts to make two machines. “And their friends, too, and then we’ll all be using the same shampoo”, to quote a somewhat relevant TV advertisement.


  4. Julian-

    Excellent work for a senior design project as an undergraduate. It may not be the ultimate form of a reproducible milling machine, but it definitely shows thinking outside the usual knee-mill/Bridgeport box.

  5. In 1997 I built two self-replicating four axis milling machines in cast iron and steel and sold on in the Want Ad for $300. I still have the sales receipt. The cost of two drill presses, two cross vises, two collet indexers, and two four jaw chucks, all from Harbor Frieght, was about that $300, proving the self-replicating part.
    “Kinetic Self-Replicating Machines” mentioned this feat on page 40.
    I am working on self-replicating a Smithy Super Shop by adding a centerless grinder to grind all the way and lift tubes. That’s the only component of itself that it can’t make “trivially” (with expertise and time).
    I’d love to participate somehow in this discussion or in Julian’s development.

    Doug Goncz
    Replikon Research
    Seven Corners, VA 22044-0394
    Alexandria, VA 22314-3659

    P.S. I left MIT in 1979 to build a self-replicating home. This would be the core of that home. The idea was to either rehouse humanity or colonize a terraformed planet.

  6. Above, Julian writes:

    “…a few frame components that need redesign to fit within the work volume of the machine).”

    I have thought long and hard about the work envelope problem. Consider the woodworking jointer; it surfaces a board flat and can make it square to an adjacent face, making a rectangular beam precisely. Consider the centerless grinder: round stock feeds in, and precise round bar feeds out. These are *open* work envelopes.

    What I describe above is the addition of an open work envelope to the Super Shop to make the only major frame piece that with time and expertise could not otherwise be made, the bed rails and table lift columns.

    In general, a machine tool with multiple configurations having both closed and open work envelopes is more likely to achieve symbiotic auxilioproductive self reproduction in a sea of metal bar stock and trained human operators than one missing either the open or the closed envelopes. Alternatively a workcell specializing each of those configurations has good potential for 100% matter closure and self reproduction by an operator with access to sufficient energy to drive it.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.