A Six-year Adventure Into The World Of CNC Fabrication

Hackaday doesn’t always get the entire back story of a build. The usual assumption is that someone decided to build something, and with just a little bit of effort the project makes it into the Hackaday tip line. This doesn’t do justice to the builder, with skills honed after years of practice and experience. A 200-word summary is deceiving, and makes everything look almost too easy. [Michal] decided to buck that trend and sent in his half-decade long adventure of becoming one of the best micro-scale machinists we’ve ever seen.

In 2006, with years of robots made out of hot glue and cardboard behind him, and the quality of 3D printers not up to his exacting specifications, [Michal] snapped. He sunk the better part of $3000 into a Roland MDX-15 desktop mill. After several months of futzing about with acrylic sheet, [Michal] came across the wonderful machining properties of modeling board.

Determined to do something useful with this modeling board, [Michal] started looking into resin casting. Casting in resin is a common technique in the artist and model maker communities to mass produce small plastic parts. After getting his hands on eight liters of polyurethane resin, [Michal] made a useful part guiding the direction his skill set would grow in the coming years.

After years of experimenting with techniques, materials, and mediums, [Michal] eventually honed his craft and was able to finally start building real robots. These projects were a far cry from the cardboard and milk jug contraptions made earlier in his career. [Michal] was now producing incredibly precise gear assemblies with accuracies within 0.002 mm.

You may remember [Michal] from his robot with pivoting wheels we showcased last week. He got a lot of email from people wanting to know how to start delving into his unique blend of artistry, engineering, and craftsmanship. The good news is you can now learn from his mistakes, so a planetary gearbox shouldn’t take more than a few months to finish.

51 thoughts on “A Six-year Adventure Into The World Of CNC Fabrication

  1. As a (good) 3D printer owner I’m amazed by the detail and quality of his parts. I can get high quality prints, but these resin casted parts are way out of my league. I think to get the same parts, but usable from a 3D printer like the Ultimaker, you need to scale every part up by 300%, maybe even more.

      1. Hey Alex, have a look at this: http://imgur.com/YkORr

        I downloaded your “proof” and zoomed in as far as I could. See that little yellow line in the upper right corner? Going by his 1mm scale bar, that line is 20 microns long (10 times the claimed accuracy!)

        What exactly do you think that picture proves?
        I think you are the one with nothing worthwhile to add here.

    1. if someone can build a microtome out of lego which frankly arent the most acurate mechanic pieces out there and still hit a 250 micron acuracy i have no problem believing that a desktop mill can achieve 2 microns with a little bit of fiddling.an industrial high precision mill will have an accuracy in the sub micron range. so i really dont see the trouble here

      1. Actually barsooma is correct, claiming two micron accuracy is absolute bullshit.

        I’m not trying to start any fights, but I feel like maintaining factual accuracy in the comments.

        To begin, there is a difference between repeatability and accuracy. Secondly, a commercial mill capable of two micron accuracy (or even repeatability) would cost upwards of 200K$ and would require full temperature control.

        As an example, the aluminium table many of these mills use will expand upwards of 30 microns given a 10c temperature change. The bearing runout under load will be well over two microns. In fact, everything from the tolerances on the (rolled) ballscrews to the play in the gibs will be more than two microns.

        If you walked into a machine shop and asked them to mill you something to two micron accuracy they would laugh in your face. A mill is the wrong tool for that kind of job.

        Even 0.01mm is impossible on a mill <50K$.

      2. Listen to Ryan, kiddos.

        The confusion may be that of resolution vs accuracy. My home CNC mill, a Taig conversion, is addressable to 1/40,000″ (0.00065mm) – the wavelength of red light. In reality, working accuracy is somewhere between 0.0002-0.003″, depending on who is cutting what, and how carefully planned and executed the operation is.

        When I first started reading up on benchtop milling, old timers would consistently advise that chasing specs (and high precision mods like ball-screw conversions) was mostly a game for chumps. They were right on. Know you’re mill’s limitations and how to work with them and you’ll get the job done. Ignore those and the best VMC in work will produce junk.

        Anyway, nice job to the author. There’s a lot of time spent designing, building and casting that we can all appreciate. One question, though, why not cut gears directly out of aluminum and bronze?

      3. @ryan if you read what i’ve wrote previously you noticed i wrote INDUSTRIAL HIGH PRECISION wich mean you wont find this setup in just any mechanic shop. but a multimillion company having a shop to produce whole machines and machine part are not going to cut corner on their own machines/mills/laser cutters etc. if a machine is requires with that amount of accuracy the cost wont matter. to be honest im not a machinist myself but my brother is (working for that same aboved mentioned multimilion company) and i distinctly remember him telling me something about 0.5 micron part tolerance.
        on another note im a lab tech and work everyday with microtome/ultramicrotome those are relatively simple mechanical machine but still i can cut sample as thin as 1 micron with pretty good repeatability and using an ultramicrotome cutting 45nm (yes thats nanometers as in 0.045 microns!!) is no big deal. so sorry tu burst your bubble but you should get your facts straight.

      4. Oh my…

        I’m just talking about how well the (tiny) part conforms to the intended dimensions.

        There’s an obvious difference between having a production setup that always gives you that, in any material, for a piece of any size, and even in extreme temperature swings; and just being able to do that for a 5 mm gearwheel machined out of plastic.

        I have a decent inspection microscope, a micrometer, and a 2 micron dial indicator, and that’s it. It may be that I’m confused or lying to make friends on the Internet, but at this point, my educated guess is that I’m not very far off.

      5. Ken: as for the choice of materials, it’s outlined on the page. Plastic versus metal is a more existential debate of course (let’s just say that plastics are cheaper / simpler to work with at home), but the casting process is a bit more efficient and flexible than directly machining a necessarily much larger workpiece.

      6. @ryan: read the articles I’ve listed above for information about sub-micron resolution and repeatability.

        Re: table flex – aluminum’s coefficient of thermal expansion is roughly 23 microns per meter of material per degree C. To get 30 microns of upward expansion from a 10 degree C change in temperature, you’d need a table 10-15cm thick. Even giving you that amount of change, it’s only relevant if the table’s temperature actually changes that much during the cutting process.

        Re: .01mm accuracy on a mill that costs less than $50k, it’s clear you’re not a machinist and have never tried. .01mm is about .0004″, which is no huge deal. A medium quality ($30-50) dial indicator has increments of .0005″. A set of shop quality gauge blocks (the cheap ones) allows you to stack reference measurements in increments of .0001″, and will be certified to an accuracy of a quarter micron.

        It takes some experience and technique to work plus or minus a tenth (cut in stages, each with a finer tolerance, never try to cut more than 10x the stage tolerance, take it slow, miss entirely rather than overcut, measure like hell every step of the way), but it’s no big deal.

        I regularly work to +/-.0005″ (.0127mm) when I actually care about a part’s dimensions. +/-.005″ (about the thickness of a piece of paper) is my standard moderate tolerance, +/-.015″ is my “don’t give a damn” tolerance, and +/-.1″ is my tolerance for cutting by hand with a hacksaw.

        The general rule of machining is never to try for a tolerance less than 10x your machine’s best. If a mill can’t make a light cut smooth to within a tenth, that mill is a piece of crap.

  2. WOW!

    I forgot I had bookmarked his page a few years ago, I was always impressed with the detail of his work.

    Good to re-read his page, tons of great information and links to good suppliers!

  3. Actually the fine of accuracy is quite obtainable on a mill. And has been for years. It just depends on how good of a mill you have. My grandfather did that kind of accuracy on a mill in his garage. And that was in the late 1970’s.
    There is a gentleman in Illinois that is making brass parts in HO scale that are equally incredible. And he uses a modified Bridgeport mill w/CNC added on.

    1. The article claims accuracy to 0.00008 inches.

      Sorry, I don’t believe that your grandfather was doing that on a manual mill in the 70s.

      Even today its going to be hard to find a shop willing to do that.

      These scales are barely measurable by hand. When/if you decide to specify a tolerance like that, you need to also spec a temperature- if you hold that part in your hand for a few minutes and remeasure, you’re going to get a measurement difference larger than the tolerance.

  4. No kidding, I truely envi him for his machining skills. His gears are beautiful!
    Although the Roland mill’s specs tell:
    software resolution: 0.025 mm/step
    mechanical resolution: 0.00625 mm/step

    So neither the milled parts nor the casted ones (shrinking?) can reach 0.002mm accuracy. But who cares? Even 0.05mm should be enough for plastic gears.

    1. You’re looking at the wrong mill, I think? MDX-540:

      Mechanical resolution: 0.00004″ (0.001mm)

      Software resolution: NC-code mode: 0.00004″ (0.001mm)

      But really, that’s not the point of that page :-)

      1. Uh… You do realize that “mechanical resolution” is an non-relevant parameter? Your positioning accuracy is two orders of magnitude larger (0.1mm, or 0.004″) with is typical for an general purpose benchtop milling machine. Working down to “tenths” (0.0001″) is possible with careful positioning, measuring, locking of axes, and cutting strategy, but the 0.001mm resolutio is essentially a fiction maintained within the controller and only roughly aproximated in practice.

      2. Plus, to be honest, I’m not sure where we’re going with that…

        I am well aware of the meaning of “resolution”, “repeat accuracy”, “repeatability”, etc, and I know what the spec says for my mill. I have a lengthy discussion of these parameters in my CNC machining doc.

        I also know how to use a micrometer, a dial indicator, and a measuring microscope, and I pretty consistently see what results I am getting for my work (which doesn’t involve large parts, hard materials, or significant temperature swings). I used to see +/- 0.05 mm, too, I can tell the difference, and I know that 0.05 mm wasn’t “good enough”.

        I don’t have a particularly good reason to lie, I think – but on the Internet, you never know :-) I am also fairly certain that I’m not just mistaken about what I’m seeing.

        You have every right to doubt that, and if so, I think it’s best to just move on.

        I did take some offense in having the entire write-up treated dismissively and the number called “bullshit”; and I also wanted to object to citing specs of a different mill as a “proof” that I’m full of it; but that’s it.

        For most part, I just wanted to share something interesting and encourage others to try.

      1. Sorry Michal Z. I never wanted to question what you saw!
        I admit, i haven’t read the whole story on your page, just followed the HaD-link to The Roland MDX-15/20 page, read the specs and had a look at some of your pictures. If i understand right, you measured out the influence of the mills stiffness and compensated as good as possible, achieving better results then the mills specs should allow. Thats a good piece of work.
        As you said, on the Internet you never know if people are telling the truth ;-) please don’t take offense.
        Take this discussion as a sign of interest in whats really possible.

        “0.05mm accuracy should be enough for plastic gears” In this case i was thinking about diameters around 10…40mm and modulus of 0,5 or 1 not diameters of 0,4mm!
        .. just curious: Where do you need such small gears?

  5. i have like 50,000,000 BMC brushless hub motors(for ebikes) with stripped planetaries. if i could just get some gears made out of less shitty material.. who wants hub motors? lol i’d gladly get rid of them for the price of shipping.

  6. Somebody just didn’t listen. From the Roland website: “MDX-15/20: X/Y-axis direction 0.002 to 0.197 inches settable in steps of 0.002 inches, Z-axis direction 0.001 inches.”
    Just to be sure I googleconverted it – 0.002 inches are 0.05 mms – which makes perfect sense.

  7. Wow! I poured over the previously linked pages by this guy and was impressed but after reading that he had a kid, up and moved across the Atlantic and managed to continue his weekend hobby really inspires me.

  8. The linked article indicates that he started on a MDX15, but when he really “snapped” was when he bought an MDX540. That mill lists a mechanical resolution of 0.001mm.

  9. His work is still really worth looking at.
    I know I can’t do that kind of work, even in my “best Days” as a model maker.
    My hat off for this beauty of machining skills.

    And who cares what or how he did it, I don’t really care if he used a hammer and chisel on a rock in the back yard, his work is really quit amazing.

  10. It’s fascinating to me to read about how accessible molding in plastics is now. I’m very tempted to continue investigating, as this is something I’d really enjoy trying.

  11. Renshape! I hear that a lot at work! :)

    I admire Michal work! He is the one that got me interested in Roland MDX-540A! :)
    You have to decide between a car or a CNC! hehe easy choice :P

    One question I have for Michal is this:
    Could you post picture of your vacuum pump setup when you do negative molding?
    When I look at this picture
    I wonder how you vacuum pump those parts!? Do you put a cover with a sheet of plastic and clamp it with C-clamp, the put the whole thing inside your vacuum pump? I thought you could of use dowel pimps for alignment.

    Instead of making mold, could you use the rotary axis on the MDX-540 and get your part in a more automated way? When one side is finish, it flip and machine the other half.

    I guest I have to finish the guerrilla guide to CNC!

    you work for google as a security expert?


  12. >You may remember [Michal] from his robot with
    >pivoting wheels

    umm no, I remember Lcmtuf from his white hacker work, Fuzzing browsers to discover multiple 0-days for example. I had no idea he was into robotics :o
    Great stuff.

  13. The more I hear, the more impressed I am. Each step shows more detail in the data and engineering decisions that led him down the path he developed. Once you get the molds done and your design validated, you have a whole range of possibilities that open up for you. I mean, you can strain at knats and argue about how many angels can dance on the head of a needle. But what keeps impressing me is the process that was developed by a man (with a kid-she’s a little cutie but still needs a major time investment) that was mainly self taught. He’s made the mistakes and is posting them out there as a warning to others. And the results are just awesome. I’m just more impressed than before and am left with more questions. Does Blender still play a role in the design process now that he has a “real” CAD program? Were opensource alternatives like FreeCAD considered? Is Blender being used for motion analysis like you would have with Alibre or SolidWorks?

  14. Thank you Michal for this insight into CNC milling.

    I have tried to wrap my head around the thing and this the best writeup about small part milling I have seen so far.

  15. Accuracy – Resolution.

    I am pretty sure the accuracy of my Legos is pretty damn precise, but the resolution leaves a little bit to be desired. I don’t think 8mm resolution will win me any awards…

  16. Oh, and at the author. This is perfect. A friend of mine is interested in getting into this kind of thing thanks to his quality control job he has now. Looking at buying a personal mill himself, and this page will be a great help and inspiration for him.

  17. To Michal – Very cool page. Thank you so much for sharing!

    And now for the rant.

    And so HaD descends once again into the Stage For Pointless Bickering. I thought there was a general agreement that direct negativity should be avoided?

    Anyone think that the article was fairly awesome? Yes? Post!

    Don’t let HaD be taken over by the negativitron! If you disagree with the author there are ways to say it. “bullshit” is not one of them and WILL drive away creative people.

    If you want HaD to be filled only with “I blinked a light on my Rdweeno” posts, that’s the way to get it – Interesting people can be easy to offend and next time they do something cool – They’ll keep it to themselves.

    Anyway – To the interesting people out there – Keep doing cool stuff and POSTING about it! :)

  18. I love the write up, and details. My hobby in DIY Micro Flight engineering just reached a new horizon. I’m really jealous of your skills, and can’t wait to Ride your Coat Tails.

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.