Part 10 [1] of his prototypes series also features incredibly impressive precision. He starts off with a $10 piece of aluminum that took maybe an hour or two to make, and is completely accurate down to nanometers. I worked in a nanotech lab with an electron microscope, and the idea that you could make a mechanical device that could accurately move a few pixels on my screen blew my mind. I thought I was hot shit with all my electrons and magnets... nope.
Dan's flexure is so sensitive that radiative heat from his hand moves the end stage 100x more than mechanical input. You could move it by shining a laser pointer at it. Insane.
The final part [2] is also very neat, and starts with a really remarkable demonstration of stiffness.
I don't understand how it's accurate to a micron - that grinding wheel wears after use. Wouldn't that throw off the accuracy?
I was expecting some sort of laser calibration during cutting but he didn't say anything like that. (i.e. don't make the machine accurate to a micron, just measure accurate to a micron and adjust the cutting real time.)
That's why grinding wheels have such huge diameter (inner grinding is much harder because of that), clamps are typically non-distorting (magnetic) and the force involved is a very small fraction of what happens during normal turning/milling.
In finish grinding after your setup the trick is to remove very little per pass so the workpiece does not warm up which would cause it to expand removing too much material. That's why you use coolant and not so much a cutting fluid (this also helps to flush out the residue from the stone and the material being ground). A single pass will remove a fraction of a micrometer (100 to 500 nm) and you count passes.
Another thing to keep in mind is that a lathe with an accuracy of motion of 1u produces workpieces with a diameter accuracy of 2u.
Doing this repeatably is really hard, and doing it in absolute terms is witchcraft of a higher order.
Usually accuracy of a ground product is not only about the diameter or some other dimension of the workpiece but about flatness of the surface or roundness (runout).
Witchcraft is right. I tried really hard in college in the machine shop to accurately make parts, and failed pretty badly. The machinist who ran the shop would sometimes take pity on me, eyeball my setup, touch the tool bit to the grinding wheel, turn a knob a hair, and turn out perfect parts.
It was maddening!
A 4 jaw lathe chuck was my enemy. No way in hell was I ever able to center a part in it.
I know your misery. I've got a Logan 820 lathe (quick change gear box, does 8-224 threads per inch; change one gear and you can do 4 threads per inch). The ways are worn so to hold a 10 thousands over 5 inches I have to twist the bed. I twist it with a level like this on the carriage:
I didn't get that one, $863 is spendy. I got a Russian copy for $150.
Anyhoo, you put that level on the carriage and then crank the carriage back and forth over the range you want straight, adjust the levelers on the legs.
I can turn mild (cold rolled) steel and hold a 10,000 of an inch for almost 5 inches and then it goes to hell. And the head stock bearing is crap, the originals had some preload but the replacements do not so I have to use a live center in the tailstock to put some load on them or it doesn't work.
(I always find the overall breadth of technical knowledge on HN impressive; seems like there’s always someone who knows an engineering area inside out)
I've been out of the industry for more than 20 years now, a lot has happened in that time though that world does not move as fast as the software world I definitely would not say I know it 'inside out' (and probably never did in the first place). Machinists dedicate their lives to the work, I merely passed through and learned just enough of the ___domain to be able to do my job.
The customers were super interesting, from a lathe that worked gold for watch cases (where the repeat accuracy was probably less important than the degree to which the machine managed to recover the cuttings) to one that turned wheels for harbor cranes and everything in between.
I don't think he was specifically talking about grinding applications per se, but the overall build accuracy of the machine. If you had a perfect grinder or cutter, most machines would have some twist or lack of rigidity or distortion of components that would lead to inaccuracies as the part is turned. Because of the way this machine is built, it is capable of achieving 1 micron accuracy. It most certainly takes extremely careful setups (and accounting for wear) to achieve this accuracy, though.
I did seem to get the impression that he could start at a reference point. He also mentioned zeroing out the set up. It also seems very possible even without calibration that once the apparatus was in operation that subsequent operations would be relatively accurate to each other within a micron. But yes, overall it seemed to me that absolute precision was potentially a whole other can of worms.
Grinding wheels wear from left to right. You dress it to a certain diameter and then particles come off the left part before they come off the right. That way SOME fraction of the width is at the proper diameter.
This is why you never ever grind full width. Of course, once you grind long enough that you wear the whole width, game over.
Touching off on the work piece with a sufficiently rigid setup is accurate to a micron... but there are tricks to it once you get down to that kind of precision.
Very cool. I've been wanting to get into machining for a while (haven't had the time yet) and I was planning to do that by building my own machines (foundry, precision lathe, milling machine, drill press,..etc) with the aid of Dave Gingery's books [0] but I am still not sure about them (they use imperial units, which I loathe and they might be outdated).
Unfortunately the key to this design is you have to force air through it constantly. The weight of the inner part is pushing a miniscule amount of air through the gap constantly, but you need a pump for extended operation. Still, its possible, if you're willing to put down a few grand for it.
It still works without forced air, just not nearly as well. You really need to force the fluid to get the high carrying capacity, or use a higher viscosity fluid so it doesn't get displaced as easily, which is actually how standard ball bearings work. A moving ball bearing has no contact between any metal parts.
Usually there is some kind of 'runout correction' after the mounting. Reverse the process and it's the wheel that's ground, so simply running the newly minted wheel against a stationary one should do it.
at this precision? Very, very carefully. The chuck, the part that holds the workpiece on the shaft, has two runout adjustment screws that can move the workpiece in the x and y relative to the shaft. In order to get micron level precision, you have to adjust those screw fractions of a turn, and the screws have to be on a very tight preload.
This is a very unusual design for a chuck. The overwhelming majority of chucks follow two (or 3-4) designs.
Thank you, djoldman. This video is incredible. I'm now watching all of Gelbart's YouTube videos on my phone. The depth of expertise shared in every one of them is unequaled. I hope he gets a better camera setup soon.
Dan's flexure is so sensitive that radiative heat from his hand moves the end stage 100x more than mechanical input. You could move it by shining a laser pointer at it. Insane.
The final part [2] is also very neat, and starts with a really remarkable demonstration of stiffness.
[1]: https://www.youtube.com/watch?v=PaypcVFPs48
[2]: https://www.youtube.com/watch?v=MtxA20Q-Uss