Monday, June 23, 2014

New and Improved 3D Printed Clock

Third time's the charm! The third version of my 3d-printed clock is working great:

The biggest change was the winding system. It's a Huygens endless-chain maintaining power, a nifty idea from 300 years ago. The main weight and a tensioning weight both hang on pulleys, and the chain is a loop that goes over the great wheel (in my case, the hour wheel) and also a wheel with a ratchet. You can wind it up without affecting the driving force at all. It also simplifies things by putting the ratchet-and-click on its own wheel.

I've also made some nice-looking weights. The main weight, about 4 kg, is a 2x10 inch piece of stainless steel bar stock, the kind of thing you'd use if you were machining something on a metal lathe. They only machining I did, though, was to drill and tap some holes for a hook. And I'll have you know, I only broke 1 tap! It's still there, in the bottom of the main weight. The tensioning weight is the same idea, but exactly half scale.

The chain I'm using is the same sort of thing that's used for dog tags, #6 ball chain. You can make it into loops with a special connector piece. I've tried running the clock with that connector piece running through the pulleys, and it works, but we can do better. With a needle nosed pliers and some patience, you can form the chain into an endless loop, with one of the balls only slightly mangled. There is also a $100 specialized tool you can get to do a better job. I've made a 3d-printed version of this tool for much less, and it works fine - I used it to form the loop you see in the video.

Assembling this one was a bit difficult. On the previous version, the bearings fit loosely, so this time I added 0.2 mm of interference to the fit. That was way too much, though, so I needed to do a lot of sanding. This made the shafts not quite true, which gave the gears a "preferred" orientation, so I needed to sand even more so the shafts fit loosely. Also, the frame flexes considerably under the total 4.5 kg weight. I used some wire to brace it, but even still, the bearings would work themselves loose until I superglued them in place. The motion work was also a bit fragile, and I ended up gluing together the parts that are supposed to slide when setting the clock. (You can still set it by removing the pendulum and letting it run freely.) After all that gluing, though, it's been holding together fine (knock on wood).

Next version I make, I want to do away with the ball bearings entirely. (See my previous post for some thoughts about that.) The goal is, send a model in, get a working clock out, with almost no assembly.


I've been experimenting with 3D printed mechanical clocks, and I really want to print one that is fully assembled. Just take it out of the machine, dust it off, hang some weights, and have it work. But to make an efficient clock, you need to minimize friction - how hard is this? To find out, I've measured the friction of a plain bearing made in three of the materials that Shapeways offers.

Here is my experimental setup:

The horizontal beam is attached to a 5 mm shaft, and it can pivot by sliding in the holes on the upper and lower arms. By hanging a large load weight from the lower arm, measuring how much weight at what distance will make the beam move, I can estimate the friction in the bearing.

The results: Alumide was best with 0.14, followed by Strong and Flexible at 0.17, and Frosted Detail at 0.30. With a drop of oil, the Alumide got 0.07, the Strong and Flexible got 0.11, and the Frosted Detail got 0.20. I tried loads as high as 1kg and as low as 100g, and in all cases it was linear.

The alumide showed an interesting effect - it started out at a considerably higher friction, but after spinning it for a minute or so under a 1kg load, the friction was much reduced. I suspect that's because it starts with a rough surface, but after a short time, it polishes itself, and you have a somewhat smooth aluminum bearing surface.

Both the Alumide and the Strong and Flexible had a "textured" feel to their motion. Especially after the oil, the alumide was a little inconsistent in its motion - a weight would make it go a short distance and stop, but if you pushed it past that it would move further. The Frosted Detail, especially under higher loads, had a grabby feeling to it - similar to when a sliding a wet finger makes a squeaking noise, but lower frequency. The oil didn't help this much.

The oil I used was "Liquid Bearings" synthetic clock oil. I applied one drop to each of the 4 bearing surfaces, and waited an hour or so. The Strong and Flexible and Alumide plastics are both very porous, though, so I'm curious how long the oil will last. I'd also like to try other lubricants, such as silicone grease or graphite powder.

Version 1 of my clock used printed bearings, but it wouldn't run due to gear clearance issues. Version 2 and 3 both used ball bearings. The ball bearings work well, and probably have a much lower coefficient of friction. But they're expensive, annoying to assemble, and don't tolerate any misalignment. For heavily loaded joints, the ball bearings probably win, but in lightly loaded joints, the constant drag of the grease may be more than the linear drag of a plain bearing. Anyway, my calculations (and some experiments) show it should be feasible to make a clock with only plain 3d-printed bearings, so I'm gong to try that next.

My first 3D-printed clock

For the past few months I've been working on a 3d-printed clock.  I want to challenge myself to make something functional and even reasonably accurate, within the constraints of 3d printing.  Here's the first working prototype.

 This video is actually a few weeks old - stay tuned for the second prototype, and some experiments.

The first version had a few issues.  The pallets came very close to one of the gears, and would rub unless I ran the gear slightly out of its bearing.  The ball chain pulley wasn't quite as good as I'd hoped, and it needed a very large tensioning weight to avoid slippage.  And there was no way to wind the clock without running it backwards or slipping the chain by hand.  The next version fixes all those problems.