Crazy Clock

After a couple of iterations, I believe I have a pair of gears designed that work.

Printing these is fairly difficult. As I noted originally, there's no way an FDM printer would be able to get these right. With SLA, the issue becomes how you support them. You can't tilt the gears at all, because doing so will result in the need to support the tips of the gears on one side. The problem with that is that the gear points are going to be smaller than the touchpoints. This means we have to print the gears flat. We could print them flat on the build plate with no raft at all, but then removing them without damaging the teeth is problematic. I've printed a few with carefully placed supports, and that works, but even removing supports that are kept behind the inner plane of the gearing, it's still possible to break the teeth off if you're not extremely careful.

I'm going to label these two gears of interest the "inner" and "hour hand" gear. The hour hand gear is the one with the stem riser that winds up accepting the hour hand. The inner gear meshes on its outside edge with the minute hand shaft and on the small raised inside nested gear with the hour hand's outside edge gearing.

The outside of the inner gear is the same for both the 12 and 24 hour models. There are 45 teeth. Where things begin to differ is on the inside gear of the inner gear. For the 12 hour model, there are 12 teeth that mesh with 48 teeth on the hour hand's gear. For the 24 hour model, there are 8 teeth on the inner gear that mesh with 64 teeth on the hour hand's gear. If you do the math, you can readily see that the 24 hour gear ratio is twice that of the 12 hour model (12/48 = .25, 8/64 = .125).

It doesn't seem like there's any optimizations to this that we can conveniently make. If we attempt to reduce the number of points on the inner gear we have to make its diameter smaller, and it's already got no room to do so given that it has to sit on a spar in the chassis that's a certain diameter itself.

Well, I tried to ask nicely, but Primex told me to pound sand.

So what is the difference between a 12 hour and a 24 hour clock movement anyway? Well, clearly the gear ratio between the minute and second hand is twice as slow. But what does that physically mean?

Here's a 12 hour movement on the left, 24 hour on the right. Notice two of the gears are grey on the 24 hour movement. Those are the two gears that transmit the minute hand movement to the hour hand. Cutting to the chase, it turns out that nothing else is different, and simply moving the two grey gears over to replace their counterparts on the 12 hour movement work just fine.

Well, can we 3D print those two grey gears?

First, if I am even going to attempt it, let me say at the outset that I am glad I have a Form 3. I have serious doubts that any FDM printer would have any sort of shot at it. SLA may not even be good enough. I may have to get these done via SLS at Shapeways.

But first thing's first. I have to model them in Fusion 360. I've never attempted anything this critical before.

Time to get out the calipers...

Primex/klockit seems to have discontinued the Q-80 24 hour movements. I still have a half flat of them, but when they run out, there will only be 12 hour movements available. Sorry.

For the life of me, I don't know why I've been using such a huge inductor all this time.

I got a rev 2.6.1 board fab'd with an 0805 footprint for the inductor and tested it out, and it works just the same as the 2.6 version.

My inductor choice was a Taiyo Yuden LBR2012T100K 10 µH 360 mΩ inductor. As before, the boost converter had a 10 µF input cap and a 22 µF output cap. The rest of the circuit was the same (based on the ATTiny44A).

EDIT: Ah. Now I see. The LBC3225T100KR 1210 inductor I have been using has about half the resistance and a reel of them costs 25% less.

Well, the XC9142 was on a par with the TI chips that I tried - and about an order of magnitude higher than the XC9140. I've reached out to Torex to ask them about the future availability of the XC9140, but I really don't see any parts out there that are any better for this application.

I've just had a pile of boards manufactured with QFN ATTiny45s and with the XC9140 boost converter, (they're not yet arrived, though), but also ordered some prototype boards for an ATTiny44A variant.

After a little bit of hacking about, it works just fine. It takes maybe 10% more power, but that's really a difference between ~18 µA and ~22 µA - not really outside the margin of error anyway.

The biggest difference is in the calibration code. The feature pin mapping is (as you'd expect) quite different on the 44 versus the 45, so the pin I was using before changed from being OC0A to OC1A. So the calibration code for the Tiny44 has to use Timer 1 instead of Timer 0. That's not really significant, since the calibration code doesn't run from a battery (or even installed in a movement).

The only changes were in calibrate.c and base.c, and all of it got handled by using the AVR processor type macros, so as long as you set the mmcu directive on the compiler properly, you'll get functional code.

Meanwhile, I've ordered some XC9142 chips, which appear to be an updated version of the XC9140. They at least appear to be somewhat more plentiful on DigiKey, so they may be the way to go. I'm going to try swapping some of them out on my test boards to see what the impact is on power consumption.

Hilarity ensues.

The ATTiny44A QFN variant, it turns out, is cheaper than the ATTiny45. Go figure. Is it volume or something? Who the heck knows?

I'm going to try a board variant with that to see if it works. It's silly though - I only need two pins, so most of the rest will be NC.

Huh. Turns out the ATTiny45-20MU is around 30% cheaper than the ATTiny45-20XUR, depending on how many you buy. Even buying them in a tube doesn't save you the kind of scratch that the QFN package does. Even if you were to buy them a reel at a time, the QFN package is 20% cheaper (of course, then you're buying 5 or 6 thousand of them).

Well, it's all the same to Bob, of course, so I'm going to try a board with the QFN part and see how that works out.

On a whim, I built a board with a 10 µH inductor instead of 4.7 µH, and a 22 µF output filter cap instead of 10 µF. Now the draw with the calibrate sketch is around 24 µA! Ship it!