Saturday, June 25, 2022

Y Size Limit of the MK3S

In an earlier post, I wrote about splitting large gears into parts to print them on a Prusa MK3S. My current project includes a gear which is strictly larger than the official size of the print bed, but by using a trick you can still make it work.

The gear is 281.33mm in the Y direction from tooth tip to tooth tip. The Prusa MK3S print bed is 250x210mm. However, you can go beyond these limits. First, in the PrusaSlicer printer setting, change the bed size to 250x230mm with an origin offset of 10mm in Y, 10mm being half the difference 230 and 210. Now load the model, move it until it is within the printable area and slice it, with skirt turned off. If you just go ahead and print it, the printer may come up against its Y limits. Nothing terrible happens at this point. You can hear a slight "clunk" as it reaches the limit, and the print it truncated, like this:

The trick is now to adjust the Y position of the model so that we minimize the truncation at both top and bottom. I found that for this model, 103.5mm was the best compromise, leading to a very small amount of truncation:

This is probably OK as the very tips of the gear teeth are not doing much.

Note that you have to peel off the pressure release strip as soon as it has printed, otherwise the print overlaps it.

This looks to be about the limit of what is achievable. With a slightly smaller model you could avoid any truncation at all while still exceeding the official limits. I guess 217mm would be OK.

Sunday, June 05, 2022

Swingtime update

In my previous post about the Swingtime clock, I noted that the remontoire occasionally goes mad. The motor keeps running until the motor arm collides with the third wheel, long after the tilt switch should have turned it off. I was using a miniature tilt switch of the sort that has two metal balls inside it (like this). I replaced the component I used at first with a second one and got the same result after a day or so of running the clock. The original design called for a mercury tilt switch (like this) and so I replaced it with this one from Amazon. It's a little big for the compartment in the motor arm, but I found that if you open up the plastic package, the actual switch is only about a third of the size. I mounted it and the capacitor on a small piece of veroboard:

(I know its hard to see. You get the idea and the scale.)

It's too early to say whether this works better. However, I've run it for more than a day with no signs of problems. With the original tilt switch, even before the overrun and crash into the third wheel, I would sometimes see it running for longer than expected, and I have not noticed that happening at all with the new switch.

One problem remains with the clock: it occasionally squeaks. It is always on certain teeth of the escapement, though it does not squeak every time. Some polishing or lubrication should help. To find when it is happening, I've been taking video at 1/8 speed. Here is an example:

Update: after further examination, I think there was more than one source of the squeaking. It got less when I polished the tips of the escapement, and then went away completely after I lubricated the escape wheel arbor. (For now, at least...)

Saturday, May 28, 2022


The Swingtime clock is a design by Clayton Boyer. I previously made his Toucan clock. Like the Toucan, the Swingtime is designed to be made out of wood, and I adapted it for 3D printing. I highly recommend watching this video to see how it works.

It's a large clock and this posed some challenges. The main wheel is slightly too large to print on my Prusa MK3S, and so I split it into two pieces, using the technique I described in a previous post. The hourglass-shaped pendulum itself is about 60cm from one end to the other.  For this, I printed the center and the two ends, and then joined them together with wooden dowels. One end was still too large to fit the print bed, and so I also split it into two pieces. Finally, the frame is again too large. I found that by reducing it to the minimum size needed, roughly the distance from the escape arbor to the main arbor plus a little extra, it will just about fit diagonally on the print bed. Very long straight objects with a low contact area are one of the hardest things to print without warping. I got lucky and there was only the tiniest lifting from the bed at one end. The whole thing is mounted on a stand of 2020 aluminium. I have lots of this lying around from previous printers, and while it's not as elegant as a wooden stand it does the job. One concern I have is that the very top of the frame is unsupported and the escape arbor sags slightly as a result. I think it's OK, but it will need to be watched over time. I don't recall the amount of weight in pendulum bob, except that it was a lot less than the 140g suggested in the original design. I used some BBs, glued into the bob with wood glue so that they don't shift as the pendulum swings; it doesn't the operation if they do, just makes a slight noise.

Here's a few pictures and some video.

The original design uses a daisy wheel for dividing the minute rotation to the hour rotation. I used this in a previous clock, and I don't much like the motion it gives. As I noted in the previous post, there are two variants of the daisy wheel design. Clayton Boyer's design uses what I think is the less good one, as it causes the hour hand to move eccentrically. You can see the tip getting closer and further away from the clock face as it rotates. You can see this in another build of the clock, around the 2:50 mark. The daisy mechanism can also give a non-uniform movement in the sense that its angular speed changes as the minute hand advances. One alternative I considered was to use Ferguson's mechanical paradox, as in the William Strutt epicyclic clock. However, after trying several prototypes, I was still unsatisfied with the smoothness of the motion from it. In the end, I used a very conventional 3:1 and 4:1 gear train between the minutes and the hours. It floats freely on the back of the clock face.

The arbors are 5mm brass, with a 6mm brass tube for mounting the assembly containing the clock face, hands and minute/hour reduction. Each arbor has a small cap. As well as improving the appearance, it stops the gears from gradually sliding forward on the arbors. This was definitely necessary for the third wheel. The pallet rests on has two MR105, chosen because I had some in my parts stores from a previous project.

The clock is driven by a weight attached to an arm behind the minute gear. The arm contains a motor and a tilt switch. When it drops below a certain angle the motor engages and move the arm up a little. I didn't quite believe that this would work until I saw it happen: why doesn't the force of the motor just drive the minute wheel round rather than raising the motor arm? I think it's that there is enough resistance from the rest of the mechanism, or maybe the escapement is locking everything in place for the short time the motor runs. Or maybe I just don't understand physics. I found that the motor sometimes moved itself up then dropped back down again, similar to what you can see in the first few seconds here (same video as before). The comments thread for that video mentions that Clayton has a modification using a detent to stop this. I had considered something similar. An alternative is to add a capacitor to the motor. This charges up while the tilt switch is engaged and then keeps running the motor after the tilt switch turns off. The motor arm then lifts up a bit further until the capacitor has discharged. It has two beneficial effects: the motor arm raises further so it is longer before the motor has to run again; and as the charge on the capacitor decreases, the motor comes gradually to a halt, which seems to make it hold better. I used 4700uF. The tilt switch is a miniature kind, from here, and the motor is this one. The distance the motor travels on activation is very variable, anything from moving the motor arm through about 10 degrees to as much as 45 degrees.

If you look at the end of the video, you can see the remontoire going mad, and driving the motor arm until it collides with the third wheel. At first I though this was because the capacitor was too large, but I now think it was due to the tilt switch getting stuck. The switch contains a small metal ball which joins the contacts, and it's possible that I slightly squished the casing causing the ball to get stuck. It doesn't do this every time, for example the first time I saw it was after a couple of hours, and it clears after a short time. I'm currently trying a replacement tilt switch to see if it helps.

I like this clock a lot. Clayton's designs are elegant and work well. It looks wonderful and the broad slow swing of the pendulum is lovely to watch and listen to. I'm also emboldened to try some large designs that I would previously have rejected.

Saturday, April 30, 2022

Printing gears in parts

Sometimes I need to print objects which are too large to fit on the print bed by splitting them into parts. I've used this in particular for the frame of some clocks such as the Thriecan. I have been reluctant to use this for parts such as gears where the dimensions need to be precise. A project that I am working on calls for a gear too large for me to print, and so I decided to conduct an experiment to see if I can split it and still get a working gear. As with other split parts, I make some 1mm holes in the part and then use metal pins to get the parts into alignment, and glue them with a gel cyanoacrylate. I have been using Loctite UltraGel.

I made a test piece consisting of a gear about 50mm in diameter, with the split in the gap between teeth. This is the least critical place for functioning of the gear as it never comes into contact with any other gear. I did two versions of the split gear, for reasons which are explained in the "aesthetics" section.

The results look like this, with the unsplit gear at the top. First, the right way up, and the upside down, with the the side that was in contact with the print bed facing up. I did no finishing on the gears, as you can see from the stringiness. If you enlarge the pictures, the pins are visible. Normally I would trim them to size so that they are hidden.

Dimensional considerations

The most important factor is whether the size of the gear is correct, defined as similar for the split and unsplit versions. I measured the distance from the end of a tooth to the one diametrical opposite, across two axes at right angles. I chose the axes like this

so as to emphasize any effects from the split. For the unsplit gear, the axes measured 50.75 and 50.82mm; for the first split gear, 50.71 and 50.83mm; and for the second split gear, 50.58 and 50.89mm. The differences are negligible at this size of gear. They are likely to stay constant in the size of the gear, so by the time we get to something where the gear needs to be split to fit on the print bed (roughly 4 times the size here), they will be even less significant.

Mechanical considerations

A second consideration is whether the split will cause weakness in the gear that could affect its operation, either by causing it to deform during use or even to split apart altogether. I don't have any good way of evaluating this. For building clocks, it probably does not matter, as there is not usually a lot of mechanical stress extending across the halves of the gear except near the power source (drive weight, spring, etc.). Even so, the bond is strong. I was not able to pull apart the two halves of the gear except in one spot where I had not applied very much glue.

Aesthetic considerations

The split can look a bit ugly, or at least noticeable. You can see this quite clearly on the middle gear in the second picture. The third gear shows a possible solution to this, but adding a notch to both the spoke with the split and the other spokes, so they all look similar.

Tuesday, March 29, 2022

Clutch prints

I think this post is mostly directed at my future self, as a reminder of what I did. Here goes anyway.

I have roller blinds in my house made by Mechoshade. They are the sort where you use a metal beaded cord to raise and lower it. Inside the mechanism there is a clutch, between the indented wheel which the cord engages with, and the hex-shaped shaft of the blind itself. The clutch for one of the blinds has failed multiple times. After the original failure, I printed a replacement in PLA:

This was in October 2015. I have not tracked how often it has failed since. It's probably two or three times. Usually I print the replacements at 0.2mm layers, 100% infill (if I remember), using PLA. I have tried PETG, which failed immediately. The most recent replacement was a few days ago, and I hope it will hold for a year or two. However, this got me thinking about other materials. Ideally you want something which is strong but not brittle. Strong, because the blind is heavy and I am not always gentle with the cord. Brittle materials tend to fail suddenly after multiple stress cycles, which seems likely to be the case here. PLA has the strength, but is also brittle. PETG didn't work because it is not very strong. A chart on Taulman's web pages gives a good guide to the material characteristics. I decided to look at nylon and two related filaments.

For Nylon itself, I bought a 200g roll from Gizmodorks. I didn't note the exact print settings I used. The results were not great. A first attempt had a poor finish with blobs all over the surface. I think this was at 250. A second attempt with a lower temperature had a better finish but the print delaminated. It is possible that the filament had not been kept well. Nylon is hygroscopic and some sites say that even a few hours exposure to the air will make it damp enough to cause problems. The print surface was blue tape with glue stick. Blue tape on its own was not enough, and when using an unknown filament I prefer not to use PEI sheets so as not to risk damaging them.

Next, I tried Taulman PCTPE. As I understand it, this is a mixture of nylon and a flexible filament similar to TPU. I used it once before, a long time ago, to make a headphone holder and found it to be very strong. For the print settings, I started with one of the flexible filament profiles in PrusaSlicer, and increased the volumetric flow rate to 5mm^3/s. Initial attempts on blue tape warped. Blue tape with glue stick worked. I also started with 240/50 and then reduced it to 220/50 for the successful print. You can hear bubbles popping a higher temperature and see water vapor coming off it. The result is quite flexible, possibly too much for the clutch.

Finally, I tried Taulman 910. This is supposed to be very strong under tension, while still having some flexibility. Some people say it is hard to print with and needs a high temperature and an enclosure. For met it worked fine at 240/50 on blue tape with no need for glue stick. The result is a little more flexible than I expected. It definitely feels strong; I have no way of evaluating this more precisely.

Warped PCTPE:

PCTPE and 910:

(Sorry for the poor pics. My camera was having difficulty focussing.)

I have not fitted the new clutches yet. It's a nuisance to take the blind apart so I'll wait until it fails again. This might not be for a year or two.

Tuesday, March 22, 2022

Favre's Clock 24

Jacques Favre has designed several interesting clocks, the designs for which are available on myminifactory. I built his Clock 24 design. It is weight driven, and uses a Graham escapement. The run time when it is mounted at a reasonable height (5 feet or so) is about 24 hours. This could be increased by using doubling the weight cord through a pulley.

One striking thing about Clock 24 compared to all of the others that I've made is that it is huge. The scape wheel and largest gears are almost twice the diameter of the ones in the first Peterson clock and the frame is corresponding larger. It uses arbors of 5mm and 2mm diameter, compared to 3mm and 1.5mm in the Peterson clock. I'm not sure of the consequences of this for ease of getting it going. I think it probably makes it a little less sensitive to printing tolerances. Does it make it more or less sensitive to friction? The contact area with the arbors is larger (more friction) but bears less force per unit area (less friction). I just don't know. As with previous clocks, I used brass rods for the arbors as it's easy to cut them without power tools.

I found it quite easy to get going after a made a few small adjustments; mostly adding some extra washers to keep some of the gears clear of each other. There are a few variants for the design, the most interesting being in the mechanism which goes between then main gear train (known as the going train) and the winding gear (which carries the weight). This can either be two gears with a simple ratchet, or a more complex design with a spring to act as a power reserve for the going train during winding. I elected to go with the simpler option. The escape wheel also has two variants, one with full depth teeth and one with tapered teeth which may have lower friction. I started with the tapered version, but found that the anchor tended to wobble as it only has a small area of contact with the escape wheel teeth. The non-tapered version worked better.

One interesting design choice is the clutch. In the Peterson clock, this is done by a spring which holds a gear and a spacer in contact. The Clayton Boyer design that I built uses a small pad of leather held against the arbor with set screw. Favre's clutch sandwiches the arbor between two metal rods with screws to adjust how tightly it is held. I didn't take a picture before assembling the clock but perhaps you can see it here:

Or perhaps not.

I am still testing the clock. I have the timing quite well tuned now. The weight is about 1.3kg; less might work.  As I write this, it has been running for a bit over 12 hours continuously. A couple of previous runs stopped after 1-2 hours and it seems that something was binding. Some of the previous clocks have stopped when friction has consumed too much of the power of the clock. You see the pendulum losing more and more of it swing and finally dropping below the amount needed to engage with the escape wheel. When Clock 24 stopped, it seemed more like something in the gear train had locked up, and it took a nudge to free it and get things going. It may be that the gear can't move freely enough on the arbors (not enough endshake maybe). If it binds again, I'll see if I can get a better idea of what is going on.

Thursday, March 03, 2022

Two minor clock projects and an update

I recently completed a couple of new clocks, both brief experiments.

Neopixel clock

The first uses a circular array of WS2812 LEDs, sometimes known as neopixels. It's hard to take good pictures of it, as the brightness of the LEDs overwhelms the camera on my phone. This will give a general idea:

The LEDs are this product. It consists of several concentric rings which different numbers of pixels. It's a nice, cheap product and comes with connectors so that you can use just some of the rings and address them in any order. The only alternative I found was an overpriced product from Adafuit.

The outermost ring has 60, making it suitable for minutes and seconds, and the next has 48, meaning you can displays hours down to quarter-hour resolution. I use the next ring, with 40 pixels for a temperature display, although that part of the software is not complete. The controller is a M5StickC Plus. Strictly speaking a level shifter should be used between the 3.3V output of the ESP32 in the M5, and the 5V needed for the control pin of the pixels, but it seemed to work OK without one. The only problem I had was during prototyping when I had it connected up through leads with alligator clips. For some reason, the signals got corrupted in this case. Using a breadboard or soldered connections works fine. The hours and minutes are displayed as a sort of swoosh with several LEDs at different brightnesses. The case is 3D printed in clear PLA. This does make the pixels a bit fuzzy, though most of the fuzziness in the video is from the camera.

The neopixel clock is not very practical. It's surprisingly hard to read the time without hands to direct your eye. I hoped the swooshes would help, but they really don't: every time I look at it, it takes a few seconds to read off the time, when it ought to be near-instantaneous.

Favre's Full Clock

The second recent build is Jacques Favre's "full clock". It's a straightforward clock mechanism driven by a small stepper motor, the widely used 28BYJ48. I used the ULN8003 driver board that came with the one I bought instead of the L293D preferred by Favre. The clock needs two 608 bearings, and there are printable versions. I had several 608s in stock, so used them instead.

The clock runs silently and smoothly. I like this design better than Steve Peterson's stepper clock (see this post), which is not quite as quiet and slightly jouncy in its movement. One concern is whether the 28BYJ48 will stand up to continuous running as they aren't really designed for it.

I am thinking of making some of Favre's other clocks, and this short project was meant as a way of learning more about his design style. Almost all of the parts worked fine. The only modifications I made were to shrink two parts with an inner thread by 95% for a better fit, and to make the hands a bit less boxy. There are some shafts glued together from two halves and I made 1mm diameter holes in them so I could use a pin or piece of wire to keep them aligned while the glue set.

Peterson 10 day clock update

Steve Peterson's 10 day clock was the first one I made. One problem I have had with it is that after a few weeks of running, the pendulum amplitude gets less and less until it stalls. Reading the comments thread on his myminifactory page, this is a problem which several people have had. The proposed solution is always to find ways of reducing friction, and I have been through several cycles of doing so: re-cleaning the bearings, polishing the arbors, reprinting the gears in regular PLA instead of silk PLA, lubricating with white lithium grease. I also tried adding more weight. The pattern has been the same every time: it seems good at first, but after 4-6 weeks of use, the problem recurs. I'm now trying ceramic bearings to again try to bring down the friction. We'll see what happens.