Sunday, August 07, 2022

Prusa extruder axle problems

I've been round the following issue several times now, and so I thought I would document it, even if only for my own future reference. If you find that a Prusa MK3 is manifesting problems like:

- blobs (like little nodules) on the surface of prints; see the right hand example below.

- under extrusion and poor layer adhesion, especially when there is a lot of retraction

- slight extruder clicking

- in extreme cases, extruder motor overheating

then the problem may be that the axle for the extruder idler has slipped. This has happened to me several times now. Open up the extruder cover and see if it turns freely. If it feels graunchy, see if you can push the axle in slightly. The movement may then get smoother. If you push it too far, you won't be able to close the door over the idler and can pull it back a bit. This has fixed the problem for me now at least three times, including one case where the axle was slightly bent and needed to be replaced.

Friday, August 05, 2022

Clock prototype #3

The latest prototype gets a bit closer to being a real clock. The frame is much more robust, and has a weight drum connected via a ratchet. Somehow I got a whole lot of things wrong in this version. It simply didn't run for more than a few seconds at first. The pendulum would swing for a short while and then stop. I spent a long time searching for sources of friction, and removing some parts of the mechanism such as the hour train and reprinting or drilling out others. This revealed a few things that could be improved. For example, the bearings for the fork arbor were very tightly fitted, and so unless they were exactly square, this put extra friction on the arbor. However, they real reason turned out to be that I had the fork inverted. The sketch I had constructed from it in Fusion 360 was for an escapement which turned the opposite way. In the earlier prototypes, I had remembered to invert the part but this time I didn't. As a consequence the pallets were oriented wrongly and didn't receive any push from the escape wheel teeth and the pendulum ran out of energy.  A definite case of a short circuit between the ears. Once I had fixed this, things worked a lot better - still stalling sometimes, but I have had a test run of 4 hours with no problems.

Here's a video.

Now you might think this is recorded at half speed, but it isn't! It seems that I got the gear ratio between the escape wheel and the second wheel wrong. The pendulum is going at the right rate, but everything after that is wrong. The brain really wasn't working well the last few days.

A couple more pictures:

There are still plenty of things to fix or add:
  • the weight drum sometimes slips forward and rubs against the hour wheel. This maybe the cause of the stalling.
  • although there is a ratchet on the weight drum, I didn't arrange a mechanism for rewinding the clock, and some changes to the frame would be needed for it.
  • I included a clutch on the minute arbor, but didn't get it right. The clutch should allow you to turn the minute hand and have it also turn the hour train. This entails having the minute pinion securely fixed to the arbor and the main minute wheel able to slip. In yet another moment of vagueness, I got this the wrong way round.
I think a lot of this can be fixed using the same frame (or something very similar), by rearranging the position and orientation of some of the gears. The decision to arrange that all the gears have the same distance between their centers helps here.

Friday, July 29, 2022

New clock experiments, part 2

For the next round of work on the new clock, I changed the configuration to have a frame which is shorter overall, and with the escapement set off to one side.

There is nothing holding the end of the side piece in place so it tends to splay out a little, but will do for this phase. (The illustration shows one version of the design. I used different minute wheels and weight drums in the experiments.)

For the first test, I made a minute wheel with a winding drum built into it, and set the clock going. With a weight of 560g, it ran for a few minutes and then stalled. Raising the weight to 680g, it ran until the weight hit the floor. The weights were not chosen with any care: I use a water bottle, and it's just a result of how much water I put in it.

Theory says that for a 50mm drum, you should unwind the string at 15.7 cm per hour if the motion is continuous. This means that for a typical 1.5 m drop, you would get a run time of around 9.6 hours. It's not enough for a practical clock, but will do while I am experimenting. I also added a weight drum on a separate arbor with 3:1 gearing.

(Yes, I know the string is tangled round the gear. It's the only video I took before disassembling it and I noticed it too late. It wasn't tangled when I did the tests.)

The 3:1 gearing triple the run time to 28.7 hours. However, I didn't get this arrangement to run reliably. It would usually stall after 10-15 minutes, though I did have one run lasting an hour. The weight I was using was 1340g (the heaviest I could get with a water bottle), and this probably explains it. You would expect something more like three times 680g, that is about 2kg, would be needed. I think the frame may also have been starting to distort slightly. Initially when I tried using the 3:1 gearing, I added a 20 tooth gear on the minute arbor, such that both it and the minute wheel were held onto the arbor with set screws. I could not get them to hold tightly enough and one or other would slip. In the end I glued the two gears together, and if I use this in future, I'll print them combined.

Finally, I put together a rough version a weight drum with a ratchet. I thought this would be straightforward, but ended up going through several variants. Looking at other clocks, I see three main styles:

  • the drum (which in all of these is within the diameter of the minute wheel) has a ratchet on on end, and there are pawls freely pivoted on the minute wheel. Gravity makes them drop into place. A lot of wooden clocks use this.
  • the drum has a ratchet on the end and there are sprung pawls attached rigidly to the minute wheel.
  • the ratchet is inside the wheel and the pawl are pushed outwards by spring. Used in Steve Peterson's SP5 (on a separate arbot).
The three styles are illustrated here:

(Credits: Jacque Favre Clock One, TheGoofy design on Thingiverse, Steve Peterson SP5).

I played around with the gravity approach for a bit, but found it hard to get the pawls to drop into place at the right time, and settled on using sprung pawls. They work at any position at any orientation and work well given that you can print thin springy plastic.

With a 50mm drum, I measured a weight drop of 8cm in 32 minutes, or about 15 cm/hour. There is something puzzling here as some other measurement showed quite different values. I also tried a drum 25mm in diameter. As expected, this halved the drop per hour. You would expect to have to increase the weight, but I got away with the same 680g, albeit with a rather weak tick. It would probably have stalled if I let it go on for longer.

I put the weight drum directly on the minute arbor for the tests I've just described. It is not a good way of doing things, as you really need the arbor with the weight to be supported at each end. That doesn't work well with some configurations of the hour train and works even less well if you are going to bring out a seconds arbor. As I mentioned before, a separate weight drum arbor with a 3:1 reduction didn't run without stalling, and so I decided to try 1:1 gearing. There should not be any problems with this, and indeed it worked fine. So this gives me the configuration I want to use for the next version, with the option that the 1:1 gearing could be changed.

Incidentally for some of these prints, I used a 0.6mm nozzle with an Arachne-based slicer (PrusaSlicer 2.5.0 alpha). Allegedly this gives as good precision as a 0.4mm nozzle. See Thomas Sanladerer's video for details. It seems to work well, though I still have a little bit of tuning to reduce stringing and blobbing, and it is more prone to producing elephant's foot. For now, it's good for faster prototyping, and I'll stick with 0.4mm when I want better accuracy.

Friday, July 22, 2022

Designing a new clock

It's about a year since I started making 3D printed clocks, with Steve Peterson's SP5 clock, and I decided it was time to try designing a clock of my own. I have modified some of the clocks I've made in small ways (adjusting fit) and in larger ones (replacing the motion work in the Swingtime clock). The only complete design I have done was the William Strutt epicyclic design, and even them much of it came from a published diagram. Now it's time to do something from the ground up. I don't know if I will carry this all the way through to a reliable design; it may turn out I don't have the skills or the patience to do so.

The starting point is a basic Graham escapement design with a rather conventional wheel train and motion train. I have a few ideas for ways I would like to refine it over time. The gear tooth counts are taken from this design on Thingiverse, itself remixed from a design by Thingiverse user TheGoofy. Having said that I wanted to design the clock myself, it might seem inconsistent to take the gearing from an existing design, but there are only so many possibilities which give you the right ratios and a number of teeth that you can manufacture. I like this wheel train as it allows you to add a seconds hand easily. Assuming a pendulum with a 2 second period (so about 1 metre long), you get this:

  • Escape wheel: 30 teeth, pinion 30 teeth.
  • Seconds wheel: 60 teeth, pinion 9 teeth.
  • Intermediate wheel: 72 teeth, pinion 10 teeth.
  • Minute wheel: 75 teeth, 16 teeth.
  • Reduction wheel (minutes to hours): 64 teeth, pinion 20 teeth.
  • Hour wheel: 60 teeth.
You have some choice about the gearing from the minute wheel to the weight drum. My initial design has an extra 20 tooth wheel on the minute arbor engaging with a 60 tooth wheel on the weight drum for a 3:1 reduction. If the weight drum has a diameter of 50mm, then it means that for each 3 hour rotation, the weight drops pi times 50mm, so that in 19 hours the weight drops by 1 metre. The pendulum length is something I will reconsider later.

I did the design in Fusion 360, using the standard add-in for generating the gears. The add-in only generates involute gears. I feel sure that I read something which said that cycloidal gears are better in the train from the weight to the escapement where each step increases speed and reduces torque, and involutes are better in the train from the minute wheel to the hour wheel, with the opposite characteristics. However, I could not find the reference again and there are many places which say to use involutes unless the teeth are so small they become fragile. This reference concisely summarizes the debate. If I wanted cycloidal gears there are tools like this one which output a DXF or SVG. I used a 14.5 degree pitch angle and a small amount of backlash to make the spacing between the teeth better. All of the pairs of gears are design to have their centers the same distance apart (60mm) so that I can stack several on the same arbor if I choose, leading to various different modules from 1.33 to 1.5. For the escape wheel and anchor, I used a parametric design a created a few months ago in Fusion 360, based on Jacque Favre's tutorial.

(Side note: what I actually did for the gear was use the add-in, and then project the profile into a new sketch. I could then project the pinion profile into a new sketch. This makes combining them a little easier.)

Here is a picture of a prototype, with a couple of minor parts missing:

The purpose of the prototype is to check everything seems to fit together. The weight arrangement is unfinished. It has no ratchet to allow for rewinding. The frame is very flimsy as I designed to print quickly. There are no bearings or attempt to reduce friction in this version, and the anchor is designed so I can experiment with different positions for the pendulum.

I printed part of this: the frame and the wheel train from the escapement through to the minute wheel, modified to include a weight drum. Originally I wanted to try this with various weights to get an idea of what weight I might need in the final version, but I had cut corners in the frame. So in the end it only ran for a few seconds with a weight hooked up. However, by applying force to the minute wheel by hand I was able to see it roughly working:

The beat isn't right and it takes quite a lot of force to make it run, but it gives an idea that things are generally correct.

More, hopefully, to follow.

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.