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.

Saturday, November 27, 2021

The Epicyclic Gear Clock Completed

Last time, I wrote about the design considerations for a clock adapted from William Strutt's epicyclic gear clock. Now it is time to complete the design by adding the frame and the driving/timing mechanism. There is one small change I made to the design documented in the previous post. The planet gear is mounted on a carrier pivoted on the minute arbor. This works better if you add a counterweight opposite the planet gear. A US quarter seems to be about right.

A few pictures and video of the completed design first, and then I'll fill in a few details.







The Frame

The frame for Strutt's original design (illustrated here) is rather ornate and a little too fancy for my tastes. I like the frame seen on a GrabCAD version and used this as the starting point for my own. The overall frame has to be taller than will fit on the print bed. In previous designs, I've looked for a point where I can split the frame and then joined the pieces with glue and pins. For my design, I decided to print it so that the very top part attaches with a couple of screws. This also allowed me to defer the decision about the drive and timing mechanism. Depending on what I decided, I could print the top pieces with different dimensions.

Drive and Timing

Most of the remaining design decisions concerned the drive and timing mechanism, that is how to get power into the clock and how to make it run at the right rate. The original clock used a spring, but I was not sure the plastic design would hold up to the stresses from it. Another option was to drive it with a weight attached to the minute arbor. The works OK for a wall mounted clock, but is not suitable for a desk clock. I toyed with using a stepper motor, but again did not like this. In the end I settled on the same drive as my two previous clocks: an electromagnetic pendulum. As before, the pendulum rotates a cam, causing pawls to engage with a toothed wheel. I'll call it an escape wheel, though this might not be an accurate use of terminology. The escape wheel then drives the ring gear via a pinion.

There are several design considerations. The number of teeth on the pinion and escape wheel must be chosen to drive the ring gear at the right rate; details of the calculations are in the previous post. The period of the escape wheel then determines the length of the pendulum, which must be less than the height of the frame at the pivot point of the cam. Finally, the teeth on the escape wheel must be large enough for the pawl to engage with it reliably. I considered reusing the exact escape wheel dimensions from one of the previous designs, but the escape wheel looks large and out of proportion. A smaller escape wheel is possible, but it must then have fewer teeth so that they are a reasonable size. After some playing around I decided on an escape wheel about 80mm in diameter with 40 teeth, with an 8 tooth pinion. This is about the smallest size of pinion that I was willing to trust. I used a trick I learned from Steve Peterson for the pinion. As only one face comes into contact with the ring gear, you can fatten up the teeth and make them stronger by displacing the trailing face.

With the gears I chose before, the ring must rotate once every 754.49 seconds, and the escape wheel then rotates each 8/168*754.49  = 35.928 seconds. Each complete swing on the pendulum advances it by one tooth, so the time per swing of the pendulum is 35.928/40 = 0.8982 seconds. An ideal pendulum for this period would then be almost exactly 200mm long, which fits well with the size of the frame.

Cam and pawl dimensions

I'll come back to the pendulum design in a moment, but first there is the question of how to design the cam and pawls given the escape wheel. A reason for wanting to reuse the previous designs is that I knew they worked. Unlike designs for standard escapements (such as the Graham escapement), I couldn't find any guidelines for working out the geometry. To solve this, I set up a sketch in Fusion 360 to try to make sure it would all work. Here is an annotated version:

Circle A represents the circle at the base of the teeth, B is the pivot point of the cam, and C and D are the pivot points of the driving pawl at the limits of the pendulum's motion. p and q are the points where the end of the pawl would contact the teeth. The angles of lines BC and BD are set by how far the pendulum swings. Experiments with the previous clocks suggest it is about 20 degrees. We can freely choose most of the other dimensions: the escape wheel diameter, the position of the cam pivot, the length of the cam, and the length of the pawl (Cp and Dq). I could then measure the angle between the pawl position at the extreme ends of the pendulum swing: the angle between the radii to p and q. This must be more than the angle between two teeth, 9 degrees for a 40 tooth wheel, but less than the angle between two teeth, as we don't want to advance the wheel too far. Based on this, I was able to choose dimensions which appeared to work, and validated them with a quick and incomplete print.

The pendulum

I mentioned that an ideal pendulum would be 200mm long. The actual pendulum for this clock deviates from ideal in multiple ways. The period calculation for a pendulum assumes the angle through which is swings is small (a few degrees), while the electromagnetic pendulum in this clock swings by something like 30 degrees. Secondly, in the ideal case, there is a mass just at the end of the pendulum. We have to have the magnet at the end, and the position of this is fixed so that it is close to the drive coil. To make it possible to adjust the timing, there is also a moveable weight bob. In the last two design, I made the pendulum shaft from a brass rod. The weight bob was held in place with a set screw, making fine adjustment tricky. This time round, I printed the pendulum shaft with a thread cut in it, and made both the weight bob and magnet holder similarly threaded. Adjustment is then much easier, and a lock nut can be used to hold the weight bob in position once the timing has been set. The weight bob is just a small printed part with a couple of M5 bolts attached to it. There are two 12mm x 3mm neodymium magnets.

The electronic circuit is the one I described a while back. It is controlled by an Arduino Nano, and this has the nice property that the code can measure the pendulum period and report it over the Nanon's serial connection. The electronics enclosure is a bodge, with various things held in place with blue tape and an opening for the USB port which is far too large. It's held to the frame with a rather flimsy bracket. One day I will come back and do this properly. Maybe.

The pendulum moved very vigorously - so much so that the frame rocks slightly. If I did a redesign, I would make it a bit heavier. The clock is very quiet compared to the two other electromagnetic clocks.

Wrap up

This is my eighth clock and the first one I have designed entirely, other than drawing the initial inspiration and (initially) the gear ratios from Strutt's original. I went though multiple iterations both in silico with Fusion 360 and in the printed parts. It's a cliche to post a picture of your box of rejected parts, so I won't. I'm happy with the end result.


Monday, November 01, 2021

William Strutt's epicyclic gear clock: design and prototype

In the early 1800s, William Strutt designed a clock based on an epicyclic gear train. There is a good description of it in an edition of the Horological Times. The key elements of the gear train are shown in the following illustration. The frame, escape wheel and driving force are omitted.


To understand how this works, start from the minute arbor. It is attached rigidly to the planet carrier (white). As the carrier revolves, it moves the planet gear (green). The blue gear is one of two sun gears and is fixed to the frame (not shown). The movement of the planet gear has two effects. Firstly, it turns the ring gear (red). At the top, you can just see a small pinion (also green), which would be attached to the escape wheel. This therefore regulates the time. The period of the escapement and the gear ratios are chosen so that the planet carrier rotates once per hour, as required for the minute arbor. The final element is the hour gear (yellow). It is also a sun gear, and is free to rotate on the minute arbor. It has the same diameter but a different number of teeth is different to the fixed sun gear. This is an implementation of Ferguson's mechanical paradox. The rotation of the planet gear causes the free sun to rotate at 1/12th of the rate of planet around the fixed sun, providing the rotation for the hour hand.

There are some existing designs based on Strutt's original, for example one by Clayton Boyer, one by Brian Law (without the paradox), and at two on GrabCad (1, 2).

The gear ratios work as follows. Let:

  • A = teeth on fixed sun gear
  • B = teeth on planet pinion
  • C = teeth on planet gear
  • D = teeth on inner side of ring gear
  • E = teeth on outer side of ring gear
  • F = teeth on escape pinion
  • G = teeth on escape wheel (not shown)
  • H = teeth on free (hour) sun gear
The period of the planet carrier (and hence the minute arbor) divided by the period of the ring gear is 1+AC/BD. The period of the escape wheel divided by the period of the ring gear is F/E. We'll come back to the hour gear in a moment.

In Strutt's design, A=66, B=8, C=68, D=144, E=168, F=6, G=34 and H=72. Thus, if the period of the planet carrier is 3600 seconds, the period of the ring gear is 3600/(1+(66*68)/(8*144)) = 735.3 seconds, and the period of the escape wheel is 735.3*6/168 = 26.26 seconds. As there are 34 teeth on the escape wheel, the period of the pendulum must be 26.26/34 = 0.772 seconds, implying the pendulum is about 14.8 cm (5.8 inches) long. This makes the mechanism suitable for a desk clock, as in the example shown in the Horological Times article.

I don't fully understand how the Ferguson's paradox works, but I can give some hand-waving reasoning about why it gives the right timing. Essentially, each turn of the planet about the fixed sun (66 teeth) advances the free sun by its number of teeth, 72. This turn takes one hour, so in that time, the free sun has advanced by 72-66=6 teeth relative to the fixed sun. This is 1/12th of its total number of teeth, hence making it rotate once every 12 hours. Note that in order for the free sun to have the same diameter as the fixed sun, it must have a different module (ratio of diameter to number of teeth), thus breaking the normal rule for gears to engage correctly.

Design decisions for a 3D printed version
I wanted to take this design and adapt it for 3D printing. The hardest part of this is finding a size which will work, by picking a suitable module for the gears. This then constrains almost everything else. We need to be able to print both a very small gear (escape pinion, 6 teeth) and to fit a very large one (ring gear, 168 outer teeth) on the print bed.

There is one other constraint. The tips of the planet gear teeth must not come too close to the minute arbor:

The distance from the center of the minute arbor to the tip of the planet gear teeth is (A+B-C-2)m/2, where m is the gear module. This follows from the center of the planet and planet pinion being (A+B)m/2 from the center, and the outer radius of the planet gear being (C+2)m/2.

At a module of 1.2, the ring gear is 204mm in diameter, and will just fit on the bed of a Prusa MK3S. The escape pinion is tiny at this modulus, with an outer diameter of just 9.6mm and teeth only about 1mm across. We actually do have some freedom to use a larger planet pinion, which in turn changes the size of the pendulum. For example, with 10 teeth (and hence 12 mm diameter), the pendulum needs to be 41cm long. You can somewhat compensate by adding more teeth to the escapement wheel: if we change it from 34 to 40 teeth, the pendulum needs to be about 30cm for a 10-tooth pinion.

The spacing between the axis and the tip of the planet gear teeth is 2.4mm, meaning the minute arbor diameter must be under 4.8mm in a world where everything is perfectly sized. In practice, you have slightly more leeway as the printer will round off the very tips of the teeth, but you also need to allow for slight misalignments and wobble as the mechanism moves. There is one further issue associated with this. The free sun (shown in yellow) is loose on the minute arbor, so seen from the side it looks like this:

In this illustration, the minute arbor diameter is 2mm, about as small as possible. How do we keep the free sun in its position along the shaft? One option is to add a shaft collar just underneath it, rigidly attached to the shaft. Another would be to add a spacer, but its hard to find a diameter which can both be printed reliably and won't interfere with the planet gear. We could also use a piece of thin tube as a bushing; for example, a 2.5mm tube with a wall thickness of 0.225mm.

Another possibility is to reduce the number of teeth in the planet gear. At 66 teeth, we have 3.6mm radial space instead of 2.4. At 64 teeth, we have 4.8mm space. The number of teeth on the inner side of the ring gear must decrease to compensate, and the pendulum needs to be slightly longer. Making this change in no way alters the Ferguson's paradox, as it is only the fixed and free suns and the planet pinion which participate in this. With 64 teeth, there is no need for a shaft collar and instead the planet carrier can be modified:

It is possible to use a 3mm minute arbor in this configuration. The inner ring gear now has 140 teeth and a slightly longer rotational period. With the 6/34 escapement, the pendulum would need to be about 1 cm longer than before.

Tooth profiles
Most gears use an involute tooth profile: the classic shape with a narrowed "waist". Cycloidal gears are an alternative that has been used in clocks, and it works well for 3D printing as the teeth have straight sides with no waist. Once the teeth are above a certain size, the sides are parallel and so fewer small gap fill movements are required from the printer. For small teeth, the sides are not parallel, though it is possible to adjust a bit from the strict profile to avoid them becoming too fragile. Fusion 360 and Blender both have add-in gear generators, but they only work for involute teeth. I was able to find a cycloidal gear generator as downloadable software. There is an online version as well, but I prefer the downloadable version as it can generate SVG files, which are more convenient for converting into sketched in Fusion 360. The SVGs need to be scaled to the correct size after loading them up. One thing the software lacks is a way of generating the inner teeth for the ring gear. Some people suggest creating an outer gear and then using it to cut away the inner part. It is approximately correct for involute teeth, but does not work for cycloidal ones. My approach was to load the SVG file, then flip the lines making up one tooth about a chord drawn on the pitch circle:

Another problem with the output of the gear generator is that the arcs at the base of the teeth do no quite line up with the edges of the teeth. I solved this by adding a base circle to the teeth. To generate the gear in Fusion 360, I extruded one tooth, copied it with a circular pattern and then added the body of the gear using the base circle.

Drive and timing
Strutt's design used a pinion plus escape wheel for the timing and was driven by a spring coupled to the minute arbor. You could also drive the minute arbor with a weight, if the clock is configured as a wall clock. The gears move quite freely and so could also be driven by a stepper motor in a similar way to Steve Peterson's desk clock. Another option is to use an electromagnetic pendulum. I can't find any examples of exactly this, though there is a somewhat related design by Nigel Climpson.

Prototype
I wanted to validate that the basic mechanism works before going further. It's a bit hard to video it (not enough hands!), but this clip shows that it moves quite smoothly.



If you look carefully, you can see the hour wheel advancing. For the next stage, I need to settle on the drive mechanism. The planet carrier also needs a slight redesign so that the circular piece counterbalances the planet gear.

Tuesday, October 26, 2021

Cycloidal gears in Fusion 360

Fusion 360 has tools for creating involute gears, including its own spur gear add-in and GfGearGenerator, and they work well. However, if you want cycloidal gears, it's not so easy to find something that works. Here's one approach, in case this turns out to be helpful to anyone else.

Start by generating the gear in DXF form using Rainer Hessmer's Cycloidal Gear Builder. Make sure to use the highest quality level. It's useful to include a hole in the middle so you can identify the center. Download the DXF file. If you load this DXF into Fusion 360 (Design > Insert > Insert DXF) you will get an unhelpful error message. The DXF file isn't in a form that Fusion 360 can handle and we need to fix it.

You can fix the DXF by importing into FreeCad and then exporting it again. Another option is to import it to Inkscape and save it with "Save As". The DXF should then import into Fusion 360. However, if the gear is large, Fusion 360 may sit for hours processing it. It might never finish. Selecting "One sketch per layer" when inserting it sometimes helps, but generally does not. So another option is save it as a SVG from Inkscape and insert that instead. It's still slow, but does work. If you are lucky, you might be able to extrude the result and create the gear from it. Or sometime Fusion 360 will just abruptly exit.

The Cycloidal Gear Generator is supposed to have an option to output to SVG but it was missing when I looked for it. However, instead you can download a desktop version of the app from here. This will give you a SVG with much better segments. Note that you have to specify that you want a pinion. In the web version you can omit it. The desktop version raises an exception if you try. The result will load into Fusion 360, but it won't work as the segments don't join into a closed curve. However, we can use the sketch as a starting point.

First note that the imported SVG won't have the right size. We need to scale it. To find the scale factor measure an element of known size. For example, if you created a 6mm hole in the middle of the gear, measure its actual diameter and scale by 6 divided by this. Measuring the radius is usually easiest, and so then you would scale by 3 divided by the measurement. To scale the whole sketch, exit sketch mode, go to Modify>Scale, select the sketch from the browse list, and enter this factor.

Now we want to go into edit sketch and delete everything except for the center hole and one tooth. The tooth will consist of two lines and two arcs. You ought to be able to create a circular pattern with these and the arc joining them to the base of the next tooth, but you don't get a closed path if you do. The problem is that the exact end points of the lines aren't right. So change everything to a construction line (with x), then change the two lines and two arcs back. Create a center circle with circumference on the ends of the lines. Now create a circular pattern with the two lines and the two arcs about the center of the gear, with a number of elements equal to the number of teeth you want. You should now be able to exit sketch mode, select all the teeth and the circle you just created, and extrude your gear.