Geared, Bi-Directional, Backdrivable CVT

I haven’t devlogged in a while cuz I was busy with outpost/opensauce, but I almost finished the build. I had terrible printing problems (my parts were warping all over the place) and I basically had to print/build the entire thing in 2 nights. As a result i have rlly bad friction on the gear shifting mechanism. After talking to some people at outpost and opensauce, I believe it is possible replace the clutch with a system of one way bearings, so I will spend most of my time brainstorming solutions for a v3 or v4.

I finished printing all the parts and tuned all of the tolerances. I ran into far too many printing issues (I probably wasted as much filament as will be used in the project), especially with bed adhesion. It was to the point where even a 10mm brim with 0 brim-object gap and increasing the bed temp by 5 degrees and washing the bed couldn’t stop warping. But at the end of the day I thugged through and got all my parts. I also ran into some cooked tolerancing with the bearings and even the screw holes for some reason. But all of that is behind me now cuz I can finally start building.

I ordered all the parts and set stuff to print to test the tolerances and I ran into some pretty annoying issues. Here’s some of the issues I ran into:

• There was only 4 packs of 2 3x6 linear bearings in stock so I had to redesign the lead screw gear shifter to use 6x12 ones.
• 32x39 bearings are literally 12 bucks per and there’s only one singular vendor for it so I switched to 30x37 bearings which are way cheaper. In doing so I ran into really sketchy wall thicknesses.
• The slots for the 28mm dowel pins were way too close to the walls of the gear, so I had to change up some of the geometry so it doesn’t just print as a half circle.
• I spent a good amount of time tuning the tolerances so that the bearings would actually pressfit and not slide off easily. Same thing with the dowel pins.
• The biggest problem I ran into was that apparently you CAN rotate linear bearings around a shaft very smoothly. Every google search made it seems like I would break my wrist and explode trying to rotate a linear bearing, and it was pretty shocking to see it spinning smoother than a fidget spinner. This completely cooks my entire design and I still can’t figure out a good solution
• the second biggest problems was that, when switching from 32 to 30mm bearings, I forgot to change the dowel pin size so now they don’t fit in the bearings. I might just end up 3d printing pins if they’re too expensive/take too long to ship.
• a lot of the parts were really hard to 3d print so I split them into multiple to screw in.

Apparently the maximum devlog time limit is 10 hrs. I had 22 hrs logged because I wanted to have some aura post with 40 hrs where the entire thing was built, but I ended up losing 12 hrs instead.

another great lock in, the cad is done (finally). V2 seems like it could work IRL so I’m gonna actually order parts and build it. Today, I split the gray gears (with the tube) into two parts so I can actually assembly it, added countersinks to parts that need it, and fixed up some tolerances.

I don’t have too many new ideas for the V3, but hopefully after I build/test V2 I can notice some key problems. Here’s the general gist of I’m aiming to accomplish:

• Obviously, I need to actually lock in with elliptical gears for the V3. This is the biggest problem with the V2. Because of the sinusoidal motion, the output gear isn’t constant speed (it oscillates slightly too).
• The whole extra gear for the CAM is super annoying, so I’m gonna keep it coaxial to the input/output shaft (running three things coaxial) and gear it from the tubes instead of a separate thing. I need a huge geometry change for this which is why I’m saving it for a V3.
• I want to figure out a better way to do the dog clutch system. I’m hoping there’s a smarter way which doesn’t rely on meshing/unmeshing but so far I couldn’t think of anything.

I’m really happy with where this has come, and I can’t wait to start assembling this.

For those who want to take a closer look, here is the link to the cad: https://cad.onshape.com/documents/c369754a085e1b437744081a/w/8784f23027b47ba880d55ef1/e/7f96dc46c879863ea30c1be9?renderMode=0&uiState=6a3f9f14c9069ebe4a356101

insane lock in, I finished the whole linear motion, and made tons of small changes. Here’s some of the bigger stuff I did:

• I was originally planning to use thrust bearings, but they were way too big. Instead, I just decided to use 3x6 bearings sideways (ball bearings are designed to handle radial load anyway). The packaging was cleaner than thrust bearings anyway.
• for the linear motion, I had to add more dowel pins - linear bearings to stop the thing from rotating with the lead screw.
• I ran into some really stupid onshape problems where everything was off by 0.088 degrees for some reason. I was messing with random parts and doing random mates for like 30 minutes until suddenly it worked. I have no clue how, but that doesn’t matter.
• The linear cams were hitting the 3mm dowel pins inside the gear so I cut out part of it.
• The linear cam seemed really weak, so I connected the bottom of the ramp to the main assembly. This was a huge pain because there’s barely any space, so tolerances had to be tight.

Just one more day of cad and I will be done with it.

Huge lock in, finished up the linear CAM so it should work now, and started the worm gear for the gear shifter. I realized I was doing the linear CAM wrong, and it has to be mirrored instead of just a 180 degree turn, so I fixed that and made a new part for the other side. I also added the lead screw and bearing for the gear shifting. Turns out everything packages nicely (i love uxcell) so I don’t have to do cooked stuff. Just another day or two of CAD and I should be ready to build.

Did tons of changes to the design and continued procrastinating my devlogs. I realized the whole dog clutch thing was too cooked with that little space because it means less space between the bearings for the perpendicular cam. It was also better to change the angle of the linear cam to 45 degrees, and doing so gave me more space. I finished up the dog clutch stuff. I also spent some time cadding the perpendicular cam, and I ended up spending a couple of hours learning surface modeling in Onshape (it was super hard to get it to work). I also put tons of time into thinking about a better way to power the cam: having a whole separate gear is messy, so I wanted to keep it coaxial on the output shaft, but it turns out there isn’t enough space. Hopefully I can switch to this in a V3. I’m also saving elliptical gears for a V3 cuz I have to get this done by outpost. I just need to do some final touches and a full assembly check before I can start building.

I added the dog clutches and realized there’s some big spacing issues. I mated the dog clutches and switched from a 19mm to 18mm length linear bearing. There’s some really bad spacing issues with the linear bearings: not only is it super long, but there also needs to be space between the two separate shafts so that the linear bearings don’t contact two at the same time. However, this distance can’t be too much or else the distance between the clutch and the gear in disengaged state will be too small. I spent some time searching for that balance only to realize that my linear bearing was contacting the flanged bearing, so I changed the lengths up but still no luck. I’m probably going to just make the whole thing a bit wider (because the angle for the linear motion is too steep right now anyway) which should make everything a lot easier.

Finished around half of the assembly. I assembled some of the parts in onshape and realized there’s a few intersections and some of the spacings were off so I changed those. I didn’t run into much problems, but I had a few better ideas that I’ll try implementing in the future:

• It turns out I was being stupid and you need two separate dog clutches for each side. This might actually be better because since they just have to rotate differently but the spacing between them is constant, I can make that outer ring much thicker. This gives me space to bearing sandwich the cam instead of doing some other cooked stuff. However this also means you need two separate shafts which are cantilevered. I need to figure out a way to support the shafts from the inner end.
• While I was assembling, I realized I had done the linear cam (for the gear shifting) wrong because everything can be one part. This will make the entire thing thinner.
• I had a stroke of genius and realized that instead of using that fifth bottom gear which makes everything bigger and ugly, I can gear the perpendicular cam (for the dog clutches) to the gear shifting mechanism, making the entire gearbox so much cleaner. This is kinda cooked because it means the gears will be sliding on each other and because it relies on the tolerancing of 3d printed parts which I don’t trust too much. But if it means making the gearbox look much better, it may be worth it (aesthetics are the most important consideration when designing functional parts).
• I realized I could switch everything from mod 1.25mm to mod 1 or even mod 0.8 because there’s no more cooked gear meshing/unmeshing. However, it may be too late to do this for the V2, so I’ll probably save it for a V3.

The gearbox is not only thinner than the V1, but is also much less cooked. And (probably best of all) the packaging will be really really clean. It does take up more space radially, but that can easily be fixed by switching from mod 1.25 to mod 0.8.

I’ve been procrastinating my devlogs, but I finished up basically all the parts and fixed a ton of problems. Now I can get started with the assembly. I’ve been procrastinating way too much with devlogging because every time I finish recording its late at night and I’m too lazy to devlog but here’s a summary:

• I tried implementing the linear cam (which is just a fancy name for a sliding ramp) but I quit halfway through because there was barely any space and I was too lazy to make it work. I then implemented my old idea (the linkage) but when you mirror it, it takes up 50mm which is huge. I then decided to go back to the linear cam. I added two ramps and two slots instead of just one so its more stable. I plan to add a screwed in 3d print plate underneath so it doesn’t bend. Now it only takes up like 35mm which is slightly better.
• I mirrored the gear and realized the plate that switches the gear ratio will be hitting the outer bearing of the dog clutch, so I spaced out the gears a bit more. I’ll probably be switching away from the outer gear on the dog clutch because it takes up tons of space.

The parts are basically done, I just have to see how everything works in an assembly.

Figured out perpendicular cam geometry. I was running into quite a few issues and managed to find a solution that fixed most of them. I pivoted from 6mm ID bearings to 32mm for the second stage of gears (the oscillating ones). I also realized that instead of moving the dog clutch directly, I have to mount another bearing around it with a 3d print mount. This not only allowed me to save tons of space, but now I only have to use one perpendicular cam because I can make the 3d prints 90 degrees offset, preserving phase. Here are some problems I ran into and how I tried fixing them:

• The 32mm bearings were hitting, so I changed up the main geometry and will be alternately adding flange space on each side of the main plate. There isn’t enough space for 2 flanges to be next to each other, but there’s enough space for 1 flange and a normal bearing to be next to each other (0.1mm of space).
• With my old design (screwing a sidways bearing directly on the dog clutch), I had to use two separate cams because the phase was not 90 degrees offset. However, with another bearing around the whole thing, I can change the angle to not be radial, making it exactly 90 degrees.

Looks like V2 will take the shape of a pentagon, and will soon look like a hexagon once I add the linear motion for the gear shifting mechanism.

I did tons of brainstorming and came up with a better replacement for the meshing/unmeshing gears (I actually had 2.5 hours more recorded the night before but lapse ran into a problem so they aren’t counted for here). I realized I could keep the gears in contact the entire time, and instead use a perpendicular CAM (idk what it’s called, it’s basically a CAM that has its offsets parallel to its axis of rotation) to engage a dog clutch onto the gears. The dog clutch basically has 0 chance of bad alignment because the clutch will self align. This hopefully makes everything much more reliable and efficient. To power this, I need a third gear directly connected to the input gear that takes the linear CAM below the input shaft (you can’t run it coaxially). To make it fit, I changed up some of the gear ratios. I soon plan to switch from 1.25 mm to 1mm modulus.

Continued sketching my V2. I spent most of my time visualizing the whole thing (I spent almost two hours sketching the thing out and doing the math). Here are some of the problems I ran into (and how I fixed them):

• I need the inputs on both sides to be powered, but I also want the output to be a pulley on the side (not in the center). This is impossible with a normal coaxial setup. Instead, I linked the first stage of gears (the one that has the ration changing mechanism) and will be changing up the linear motion mechanism.
• When connecting the first stage, it hit the output gear, so I had to change some of the geometry.
• As mentioned before, the radial force by the gear changing mechanism of V1 will probably cause the linear bearings to bind, so I’ll be changing this to a static linear-CAM type of motion for a constant force throughout the motion. Because I’m linking the first stage, everything will basically be inside a tube.

Did some math for the V2. I want the output to be at a constant speed, which means the gears have to be noncircular. The output right now actually isn’t even sinusoidal, and Claude and I had to do tons of weird trig to find the angle as a function of time. After graphing everything on Desmos, I figured out the ideal shape of the gears. Next, I’ll do more mastersketching, and export the shapes from Desmos into Onshape. In the image, the black curve is the position as a function of time, and the blue and red curves are the ideal gear shapes. Actually, now that I think about it, this is wrong and I need to change the whole thing.

I made some minor changes and I’m almost ready for a v2. I fixed up some of the spacers, changed the direction of the linkage that changes the ratio, changed some of the bearings. I also changed the coloring, and it looks much cleaner (I chose pink because I have pink filament loaded in, if you guys have any recommendations please tell me). Here are some fundamental problems with the current version and how I’ll fix them in the v2:

• The biggest problem is probably gears meshing. The entire thing has to be perfectly synchronized for the gears to mesh perfectly every time, which is obviously impossible. The only thing preventing jams right now is my prayers. However, backlash does give it some room for error. I’ll probably find a better gear tooth profile for my V2, but this is sort of a fundamental problems that will exist.
• The second biggest problem is that the gears oscillate from max speed to 0 speed every half-cycle, which means the speed will be constantly oscillating. This is easily fixable: in my V2, I’ll use more driving gears (probably 4) each active for a smaller fraction of their period. This combined with elliptical gears to counter the sine waves should make the output basically constant speed.
• The linkage for the gear changing has basically zero torque, and backlash will completely fry it. Also, this adds weird thrust forces on the linear bearings which will probably cause them to bind. In my V2, I’ll probably shift from a linkage to a linear CAM type of thing (I still have to figure it out).
• This is really big and wastes some space. My V2 will (hopefully) have better packaging and be smaller. This is achieved by putting the second layer of gears on the inside of the plate and praying there is enough space.

The uniform coloring makes it look way better. Once again, if you have any colors you think would look better, please tell me.

I basically finished up the linear motion for the CAM. All that remains is some minute adjustments.

Here’s some problems I ran into:

• The geometry was incredibly hard due to the lack of space, and I ended up just trying a bunch of possibilities and landing on one that worked decently. While I was planning to use belts, I switched to gears after I saw how little space there was.
• Screws were hitting the servo for the gear changing mechanism, so I rearranged it so the lip is on the outside of the plate, giving me a precious 4mm to work with.
• It was hard to come up with a way to ensure the full range of motion always happens. There’s nothing pushing the gears up after the CAM pushes it down. Initially, I was planning to do some complex linkage relying on the definition of an ellipse (the sum of distances from the foci is constant). I tried sketching it out but there was literally no space. At the end, I thought of tying a string to both gears and wrap it around an idler to keep the constant sum of distances.

This thing’s gonna look clean once everything is colored. All that remains is some minute adjustments and I can move on to a V2.

Started the linear motion for the CAM. I was initially thinking about using MGN3 linear rails until a realized that each rail is over 100 bucks. Instead, I decided on linear bearing based motion with 3mm rods similar to how the ratio changes. Here’s some problems I ran into:

• Cheap MGN rails were way too big so I decided on the linear bearings (I spent way too long searching for cheap MGN3 linear rails)
• I initially wanted to put the rods between the yellow and blue plate, but after sketching it I realized the tolerances were too sketchy.
• Initially I was going to do one main CAM in the center, but I realized when the CAM is pushing the shaft to the outside, the angle is weird so the force is basically transmitted perpendicular to the direction of the motion, which really cooks efficiency. So instead I decided on 2 smaller CAMS that are belted to the main gear.

some future/long term problems I need to fix:

• I don’t think there’s enough space for the belts and cams, so this whole design might be fried
• The 3mm rods aren’t constrained on their ends, so I have to figure out a way to do that (right now its only held by friction).
• The motion of the driven gear will oscillate a ton because of the CAM. One potential solution is to use elliptical gears and have the gears engage only in a certain phase where its speed is high (so its not stopping)
• There’s something similar to the coaxial effect from swerve drives in this gearbox too: I’m gonna call it the CVT effect. When the gear ratio changes, the linear motion makes the gear change phases. Something similar happens because of the CAM, but this isn’t that big of a problem. But when in doubt, assume software can account for it.

Mounted the CAM gear, added a front plate, and added screws. I ran into a problem where one of the hubs hits the belt to the lead screw so I changed up some of the geometry. The gearbox is starting to look pretty complicated and dense without the second most complicated part (the CAM). I am holding off coloring everything and pocketing till the very end because it’ll be very satisfying.

Finish up Linear Motion for CVT. Mounted the linear bearing, added the servo, and finalized the linkage so the actual CVT part now works. Now, I have to work on the CAM mechanism.