Saturnalia: A rotating Christmas tree

2019-12-08 esp32 / silly / robots / 3dprint / micropython
Apparently I have a little too much time on my hands and maybe a little too much enthusiasm for the holiday season

So we have an enormous, fake Christmas tree. An eight foot tall monstrosity in plastic and steel, which for 11 months lurks under the house in a giant bag, waiting for December to come around so it can loom over our living room.

Rotating Table

A couple of years ago I built a rotating table for the tree to sit on, out of a cheap lazy susan bearing and a couple of big circles of MDF. This makes it really easy to wrap tinsel around the tree, and to decorate it evenly all over, but one side always ends up getting left against the wall.

Which made me think: what if the tree could very slowly rotate on its own?

Like most silly ideas, it sat and fermented for a while until this year, finally, I had a chance to put it into action. Late last year I bought a cheap 3D printer and I have a bunch of random motors and microcontrollers laying around so I set out to make it work as much as possible with junkbox parts.

Motor Drives

The first thing I needed was a way to drive the base. I thought about belt drives or adding some kind of gear right around the edge of the platform, but keeping things aligned seemed like a challenge and I wanted to keep the side load on the bearing to a minimum.

There’s a 10mm gap between the base and the platform though, so there’s a little room in there for the gears. The bearing has a nicely formed metal edge at 305mm OD, so it seemed natural to use that as a guide and build the gear around it.

The turntable platform The turntable platform

I quite like working in OpenSCAD so I grabbed the excellent openscad gears library and experimented a bit with involute gears to pick an appropriate gear modulus and number of teeth.

Big Gear Segments Big Gear Segments

After some fiddling around with the design, I ended up with a tooth diameter of 335.5mm and since the required gear is way bigger than the 120 x 135mm printable area of my printer, I picked 9 segments of 13 teeth each for a total of 117 teeth. I printed these 3 at a time, and after a couple of bad prints due to broken and jammed filaments I finally had my full big gear. There’s a real thrill to lining up printed parts and discovering that they fit perfectly!

Printing big gear Printing big gear

It fits exactly around the 305mm outer diameter of the lazy susan bearing, with three screws holding each segment in place.

Big gear on base Big gear on base

pictured: two screws per segment

not pictured: yet another trip to the hardware store

I wanted to have the motor components all accessible from on top of the base in case I need to fix anything. So the big gear is actually attached to the stationary base and the motor is attached to the rotating platform. The motor mounts through a 51mm round hole in the platform, cut with a hole saw, which allows the smaller gear to be on an eccentric so it can be moved in and out a little to adjust the gear clearance.

Eccentric mount for 28BYJ-48 motor Eccentric mount for 28BYJ-48 motor

First I tried a NEMA-23 stepper I happened to have in the junkbox, directly driving a 7 tooth pinion. This worked okay and had plenty of power, but the aim was to turn the tree quite slowly and at low speeds it was jerky and not terribly effective.

I also tried out a small 28BYJ-48 stepper, which has a 1:64ish ratio gearbox built in. While it did an okay job at very low speeds, it had barely enough torque to do the job and stalled a lot once the platform was weighed down a bit. It would probably have been fine for a more modest tree though.

So with yet more OpenSCAD work I ended up with the following:

Gear Train Gear Train

The NEMA-23 motor turns a 7 tooth pinion (‘A’), which engages with a large 39 tooth gear (‘B’) on the output shaft. There’s another 7 tooth pinion on the lower end, which drives the 117 tooth gear on the base. So overall there’s about a 93:1 drive ratio between the stepper and the platform.

The output shaft assembles using a square end and a square socket in the primary output gear. By making the square end’s diagonal just a smidge smaller than the bearing ID it can be assembled through the bearings. and the shaft then retains the bearings in their seats.

The primary input gear is currently just press fitted onto the stepper motor shaft and retained with a little superglue, but i may need to come up with somethng better if it comes loose again – perhaps a metal collar and grub screw or a keyway ground into the stepper shaft.

The whole thing is sized to use every last mm of the printable area of my printer. The teeth are a slightly smaller pitch than the secondary gears, but still pretty chunky.

Printing small gears Printing small gears

The output shaft is supported by a couple of cheap 6901Z ball bearings and has another 7 tooth pinion (‘C’) on the lower end, which drives the 117 tooth gear on the base. The stepper is a 1.8⁰ per step or 200 steps per revolution of its output shaft, so overall that’s about 18624½ steps per revolution of the platform.

Geartrain in base Geartrain in base

Base mounted on platform Base mounted on platform

The gears need a housing to support them, and that’s 3D printed as well. The lower part contains the eccentric and the seats for a couple of cheap 6901Z ball bearings. These are a nice size to use because they are quite compact but the 12mm ID is big enough for a 3D printed driveshaft.

6901Z bearings 6901Z bearings

The upper part contains the motor mount for the NEMA-23 motor. A couple of wood screws hole the whole thing down the the platform.

Gearbox with cover & motor mount Gearbox with cover & motor mount

The secondary pinion is very close to the stationary base, which it is moving relative to, so I’ve added a socket for a 8mm ball bearing in the end of the pinion. This prevents the teeth of the pinion from touching the base and allows the pinion to glide smoothly over the base surface. The ball is pressed into place in a vice and then loosely retained by the socket.

The 3D Models are on GitHub although there’s still some work to be done. (Since printing it, I’ve made the lid a little heavier and added some mounting ears just to keep the cover from buzzing)

Interestingly, just as I was finishing this article there was a subtle “pop” and the tree ground to a halt. On inspection, the output shaft had snapped right where the gear meets the shaft:

Snapped Output Shaft Snapped Output Shaft

This looks like it was due to poor selection of infill: the thick walls and open grid infill which worked fine for the primary output gear wasn’t suitable for the much smaller parts. So I’m reprinting that with a much denser gyroid infill.

You might also notice I’ve started printing version numbers on everything. This makes it a lot easier to keep track of which print is which. Currently the tree has a v5 base, a v6 cover and I’m now printing a v7 driveshaft. Not much has changed, but its good to keep track of it.

Slip Ring

The motor and controller need power to run, and getting power to a continuously rotating platform isn’t easy.

There’s a device called a “slip ring” which allows this to happen using sliding contacts, and while these are pretty cheap on Ebay I thought it wouldn’t quite fit the spirit of this junkbox project, so instead I figured that any concentric connector would do in a pinch.

The most suitable connector in the junkbox was a 6.35mm / 1/4” phono connector, as used in guitar leads and similar audio applications. This isn’t rated to any particular voltage or current but they’re very cheap and the contacts are pretty well formed for this sort of thing.

Phono plug slip ring socket Phono plug slip ring: socket

Phono socket and plug Phono plug slip ring: plug

Power is supplied from an old laptop charger at 15V, through the slip ring and then power goes to the L293D based motor driver and to a couple of 5V/3A buck converters. Running the slip ring at a higher voltage lets us run it at a lower current, hopefully reducing losses and increasing its life.

Lights

The tree lights also have to be powered the same way, and so a 5V/3A buck converter module is used to supply power to a chain of 300 WS2811-driven LED lights. These are individually controllable from the same processor as the stepper driver using the MicroPython NeoPixel library.

WS2811 lights WS2811 lights

The LED lights I bought on Ebay and so far I’m very happy with them. Each light has its own tiny PCB and controller. The strings can be connected together but the wiring is far too flimsy to run more than one string at high brightness … later LEDs turn yellowish as the blue LED runs out of voltage to function. But if you’re only driving a few LEDs at a time, or driving them at low power, the whole chain works nicely. There’s also spare wires at the start of each chain to attach extra power connections to boost the voltage in the chain.

Prototype Wiring Prototype Wiring

Software

This is still very much a work in progress, or to put it another way, it’s rather janky.

Code in MicroPython/ESP32 controls both stepper and lights.

A simple web server using PyCoSe lets the tree be remote controlled from devices on our home WiFi. Yeah, PyCoSe is a bit experimental at this point and I’m mostly using it for this project to eat my own dog food.

Further Work

The software needs a lot of work. I think I might have just dislocated a pinion. And the lights and baubles are not yet hanging (some of the lights are still in the mail) and the videos are yet to be shot. Also I want to add in some more explanatory text about how the motors and gears work.

The biggest problem I’ve got is that both the lights and the motor need real-time control, and the time taken to update the lights is greater than the interval needed between microsteps of the motor. So either I have to have the motor running less smoothly, or break the lights into smaller chains.

The ESP32 is a dual core CPU so in theory one core could handle each task but there’s also the WiFi to compete with: in practice I might end up using a secondary CPU for motor control, or maybe just a whole ‘nother ESP32.

I’m pretty happy with how the gearbox is working now, but the gear ratio could be bigger and if I was starting the design process over again I’d consider using a Cycloidal Drive instead. It’d also be nice to have some kind of clutch mechanism so that the tree can still be turned by hand when the motor is off.

The tree is a lot heavier now all the lights and decorations are on board, and thus has a much higher moment of inertia … I’ve added some code to gently accelerate and decelerate the tree to reduce load on the geartrain, but it’s interesting to see how much movement there is in the decorations as the tree wobbles very slightly in its orbit.

If you’re interested in following along, follow me on Twitter and I’ll post updates and videos and so on there.

MQTT

I mention just above that having two real-time functions plus the web server all competing for CPU time wasn’t working real well, and I was considering adding an extra secondary CPU … well, instead, I added a whole ‘nother ESP32 and switched to using MQTT instead of HTTP.

How this works is that the two controllers each listen to their own MQTT topic, and wait for a relevant message to set either the speed or the light mode. Obviously this is pretty wasteful and flagrantly insecure and if you’re connecting to a public MQTT server then anyone in the world can spy on your Christmas tree and even seize control of it by sending their own messages.

But it’s pretty amazing how much neater and cleaner the code is, and how easy it was to set up. There’s no need to worry about DHCP addresses or anything like that, the tree just configures itself and listens. This ties into my L2IoT protocol concept as well.

(Update: after a certain amount of messing about it turns out that you need to provide a keepalive ping every 30 seconds or so to stop the server vagueing out on you. It seems odd that the library doesn’t do this for you, but thankfully it’s easy to set up a timer for this purpose)

More work

Depsite using microstepping, the motor is noticeably “ticky” as it jumps from phase to phase. The motor control loop takes about 3.5ms to run, so it’s not that. I think I just need to fiddle the parameters a little to get a smoother action.

Plotting 3D position

Plotting the positions of lights on a tree Plotting the positions of lights on a tree

This tweet by Lorraine Underwood discusses determining the location of the lights on a tree in 3D space. I’d been thinking along similar lines, using the rotator naturally, and now I’ve got my full complement of lights on board and an ESP32-CAM board working it should be a fun project. The aim then will be to synchronize the lights and motor movement, so that for example the lights could be set to a vertical striped pattern which stands still even as the tree rotates ‘under’ it.

Acknowledgements

Thanks to my family for inspiring and putting up with this strange project and the terrible state of the loungeroom while I was assembling it …

messy loungeroom

Merry Christmas (etc) to all of you out there in Internetland and all the best wishes for 2020 :-)

VIDEO

Everyone’s been asking for it, so here it is:

Time-lapse video

(I just assumed that YouTube would handle vertical videos more elegantly by now, but apparently not.)

UPDATE BOXING DAY

The gearbox survived through Christmas! I’ll pull it apart when the tree comes down and have a look at the wear on the teeth and the phono plug slip ring.

FINAL UPDATE: JANUARY 2020

After a few days’ runtime, wear on the gears is negligible, even on the smallest pinion on the motor shaft which turns 93 times as often as the tree itself. Surprisingly, the phono plug slip ring is showing no signs of wear either, although I can’t see the internal socket contacts so they may have suffered. The re-printed output shaft was much stronger, and didn’t break again despite the tree getting stalled a few times by coffee tables, dogs and other obstacles.

My favourite way to run the tree was at very low speed, so that the movement was barely perceptible. The leaves vibrate slightly as if it a breeze, and if you stare at it long enough, the tree seems to stay still and the rest of the room seems to move slightly.

Everyone else found this horrible and disturbing and preferred the tree to whisk merrily around. This was a bit beyond the capacity of the existing motor so I suspect for next year we’ll have an improved motor drive capable of handling 1 RPM or so, possibly using a toothed belt drive to minimize vibration. Probably a proper slip ring too and the lazy susan bearings might not be the best choice.

The old-school L298 motor driver I used wasn’t a great choice, running way too hot due to the voltage drop across the bipolar transistors. It also has quite limiting PWM capabilities due to slow switching (~1 µs). I’ll switch to MOSFET based drivers if I’m using a stepper again.

I also need to run more power wiring for the LEDS. The chains plug togther, but voltage loss over more than one chain is way too high, leading to color shifting and very limited brightness overall. Attaching a separate pair of power wires at each end of each chain would help this a lot.

The neopixel timing was also annoyingly glitchy due to timing issues. New work on the RMT peripheral in Micropython/ESP32 has made it possible to control neopixels without glitches. Alternatively, a little picopixel driver board could be used to take the load off the main CPU.

I’d also consider setting up a separate small MCU just to handle the stepper timing: dealing with that in multi-threaded code is a bummer. This would make a really nice stand-alone board anyway: a little board with a ATTiny or similar and a couple of MOSFET drivers, communicating to the host CPU on I2C much like the PicoPixel does.

Or maybe this is a good case for having a ESP32 + FPGA board?

Last but definitely not least, in my haste to get this project ready in time for Christmas Eve I ended up falling back on all the IoT antipatterns I like to complain about … the tree just listened to unencrypted messages on a public Mosquitto server and I even published the details in my github repo. I never noticed anyone fooling with it, but it’s still not a very clever way to go.

But anyway: Happy New Year, and may 2020 be a happy and creative one for all of us.