The Orange Involute

Big Data needed more drive­train power. How do you get more of that sweet, sweet sensored brush­less power?

Portescap 22BL-8B-P17

More motor.

These are eBay-supplied Portescap 22BL-8B-P motors, precur­sors to the current Portescap lineup of 22 mm motors. With differ­ent magnets on the rotor and a higher ther­mal resis­tance of 13 °C/W, they are rated for 20 W contin­u­ous rather than 70 W, but that’s still much more than a beetle weight (3 lb) robot needs, espe­cially since drive­train power comes in short bursts.

22BL drawing

The mount­ing pattern in the diagram doesn’t quite match that in the photo. I’ll deal with that.

The more press­ing concern is the insane speed. The hobby R/C world would describe this motor as a 1500 Kv “inrun­ner” with a no-load speed of 36800 RPM at 24V. With a contin­u­ous stall torque of 21.6 mNm and assum­ing 70 mm diam­e­ter wheels, each motor would have to be reduced at a ratio of 1:10.8 to be able to win a static push­ing match against another beetle.

Even with that much reduc­tion Big Data’s second coming would still have an esti­mated top speed of over 10 m/s (20 mph). The max power is limited by heat, not wind­ings, so for short bursts the motors can drive Big Data to a full 1 G of hori­zon­tal accel­er­a­tion at any speed within 90% of their top speed. Remem­ber that this is the fastest that a robot can accel­er­ate within the normal limits of trac­tion (i.e., no tire adhe­sion, no magnets, no suction, etc.).

Gear train design

This design calls for at least two stages of gear reduc­tion. I decided on 8:42 on the first stage and 28:68 on the second for an over­all reduc­tion of 1:12.75, using spur gears with a 0.8 mm diame­tral pitch (“Module 0.8”) and a 20° pres­sure angle.

I didn’t even know they made 8-tooth pinion gears, but there they are from Stock Drive Prod­ucts1. Unfor­tu­nately, they don’t make a stepped spur gear with 42 and 28 teeth, so I had to make my own.

Planning mill tool paths

First I drew up the invo­lute gear profiles as equa­tion driven splines in Solid­Works and designed a machin­ing solu­tion using HSMX­press, my preferred computer-aided machin­ing (CAM) pack­age.

Roughing positive mold with a 6 mm tool

Then I milled the posi­tive mold cavity from Free­man Supply’s “purple machin­able wax,” which is basi­cally hard­ened and plas­ti­cized paraf­fin. The machine is a Roland MDX-402 on loan to the Garage, the maker­space on Google’s Moun­tain View campus.

Finishing wax mold with a 0.4 mm tool

It gets pretty scary when I spin the 0.4 mm four-flute end mill to 15000 RPM and finish the teeth at over 250 mm/min. The high feed rate keeps the tool from melt­ing the wax. Molten wax chips tend to stick back onto the mold and are tedious to get rid of.

No tools were broken in the making of these gears.

Degassing silicone in wax mold cavity

Next I mix up some Quan­tum Sili­cone QM 270 and drop it in the posi­tive mold cavity. Under a strong vacuum, the air trapped under the sili­cone in the mold’s nooks and cran­nies balloon up and writhe to the surface, where they pop rather satis­fy­ingly (and silently!).

silicone ready to demold

I kept the back­side of the nega­tive mold flat using a sheet of acrylic. Sili­cone doesn’t stick to anything, so this worked well.

Silicone mold and polyurethane gears

Now the nega­tive mold is out and I have a reusable, flex­i­ble sili­cone nega­tive for plas­tic gears. I dump in some Inno­v­a­tive Poly­mers IE-3075 and a lot of the orangest dye I have ever seen. Unfor­tu­nately, polyurethane does stick to acrylic.

gear separation

BAM, I got some gears. I love that the machin­ing marks from the orig­i­nal posi­tive wax mold are preserved through the process to the final part.

Geartrain test fit

Seems to fit.

Thanks Dan for loan­ing us the Roland mill and thanks Michal for the superb CNC and resin cast­ing guide.

  1. I’m not link­ing to them because I’ve always had such a terri­ble expe­ri­ence with them. I’m finally taking my busi­ness away from SDP/SI for good. []
  2. Not MDX-40A, which takes G-code, but the older model that takes Roland Machine Language (RML). []