A year of adventures

This isn't one of my usual project blog posts, but I owe my dear readers some context for the lack of project content.

The last year did not go like I expected.

Part 1: I ride my bike a lot

In mid May last year my friend Sam and I decided to ride the Great Divide Mountain Bike Route, a mostly off-road route from Banff, AB to the Mexico border in Antelope Wells, NM.  We'd been talking about it for years (I first learned about the route in 2014 from Shane Wighton, who had raced the Tour Divide, when Sam and I were interns at Formlabs), and the opportunity finally arose, between Sam switching jobs and me having accumulated a boat load of PTO over the last 5 years.

The decision was fairly last-minute as things like this go.  Mid-May through July was consumed with planning, getting gear, building up a bike, riding my bike even more than usual, and also building some cool robot grippers.

Here's a view of the route, down the Rockies through Alberta, British Columbia, Montana, Idaho, Wyoming, Colorado, and New Mexico.  We did the Tour Divide race route (rather than the ACA version of the GDMBR), which has some "fun" alternates that most people touring the route seemed to avoid.

Here's the bike & setup, and how I stored all the stuff on my bike.  If I did it again I would absolutely use a suspension fork.  The Cutthroat can take a 100mm travel fork, and I thought about getting one, but figured "I've seen all these pictures of people doing it without suspension, it'll be fine, right?".  No, it was not fine.  Sam, meanwhile, rode it on my old full suspension 2014 Santa Cruz Tallboy LT.  Definitely not your typical Divide bike, but it made it  - with many creaks and a seized on cassette - but it made it.

The bike:  2019 Salsa Cutthroat frame bought used, Hunt XC Race wheels, Redshift suspension stem (came with the frame), GRX 1x12 groupset, 2.35" Vittoria Mezcal tires.

The setup

It was amazing, and there were some really hard days.  

Here are some roughly chronological pictures

Spray Lake, on the first day

Lugging my bike up the scree field that is Koko Claims pass.  My upper body was not prepared for this:

Holland Lake, MT

Ascending Fleecer Ridge

And hiking down Fleecer Ridge once it got too steep

Montana is sometimes a desert

Idaho - the route only hits a little corner of the state

Jackson Lake, WY

Old mine just before Atlantic City, WY

The Great Divide Basin, where we got blasted by brutal headwinds for ~200 miles until we hit Colorado

Sleeping by the side of the road half way up a pass in CO.  It's hard to get out of Brush Mountain Lodge early in the morning when Kirsten makes is making you pancakes

Descending toward Steamboat Springs

Nice welcome before Steamboat Springs, CO.  Apparently the local farmers hate the Steamboat gravel race

A steep, loose climb towards Salida, CO

View during the incredible descent to Salida

Deer in the aspen

Descending Carnero Pass towards Del Norte

Ran into Joe, another mechanical engineer from Boston, and rode with him through a hailstorm to Del Norte, CO

Leaving Platoro, CO

A rocky and exposed section of CDT in New Mexico.  Should have taken more pictures, the view was amazing but it was brutal to ride on a loaded bike

Leaving Abiquiu, NM, beginning one of the longest climbs of the trip.

Cool rocks on the long road stretch to Grants, NM

Long stretch of road from Cuba and Grants, NM

Pie Town to Silver City

More slow-rolling CDT out of Hachita, NM, ~40mi to the border.  The road disappeared and it turned into winding through cacti and other pointy plants

At the Mexico border!

I tried to record many of the descents on a GoPro - here's one hour (out of many) of riding down mountains:

We finished the trip in 33 days, not taking it easy by any means, but there were some short days when we had to bike repairs done (Sam did a rocky descent before Pinedale in the dark with no rear brake) or really wanted to hit a town and sleep in a real bed.  The typical touring speed seems to be a fair bit slower, judging by the dots surrounding us on Trackleaders and the number of people we passed, and the record is a mind-boggling 11 days, 19 hours, set during the race this year.

Mechanically my only casualty was one derailleur hanger (I brought a spare), otherwise everything worked great.  We did get super lucky with weather and had almost no rain and no scorching temperatures - minimal rain and mud definitely kept the drivetrain alive, I've heard stories about needing to replace chains and cassettes half way through when it's wet and muddy.  Every evening I wiped down my chain and lubed with Silca Synergetic (one of the the best non-wax chain lubes around), and by the end I was only 2/3 the way through the allowable chain wear.

I got no flat tires the entire trip. When I got home I pulled many thorns out of the tires, but the sealant did its job.

Would I do it again?  Maybe - I'd definitely do another ride like this again somewhere else, but watching the race play out this year got me thinking about going back.

Some stats according to the logs:

Miles: 2,722.68

Elevation Gain: 178,717 ft

33 Days:  82.5 mi/day, 5415 ft/day

Average moving speed: 9.58 mph

Average hours “out” per day: 11:12 

Average moving hours per day: 8:36

Top speed: 48 mph

Part 2: you don't need a shoulder anymore

The last couple days of the Divide my shoulder started bothering me.  One day it hurt while riding, but it was a flat day with lots of tarmac, so I could hunch over in the aero bars and ignore it.  The night after we finished, I couldn't sleep from the pain.  And it continued for almost a month - hurting but tolerable during the day, and the worst continual pain I've ever experienced at night.  

A nearly sleepless month, a x-ray, MRI, and EMG/ECG and some physical therapy later, I finally got a diagnosis - Parsonage Turner Syndrome - mysterious damage to the brachial plexus nerves.  It's not really clear what causes it, other than that some risk factors include getting a viral infection (which I'm pretty sure I did not), and "strenuous exercise" (which I certainly did).   Lack of nerve signal feeding the deltoid caused it to almost completely atrophy, and if I tried to raise my arm to the side, nothing at all happened.

that's not supposed to look like that....

After 8 months it finally started showing signs of life, and a year later I have maybe 1/2 of the deltoid again and can raise my arm. 

Part 3:  I moved to California

Just the TL:DR version: I left the Boston area after 13 years, moved out to the Bay Area, and am working on robots with the awesome folk at Physical Intelligence.  The shop came out here with me, fortunately now in a single-car garage, not in a spare bedroom.  

After an eventful year with the least personal project progress I've ever had since I went to college, I'm finally back in action.  More kart updates and other new stuff coming. 

Go Kart Revival Part 1: Drive Units

 As I've mentioned in previous posts, robot land has an annual car-and-other-vehicle show.  Last year I brought the trike, but the year before that I brought the electric kart I worked on with Bayley.

Picture from Bayley

Before the event, the kart hadn't been touched since the pre-Covid days.  Bringing it back to life and ripping around the parking lot was a great reminder of two things:

• This thing is so much fun to drive.  
• Thing is so janky

A few parts of the kart were mechanically nice like the polychain pulley system, but the rest of the kart mechicals were bodged together out of literal garbage from the miters floor (not my fault, honest).  Not even joking, the tube holding up the inverter is a piece of electrical conduit.  

Anyways, this got me motivated to make the cart not awful - which basically means re-doing just about everything.  The drivetrain isn't the most offensive part of the kart, but I'm starting there, since it involves motors and gears.

Step 1, actually make a CAD model of the kart.  I sketched out the chassis dimensions and solidworks'd up a model.  I brought the fiberglass Tillett seat into work and got a 3D scan of it with Bill's help.

One feature I wanted with this upgrade was independent control of the rear wheels rather than a solid rear axle - so I can do fun traction control stuff in the future.  There are some complications making this work well on a kart chassis (which are designed around solid axles and jacking up the inner wheel around corners), but I have plans to deal with that I'll explain in some other post.  I explored a few different layouts for independent drive units - sticking with belts and single-stage spur or helical gear reductions were the two main other candidates - but a pair of planetary drive units in-line with the rear axle looked the most promising.

For motors, I'm using the hybrid starter generators (HSG's) from the 2017 Ioniq hybrid.  The kart originally used a pair of 2012 Sonata Hybrid HSG's.  2 years ago I got one of the Ioniq HSG's to tear down and take some measurements on (this project has been brewing for a while), and found that the terminal measurements (resistance, inductances, back-emf) were identical to the older Sonata HSG's, but the motor was slightly smaller and a couple kg's lighter.

V-magnet arrangement rather than flat magnets like the 2012 HSG

Carefully freeing the stator from its housing with a hacksaw:

I looked into a few different options for planetary gears and ended up buying some chinese industrial servo planetaries in the hopes I could modify them to work.  Assuming this would work it was dramatically cheaper than using off-the-shelf or custom made gears.  They look absolutely enormous as they came, but I was pretty sure these had a lot more steel than they really needed.

I took apart the gearboxes, modeled them, and figured out where I could hack away material.  I was able to substantially reduce the length of the gearbox, and shave 5.6 kilograms off its weight as-received, getting it down to 2.6kg.  Not as quite as light as the old belt system, but the overall assembly should end up a fair bit lighter than the old motors + pulleys + mounting brackets.  A lot of the mass and size savings was from going down a cross section size on the output bearings - the stock bearings were much beefier than I needed.

For re-housing the motors, I tried out PCBWay's aluminum 3d printing services.  Printing the housing let me integrate a waterjacket into a monolithic housing, and was cheaper than getting the housings machined from multiple parts out of billet.  

For the waterjacket I did a counterflow design, mostly for mechanical reasons so the inlet and outlet would be right next to each other.  I added ribs down the water channels to add surface area and reduce the pressure drop around the 180 bend (top left in the picture below).  Not that I actually did any heat exchanger math or CFD to back up this design.  If the stock waterjacket was adequate  this should be good enough....

The yellow surfaces on the model below were post-machined.  These all had 1.5mm-2mm extra material added for machining away

Cross section of the full drive unit:

In context on the kart:

The PCBWay printed housings turned out great - they must have done a ton of hand finishing work, there were no traces of support material left on the build platform side of the part.  They were designed to be printed motor bore up, so all the circular features were in the print plane and ended up very concentric and close to nominal size.

One distortion showed up in the prints - there was some shrinkage in the outer diameter of the part where the ribs in the water channel are.  Not surprising given the thin outer wall (1.5mm) and the sudden change in cross section at the ribs.

I didn't order any spare housings (they were cheap for a metal 3d print this big but not that cheap), so I had no opportunity to screw up the post machining.  

Post machining step 1 was the front side of the motor housing - the gearbox interfaces and front bearing bore.  To hold the housings I made an expanding mandrel to grip the un-machined motor bore.  The expanding mandrel is basically a collet with a cylindrical OD and tapered ID.  The collet is pushed axially down a tapered shaft to expand and grip the ID of the part.

Mandrel 3d-printed on the Bambu printer.  Printed in PETG with solid 5mm thick walls.

The tapered shaft part of the fixture

To get the fixture to run true I skimmed the outside in-place on the lathe.  A single M8 bolt and a big aluminum washer clamp the mandrel axially to expand it.

Housing mounted - this is actually after machining, somehow didn't get any pictures before finishing the turning.

Runout on the un-machined surfaces turned out surprisingly good for fixturing and measuring off printed surfaces.  

First operation finished on both parts:

To machine the stator side of the housing, I made a fixture that located the parts on one of the features turned in the last op.  Held in the 4-jaw chuck so I could indicate the fixture back in true if needed:

Runout on the as-printed motor bore after fixturing.  After post-machining there would be some nominally 1 and 1.5mm thick walls, so I didn't have a ton of wiggle room:

Large thin-walled parts like this are prone to ringing during machining.  To damp out vibrations better I wrapped the housings in a bicycle inner tube - see video below of before/after ringing.

All the turning finished:

Some video from the turning process:

 The milling was pretty simple - nothing tight tolerance, just a bunch of tapped holes and a few faces milled flat.  I set up the housings on the Bridgeport using the same fixture as the previous turning step:

Indicating it in for the hole patterns on the other side:

For the side operations I held the housing in the vise with some over-sized vise jaws.  The side features were aligned to the previously drilled holes, so I could use the holes for alignment rather than having to indicate the part in level:

Finally, tapping the threads for water cooling fittings.  There wasn't much clearance in the water channel, so I had to grind off the tip of the tap in order to tap to full depth.  In retrospect I should have made the bosses for the fittings a millimeter or two taller.

I preemptively helicoiled all the structural M4 threads on the part - the printed aluminum alloy (AlSi10Mg) is similar strength to 6061, so not particularly strong.

I got even more use out of the stator fixture for installing the stators.  I added a guide rod to the fixture, concentric to the motor housing:

I 3d-printed a flexure to guide the stator into the housing - the flexure is a light press into the ID of the stator, and designed to be stiff in tilting but compliant in X/Y.  This keeps the stator axis parallel to the housing axis, but lets the stator center itself on the housing bore:

The stator was a light interference fit into the housing and secured with retaining compound.  I heated up the housing to around 100C with a wrap-around heater to open up the housing and slip the stator into place.  I did the assembly in a pneumatic-driven hydraulic press, I could send the stator home with the press if it got stuck half way down:

Most of the work modifying the gearboxes was machining the housing/ring gear.  The bulk material removal was done on the lathe at work - I tried at home but my little Clausing wasn't stiff enough to handle interrupted cuts in tough steel on a 6" diameter part.

After turning there was some profiling, hole-drilling, and counterboring on the tiny 5-axis mill.  It made terrible noises but did the job:

Both ring gears after some hand-tapping and deburring

Gentle warming with a heat gun and it slips onto the front of the motor housing:

I made a new rotor shaft for the Ioniq motor to interface directly with the sun gear. These were turned out of 4140 HT bar  (pre-heat-treated to ~32 HRC) on the Clausing:

I roughed with inserts for steel, but the polished insert for aluminum left a fantastic finish on light finishing passes

Nice

The front is threaded for a lockring that clamps the front rotor bearing:

Bearing and lockring test fit:

Here's the new shaft next to the original.  The original used a straight knurle press fit to transfer torque from the rotor lamination stack.  I went with a straight press-fit with retaining compound since it's simpler and not a super high stress interface.

The sun gear is held in with a heavy press fit + retaining compound.  I copied the amount of interference from the original gearbox shaft:

I put a generic M6 tap + 15mm bore interface on the back of the rotor for encoders.  Eventually I'll probably switch to a magnetic encoder, but right now I've attached the stock resolvers since that's what the kart's electronics are set up to deal with

I may get the rotors re-balanced, apparently it's not to expensive and the balance has probably changed with the new shafts.

Rotor installed in the drive unit:

The output of the gearbox clamps directly on the keyed go-kart axle.  I shortened the stock output shaft on the planet carrier, tapped a hole pattern in it, and machined some clamping shaft couplings that bolt to the face of the carrier.  This part does double-duty clamping the inner race of the bearing on the front of the gearbox

Planet carrier milling.  Planets taped up so swarf doesn't get in sensitive places

I broke out my old lathe-broaching tools and single-point broached the 6mm keyway on the lathe:

Gear box re-assembly with new bearings.  I splurged a bit on legit NSK brand bearings:

Gearbox assembled with the new front plate (sent out to have this machined and anodized black) and output coupling:

And there it is, an assembled drive unit:

Giving it a spin with one of my little motor drives:

I was hoping to get the kart driving again this summer, but ended up doing (and preparing for) other things.