Jack Crossfire's blog View Details
Archive for June, 2021
Posted by Jack Crossfire | Jun 29, 2021 @ 11:29 PM | 7,469 Views
Another 6 miles at servo & broad yielded the same problems with both Trackstars. It definitely looked more like a fragged pot. According to the blog record, the 1st Trackstar arrived on July 2, 2018. Its 1st failure was Sep 06, 2018 after 400 miles. It seemed to get somewhat more reliable after swapping the BEC. The 2nd Trackstar arrived on April 2, 2019. They were the cheapest, at only $60. They always had intermittent glitches, but nothing serious until the 1st heatwave of 2021.

Another theory was they were really bad at water resistance. Any bit of water would ruin the pots, which was problematic for a truck. A 1 year lifetime per servo was a long time compared to the 6 months of a brushed servo.

Fortunately, the goog retrained itself & today was happy to spit out a $35 find, the mighty INJORA SPT5835W.


In the mean time, the decision was made to put in a bottom line, plastic geared, brushed Hextronix HX5010. A bag of these was ordered in 2018, when the lion kingdom resigned itself to a life of swapping servos every 6 months. They definitely won't last long.

Who would buy such cheap servos after the experience of the last 5 years? They're used in animatronic robots, camera pointers, things that don't fly, where quantity is more important than quality.

Anyways, the lion kingdom finally printed a socket wrench, because those locknuts are really hard to reach for a lion. Locknuts don't need as much torque as threadlocked nuts, so PLA is enough to make a tool. This was a huge improvement over the Durabuilt nightmare of the last 20 years.
Posted by Jack Crossfire | Jun 28, 2021 @ 11:38 PM | 10,990 Views
The servo got to failing every mile. It would intermittently turn. After a power cycle, it would freeze completely. Only powering it off for a few seconds to let the caps drain would restore it. It wasn't hitting it or shuffling cables, but the long power off which restored it. Changing from a 3A to a 5A BEC didn't improve it. It was definitely a progressive failure in the last 2 weeks.

Since it was obviously the servo rather than the ARM, a cable or the BEC, the decision was made to swap in its twin. The teardown showed nothing broken. The only thing which came to mind was the 47uF 16V tantalum cap going bad, but 3 tantalums in parallel aren't going to die & this board never saw any stress which could crack a solder joint. The motors are well & truly anchored to the case.

Abusing the board to get the motor out revealed some more caps, an old fashioned 5k pot which worked perfectly, & the brushless beast which spun freely. The pot was anchored but good.

Hobbyking doesn't sell any more brushless servos & the goog only shows nosebleed prices starting at $90. Brushless servos probably weren't a big enough win for the typical use & DJI moved the world to quad copters which don't use servos. The typical user doesn't drive 1000 miles in 6 months while autonomously swinging the servo arm to fight every slight bump. The trackstar was being marketed as a tail rotor servo, which definitely isn't being used anymore.


That would be the most lions spent since their 4k monitor.
Posted by Jack Crossfire | Jun 25, 2021 @ 01:00 PM | 28,565 Views
The servo died again in the same part of the route, same direction, same distance, but cooler weather. It could be the particular vibration in that spot & speed. There were some full deflection glitches which were blamed on the paw controller getting sweaty, but hall effect sensors have such low impedance, it's unlikely for them to glitch to full deflection. There could be a case for rebuilding all the wiring.

The lion kingdom could reprogram the microcontroller on the servo. There's also 1 more brushless servo with a weird size. It would require printing a new steering section. The idea of a brushless servo with a brushed angle sensor never did appeal.

The theory has always been driving thousands of miles with automated steering has worn them out faster than usual.
Posted by Jack Crossfire | Jun 17, 2021 @ 08:59 PM | 12,427 Views
The journey began as the apartment weather station showed 112F. The more accurate weather stations showed 104-106F. It was the hottest weather a lion ever ran in & the longest distance in such weather. The average speed had to be reduced to 5.4mph for the lion, yet the lion was still wiped out after 8 miles. There were some brief 6.6mph segments. Amazingly, the traction motors didn't die. What did die after 2.7 miles was the trackstar brushless servo. After power cycling it a few times & letting it rest, it managed to complete the run. It might have overheated.

It managed to outdo the mighty lunchbox. The lunchbox tires expanded & slipped off in heat.
Posted by Jack Crossfire | Jun 15, 2021 @ 01:02 PM | 24,633 Views
A rare moment when Elon photographed it as something besides window dressing behind his girlfriend. She's only easy on the eyes in his own mind. Engine positions are clearly visible. At times, it seemed the 20 outer engines might all mount on the tank dome, but now they're clearly mounting on the ring with the nozzles protruding outside. The feedlines attach to the thrust puck. Inside the aft section is a forest of tubes bringing methane from the downcomer to all 29 engines.

What look like gridfin holes are actually reflections. The left section goes on top of the aft section & stores oxygen. The completed ship on the right is probably going to be torn down.
Posted by Jack Crossfire | Jun 12, 2021 @ 06:36 PM | 46,010 Views
The motor encoders proved to get dislodged quite often, requiring field calibration, so it was time to finally make a filesystem. To keep it simple, it's a linked list of files. The filenames are just numbers. It just appends every new file to the end of the list. When it runs out of space, it erases all the flash & rewrites the unique data to the start. The cause of running out of space is normally a motor table, so the file ordering usually ends up being a config file followed by a motor table.

The trick is the unique data has to be smaller than RAM for the erase to succeed. The RAM is currently configured for 81kbytes & the flash is configured for 16kbytes. There's no way to erase a single file at a time. If the list is too large to be traversed before the watchdog timer expires, it could get stuck. Corruption with 0 length files is quite easy, but it gets the job done.
Posted by Jack Crossfire | Jun 01, 2021 @ 12:53 AM | 27,256 Views
The apartment balcony showed 103F. The billionaire homeowner weather stations showed 98F. After 5 miles, the left motor was dead.

The motor mount was liquid & the heat sinks were warm. What seems to happen in hot weather is the motor mounts deform just enough to make the encoders more prone to hitting something. It hits something during a bump & gets shifted out of alignment. Then the fouled motor spends a lot more time stalled, but gets pulled along by the functioning motor. The stalling causes rapid heating until the motor mount softens enough for the farsteners to come out. The deformation of the PLA is very slight, but enough to get the farsteners to come out. Once the farsteners come out, the motor won't turn anymore.

Strangely, the left motor's retaining ring was still in place after the encoder got shifted. There might have to be larger gaps near all the tight tolerances.

The next step was to finally support motor calibration in the field. It creates a memory resident lookup table. The table is lost when it's powered down so a firmware update is later needed. Embedded filesystems are very unreliable & complicated, so the memory resident table was the most useful alternative.

The motor mounts definitely need to be printed in PETG.