Jack Crossfire's blog View Details
Posted by Jack Crossfire | Sep 16, 2021 @ 11:00 PM | 6,923 Views
So the lion kingdom finally got a set of retaining rings & the tool for ending the motors slipping off. It should now be possible to use wide offroad tires. 1 fell swipe of the full faith & credit ended a year of problems. Full faith & credit is sometimes worth a lot more than being creative.

The tool is too big to fit in the 6mm retaining rings, but is necessary to put them on. Once on, the tool can easily fit to take them off. There is a choice between angled tips & straight tips, but they're all different sizes. The smallest tip is only straight. What helps is using a metal shaft to twist apart the retaining ring just enough for the tool to get in.
Posted by Jack Crossfire | Sep 15, 2021 @ 09:25 PM | 9,513 Views
Pressing down on the board in just the right way got the servo to glitch reproducibly. Only after thoroughly cleaning the PWM trace, the problem finally appeared.

There was a micrometer scale break in the trace, completely invisible when it was dirty. It's believed to have been there since the board was made. The board was made with a dedicated buck converter for the servo. An xacto nicked it while cutting a wire to the buck converter. The oldest photo is too small & dark, but some grotesque unsharping reveals 3 slightly brighter pixels in a row where the break was spotted later.

The break only cost $35 in unnecessary new servos. The previous $150 of servos were bought during the lunchbox days to solve common wear. Fixing it required soldering a jumper wire, despite such a small break. Merely tinning it wouldn't do the job.

With the electronicals working again, it became clear that the steering is still loose. This PID controller might need a highpass filter.
Posted by Jack Crossfire | Sep 13, 2021 @ 03:42 PM | 11,536 Views
The SPT died the same way the Hextronix died, a sudden full left lockup followed by intermittent restoration after wiggling the cable. Power cycling it did nothing. It's been 7 years of chasing erratic servo behavior. Most of it might be pots & motors wearing out. The last few have been converging on the wiring.

The cable has been already replaced. Replaced it again. This one is coming down to a loose solder ball or dry solder joint. So that's 2 days of not going anywhere because of a retaining ring, dead servo, dead 3D printer.
Posted by Jack Crossfire | Sep 11, 2021 @ 11:41 PM | 11,012 Views
Sharpied the final designs on. PLA+ is shinier than PLA, making it look slightly more upscale, but PLA+ is expensive so the lion kingdom didn't order any more after the container.

Sadly, the latest retaining ring popped right off. 1 problem is it can't be inserted bed side up, so it has a bevel which just slides off. After a moment of grief & debating destructive shaft modifications, the decision was made to spend $25 on a set of steel retaining rings & a retaining ring tool. It was as much as a new motor.

The problem with modifying the shaft was how a more robust PLA retaining ring would slide on. It would entail removing the encoder. Getting the right tool for the job was the only way.

The quest to print a new retaining ring with the right bevel revealed the hot end was stripped. It's aluminum & lions turned the nozzles as tight as the stock nozzle. The stock nozzle had red threadlocker. Heat blocks are sold in bulk because apparently they do strip all the time, but the lives of the grub screws & ceramic heater are sketchy & it becomes exponentially more painful to break down the parts in order to replace less.
Posted by Jack Crossfire | Sep 11, 2021 @ 01:10 AM | 12,715 Views
After several days, all the panels were done filament depositioning. None of the 228mm ones had any showstopper defects, though they improved over time. The last panel was a narrower one with no bed heating, but the 228mm ones definitely required it.

There was an attempt to evolve joining the segments to use PLA zigzags. This was the 1st attempt at expensive printed tape. The mane problem was clamping it to get the adhesive to stick. Magnets were terrible. It really needs a big, expensive jig. 1 piece goes on the flat side. The other piece needs pins pressing on all the facets of the other side. The final solution was spraying water to accelerate the glue, but water leaves a white film.

Despite the artifacts, the zigzags were visually lightyears ahead of duct tape. The mane issue is how they look when light shines through the panels.

The corner bumpers had a lot of delamination. They definitely needed 2mm walls & .32mm layers, but the TPU absorbed a lot of water from the lion kingdom's air conditioning.

Lighting used a slightly improved hot glue process & no solid core wire after all the solid core wire broke in the last container. The 3D printed containers have a lot more flexing near the corners than coroplastic.

The standard storage compartments went in with duct tape since duck with zip ties has been very robust....Continue Reading
Posted by Jack Crossfire | Sep 09, 2021 @ 12:04 AM | 18,290 Views
The lion kingdom's oldest battery finally succumbed after 12 years. It was $90 in a Hobbytown retail store. After its flying years, it must have rolled 1000 miles. It had a self adhesive velcro for flying which was ripped off at great expense before its driving career began. It spent 2 days draining into some LEDs, then got the short circuit of death.
Posted by Jack Crossfire | Sep 07, 2021 @ 11:54 PM | 18,858 Views
Blender has proven essential for doing certain final pass booleans & transformations where Freecad falls over. The models from Blender can be printed the same as models from Freecad, but it's another moving part.

It feels like the lion kingdom would save a lot of time & effort by using Fusion 360 instead of Freecad with a pile of workarounds in Blender. There's just the memory of Blender becoming the world's standard 3D animation program after lions used Maya for years. Then kicad became the world's standard EDA program after lions used Eagle for years. Arguably, Altium Designer is now the world's EDA program but kicad got a lot of publicity a few years ago.

Lions have come to rely on being able to access a lot more than 10 CAD files at any given time. These CAD files stay in the same directories as software & electronical files. They're all managed by the same git repositories with synchronized revision histories.

It would be another hoop to have the CAD files exclusively online in an autodesk server while all the other bits are still on the gits. Then, these CAD files would have to be constantly toggled read-only to get beyond 10. The revision histories would have separate checkins. The lion kingdom uses Comca$t internet which constantly goes down for days at a time but is never quite worth the hassle of switching out of.
Posted by Jack Crossfire | Sep 06, 2021 @ 08:29 PM | 17,752 Views
400 miles ended with a shredded tire, burned out servo & cracked rod end. The Hextronix died after 394 miles. It was a motor burnout. In went the SPT.

The decision was made to make a helical front tire like the helical rear tires. Because of the round tread, it couldn't be a simple C program that replicated & rotated an STL file. This one performed 96 boolean operations in Blender to create individually rotated layers, over 6 minutes.

Freecad is so bad, Blender has proven useful for final pass booleans, but it doesn't do parametric models, can't import a scene graph from FreeCAD, or have a model history.
Posted by Jack Crossfire | Sep 05, 2021 @ 03:47 PM | 24,582 Views
The lion kingdom's 1st 228mm wide part was a historic moment in Ender 3 lore. The Ender 3 was only marketed with a 220mm printable width. It was made possible by a 3dtouch & a lot of firmware hacking. The truck's container can now be printed in 8 segments instead of 12. If only lions could afford an Ender 5 Plus, the container would require 5 segments, but lions only have 1 part which requires such a large printer. An Ender 5 buys a lot of ramen noodles for someone in Shenzen.

The other new part was a new retaining ring. This retaining ring is more rigid, but it still can't compete with metal....Continue Reading
Posted by Jack Crossfire | Sep 02, 2021 @ 01:45 AM | 27,998 Views
So the 60mm treaded tires made it up the test trail in 4m33s, the same as the 30mm tires. There was subjectively no obvious improvement in traction, but speed highly depends on how much gravel the driver goes over. They didn't materially affect the steering. They might have made the very fastest turns slower.

The mane problem is they had so much mass, they instantly popped the motor's retaining rings off. There are ways to make the retaining rings permanent, by drilling into the motor shafts.

The mane evidence of improved traction is they left much less of a mark than the 30mm tires, so they were pushing forward more instead of throwing dirt back. Another section of trail with less gravel showed them going much faster than the 30mm tires, but it was bugger all distance for estimating speed from GPS. A shift of a few GPS readings makes the speeds equal.
Posted by Jack Crossfire | Aug 31, 2021 @ 09:35 PM | 24,796 Views
After suffering with a paw controller which was barely audible for 8 months, the lion kingdom finally reprogrammed the sounds to increase the loudness. Aural cues are the only way to know the auto throttle setting. The original sounds were centered around what was thought to be a single resonant frequency. The problem is loudness drops off catastrophically away from the resonant frequency.

In reality, the piezo buzzers in toys have 2 resonant frequencies. The one in the paw controller has a fundamental of 4264Hz & an overtone at 8528Hz. Instead of centering the sounds around a single resonant frequency, they need to be centered exactly between the 2 resonant frequencies & bounce between the 2 to be reasonably loud.

Bouncing between the 2 resonant frequencies made things a lot easier. There's just enough of the inner tones to convey some meaning while most of the meaning is conveyed by the order & timing of the 2 resonant frequencies.
Posted by Jack Crossfire | Aug 29, 2021 @ 12:14 AM | 21,351 Views
The lion kingdom finally upgraded to colored filament. A high vis paw controller was seen as the biggest win from colored filament. Orange was the closest PLA color since there are no high vis PLA colors.

Gootube videos extoll the fluorescence & shine of a $60 filament vs a $18 filament, but lions seriously doubt there's any difference in fluorescence. It's going to be covered in sweat.

It was time to model in all the lessons of the last 6 months. It got a few more M2.6 screws. The decision was made to skip M1.7 screws, since the amount of PLA required to make a standoff would nullify a smaller screw.

The original grey controller was quite soiled.

This was the 1st deployment of the clicky momentary pushbutton. What a great click it is, but it was absolutely worthless when running.

The electronicals had extreme water damage. It surprisingly only entered the clamshell facing away from the lion, a retrograde motion of the normal sweat path.

A trace perished & got a bodge wire replacement. The other traces got a tinning. This board is in the twilight of its years.

Then it got a hot glue potting.

A few wires got replaced. There were 3 hours of damaging a wire to replace a wire, going around the board.

The mechanicals were an epic disaster. It seems the orange build series filament does better at 200C than 210C. At 210C, the layers were expanding randomly. Tolerances were also way off, from bed leveling deviations....Continue Reading
Posted by Jack Crossfire | Aug 27, 2021 @ 01:34 PM | 21,759 Views
The next container update features TPU corner bumpers, updated headlight areas, & many other tweeks. Every recompute now takes 40 seconds. Using sketches instead of discrete polygons didn't make any difference. The lion kingdom still wonders what benefit printed containers have over the ages old coroplast containers besides looking better.

The corner bumpers introduce M1.7 self tappers. They can always be doubled in thickness. The bottom panel can be reused. Only new side panels have to be printed.

Making the M1.7's possible & optimizing the mass will require automated bed leveling, possibly the most overpriced accessory in the 3D printing mafia. The lion kingdom found a slightly cheaper copy of the bltouch.


It's just a solenoid, hall effect sensor & magnet. It would be $5 in any germane phone, but it's $32-$50 because 3D printing is an agile methodology, trendy startup, 5 trillion doll hair industry. The hall effect sensor gives a very precise height measurement of a magnetic pin. The solenoid alternates between retracting the pin & extending the pin. Installing it will include finally ripping out the cooling fan, making a new cable bundle.
Posted by Jack Crossfire | Aug 25, 2021 @ 11:20 PM | 11,092 Views
After 48 hours of printing, the 1st 60mm wide wheels arrived. If these can't get around on gravel, nothing below a scooter can. It immediately became obvious that they're hard to install. There's no way to see the motor holes down those wide hubs.

There's no easy way to detach the hubs from the tires so they can't be glued. They're so wide, it's going to be really hard to steer. The original lunchbox was 290mm across. These are 320mm. The next tires should be 45mm wide, at most. They should also have 5mm more TPU, since the tread pattern is not flexible.

These use a 10mm honeycomb + 5mm tread to allow the same hubs to fit the smooth tires. It was later decided printing all new hubs was more convenient than changing the tires on the same hubs. Obviously, they're only going to be on when an offroad drive is planned.

The original lunchbox tended to have its hubs crack where the right angles are. The hex hubs have a lot less material in the key right angle, but benefit from wrapping around the motor. The hex could be extended all the way to the bolts, but it would be expensive.

It might have been easier to stick the original lunchbox tires on, but they had less range & were expensive. The worst case 3D printed tire is cheaper than a $5 lunchbox tire.
Posted by Jack Crossfire | Aug 24, 2021 @ 12:54 AM | 13,301 Views
It went up the test trail in 4m33s. The smooth tires went uphill in 5m11s. The smooth tires went downhill in 4m9s. In exchange for this slight improvement, they were much noisier on pavement, despite the helical treads. They actually handled bumps & hit the programmed speed about as good as the smooth tires. The 5mm treads acted just like another 5mm of tire radius.

They dug a lot more into the gravel. The next step is to widen the offroad tires & finally update the paw controller to address some teething issues.
Posted by Jack Crossfire | Aug 22, 2021 @ 04:44 PM | 19,099 Views
Bay bridge tempo run (6 min 39 sec)

A trackstar finally began its new life as a 360 stepper motor.

As a stepper motor, it needs to commutate really fast to get 1 revolution every 4 seconds.

At 5V, it burns 500mA. At 6V, it burns 1A & gets hot. Its key advantage is it doesn't need to be powered to stay in 1 place. Timelapse mode still requires the full current, so it can't be given the 1A. The standby current when it doesn't move is 50mA. It doesn't have as much torque as the stock servo.

New mounting hardware was printed. The servo must be screwed in before the carbon fiber. There should be enough torque for clothing to not jam the rotation, but you never know.

While this turns very smoothy with very little power, it's still a horrible design. It uses an L6234 for the previous motor to directly drive the servo motor, even though the servo already had the electronicals for driving the motor. It would save a lot of space to reprogram the servo to do all the motor control & have a much smaller board outside just read the radio. The mane compromise was to give up empty space for less time programming.

The C8051F330 in the servo can be programmed with an arduino & the pads are pretty obvious.

It easily did 7 minute miles going downhill, with the wind. The lion was wiped out after only 2 miles at this speed. The full 20 minutes considered a tempo run were at 7m34s/mile downhill, with the wind. It was...Continue Reading
Posted by Jack Crossfire | Aug 20, 2021 @ 12:36 PM | 35,503 Views
After much fighting with Freecad's lofting tools, a better way emerged. Extrude a single layer, then make a discrete model rotated for every layer. The problem with lofting to get a helix is it constricts the center like twisting a sheet of paper. The 96 discrete models require manipulating an STL file in C. To retain some layer adhesion, it had to use parallel treads instead of V's. Getting a V tread would require moving the treads much closer together.
Posted by Jack Crossfire | Aug 19, 2021 @ 02:40 PM | 24,667 Views
There's not much layer adhesion when .5mm walls are diagonal. .2mm layers get better overlap but double the print time.
Posted by Jack Crossfire | Aug 18, 2021 @ 02:05 PM | 28,443 Views
The answer is no. You can't 3D print a water pump & expect it to last. Someone had to be the 1st to document it.
Posted by Jack Crossfire | Aug 15, 2021 @ 05:40 PM | 20,053 Views
A 4 hour cleaning job revealed the motors did a better job sucking rocks from the trail than a street sweeper. The leading theory is most of them got in through a wire passage. Adding tape around the motor assemblies might keep some rocks out. Flexing of the motor shrouds may also have let smaller rocks in.

1 screw had worked loose. After permanently farstening a screw by CA gluing it, the next attempt at thread locking a screw to PLA was to apply just a drop of CA glue.

A tire with a defective layer split in 2. There's no layer adhesion in .5mm thick walls of TPU, so the tires are held together by their 1mm thick treads.

There's no doubt driving on trails entails a lot of cleaning, no matter what. The container shed a few pounds of dust. There's no way to clean the radio panel without disassembling all the electronicals. If the electronicals get wet, the connectors corrode.

Since it's used for transporting food, it might be worth fabricating a bag for it, with a hole for the camera, a flap for accessing inside. The bag needs to be taped on the underside or it needs some kind of bungy cord on the underside. Dust gets in from the rear, where the tires throw it up.

There's a lot of material coming off the motor rotors where the PLA hubs flex in & out. It may be dirt getting between the hubs & motor, acting like sand paper. A leading idea is to make the hubs out of TPU but still have separate hubs & tires. There are implications for the place where the hubs bolt on the motor.