11/03/2018 at 01:01 •
My base is way, way too flexible.
I can get about 1 cm of variation in the end effector position just by lifting a corner. This explains a lot about the issues I've been having.
My nice looking 1/2in oak faced plywood should have been my 3/4in plain Jane MDF apparently.
I have some new 3/4in MDF parked by the door, and this time I'm going to do two layers with cross beams. I need to route out the cable holes through, which in my current 'borrowed shop' situation, requires tools I need to use outside.
It's raining here in Montana, and has been for days. MDF really doesn't like moisture. I'll see if I have a window this weekend, although the forecast says Tuesday. Wish me luck.
<EDIT> No luck, and now it's snowing. So I have cleaned up/moved stuff around my borrowed outdoor wood working area so I should be able to get this cut tomorrow rain or shine, or more likely.. snow.</EDIT>
10/22/2018 at 15:06 •
3 minute video. 4 crappy versions. Third try at uploading the last one, 69 minutes and counting to process it. This has not been my day.
Well I guess now at least I can stop focusing on the Prize and get back to more of the fun stuff. :)
10/22/2018 at 08:42 •
I cooked my SG90 mirror arm servo, and I don't have another one just like it. I set the up position too high, so it was stalled the entire time I was figuring out why I lost half my light ring.
It still works, ironically, but it gets really hot now, and moves really slowly. Like 10 seconds per 1/4 rotation slow.
I have the MG90 knockoffs which jitter terribly, and some HS-53 Hitec ones, without arms long enough.
There is some super glue attempting to make the Hitec ones work. Cross your fingers, and I might actually get a video done.
And I cooked another one. Limits were off by 2.5 degrees. Homing set the servo to 0, which was just outside of the range. Fixed in software now...
10/21/2018 at 06:58 •
Took a significant amount of time, and remade the spooling rods to be accurate in diamater to within .01mm.
That should do it.
10/17/2018 at 18:16 •
I want to use the TMC2130 as my drivers, soldered on-board. However, I don't want to consume a CS pin per chip to do it (as I don't have a lot of extra pins on the Pocketbeagle) and I can't figure out how to daisy chain them properly. (Section 4.2 of the PDF) Kind of important.
I also wanted to be able to use the end of that SPI chain to onboard a simple shift register based open drain LED driver chip. That gives me 8 more low current PWM outputs without consuming any more pins. Then, terminate that SPI chain to pins, and I could just keep going...
I don't think I'll be done in time, and actually have a working board the first time around going this route. Time to make a choice. :(
I did find a nice starting point in this ready-made cape for import as a template into Kicad.
As a bonus though, I have discovered UART based chaining with internal step generation and velocity control in the TMC5160 line though. That would be very useful for the mass of steppers I use on the M2. One serial line and power would be all that is needed to run them all. Sweet, but off-topic.
I'm going to slow down and do the Pocketbeagle board right. I'll find something else to assemble.
The consensus seems to be run the SDO to the SDI of the next driver, and bus the rest of the pins for chaining. I'm past my deadline now though, so still going to slow down and do it right.
10/17/2018 at 03:53 •
I've had some personal stuff come up which has consumed a few weeks here, and will probably consume a lot more. However, tonight I have some time.
So, I'm designing a board in Kicad for this to run on a Pocketbeagle. If I can pull this off, I should have just enough time to have the parts and board in hand.
Then, I can have this build itself for my entry video. :)
Oh.. have I mentioned I've never actually designed a board in Kicad before? I just downloaded it. I have watched a couple tutorials so this may work..
Most stuff I've done was just deadbugged or hand routed. The 'best' work I've done up to point this was routed in my head, drawn using Gimp, printed on transparencies, and transferred using some spray photoresist. That was an arcade cabinet Jamma 4 board switch.
Wish me luck..
Well that was confusing. Turns out if you install the kicad-libraries meta-package, it doesn't install the symbols, but marks them as installed. So... no parts. Hours later, removed, purged, installed each package by itself.. now downloading 360mb. That's better.
10/02/2018 at 08:11 •
Everything mostly worked, with a couple show stopping issues along the way.
The 3D printer lost extruder 4 just after power up. Don't know why yet. All of my 'ready to go' gcode which used black, in other words... all of them, couldn't be used as a result. cymKwt = Cyan, Yellow, Magenta, Black, White, Clear.
So I improvised and pulled out a 2 layer butterfly model which I could quickly slice, and then added some color codes shifting to avoid the broken extruder, with a Perl script. Probably the best accident I've had happen yet. As long as I cared not for quality.. I could print them in about 2 minutes, and so pretty much every kid got one.
The actual project got a little screwed up in transport, and I lost positioning on the A axis endstop. As the tuning is completely manual at this moment, I never got it back for the duration of the show. Just too much going on for me to concentrate on any task for more than 5 minutes at a time.
I also lost the use of the monitor which had my presentation slideshow on it. That's probably the most sucky thing, as that had my website, repository information, the project point, pretty much everything people needed to know on it.
So I just ended up explaining things a lot, and set it up looking pretty.
Me, noticing my Pi camera wasn't plugged in..
A day of handing it to people and never cleaning the optics = wow.. that's dirty... and green. Resetting the camera fixed the green part.
12mm and 8mm tapes, same LEGO mount. The top for the third block was MIA by this point. It probably ended up in the 'failures pile', from which I let people take stuff.
By this point the batteries in my wireless keyboard were dead and I couldn't locate my spares until after the end, so my exposure would only move in increments of 50 with the mouse. 7,57,107.. The light is pretty bright and the exposure range goes from 1 to 1000. It's about right at 35. Above it was set to 7.
It could have been better, but it certainly could have been worse. I'll take it.
09/29/2018 at 13:59 •
At the Milwaukee Maker Faire today and tomorrow. If you're in the area come on out!
I discovered another issue and solved it. Turns out one of my spooling rods was a slightly different diameter. During the last iteration of rebuilding a schooling rods I replace them with the carbon fiber rods I use for the uprights. I thought I remember measuring them to make sure they were all the same, but I guess not.
That was contributing to the X & Y axis not being at 90° from each other. I entered the correct value for the diameter for the oddball spooling rod and the issue went away. Yay. So I got it all tuned and perfect, and then took it back apart for transport.
So I'm bringing an unfinished project to the Maker Faire, but I got a lot of those so I guess it's okay.
Hope to see you there!
09/28/2018 at 07:34 •
Using vision has uncovered some annoying details about tuning for the C1 motion platform.
Turns out if an axis not being perfectly planar in it's travel, due to an end-stop being even slightly off, that also throws off the X-Y linearity in a big way.
So even if you have the machine tracking perfectly along Y, but you have either of the front axis end-stops off by a tiny bit, X will track an angle which is not 90 degrees from the travel you get for Y.
And so just when I have it basically perfect in all directions, I can then detect that my tripod kinematics numbers were off by like .5%. So I fix that... and it throws off the endstops and axis scale again. And then while adjusting the endstops, I break one. So I redesigned them to be a bit thicker where it needed it... reprinted them all... and then started all over.
Blew another entire day on that.
If people are going to remain sane, I'm going to have to implement visual homing.
Also, my base plywood board I picked is not stiff enough. With the camera I can easily detect the board flexing if I happen to lean on it. I'll need to glue some cross bracing in underneath.
09/27/2018 at 12:22 •
Staring at a screen of full of the tops of Lego pegs for hours... where 4 pegs fills an HD image and every spec of dirt looks like a boulder.. the slightest mis-tuning will drive you absolutely mad.
First I had to re-align the camera as the image sensor inside it was rotated by ~2 degrees. It was also tilted by about the same. I think I got that about as good as I can get it, so I glued it down. In the process I disconnected the sensor from the board a bunch of times while under power. Thankfully, I didn't destroy it.
I also had a Star Wars flashback as the camera flew down rows of mountainous Lego pegs at high speed... :)
Now.. I can tell you this.
My motion platform is reliable, aka... my errors are consistent.
And... it is still not tuned properly. :) Over 64mm, I'm off by about 1mm still.
I should be able to pick any 8mm multiple of X or Y now, and see the same image (as the Lego sheet I'm imaging, the pegs are all at 8mm spacing). They still shift a little.
Eventually, I hope to use this to automatically calibrate everything, but I'm not there yet.
Working on it..