smellsofbikes

Testing whether I can still crosspost from dreamwidth.

You want a program to start running every time the beaglebone boots.
Use systemd because it's there and it actually works okay.

In /lib/systemd/system/ make a file (programname).service
So for this example, blinker.service

It should look like this:

[Unit]
Description=blinker
ConditionPathExists=|/home/debian/programming/python

[Service]
ExecStart=/home/debian/programming/python/blinker.py
SyslogIdentifier=blinker
Restart=always

[Install]
WantedBy=multi-user.target

Then go to /etc/systemd/system/multi-user.target.wants and make a symlink to that file.

ln -s /lib/systemd/system/blinker.service /etc/systemd/system/multi-user.target.wants/blinker.service

The file itself, if it's python, must begin with #!/usr/bin/python (or wherever you have python installed) and be executable.

Then, systemctl --system daemon-reload
and systemctl start blinker.service

and your program should start immediately and also start every time you boot the beaglebone.

If your program is dependent on other services before it starts, you can add those under the [unit] group, like:
ConditionPathExists=|/home/debian/programming/python
After=network.target

That way it will wait until the network is up before running -- which, in the case of what I'm building, is nice, because it's streaming internet radio.

working on formatting and flow.
You can get (much) larger versions of most pictures by clicking on them. I will probably use larger pics in the final version of this writeup.

I have this old Triumph Spitfire. When I got it the turn signal consisted of a standard Digikey toggle switch with a ball point pen body stuck on the toggle. I bought a new replacement for the turn signal and installed that but even though it's listed as the stock replacement, it doesn't cancel when you come out of a turn.
The original owner, however, still had the old one and gave it to me.
So I took it apart to find out why it was broken.



There's a big hole burned through one part where it shorted.



The material is phenolic with copper on top, 0.031" in thickness: a standard printed circuit board, with a very specific outline.

Well, I thought it would be interesting to learn how to take a picture of something and turn that into a copy of the object. I've gone down this route in several different ways, and in this case I'm going to cover what's probably the fastest, if not the most precise or beautiful, manner for converting a scanned image to a circuit board, using Inkscape, pstoedit, and geda pcb.

I put a piece of graph paper on the scanner back, and the burnt-up circuit board on the face, and closed the scanner, so I could get a scan with a light, calibrated background. Having a calibrated background matters because inkscape (and scanning in general) doesn't always honor size information.

So, a scan.



First thing is to box it in. I draw a box larger than any part of the scan, on the graph paper, because that way I can group the scan and the box, and resize it to the size I know it is from the graph paper's scale, and know that the result is the size it should be regardless of what the scan or inkscape import process does.



At this point I zoom in on the image, add a new layer (in this case, I'm calling it 'outline') and hand-trace the outline using the bezier tool. (I'll cover various autotracing options some other time.)



In this case, I will need two layers: board outlines, and copper traces. The reason for this is that I'm going to export this in those two groups and process them individually, one as polylines and one as filled copper area, using pstoedit.

Because they have to be processed separately, and for reasons I don't fully understand, pstoedit converts them differently, I need to retain the box surrounding the features all the way into the pcb layout program, so I can align the two different groups once in pcb layout.

So, two layers. In this shot you can see I actually had three layers: inside, outside, and copper. One of the two pcb fabrication processes I use cuts to the line, and you specify inside or outside, so for that I want to identify each individually. The other fabrication process cuts right down the middle of lines, so it doesn't matter what you call them, and you have to handle cutter diameter by moving the entire line in or out. Both have their good and bad sides.



Then I click on the image somewhere, outside the bounding box I drew, and drag it off to one side, to make sure I got everything traced that I wanted, and then I delete the image.



Then, select all, group. I know that the box width is 5 inches, because I'm using Imperial graph paper, so I resize the image to 5 inches (or 5000 mils).



And I drag it down close to what Inkscape considers the picture origin, in the bottom left corner of the page, where it now looks very tiny.



And I group all the copper bits into one group.



Some tricky bits ensue. I want to export the copper traces as fills, so what comes into the pcb layout is nice solid copper chunks, not outlines. That requires closed polygons. Well, I have three closed polygons in the center, and I have them inside of a box. If I export them this way I'll get a huge solid copper box with them hidden somewhere in the middle. That is actually usable but it's not what I want. If I delete the box, I can't accurately align the copper to the pcb. So what I do is I draw four boxes, one in each corner of the bounding box.



Now I'm ready to start exporting stuff.
I choose what I do _not_ want to export, and hide it.



In this case, I do this with the inner and outer outlines and the bounding box, leaving just the copper and the four boxes I've drawn at the bounding box corners, leaving me with just the bits I want visible. I group what's left, go to 'properties', and give it a name. (Remember to hit 'set' after naming it.)



I then export that as an Encapsulated Postscript, exporting only the object/group I just named copper. That'll produce a [name].eps file.



That ends up with an image that looks like this.



I do the same thing with the outline. (This requires 'select all in all layers' then 'unhide' and then I group the outlines with the bounding box, give them a name, hide the copper and the copper fiducials in the corners, and export the outline as a second encapsulated postscript.)

So now I have two encapsulated postscript files, and it's on to pstoedit. (Which I always type as ps2edit the first time, since that's the way most unix programs are foreshortened.) I'm going to refer to the two files as lucas_copper.eps and lucas_outline.eps

You can use pstoedit to convert a postscript-like file to just about anything. I'm using the program's hooks into gEDA's pcb layout program here.

So, I open a terminal, and:
pstoedit lucas_copper.eps -f pcbfill lucas_copper.pcb
and likewise
pstoedit lucas_outline.eps -f pcb lucas_outline.pcb

Now I have two pcb files.
Open one: 'pcb lucas_outline.pcb' and you find your outline.



Go to import layout data, choose the other pcb, and it'll appear as a draggable object on top of your first pcb.



Drag them so the fiducials match in the corners, click to set it, delete all the fiducials, and drag the resulting two layers up into the top left corner as a pair.



One came in on the layer called Spare, and the other on Silk, neither of which is what we want. Save, then go into the pcb file with VI or EMACS or gedit or whatever and cut out the copper fill under the Silk layer and copy it into the Component layer, and likewise with the outlines from the Spare layer to the Outline layer. Or just swap the names in the pcb file.



So in this case, copying all the Polygon("clearpoly") entries under Layer (10, "Silk") to Layer (1, "top") will make them become copper on the top gerber. (You can leave them on 10, if you want, or delete them: you won't be using that gerber anyway.)

Then create gerbers and run it.
I milled this on my machine at home, using a 1mm router bit after an engraving bit did the copper etching.



To be fair, when I made this I did rotate it slightly in inkscape so the sides were straight lines, and a few other cleanup bits. I also put in drilled holes for the wiring connections. I also won't be using this board as my final board because the final one needs to be in thinner (0.031" rather than standard 0.062") material and should have a bunch of vias through the material to help the copper stick to the underlying substrate more tightly. But this certainly showed that I could make a complex outline fit into a complex enclosure correctly.

I wanted to put a laser scribe on the CNC mill. Since we build LED drivers at work, that seemed like a good way of controlling the laser diode. This is set up to supply about 700mA or so -- a value determined by a sub-ohm sense resistor -- to power the 120mW red laser diode module.

I put together a schematic based on the LM3402 because I know it really well since I designed the official evaluation board for it. I ran the laser diode current requirements through Webench (on the right-hand side of the page) and took the schematic it produced and ran that into geda's schematic entry tool to come up with my working schematic.

I ran that into PCB and did a layout.

I'd like to talk about that layout a little bit. This particular chip has an exposed pad on the bottom of the chip to dissipate heat from the internal switching mosfet, so the layout requires exposed ground beneath the chip. To solder this down, the best way I've found is to flow solder on the ground area so the whole area under the chip is coated, then solderwick it all back off, and add one tiny point under the chip. You put the chip down and it rocks on that point. You put the soldering iron down beside it and add a little solder, and when the chip suddenly drops down flat you know that it's down and attached, and then you can solder down the legs.

In any switching power supply you want to have the shortest traces possible on the loud and noisy nodes, most notably (in this case) the pin that switches the inductor. It's a good idea to make that just a single small copper pour with all the parts on that node as close as possible to each other so it's not spraying noise all over the place. Lots and lots of ground -- as much ground as possible, in fact, and for small power layouts you don't want thermal relief around your pads. (Which in geda/pcb means making parts with pads that have zero clearance, an experts-only move because if you don't do your layout right you'll have shorts and the program won't tell you.) I snaked the 'enable' pin around the inductor because it doesn't seem to care much about noise.

For this layout, the chip will take anything from about 7-35v as its supply, and converts that to the laser diode drive with between 80-90% efficiency.

You should have something like a 10k resistor from the enable to ground, so the chip powers up OFF rather than ON because unexpected lasers are really bad for your eyes. (In this case the controller takes care of that, but it's an extremely good idea to have it there even if the controller comes up with the IO pins grounded.) In this case the ENABLE pin is labeled DIM on the schematic.

When I run a file through a gcode processor it produces lots of Z up/down moves. I then take a look at those and use sed to change Z up moves into digital pin off moves, and Z down moves into digital pin on moves: "sed 's/Z-10.000000 F100.0/M64 P01.0/' for instance.

SO! with that done, here's a video of the result.

Need I point out that you absolutely positively must have approved laser goggles that filter the wavelength you're using? Visible lasers that are over 10mW are really bad for your eyes. Even with actual laser goggles, if I look at the burn spot for any length of time I see spots. Be careful with your eyes: you only get two and they're very delicate.

Aaaand a second try to see if I can get the dreamwidth footer in place (although I'm not sure that's necessary or even useful.)

Maaaaaan I made the best bread today. I started the yeast beforehand, and by the time I got back to them they'd gone from a teaspoon of yeast in two teaspoons of water, to completely filling a one-cup bowl. RAAR. The bread went crazy, overfilled the bread machine, and when I did the second rise on the dough in the stove, it started to creep off the edge of the baking sheet.
So we went to a soiree -- walked, as it happens -- and I carried the bread. Put it down, and nobody touched it. Two hours passed.
Finally I cut off a piece and ate it, and it was fabulous.
And ten minutes later half the loaf was gone and everyone was saying "hey, this is the best bread EVER."

People are funny.