Posts Tagged Improvements
Stripping the components from the back of a “5 W” COB LED gets it ready for action:
Jumpering the pads with nickel strips harvested from various NiMH and lithium cells restores the original contact pads to service:
A bit of bandsaw artistry produced a replacement for the OEM LED bracket:
The epxoy bonding the LED to the heatsink happens a few paragraphs ahead in this story, but the view justifies it. The 2 mm hole just to the right of the 3 mm SHCS aligns the heatsink to a pin in the machine’s frame, ensuring it doesn’t twist around under vibration.
The view from below (in a mirror on the machine’s bed) shows the COB LED just barely fits in the opening:
I screwed the bare heatsink into the Juki, applied double-stick tape to the COB LED, aligned LED with opening, and stuck it in place. Back in the shop, I traced around the LED to figure out what part of the heatsink needed removing, introduced it to Mr Disk Sander, and contoured it to match the LED.
Clean everything with denatured alcohol, put the heatsink on a glass plate, and clamp it to the height gauge:
Butter up the LED PCB with JB Kwik epoxy, having previously masked the contact pads (with masking tape!) to prevent oopsies:
Raise the height gauge, align LED & heatsink, lower height gauge to squish epoxy into an even layer, raise slightly to ensure the aluminum heatsink doesn’t short the nickel strips, and fast forward a few hours:
Peel off the masking tape and solder a cable in place:
The transparent doodad around the cable is a PET clamp snipped from a consumer electronics clamshell package, then punched and folded to suit. It didn’t work particularly well, so more rummaging will be required.
Foreshadowing: all this went swimmingly and looks pretty good (in a techie sort of way), but I’ve been running a nasty cold (stipulated: there being no pleasant colds). Building While Stupid is never a good idea, as the part of your brain in charge of telling you you’re about to do something catastrophically wrong is the first thing to go.
More to come …
A bipolar transistor version of the astable multivibrator with a yellow Pirhana LED required absurdly large capacitors for a reasonable blink rate and, seeing as how I need a demo circuit for Show-n-Tells, it seemed a good candidate for a faster blink. I replaced a 100 µF cap with the 22 µF electrolytic cap from the other side, installed a 2 µF cap (which, judging from the lack of polarity indicators, may be a film cap) from the Squidwrench
junk heap parts bin in its place, and hitched up the DSO150 because I brought it along:
Worked the first time and caught it in mid-blink! [grin]
The DSO150’s triggering remains a mystery, as it seems difficult to get a stable trace from a perfectly reasonable waveform. The scope didn’t trigger well on the astable’s original seconds-long pulses, perhaps due to a DC blocking cap in the triggering circuitry (whatever it may look like), but this waveform should be dead simple.
Having gained a visceral understanding of why MOSFET astables produce better battery life, this bipolar transistor design is just a milestone along the way.
For the record, Juki thinks this SMD LED provides enough light around the needle of Mary’s TL-2010Q sewing machine:
A detailed look at the active ingredient:
The 30 Ω resistor drops exactly 2.0 V, so the white LED runs at 67 mA.
We think it’s a glowworm, compared to the COB LED bar across the back of the arm:
Changing from 2.0 Ω to 2.2 Ω produces a noticeable decrease in light, so 10% steps around 2 Ω seem to be about the right increment. The COB LED strips claim 6 W at 12 V = 500 mA nominal, so they’re running well under the spec.
The general idea is to replace this:
Thereby solving two problems:
- Pitifully small battery capacity
- Wobbly camera support
The battery is an Anker PowerCore 13000 Power Bank plugged into the M20’s USB port. Given that SJCAM’s 1 A·h batteries barely lasted for a typical hour of riding, the 13 A·h PowerCore will definitely outlast my legs. The four blue dots just ahead of the strap around the battery show it’s fully charged and the blue light glowing through the case around the M20 indicates it’s turned on.
The solid model has four parts:
Which, as always, incorporates improvements based on the actual hardware on the bike.
A strap-and-buckle belt harvested from a defunct water pack holds the battery into the cradle and the cradle onto the rack, with a fuzzy velcro strip stuck to the bottom to prevent sliding:
The shell around the camera is basically a box minus the camera:
The shell builds as three separate slabs, with the center section having cutouts ahead of the camera’s projections to let it slide into place:
The new shell version is 30.5 mm thick, so a 40 mm screw will stick out maybe 5 mm beyond the nylon locknut. I trust the screws will get lost in the visual noise of the bike.
A peg sticking out behind the USB jack anchors the cable in place:
The front slab and center top have curves matching the M20 case:
The camera model has a tidy presentation option:
And an ugly option to knock the protruberances out of the shell:
The square-ish post on the base fits into an angled socket in the clamp around the seat rail:
The numbers correspond to the “Look Angle” of the socket pointing the camera toward overtaking traffic. The -20° in the first clamp shows a bit too much rack:
It may not matter, though, as sometimes you want to remember what’s on the right:
FWIW, the track veering off onto the grass came from a fat-tire bike a few days earlier. Most of the rail trail had cleared by the time we tried it, with some ice and snow in rock cuts and shaded areas.
Contrary to the first picture, I later remounted the camera under the seat rail with its top side downward. The M20 has a “rotate video” mode for exactly that situation, which I forgot to turn off in the fancy new mount, so I rotated the pix afterward.
A 3 mm screw extends upward through the hole in the socket to meet a threaded brass insert epoxied into the shell base, as shown in the uglified M20 model. Despite appearances, the hole is perpendicular to both the socket and the shell, so you can tweak the Look Angle without reprinting the shell.
All in all, the mount works well. We await better riding weather …
The OpenSCAD source code as a GitHub Gist:
A bag of G4 COB LEDs arrived from halfway around the planet:
Those are “5 W” and “4 W” cool white modules, respectively, with another set of 4 W warm white looking pretty much the same. There’s no provision for heatsinking, which makes the wattage seem suspect; halogen G4 bulbs run around 20 W, for whatever that’s worth.
The silicone overlay becomes nearly transparent when seen through an ordinary desktop document scanner:
Highlighting the PCB copper pours shows 18 LEDs arranged in three series groups of six LEDs in parallel:
The “smart IC” touted in the writeup turns out to be a bridge rectifier for AC or DC power:
The SMD resistors on all 15 modules measure 27.6 Ω, more or less, and seem randomly oriented face-up or face-down. I assume that one is face-down; maybe it’s just unlabeled on both sides.
Back of the envelope: there’s no way it will dissipate 5 W. The bridge drops 1.4 V = 2×0.7, the LEDs drop maybe 9 V, leaving the resistor with 1.6 V to pass all of 60 mA, so call it 700 mW.
With 12 VDC applied to the pins, the bridge drops 1.6 V, the LEDs 8.2 V, and the resistor 2.2 V, with 80 mA through the whole affair dissipating just under 1 W.
Cranking the supply until the current hits 200 mA puts 15.7 V across the pins for a total dissipation of 3.1 W, burning 1.7 W in the LEDs and 1.1 W in the resistor.
Cranking the supply to 21.3 V drives 410 mA, dissipates just under 9 W total, produces a curl of rosin smoke from the PCB, and maybe delaminates the silicone around some of the LEDs.
OK, now I have a crash test dummy.
Given complete control over the application, I’ll strip everything off the PCB and bond it to a heatsink of some sort. With 6 LEDs in parallel, 120 mA (6 × 20 mA) total current might be reasonable and 200 mA (6 × 30 mA) probably won’t kill the things outright. Plus, I have spares.
An external 18 Ω resistor should suffice. Perhaps a pair of 6 Ω SMD resistors on the PCB, with fine-tuning through an external resistor. Call it 250 mW apiece: don’t use little bitty SMD resistors.
I’ve been using YAGV (Yet Another G-Code Viewer) as a quick command-line Guilloché visualizer, even though it’s really intended for 3D printing previews:
Oddly (for a command-line program), it (seems to) lack any obvious keyboard shortcut to bail out; none of my usual finger macros work.
A quick hack to the main
/usr/share/yagv/yagv file makes Ctrl-Q bail out, thusly:
diff yagv /usr/share/yagv/yagv 18a19 > import sys 364a366,367 > if symbol==pyglet.window.key.Q and modifiers & pyglet.window.key.MOD_CTRL: > sys.exit()
I tacked the code onto an existing issue, but yagv may be a defunct project. Tweaking the source works for me.
The Ubuntu 18.04 LTS repo has what claims to be version 0.4, but the yagv GitHub repository (also claiming to be 0.4) includes code ignoring G-Code comments. Best to build the files from source (which, being Python, they already are), then add my Ctrl-Q hack, because my GCMC Guilloché generator adds plenty of comments.