Ed Nisley's Blog: Shop notes, electronics, firmware, machinery, 3D printing, laser cuttery, and curiosities. Contents: 100% human thinking, 0% AI slop.
Category: Science
If you measure something often enough, it becomes science
The HRECOS folks installed a display on the Walkway Over the Hudson that shows current environmental conditions at the river sampling station just north of the bridge:
HRECOS Display with internal condensation
Those two blurry white rectangles are paper charts taped to the inside of the case below the scrolling LED display, so I think they’re discovering what happens when you trap ambient air inside a sealed enclosure without dehumidification. Even if they weren’t opening the case every now and again to change the charts, diurnal pumping would pull outside air past any affordable non-hermetic seal.
That fancy electronics won’t last long under those conditions; I foresee several pounds of silica gel in their future…
Three years ago I installed a 1.5 TB WD Elements USB drive as an external backup for the “file server” in the Basement Laboratory. The log files show that the drive started spitting out “short reads” early in October, which means the rust has begun flaking off the platters.
Repeated fsck -fyv /dev/sda1 runs produce repeated failures at various spots, so it’s not in good condition:
e2fsck 1.41.14 (22-Dec-2010)
Backup-1.5TB contains a file system with errors, check forced.
Pass 1: Checking inodes, blocks, and sizes
Error reading block 97649088 (Attempt to read block from filesystem resulted in short read) while getting next inode from scan. Ignore error? yes
... snippage ...
Pass 2: Checking directory structure
Error reading block 104039017 (Attempt to read block from filesystem resulted in short read) while reading directory block. Ignore error? yes
Force rewrite? yes
Directory inode 26009985, block #26, offset 0: directory corrupted
Salvage? yes
... snippage ...
Pass 4: Checking reference counts
Inode 25903223 ref count is 41, should be 40. Fix? yes
... snippage ...
Backup-1.5TB: ***** FILE SYSTEM WAS MODIFIED *****
736471 inodes used (0.80%)
10173 non-contiguous files (1.4%)
9367 non-contiguous directories (1.3%)
# of inodes with ind/dind/tind blocks: 119655/12234/0
142996292 blocks used (39.04%)
0 bad blocks
3 large files
276772 regular files
459614 directories
0 character device files
0 block device files
0 fifos
10377447 links
76 symbolic links (72 fast symbolic links)
0 sockets
--------
11113909 files
Given that rsnapshot lashes the daily backups together with extensive hard links, so that there’s only one copy of a given file version on the drive, I don’t know what 76 symbolic links might mean.
It’s been spinning up once a day, every day, for about 40 months; call it 1200 power cycles and you’ll be close. The usual runtime is about 10 minutes, giving the poor thing barely enough time to warm up.
One data point does not a curve make.
The warranty on new WD Element drives seems to be a year; I have no idea what it was slightly over three years ago, although I’m pretty sure it wasn’t more than three years…
The various desktop boxes around here get powered up once a day, too, but I tend to replace them every few years and have never had a hard drive failure; a few system boards have crapped out, though. The boxes acting as controllers for the 3D printers and the Sherline CNC mill have a much lower duty cycle.
The improved platform for the M2 runs at 30 V, but the RAMBo board specs limit the max HBP voltage to 24 V, presumably because the 15 A ATO fuse won’t clear a high-voltage, high amperage DC short. While setting up the SSR that drives the new platform, I looked up the specs for the PSMN7R0-60YS MOSFET controlling the bed heater and … it doesn’t have a logic level gate.
The rDS spec is an impressive 6.4 mΩ max, but that’s at VGS = 10 V. The 1 mA threshold voltage VGS(th) = 4 V max, which means there’s only 1 V of headroom to turn the transistor on enough to pass upwards of 10 A.
The typical ID vs. VGS curve (Fig 6) shows 20 A at maybe 4.2 V, but the typical RDSon curve (Fig 8) shows the resistance skyrocketing for VGS under maybe 4.8 V; sliding that curve a wee bit to the right would cause a Very Bad Thing to take place.
A 20 mΩ resistance dissipates 4.5 W at 15 A, which seems rather aggressive for the small PCB copper-pour heatsink on the RAMBo board. It’s a somewhat more bearable 2 W at 10 A, but I think that’s still too high. Of course, the typical dissipation will should be much lower…
A good engineering rule of thumb is to ignore the datasheet’s “Typical” column and design using the “Minimum” or “Maximum” columns, as appropriate. When you depend on typical specs, getting “the same part” from a different supplier can provide a real education in supply-chain management.
I suspect tolerance stacking works well enough that nearly all the MOSFETs on nearly all the RAMBo boards run cool enough to survive, but I’d rather see logic-level MOSFETs in Arduino circuits where the maximum gate voltage won’t ever get above 5 V.
The last of the silica gel from one bulk can went into a mesh bag:
Silica gel beads in mesh bag
That kept a batch of fresh-baked crackers crisp during several humid days. It started out at 110 g net = 112 g gross (with bag and ties), rose to 115 g after a day, then to 117 g by the time we were done with the crackers. That’s about 5 g of water = 4.5% by weight, so those charts say the humidity should be under 10 %RH, which agrees with the fading blue dot on the humidity indicator card I dropped in the can.
When the bag gets up to 130 g = 30 %RH, then it’ll be time for a recharge… or, more likely, a refill from one of the remaining three cans.
I recently harvested the compressor from a defunct dehumidifier:
Harvested Dehumidifier Compressor
That ought to be useful in a DIY vacuum table that needs a good, low-volume pump. It seems refrigeration pumps can get down around 29 inches of mercury, so the net pressure difference is maybe 13 psi and I’d round it down to 10 psi. Typical small PCBs, say 1 x 2 inches, would have 20 to 30 pounds of downforce.
From what I read, the pump will blurp oil from the smaller outlet tube while settling down to pull a vacuum through the larger, rather discolored, inlet tube; adding a larger diameter vertical catch chamber with a splash plate to the outlet would be in order. I think a trash filter on the inlet, perhaps conjured from a defunct whole-house water filter with a 3 micron spun-fiber filter element, should keep dust and crud out of the compressor; the inlet already has a small filter / dryer (the lump next to the compressor body), but that probably won’t withstand an assault of glass-fiber-laden PCB drilling dust.
As far as the vacuum table goes, I think a 3D printed base with a machinable wax insert might be just the ticket: the base collects all the complexity, including hose fittings and a plenum under the insert, into a 3D model where it’s easy to duplicate and the cheap-and-simple wax acts as a moderately hard sacrificial platform. The base would have 10-32 holes around the outside to match the Sherline’s tooling plate. The wax insert could stand proud of the base and have holes only where they’re needed, so the base holds the insert in place mostly by vacuum.
You’d (well, I’d) like to cast the wax in place, but it melts around 240 °F = 115 °C and gets pourable around 270 °F = 132 °C, well above the point where PLA gets juicy and about where ABS gets gummy, so I think a drop-in slab makes more sense; cast it on a plate for a flat bottom surface, trim off the mold flash, and drop it in place with the flat side down. Then, with the vacuum turned on, flycut the rumpled top to get a known-flat-and-true surface, mill some vacuum channels, and drill holes to match the 3D printed holes in the plenum; all that would be a G-Code routine, of course. A simple hexagonal drilled pattern (big shallow holes for maximum clamping, little through holes into the plenum) might be a good starting point, at least for the simple, low-stress stuff I’m doing: PCBs and maybe edge-lit ersatz Nixie tubes.
You could gently heat the part to seal it to the wax, although that might risk losing the top surface alignment. Given reasonably flat PCB material, a custom channel pattern under the board might be just as good.
When the wax gets sufficiently chopped up that it can’t hold a good seal, toss it in the remelting bin, drop in a new slab, mill it to suit, and continue the mission.
If you do it right, everything’s parametric and you can generate a custom base with a custom insert by twiddling a few parameters that set the overall size of the thing; print up the base, drop in a wax plate, machine the top surface, done. You’d need two source files: OpenSCAD for the base and custom G-Code for the insert. Maybe the OpenSCAD script can generate and export a DXF-ish file that could produce the mill / drill code for the insert.
Part of the Curvelicious Cookie Cutter effort involved making the thinnest possible cutter blade wall consisting of two adjacent threads, because that’s about what the Afinia printer was producing (from a different model). My OpenSCAD code, based on an Inkscape model derived from the as-printed Afinia cutter, enlarges the cookie shape by a specific distance with a Minkowski sum; the model ultimately becomes G-Code directing the extruder nozzle around the outline.
Obviously, that required a bit of fiddling:
Robot Cutter Variations
The pink cutter on the top came from the Afinia, complete with raft. The red cutters, all with short blades to speed up the printing, came from my M2.
The printer mechanics determine the step/mm values for all four axes: X, Y, Z, and the extruder. The effective diameter of the “gear” driving the filament into the extruder seems subject to some quibbling, but setting it so the thinwall box comes out with the proper filament width seems reasonable. Given those four values, the slicing software can control the extruder speed to produce the proper volume of plastic as the XY speed varies.
The slicing software must also know the raw filament diameter, which seems to be consistent within a few percent for the filaments in my collection. Because a 1% change in filament diameter produces a 3% change in extruded volume, a few percent is about all you can tolerate; broad-tolerance filament may require sensors and adjustments that printers don’t currently offer.
There is one remaining variable, essentially a Fudge Factor, which Slic3r calls the extrusion multiplier. This seems to be a linear factor applied to the extrusion volume, so that increasing the factor proportionally increase the flow rate. Given correct step/mm settings and the measured filament diameter, you (well, I) adjust the extrusion multiplier to get the proper extrusion flow. As it turned out, the multiplier I’ve been using with the M2 worked out to 1.00, although I’ve also used 0.97 on occasion. Although I haven’t read the Slic3r source code to verify this, varying the multiplier by +3% should fudge the diameter by about +0.017 mm = 1% of the measured 1.72 mm.
Note that the Makergear-modified Marlin firmware in the M2 will produce different results, as they use a different value for the extruder gear’s effective diameter. More discussion on that is there.
Soooo, I set up the extrusion multiplier to produce parts with accurate dimensions, because that’s what I care about, and didn’t worry too much about perfect surface finish, because I don’t really care about that. Cookie cutters, however, need a completely filled surface that prevents dough from collecting inside, but have essentially no dimensional accuracy requirements.
The quartet of stumpy cutters bundled together on the left of the top photo explored the effect of changing the extrusion multiplier. I used the same STL model for all the cutters and varied only the extrusion factor, so the results depend only on the plastic flow rate and the M2’s impeccable mechanical stability.
A sharp cusp at 0.96 has a slight opening:
Robot Cutter – 0.96 extrusion multiplier
The cusp fills in at 1.10:
Robot Cutter – 1.10 extrusion multiplier
The handle surface is slightly open at 0.96:
Robot Cutter – 0.96 extrusion multiplier
And filled in at 1.10:
Robot Cutter – 1.10 extrusion multiplier
In all those cases, the measured blade thickness varied slightly, but not enough to matter in this application. I didn’t record those numbers and no longer have the models, but … you just tune for best picture.
I picked up five 12 V 40 W cartridge heaters from the usual eBay source for some extruder experiments and did a quick check to make sure they actually worked:
Cartridge heater test
The bench supply is good for 3 A, which isn’t quite enough to light them up all the way, but at 8 V they drew anywhere from 2.67 to 2.20 A, declining by about 0.1 A as they heated over the course of maybe 5 s, which is about as long as you want to run them outside of whatever they’re supposed to be heating.
Those dissipations are a bit lower than I expected; at 8 V you’d expect to see about 27 W = 2/3 * 40 W, not the 18 to 21 W I actually measured. Current & power don’t scale linearly, so I must gimmick up a larger block and make some better measurements when I get the LinuxCNC hardware set up.
The insulating tubes on the wires emerging from the cartridge, inside the main sheath, show the usual attention to detail I’ve come to know and love from eBay suppliers:
Cheap cartridge heater insulation
Ah, well, it keeps my toy budget under control…
There’s a story behind the dark vertical smudge just to the right of the cartridge. More on that in a bit.
The Smell of Molten Projects in the Morning 