Build log
Squirrel brain, written down. What's on the bench right now, noted as it happens, before any of it becomes a finished post.
- REV
- W44
- DATE
- Oct 26–Nov 1, 2026
- CHANGES
- 1 notes · 1 days
Brought up the first radio board for my wireless CNC touch probe. It flashes over USB and the radio both talks and listens. Found a bug in the example code that was sending oversize packets, so scanners threw the contents away.
- REV
- W43
- DATE
- Oct 19–25, 2026
- CHANGES
- 4 notes · 4 days
Got the PulseTen dial working on real hardware. The motor fakes the whole rotary phone feel, walls and spring return included, and every digit I dialled came through. The real friction fix was trimming the case where it rubbed the wheel.
Got USB working on the Gavel103 macropad’s CH32L103 under ChibiOS. The chip reports the wrong endpoint for setup packets, and my driver lost track of busy endpoints after sleep. I found both by lighting the board’s ten RGB LEDs, one per step of the USB handshake.
Wrote the first firmware for Barnacle, the little Ethernet node board, before any hardware exists. It speaks the same packet as my CNC controller. I also taught my timing tool to compare talking to several boards one at a time against all at once.
Worked out where my wireless touch probe’s receiver should plug into my CNC controller: into the motion microcontroller through a small keyed header, not the Raspberry Pi. Only the microcontroller can ever catch the touch the instant it happens.
- REV
- W42
- DATE
- Oct 12–18, 2026
- CHANGES
- 4 notes · 4 days
Built a desktop simulator for my bench supply’s new display firmware. The UI code runs unchanged in a window against a fake power stage, so I can try layouts and test overloads and dropped links without touching hardware.
Started CatsWhisker, a wireless touch probe for my CNC mill. Three piezo discs under the stylus seat feel the touch, a tiny radio chip sends it to the machine, and it charges wirelessly in a dock. Bench testing the piezo discs next.
Brought up the Canteen head boards. Power, the USB-C front end and programming through the USB-C port all worked on the first try, and the little 20-pin MCU runs USB with no crystal. Its internal clock came out 0.7% fast, so every board will get a quick calibration step.
Named my Snapmaker U1 “Ansel.” My Flashforge Adventurers are named after explorers, so the Snapmaker gets famous photographers. Now the brand picks the theme and I never start from a blank page.
- REV
- W41
- DATE
- Oct 5–11, 2026
- CHANGES
- 1 notes · 1 days
Got Remora running over the STM32H723’s built-in Ethernet: LinuxCNC drove six stepping joints over the network with zero lost packets in 600,000 servo periods. With the firmware proven, I wrote down everything it locks on the controller board, from the clock crystal to the Ethernet PHY straps, and I’m starting the schematic.
- REV
- W40
- DATE
- Sep 28–Oct 4, 2026
- CHANGES
- 7 notes · 4 days
Started on a macro version of the phone shop camera, with a small relay lens in front of the phone’s own. The first test print of the lens holder failed when two clamps came off the bed: the slicer had given the big parts a brim and the small ones none. Second try, with a brim on everything, went down flat.
Spent an evening on lighting tests for the channel. A warm bulb behind my head did nothing until it moved much closer, and the lesson was that the main light was drowning it out, not that the bulb was too weak.
Merged the solder paste stencils for six boards into one sheet, so a single stencil order covers all of them. A script checks every one of the 981 openings against each board’s own export, because a stencil that is quietly missing a pad is worse than no stencil.
Simulated the knob board’s power stage and it turned up a real problem. Plugging it in is fine, no soft-start needed. But flick the knob and stop it hard, and the motor pushes enough energy back into the supply to trip the chip’s overvoltage protection and reset it. The fix is firmware: watch the supply voltage and change how it brakes when the voltage climbs. Written, not yet tried on hardware.
Finished the layout for the smart knob’s motor controller board. Laid out one of the three motor drive stages by hand, copied it twice, and every check comes back clean. It’s ready for the fab.
Put out the channel’s first Short: a teleprompter built from a five dollar picture frame. The hardware went up alongside it, CAD, laser patterns and printable parts, so anyone can build one.
Shot the first real video with the homemade teleprompter, and it reads far easier than anything I’ve tried before. Added a pace mode that scrolls on its own but still jumps when I call out a section, and started planning a swap from laptop to tablet to shrink the rig.
- REV
- W39
- DATE
- Sep 21–27, 2026
- CHANGES
- 10 notes · 6 days
Got the new wireless lav talking to the phone I shoot with, which was a gamble since the manufacturer’s compatibility list stops one model short of mine. It works. The interesting part was proving it: I had planned to tap the mic capsule and compare loudness against the phone’s own mic, but both taps hit full scale, so the comparison settled nothing. Android will just name the active input device if you ask the audio layer directly, which is a much better answer than guessing from a waveform.
Spent the evening getting a shop camera and a lav mic working together for the first time. Most of the night went to chasing an audio problem that turned out to be microphone placement rather than gain, which is obvious in hindsight and was not obvious at all while it was happening. Came out of it with about 45 seconds of usable footage and a rig where every setting is written down.
Gave the channel intro its sound and put it on YouTube. Every sound comes out of the same Python script as the picture: the gear’s teeth tick as it rolls, stepper motors sing at the real step rate of each move on screen, and the T lands with a slam. On headphones the ticks cross left to right and the steppers right to left.
Built the channel intro. The gear rolls in and hunts like a badly tuned PID loop before it locks, then a gantry arm slams the T home. The same script that draws the logo generates every frame, and each episode gets its own drawing number.
Killed a good idea this week. Had a way to gut a phone down to just its main board and tap the internal ribbon for USB and audio, which would have made for a much tidier build. It needs a flex soldered to 0.35 mm pitch pads. I can solder that. Someone putting one together on their kitchen table can’t, and that’s who it’s for. The rule I wrote down: every attach point has to be a plug, or it isn’t a product.
Redesigned my battery fan project so one USB-C charger can run the fan and a Pinecil together, and moved to a four-cell pack so the iron gets its full 60 W on battery too. Placed all the parts in KiCad and wrote up a pin-by-pin reference for capture. Also worked out a fume-extractor mode: flip the unit and a magnetic filter goes on the intake.
Went through the knob board assigning part numbers and footprints. Found a through-hole footprint on fifteen surface-mount resistors, and two part numbers that had quietly stopped matching their parts after the values changed. None of that trips any automated check, which is the argument for reading the parts list rather than trusting the tools. Also drew a connector footprint from the manufacturer’s drawing, and got the courtyard smaller than the pads on the first try. Rendering it made that obvious in a second.
Pulled the screen off the phone that’s becoming a wired shop camera and it came up fine headless, streaming about 50 seconds after power-on with nothing plugged in but Ethernet. Found the ugly one while poking at it: if the camera drops out, the app keeps sending audio, so the recorder shows a healthy live feed with no picture. Needs a watchdog.
Spent the day on the motor controller for the rotary-phone numpad. The interesting problem turned out to be mundane: a desktop USB port only offers five volts, and this controller cannot run on five volts at all, for a reason buried in the chip’s internal regulator rather than anything to do with the motor. So it gets two USB sockets instead of one. Most of the day was the motor drive stage itself, which is now drawn and checked.
Reworked my schematic bench book so it opens as tabs instead of one endless page: overview and to-do, rails, parts, pinouts, reference. The parts list builds itself from the document, so clicking a part jumps to its pinout and its datasheet at the right page. The copy I share now opens with a short explainer of what a bench book even is.
- REV
- W38
- DATE
- Sep 14–20, 2026
- CHANGES
- 17 notes · 7 days
Started working out how to get my old South Bend lathe turning again. The plan is a salvaged treadmill motor driving the countershaft, which adds continuous variable speed on top of the cone pulley steps the machine already has. Lined up a free working treadmill to pull the motor and controller out of. Also put together a small library of old patternmaking and foundry books, since casting near net blanks to finish on the mill is where I want to end up eventually.
Spent the evening on the power over ethernet adapter board, working out whether to put a small network switch on it so the onboard microcontroller could stream a backup audio track by itself. Priced it out properly and it added about 40 percent to the parts cost plus a real chunk of the power budget, so it’s out. The phone already records usable audio alongside the video for free, which turned out to be most of the answer.
Kept going on the PoE adapter for the phone camera. Added an 18650 so it can move between filming stations without rebooting, swapped in a proper PoE+ controller, and tracked down public reference designs for the USB Ethernet chip. Also built a bench book for it: every part’s pinout next to its datasheet, one click from the page that proves it.
Set up a bench book for the haptic knob’s core board: every datasheet saved in the repo, a pin-by-pin reference that links straight to the right datasheet page, and the chip vendor’s eval-board circuits cropped beside notes on what to copy. Turned the whole setup into a reusable tool for the next board.
Spent the evening chasing the architecture for a PoE adapter that turns an old phone into a wired network camera. I nearly put an Ethernet switch on the board so a controller could have its own network port. It ended up much simpler: a tiny USB hub lets the phone itself carry the controller’s traffic.
Gave the build log a proper layout. Each week is now a row from a drawing’s revision table, each day gets one date you can link straight to, and a little orange gear splits up days with more than one entry. Tried a solid orange stripe and a square-root tick along the way and binned both.
Made the build log update as things happen instead of in a weekly batch. When a piece of work wraps up, it gets a short entry here the same day, so anyone curious can check what’s on the bench right now. Friday’s roundup just points back at it.
Renamed the portable bench power supply I have been designing. It was called PocketPD, and a small US outfit already sells a product under that exact name, in the same category, shipping right now. Anyone who found mine would have assumed it was a knock-off of theirs, so the name had to go before any of it is public. My first instinct for a replacement turned out to be another product from the same company, which is a good argument for checking rather than trusting a gut feel. Worked through a long list against what is already on the market and ruled out six more: one too close to another bench supply on sale, one a live project elsewhere, one sharing a prefix with a well known USB power brand, one belonging to a bicycle company. Settled on FieldPD, as in fieldwork, the supply that leaves the bench. PD keeps its double meaning, the standard the charger speaks on the way in and the power it hands out the other side. The rename reaches into the circuit board files themselves, not just the documents, so I re-ran the electrical and manufacturing checks afterwards and confirmed the board is byte for byte the same design it was before.
Turned the Tangent Heavy Industries website into a proper blog, built with a small off-the-shelf blog tool instead of anything homemade. Posts can have code, photos that are shrunk automatically for the web, tags and a feed. Comments come from Bluesky: each post links to a Bluesky post, and the replies there show up under the article. Visitors don’t load anything from Bluesky unless they press a button for the newest replies. It is built and tested, but not live yet.
Put the Tangent Heavy Industries blog live and tied everything together: the website, blog, YouTube channel and Bluesky account now all link to each other, and the Bluesky account uses the website’s address as its name. Also got the longer tangentheavyindustries.com address working, so it now forwards to the main site.
Set up a Bluesky account for Tangent Heavy Industries, now under the channel’s own web address as its username (@tangentheavy.com), to go with the YouTube channel: name, bio, a wide version of the drawing-sheet banner, and follows for a handful of hardware and maker accounts (Hackaday, Adafruit, KiCad, OSH Park and others) picked for being active and on-topic. Also drafted a first introduction post and a plan for using the account to help the channel grow.
Gave each drive bay on the NAS backplane a second light: a red fault/locate LED next to the existing activity light. The drive controller card already sends a ‘this drive has failed’ and a ‘find this drive’ signal for every bay, and the board’s little microcontroller was quietly ignoring both; four of its spare pins now drive the new lights, while leaving the pins that would suit a temperature sensor or fan control free for later.
Decided the Turnip lathe’s cast bed will be a slant bed, the way production CNC lathes are built, rather than a flat one. Since the bed is poured, a forty-five degree face costs one angled board in the mold, and it means chips fall into a tray instead of onto the rails. Also confirmed a small bare lathe spindle from a US supplier as the first-version spindle and put it in the cart.
Worked out how the NAS drive backplane should be powered and started, and shopped for cheaper parts. Four hard drives all starting their motors at once pull about eight amps, which is more than the small power plug on the board is rated for, so it will switch to the old-style chunky 4-pin PC power plug, and the drives will be allowed to start one at a time instead of all together. Also went looking for a cheaper version of the expensive nine-dollar drive-cable socket: the only true equivalent is a surface-mount part at about a quarter of the price, harder to solder by hand, while the look-alikes turned out to be the older, incompatible cable standard.
The homemade NAS drive backplane passed its first real hardware test. A three-terabyte drive plugged into the board came up at the full six-gigabit speed and read eight and a half gigabytes at a hundred and sixty megabytes a second with not a single error, which is as fast as that drive goes. That also proves the workaround for the board’s known wiring mistake actually works in practice rather than just on paper: a particular kind of adapter cable, plugged in backwards, cancels the error out. Most of the evening went to a false trail, though. The test PC turns out not to notice drives being plugged in while it is running, and worse, the command meant to make it look again reports ‘nothing there’ rather than admitting it did not really check. So a perfectly good board looked dead twice, and at one point I became convinced the PC’s drive sockets had been switched off in its setup screen, which was wrong. The fix was to stop trusting anything but a full restart. Only two of the four drive sockets are soldered onto the board so far, so that is the limit of what can be tested, but the result is good enough to justify ordering the corrected version of the board.
Checked the NAS backplane’s main cable connector against the official industry pinout, and it found a board-killer. The connector is wired the way the controller card’s end of the cable is wired, not the drive-backplane end. The standard cables cross the wires over, so as drawn every drive’s send and receive lines would meet the controller’s send and receive lines head-on and no drive would ever show up, and the little activity-light signals land on the wrong pins too. The part’s physical footprint itself turned out to be exactly right, so the fix is rewiring which pin goes where, not redesigning the part.
Turned out there’s a proper CNC bed mill and a Bridgeport I can use after hours at work, which reshapes the Turnip lathe plan. The headstock can now be a solid steel block with its bearing seats bored properly on a real machine, instead of a bought spare part or a bolted-together box. The slide plates and the bed’s steel top plate become evening jobs on that mill too. The only new limit is machine time, so everything gets designed for as few setups as possible.
- DWG
- THI-200
- INDEX
- BUILD LOG
- UPDATED
- AUTOMATICALLY
- SHEET
- 1 OF ∞