Shop output board

Shop Output Board

On the shelf as of 2026-08-06 — not cancelled, just unnecessary

Commercial Pico relay boards do this job for around £22, in an enclosure, on a DIN rail. This board switches a 5–24 V signal that still has to drive external relay modules, so the honest comparison was always "custom PCB + 8 relay modules + a field PSU + a case" against "one bought board". See commercial-alternatives.md for the comparison and what to verify on a bought board.

Nothing here is wrong or unfinished. The design is complete, DRC-clean and documented, and it stays the answer if a requirement appears that dry relay contacts cannot meet — a switched 5–24 V output, or genuine galvanic isolation with the field ground leaving the board. It has still never been on a bench.

8-channel opto-isolated high-side switch board, driven by a Raspberry Pi Pico over USB-C. Takes a 5–24 V field supply and sources it to screw terminals, so shop wiring carries a robust signal instead of 3.3 V logic. Built for the sender's shop outputs — dust collection, lighting — on hosts that have no GPIO header of their own.

Nothing here has been near a bench. Read design.md before ordering anything.

Where to start

State

Part rotations in production/positions.csv are in KiCad's native convention and have not been translated for any assembler. Eighteen parts — C1, Q1–Q9, U1–U8, all of them polarized — sit at a different angle under JLCPCB's convention than under KiCad's. Settle this with whoever does the placement before an assembled run. It does not affect bare boards, which is the next order anyway.

Regenerating

Everything is generated, so a value change propagates rather than needing to be chased. Run from this directory, in this order:

python generate_schematic.py
kicad-cli sch erc --output erc.rpt --severity-error shop-output-board.kicad_sch
kicad-cli sch export netlist --output netlist.net shop-output-board.kicad_sch
python verify_netlist.py

python generate_pcb.py
kicad-cli pcb export ipcd356 --output pads.d356 shop-output-board.kicad_pcb
python verify_pcb.py
kicad-cli pcb drc --refill-zones --output drc.rpt --severity-error shop-output-board.kicad_pcb

Diagrams: python generate_channel_diagram.py, python generate_power_diagram.py. Assembly BOM: python generate_production_bom.py.

verify_pcb.py reports D9's pads swapped, and that is expected. D9 was rotated 180° by hand during routing. Rotation preserves pin-to-net mapping, so the circuit is unaffected — what it means is that the board is now hand-owned and generate_pcb.py's PADS no longer mirrors it. The barrier checks in the same script still hold, and DRC is the authority on connectivity.

PDF export is flaky when KiCad has the project open. kicad-cli sch export pdf will report Plotted to ... and Done., exit non-zero, and leave a 0-byte file. Close KiCad and Pcbnew first, or export somewhere else and move it in:

kicad-cli sch export pdf --output %TEMP%\sob.pdf shop-output-board.kicad_sch
move %TEMP%\sob.pdf shop-output-board.pdf

Always check the file size afterwards. SVG export does not have this problem.

generate_pcb.py writes placement only. --route also emits traces, but that routing is incomplete — down from 191 DRC violations to 113, still with shorts and crossings. A half-routed board is worse to inherit than an unrouted one, so it is not the default.

The two verifiers, and why they exist

verify_netlist.py asserts the connectivity in netlist.md — every channel individually, and that PICO_GND and ISO_GND share no node. ERC proves the schematic is well formed; it says nothing about whether it is the circuit anyone meant.

verify_pcb.py compares pad positions against KiCad's own IPC-D-356 export rather than against the generator's arithmetic. That distinction is not academic: the generator once used the textbook rotation matrix, KiCad's is mirrored, and every rotated part landed on the wrong side. It looked perfectly correct in the generator's own output. The optocouplers' isolated pins were sitting in the logic zone until someone opened Pcbnew and read the pad names.

Before this becomes copper

Breadboard one channel and confirm it switches cleanly at 5 V and 24 V. Everything verified so far is topology and geometry — nobody has yet watched the gate clamp work.

bench-test.md is the procedure, with the expected voltage at every node and what each failure mode looks like. The headline numbers: Vgs should be −4.8 V at 5 V and −8.2 V at 24 V, and if it ever goes past −12 V the clamp is not working and the MOSFET is being destroyed. One channel draws ~16 mA at 24 V; all eight draw ~126 mA.