v10.3, 2026-10-19. The v10.2a guides with every decision made through 2026-10-19 written in. What is still open is in Pending Amendments v10.4, the only open list. Nobotic USV // CD25D Conversion. What this document is the home for. What is inside the two compute boxes (secs 2 to 4), the watchdogs (sec 4.3), the central node's relay map (sec 11) and command paths (secs 10 and 12), and electronics cooling (secs 2.3, 3.5, 3.6 and 4.4). It is also the final word on the two-island power layout, the isolation bridges between the islands, the lightning watch and the surge rules. What lives elsewhere. Every box, its holes and wiring, the sensors, power distribution (bus, fuses, breakers), the emergency bank, nav lights and comms hardware: Electronics and Wiring Plan. Every setting, limit, threshold, timer and address: System Settings Sheet. Parts and prices: Bill of Materials. How a node works inside, with drawings: Node Boxes Explained. Reasons for decisions: Design Decisions and Rationale. Terms used throughout.
| Term | Plain meaning |
|---|---|
| Island | A group of gear fed from one battery, with no wire path to the other group |
| Galvanically isolated | Two circuits pass information with no metal path between them. The signal crosses as light, magnetism, or an electric field |
| Node | A small team-built box with a microcontroller in it |
| Discrete | A single on-or-off wire carrying one bit, not data |
| PoE | Power over Ethernet: the network cable carries the device's power and its data |
| PoE class | The power a PoE device asks for when plugged in (802.3af up to 15.4 W, 802.3at up to 30 W). The switch holds that much aside, used or not |
| Web API | The switch's own documented commands, over the network. A Pi program uses them to switch a port on or off |
| SNMP | A standard way to read a network device's state. On the TSW202 it is read-only: good for checking, not for control |
| H-bridge | A four-switch circuit that runs a DC motor in either direction |
| RS-232, RS-485 | Two wired serial standards most science instruments speak. RS-232 is one device per cable; RS-485 is a shared two-wire line |
| Normally closed contact | A relay contact that is closed while the relay's coil has no power. Powering the coil opens it. A normally open contact is the reverse |
| Sacrificial | Allowed to die in a lightning strike. Nothing the boat needs to survive depends on it |
Both boxes mount inboard, aft of the settee, on existing structure, a few feet apart, with no hull holes, as far from both lightning side straps as space allows. Cable entries go on the bottom face, with drip loops (a downward sag, so water drops off instead of following the cable in) hanging 3 to 4 in below the box, below the connectors at a long heel. Pi A, glued to the hull by its back face, is the exception (sec 2.3).
Through-bolted by its back wall to the 3/16 to 1/4 in solid glass panel aft of the settee. No plywood pad. Pi B, under 5 W, needs no hull cooling.
The bilge pumps mount on the other face of the same panel, on rubber feet, biased toward the Pi B side (the Gulpers thump). Lay out both faces on one template so a pump foot never lands on a compute-box nyloc.
Pending 466, decided in Pending, guides not rewritten. 466. Motor coolant loop DROPPED; the jacket is filled three quarters with coolant and plugged (Mike, 2026-10-29: "yes, drop the loop and fill a quarter air"). The same bed glues the Pi A box, the network switch box and the VESC A box floor (the motor cooler plate is gone: item 466). The four-boxes sentence in the paragraph below is the unedited guide.
The Pi A box is glued to the hull skin below the waterline, aft of the settee, as far from both straps and bar runs as space allows, for cooling, so its wall stays near sea temperature whatever the cabin does (why: Design Decisions R-8.1C). It also sits at least 2 ft from the main battery and motor cables where space allows, for its two IMUs (sec 4.1; 2 ft is a guess at a safe distance). The back face (the base, opposite the lid) goes flat to the hull; entries and the bond stud go on the other faces, with drip loops. Four boxes are glued to the hull for cooling: Pi A, the motor cooler plate, the motor controller plate and the network switch box. The network switch box is a third CXCESNS aluminum box (sec 3.1), glued by its back face straight to the hull like Pi A, but wired to nothing (Electronics and Wiring Plan A6a). It holds only the TSW202 switch and its supply, both clipped to a DIN rail on a long side wall, not glued: the floor is too shallow, since the TSW202 is 4.8 in deep (122 mm, datasheet) and the box only 3.9 in. Their heat, about 5 to 15 W in all, runs rail to box wall to the glued face to the hull. The motor controller reaches the hull through an aluminum plate in its box floor (sec 4.4). Measure at the boat first: the hull laminate thickness at each spot and the size of the motor controller plate.
The bed carries heat: thicken the epoxy with a two-grit aluminum-oxide filler, coarse #46 or #80 plus fine #220 or #320, about 2:1, loaded to half the mix, after wetting both faces with plain epoxy. Never 3M 5200, never 404 and never microballoons under a heat path. On any hull-glued box, glue the flat back face, not a side wall, and put the glands in the side walls near the lid. The CM5 heat block and its gap pad ride this back wall (sec 3.5). Open: the grit. The coarse grit is struck from the buy list and the fine grit bought is #120, not the #220 or #320 above; either the mix changes or the grit on hand is proved on the bench before any box is glued.
The cold wall is on purpose, so the box is built for the water that forms on it:
Desiccant is a first-setup aid, not the protection. Open: whether Pi A also needs a desiccant pack once its back sits on the cold hull; the hull laminate measurements settle it.
Two identical CXCESNS aluminum junction boxes, on one drilling template. A third of the same box, ordered 2026-10-03, holds the network switch at the stern (Electronics and Wiring Plan A6a); it has its own hole schedule.
| Item | Spec |
|---|---|
| Size | 11.8 x 8.3 x 3.9 in |
| Sealing | IP66, silicone gasket |
| Price | ~$66 each, Amazon |
The four lid screws tie lid to body well enough at lightning frequencies (below about 30 MHz), so the aluminum shields the wiring inside. Scrape to bare metal under the lid screws and at the bond stud (where the lightning wire lands). Wiring to the lightning copper: the Pi A box gets one 6 AWG bond spur (a short wire to the lightning ground network) under the substantial metal rule of the Lightning Ground System Guide. The Pi B box gets none (sec 5.2).
Inside each box, a 1/8 in (3 mm) G10/FR-4 plate (stiff fiberglass sheet, an electrical insulator) carries all the boards.
Each CM5 stack is bolted through standoffs at every level: carrier and CAN HAT.
The heat path: SoC (the main processor chip), heat pad, flat aluminum block, gap pad, box wall.
After first power-up, run the CM5 hard for ten minutes. The wall over it should feel warm.
About 30 W in the Pi A box (SSD and Hailo), under 5 W in the Pi B box, leaving through a painted shell that runs 27 to 36 F (15 to 20 C) above cabin air.
| Cabin air | Interior result | Verdict |
|---|---|---|
| 77 F (25 C) | SSD near 117 F, CM5 near 136 F | Comfortable |
| 113 to 122 F (45 to 50 C, sealed boat, tropical sun) | 153 to 162 F | Above the SSD comfort line, and near the CM5 throttle at 185 F (85 C) |
The second row is why Pi A is glued to the hull. Pass/fail is the compute-box heat test in Build Guide sec 11.1: box interior at or below 122 F (50 C), with a 122 F simulated cabin and the hull spot at sea temperature.
Before anything else, hold the upright carrier plus CAN HAT against the box width and check the bracket footprint.
Pi A (every hole: Electronics and Wiring Plan A1): a CM5 (Raspberry Pi Compute Module 5) 8 GB / 32 GB on the carrier on hand, booting from its own onboard eMMC memory chip, its clock capped at about 2.0 GHz in config.txt, below the CM5's 2.4 GHz top, so it runs cooler and lasts longer (System Settings Sheet sec 10), with:
USB rule. Every USB plug in the box (the hub lead and the CAN adapter lead) hangs no weight: clamp each cable to the plate within 2 in of the plug, and stake the plug body with a dab of removable adhesive. Zip ties are fine for cables; board stacks use their screws. The hub takes a detachable USB lead, not a captive cable, so its hub end is clamped and staked the same way in the hub box. At the hub, zip ties also hold every plug in its port (Mike, 2026-10-19): the hub sits on standoffs (short spacer posts; the M3 nylon kit on hand) so the ties pass under it, and each tie wraps round the hub body and across the plug grips. The hub's screw-lock holes (threaded holes beside the ports, for cables with locking screws) are not used, and no screw-lock cables are bought. Watchdog WD-A: Arduino Nano Every (on hand), main island, on its own 5 V from a Pololu D36V6F5 (sec 4.3). It watches Pi A's heartbeat over its own UART (a simple serial line) and hard-cycles Pi A's 24 V feed through its relay after the set silence (System Settings Sheet sec 10). It sends the "main island alive" discrete to WD-B through an optocoupler (the signal crosses a gap as light, so no wire crosses). It holds the VESC kill line, also through an optocoupler, and asserts it on heartbeat loss only (the line lands on the VESC's servo plug PPM pin, active low: System Settings Sheet sec 2); it does not listen to CAN, so the kill path sits on no bus. The alive line and the kill line use HiLetgo PC817 4-channel optocoupler boards (on hand; input 3.6 to 30 V, so the 5 V Nano drives them). The default-ON rule and the restart rule: sec 4.3. The safe-mode wire. One wire inside the Pi A box from a WD-A output pin to Pi A header pin 37 (GPIO26), through a resistor pair so the Pi sees about 3 V, not 5 V: the 15k in the wire, the 22k from pin 37 to ground (swapped, the pin would see only about 2 V). WD-A sets it before the restart that starts safe mode (sec 4.3). A dead WD-A, a cut wire or no wire reads as a normal start. No new box hole. Pi A's motor kill pin, when assigned, must not use pin 37. GPS A: a hiBCTR NEO-7M module (bought 2026-09-09 in a 4-pack; two fly, two are Boat 2 stock): TTL serial at 3.3 to 5 V, SMA antenna, single-band GPS with GLONASS, good to a few yards, position only. Only the antenna coax crosses the box wall. Its wiring: the board has a soldered 5-pin male header and no bare holes, so one 4-pin Dupont housing (the RLECS 4-pin female leads, bought 2026-10-06) goes on VCC, GND, TX and RX (PPS unused; check the board's pin order first). Seat it fully, run a zip tie lengthwise over the plug and around the board so it cannot back off, tie the cable to the plate within about 1 in, and add a dab of electronics-safe silicone or hot glue. Push-on pins are banned only in release logic (Electronics and Wiring Plan sec 4), and the boat wiring standard's solder rule (ABYC E-11) does not apply, since the Dupont pins are crimped. Not done: unsoldering the header to fit screw terminals (more bench work for a small gain). IMU A1 and IMU A2: two BNO085s on the same I2C pair (I2C: a two-wire data link between boards), A2 set to the BNO085's second address, one at each end of the plate, far from the converter as space allows (the box's distance from the battery and motor cables: sec 2.3). Each is mounted flat and square to the plate, rigidly, with an arrow on the plate marking the board's forward axis, and has its own heading offset and roll and pitch level offsets (System Settings Sheet sec 7). When A1 and A2 disagree past the set limit (System Settings Sheet sec 7), Pi A falls back to the GPS track and gyro, flags it and sends an alarm home. Heading comes from the three BNO085 IMUs (bought 2026-09-08; the third, IMU B, is in the Pi B box) and the GPS track; the ZED-F9P moving-base heading pair and its UART are deleted. Not fitted: an IMU elsewhere in the boat. It would only add protection against magnetic trouble near the Pi A box, and costs a long I2C run (unreliable past about 3 ft), a new box hole and a small sealed box. Storage, solid state only: one 4 TB NVMe SSD (WD Blue 4 TB, bought 2026-09-15, due 2026-09-28) in the carrier, holding the readings log and the sound archive. Sound records at 48 kHz/16-bit (or FLAC, a science call), about 8.3 GB a day continuous. Recorded plain, the 4 TB drive holds about 16 months, less the readings log; 18 months (about 4.5 TB plain) needs FLAC, which takes less room; how much less on sea sound is not known. Neither the drive nor the Hailo rides USB. Not fitted: an 8 TB pair, spinning drives, a mirror, or a 192 kHz full-band archive on two spinning drives (do not re-propose: hot drives in a sealed box were the heat problem). Losing the archive loses science audio, not the boat. Logging: SQLite on the SSD for the structured readings (10 Hz IMU, GPS and VESC; 1 Hz environment; camera frames on events), in WAL mode, with a daily checkpoint and a monthly compaction. Compressed batches go home over Starlink; the fallback is the last 24 hr as a position-only line by Iridium SBD (short burst data, Iridium's small-message service). Power section: one small branch on its own auto-reset breaker at the busbar (Electronics and Wiring Plan sec 7.6) feeds it, wired in this order: breaker, TVS surge diode, INA226 current monitor, converter, carrier. WD-A's relay contact sits in the 24 V feed ahead of the converter (sec 4.3). Terminal strips; no MEGA fuse, no blade-fuse block. The converter is a Mean Well DDR-60L-5, for Pi A only: 5 V 12 A, 18 to 75 V in, isolated (its 5 V side has no wire connection to its 24 V side, so a failed converter cannot put 24 V on the Pi), on a short DIN rail by the terminal strips at the far end of the plate from the IMUs. Output set to 5.1 V; output negative tied to input negative. Its 5 V also feeds the USB hub, on the 3 A fuse (above). WD-A's own D36V6F5 takes its 24 V ahead of the relay (sec 4.3).
Pi B (every hole: Electronics and Wiring Plan A2): CM5 on its carrier with its own Waveshare 2-CH CAN FD HAT, stack staked, fed from the emergency bank only, through its own 5 A auto-reset breaker on the backup battery board beside the Dyness (no fuse in the box; Electronics and Wiring Plan A10; drawing Backup_Battery_Board_Layout_2026-10-15), on 14/2 into entry 1, then a terminal strip and its own potted Klnuoxj 24 to 5 V 5 A converter and TVS, referenced to the island's floating negative, with WD-B's relay contact in the 24 V feed ahead of the converter (sec 4.3). A 256 GB NVMe SSD in the carrier's M-key slot (CM5 IO Board Rev 2) holds its synced state copy. No secure element chip: the key is a file on the SSD. Watchdog WD-B: Arduino Nano Every, emergency bank, on its own 5 V from a Pololu D36V6F5 (sec 4.3). It holds Pi B's 24 V feed through its relay, runs the wake schedule (System Settings Sheet sec 10), and owns the whole lightning role: it reads the AS3935 detector (the Playing With Fusion SEN-39003 board, run at 5 V from WD-B's own 5 V, so no level shifter) over a short 5-core lead (5 V, ground, SDA, SCL, interrupt) from the detector mini-box within 3 ft of the box (sec 13.1), keeps the strike log, and promotes Pi B to full-time command if the "main island alive" discrete from WD-A goes quiet. WD-B never cuts Pi B's power for lightning (sec 13). GPS B: the second NEO-7M, emergency powered, its coax entering this box, wired the same way as GPS A (sec 4.1), so the beacon role always has live position. IMU B: the third BNO085, mounted and set like IMU A1 and A2: flat and square, a forward arrow on the plate, its own heading offset and roll and pitch level offsets (System Settings Sheet sec 7). Iridium B: linked to Pi B the same way as Iridium A (sec 4.1): its own cut RockBLOCK cable, the modem on the Iridium B box's Mean Well, the USB plug into a second DSD TECH SH-G01A USB isolator (one of two bought 2026-10-16) in a Pi B USB port. The cable enters the Pi B box at entry 7 in a split gland. Iridium B's 24 V comes from its own 5 A auto-reset breaker on the backup battery board (Electronics and Wiring Plan A10). Inter-box link: only the main-island-alive discrete from WD-A, through an optocoupler (a HiLetgo PC817 board). No heading UART between the boxes. Not fitted: an ADuM1201 isolator for either Iridium line; each modem reaches its Pi over USB through a USB isolator (secs 4.1 and 4.2).
Two watchdogs, one per island, hardware only. Both watchdog rail switches, WD-A and WD-B, default ON: the switch part conducts while the watchdog is dead, and the watchdog must act to cut or cycle power. So a dead watchdog leaves its computer powered.
Bench check: pull the Nano out of its carrier and confirm the computer stays up. Safe mode, the restart rule (Ken, 2026-10-07). When WD-A restarts Pi A for the third time within about an hour (the count: System Settings Sheet sec 10), it first sets the safe-mode wire (sec 4.1), and Pi A starts from its safe copy on the eMMC. In safe mode Pi A runs only steering, CAN, the WD-A heartbeat, its IMUs, GPS A and Iridium A; the SSD, the Hailo, the cameras, Starlink and the science lines stay off. Only a signed command from shore ends it; with no answer the boat keeps sailing in safe mode. It adds no restart and changes no watchdog timer, so if the safe copy also fails, Pi B takes over exactly when it would have (about 30 minutes for a Pi that never starts, a guess from the timers). Open: whether the CM5's startup reads the header pin to pick the safe copy; Ken's check on a real CM5, in the software repo, settles it (a fallback is written up there). Failover is sticky until a manual HMAC-signed reset arrives over Iridium (HMAC: a secret-key signature that proves a command is genuine). The watchdogs have no remote trigger, on purpose. Remote recovery is that signed reset plus reimaging over Iridium (authenticated rsync, then a watchdog-run swap that reverts by itself on a failed update). Software built to take a hard power cut. Both Pis run Ubuntu Server 24.04 LTS from local eMMC (the Pi's built-in flash) on independent OS images, with a read-only root that is never written, so a hard power cut (WD-A's rail cycle is the only way to reboot a hung computer) cannot corrupt the operating system. Overlayfs takes the short-lived writes, protecting the eMMC over 3 to 4 years. Data partitions are journaled; the boot runs the filesystem check and falls back rather than hanging. A failed image update reverts on the next watchdog cycle. Not fitted: a pre-cut warning wire from WD-A (the safe-mode wire only picks which copy starts) and a flush-on-warning handler.
| Not in the boxes | Where it lives |
|---|---|
| Central node (release channels, the four cord puller reversing relays, monitoring, the Ewellix drive and its power-cut relay, the two 8-channel relay modules of sec 11) | One plastic box near the battery, busbars and compute area |
| Release Node B (sec 12) | The stern |
| Battery watchman (SBMS0 + JK balancer, on no bus, controls nothing) | The BMS box as bought, beside the battery box (Electronics and Wiring Plan A4) |
| USB hub (fed from Pi A's 5 V), the 12 V converter for the SDI-12 sensors, the dAISy, the science serial adapters, Iridium A's USB isolator, the wing reader and the wing's CAN adapter (secs 4.1 and 10.2) | Its own small box beside Pi A (Electronics and Wiring Plan A1a) |
| FSESC box | Not fitted on either boat |
| Aft junction (passive: post-cable terminals, the motoring-white relay, the post wind meter's 5 V regulator) | The stern, a plastic box |
| Network switch box (only the TSW202 switch, with its SFP module, and its 54 V supply, both on a DIN rail on a side wall, sec 2.3) | The stern, an aluminum box beside the aft junction box, glued to the hull |
| Network surge board (the five Ethernet surge protectors and their ground bar) | A small board on a bulkhead near the deck, where the outside network cables first come inside (Electronics and Wiring Plan A6a) |
| VESC A (VESC B and its phase-transfer contactor are Boat 2 only) | Its own sealed box on the cockpit web at the motor |
| Iridium modem A | Small box near Pi A, main bank; its 5 V from an isolated converter (sec 13.3); its USB cable to a hub port through a USB isolator (sec 4.1) |
| Iridium modem B | Small box near Pi B, emergency bank, fed from the backup battery board's 5 A breaker; its 5 V from an isolated converter (sec 13.3); its USB cable to Pi B through a USB isolator (sec 4.2) |
| The wet lab box (a small node that switches each wet lab part and reads its sensors) | Beside the wet lab (Electronics and Wiring Plan A18) |
| AS3935 detector mini-box | Within 3 ft of Pi B |
| Starlink electronics and antenna | Its own mount; its Cat6 runs to the TSW202 at the stern |
| The PoE light driver (PoE Texas AT-LED-24V, bare board) | Its own small dry box, on TSW202 port 6 |
| The wing's eight slip ring wires: the trim actuator's 24 V and CAN, the tip wind meter's RS-485 and the wing compass's RS-485 | The wing box below the deck, where the wing's wires enter the hull: its three hull cables leave the can by three PG9 glands in the can wall and go down through their own deck gland straight into it (Electronics and Wiring Plan A17; the can's glands: Wing Can Spec Sheet) |
Full contents and every wall entry, and the aft junction box's size check: Electronics and Wiring Plan Appendix A (A6 and A6a for the two stern boxes). Cooling the switch and the motor controller. The TSW202 is in its own aluminum box, the same CXCESNS box as the compute boxes, glued by its back face straight to the hull in the heat-carrying epoxy of sec 2.3, the way the Pi A box is (Electronics and Wiring Plan A6a). Inside, the switch and its supply clip to a DIN rail on a side wall, and their heat runs rail to wall to the glued face (sec 2.3). VESC A (in its own sealed box on the cockpit web) reaches the hull through an aluminum plate through the box floor, set in the same epoxy (decided 2026-09-26). Size that plate before the box is drilled. The switch box and the plate are wired to nothing and stay off the Lightning Ground System Guide's list of metal wired to the lightning copper (sec 6.2).
No single electrical event, above all a lightning surge on the main rail, may kill both computers. So there is no copper power path and no shared space between the islands: each has its own box, computer, watchdog, GPS and Iridium voice path.
The 32 kWh main bank, Pi A, VESC A (VESC B is Boat 2 only), the MPPT bank, the central node and its relay modules, the rudder actuator, and every normal load. Pi A runs the boat. The rudder actuator is one Ewellix CAHB-22 at the 13 in station (Steering System Handbook sec 3): 2,250 lbf (10,000 N), 11.8 in stroke, 24 V DC, self-locking screw, built-in position feedback, no CAN, driven by the central node (sec 10.2). No backup actuator, no FSESC, no encoder B, no pull-pins on Boat 1. BOAT 2 ONLY: the Thomson Electrak HD24B026 is a CAN device; on the emergency bank its transceiver ground would tie the islands together, so steering stays on the main island.
The bank: the bought Dyness 24V 100Ah LiFePO4, 2,560 Wh, with its built-in 100 A BMS with low-temperature cutoff (the bank, its restraint, terminal boots and 30 A ANL fuse: Electronics and Wiring Plan A10). Its own solar charges it through its own SmartSolar 75/10: two 50 W flexible panels in series, glued flat like the deck array, in place of the old framed ~50 W panel (decided 2026-09-26; Build Guide sec 4.4). It powers only the Pi B box (Pi B, WD-B, GPS B), Iridium modem B, and the detector mini-box. Pi B and Iridium B each take their 24 V from a 5 A auto-reset breaker on the backup battery board beside the Dyness, which also holds the Orion-Tr and the SmartSolar 75/10 (Electronics and Wiring Plan A10); the detector runs on WD-B's 5 V. No steering actuator, no CAN device, no diode OR to the main rail. The one bridge: the Victron Orion-Tr Isolated 24/24-5, a float and monthly exercise path, run by relay channel 10 (sec 11). Channel 10 is normally open, so a dead central node leaves it off and the bank's own two 50 W panels keep the bank charged. Transformer isolated, sacrificial in a strike, no disconnect relay in its power path. The floating negative is wired to nothing. It touches the world only through its own two 50 W panels, the isolated Orion, and the isolation bridges of section 6. So the Pi B box gets NO bond spur to the lightning copper, ever; its lid screws are its only lid-to-body connection, and that is correct. Operation. Pi B sleeps most of the time, waking a few minutes each hour to copy state over Ethernet (the schedule: System Settings Sheet sec 10); WD-B can force it awake. If the main island is lost, Pi B beacons at low power over Iridium B with live GPS B position, for as long as it takes, on the emergency solar budget.
| Load | Draw | Wh/day |
|---|---|---|
| WD-B (Nano Every + its D36V6F5) | ~0.3 W continuous | ~7 |
| Iridium B, listen standby | ~0.2 to 0.3 W | ~5 to 7 |
| Pi B, duty cycled ~8% | ~3 to 3.5 W awake | ~10 to 15 |
| GPS B | ~0.3 W duty cycled with Pi B | ~1 to 2 |
| Design point, normal operation | ~25 to 32 | |
| Post strike beacon mode | duty cycled + few sessions/day | ~15 to 25 |
The two 50 W panels gather about 36 to 86 Wh/day in Southern Ocean gale weather (peak sun hours 0.5 to 1.2, system factor K 0.72) and about 108 to 180 Wh/day in a typical summer there, so solar carries the design point in nearly all weather. The bank alone covers about two to nearly three months of total darkness at ~25 to 32 Wh/day (2,560 Wh rated, about 2,050 Wh usable from 90 to 10 percent: about 64 to 82 days).
The islands talk only through galvanically isolated paths:
| Path | How it is isolated |
|---|---|
| CAN, both buses | Each HAT's onboard 5 kV isolation. Each bus is a separate barrier crossing |
| Ethernet | Transformer isolated on each port, by the standard. Each Pi runs its own Cat6 to the TSW202. The five deck and near-deck cables reach it through Ethernet surge protectors (Lightning Ground System Guide sec 7.4). A strike may still kill the switch, costing Starlink, the cameras and the Pi-to-Pi Ethernet, not both Pis |
| Inter-box link (main-island-alive discrete only) | One optocoupler on one M12 cordset (sec 4.2) |
| Orion-Tr Isolated | Transformer isolated, sacrificial (sec 5.2) |
| Storage | The SSD attaches to Pi A only. Pi B keeps a copy of the essential state (position, mission phase, recent readings), refreshed over Ethernet each wake |
These bridges close the rail path only. Antenna line surges and ground shifts are handled by the surge rules (sec 14), the metal boxes, and keeping the two Iridium chains apart.
First-setup check: on each CAN drop, meter from the far shield end to the Pi A stud for continuity (budget cordsets sometimes leave the shield out, and a shield tied at neither end is an antenna).
Every hole in both compute boxes, with its gland and drill size, is in the Electronics and Wiring Plan: Appendix A1 for Pi A, A2 for Pi B, and the rules for every entry at the head of Appendix A (Pi B's entries all on its bottom face, Pi A's on any face but the glued back face; no M12 receptacles on any box). No load current passes through either box except its own feed and, in Pi A, the USB hub's 5 V lead (entry 12): the boxes pass buses, radio, coax and USB only. Switched loads and their harness live at the central node and the aft junction.
Electronics and Wiring Plan A1. Entries 5, 9, 10, 13 and 14 are not drilled. The watchman and science lines, the dAISy's coax and the wing cable enter the USB hub's own box (A1a). Pi A takes two holes for the hub box: entry 12 for the hub's 5 V lead out, and one split gland shared by entry S (the hub's USB lead) and entry 7 (the wing CAN adapter's USB lead) (sec 4.1).
Electronics and Wiring Plan A2. Entry 7 takes Iridium B's USB cable in a split gland (sec 4.2).
The cabling rules are wiring, and their home is the Electronics and Wiring Plan. What this plan leans on, in one line each:
One straight backbone per bus in NMEA 2000 Micro-C cable, molded tees and terminator plugs, and 12 V bus power fed once per bus from its own busbar branch, never from inside a Pi box: Electronics and Wiring Plan secs 5.1 to 5.3.
Control and house, so a babbling device cannot delay steering or throttle traffic. Only the Pis touch both, each across its own isolation barrier (sec 6), and the battery watchman is on no bus: Electronics and Wiring Plan sec 5.4.
Supply and return in one jacket, one lashed bundle per route, runs along the centerline, no spare coiled in a circle: Electronics and Wiring Plan sec 7.7. They are the first lever against lightning (sec 14.2).
Genuine mix 31, only on the cables that leave the cabin (about 11 boat wide), glued down, never hanging on the cable: Electronics and Wiring Plan sec 7.5.
Every device the boat can command uses one of these patterns. No safety default depends on a message arriving: every actuator has a named node, a named bus and a safe place to sit with no power.
| # | Pattern | How it works | Who uses it |
|---|---|---|---|
| 1 | CAN with a translator node | The device is dumb (burn wire, solenoid, relay coil, DC motor) and the central node drives it by plain wire | Release channels, switched loads, the Ewellix steering actuator |
| 2 | CAN native | The controller is built into the device and the Pi talks to it in its own dialect | VESC A in VESC format, keeping its three separate layers: CAN, a UART backup and a discrete GPIO kill line (Build Guide sec 5.1); the em-trak in NMEA 2000 (Boat 2's Electrak in J1939); the wing trim actuator in J1939, from Pi A on its own short CAN link through the wing's CAN adapter (sec 10.2) |
| 3 | Direct line | Data-heavy talk on private serial, USB, or Ethernet | Iridium x2, dAISy, GPS, Starlink, the PoE cameras, the TSW202 (Web API out, SNMP back), the science instruments on FTDI serial, the wing's tip wind meter and compass on the wing reader (sec 10.2), and the SBMS0 text stream |
| 4 | Radio | A radio link to a device | Nobody: the wing's trim actuator, tip wind meter and compass are wired down through the slip ring (decided 2026-09-26) |
The watchdogs work the other way: the Pis report heartbeats, and the watchdogs hold the rails. The Pi says what it wants; devices do the mechanics. Every device has a planned answer to command silence:
| Device | Answer to silence |
|---|---|
| Ewellix steering actuator | The self-locking screw holds position; the drive has no power |
| VESC | Ramps to zero |
| Release channels | Stay disabled: enable relay open, every MOSFET off (burn-wire short relays DELETED) |
| Wing trim actuator (Thomson Electrak MD) | Stops and holds on CAN silence (Wing Trim Actuator Installation Guide sec 5). Strong wind still opens the wing tab spring link and the tab goes straight, so feathering needs no power and no command |
| Switched-load relays | Drop out with the node (sec 11) |
Three nodes are team built: the central node here, the minimal Release Node B at the stern (sec 12), and the wet lab box beside the wet lab, which switches each wet lab part on Pi A's orders (Electronics and Wiring Plan A18; Flow-Through Wet Lab Design sec 10). How a node works inside, with drawings: Node Boxes Explained. Box contents, entries and monitoring: Electronics and Wiring Plan sec 4.1 and A3. The processor: an STM32 G4 on a NUCLEO-G431KB (about $20), in a Nano screw-terminal carrier with a CAN transceiver breakout, picked for CAN on the chip and the team's Rust tools (Embassy). The central node, Release Node B and the wet lab box each get a G431KB (the wet lab box's is to buy); the wing pod's went with the slip ring. The boards are on hand (Mike, 2026-09-26): check that each is a G431KB. A G431RB in place of a KB needs an Arduino Uno screw-terminal shield instead of the Nano carrier, and a recount of the pin map below. The NUCLEO-G431RB on hand is the workbench test board: a Nucleo-64, it does not fit the Nano carrier, and its extra pins are bare 0.1 in headers, banned in release logic. The expander: the Nano header's 22 pins are not enough, so one MCP23017 I2C expander board, screw-terminal version (~$6), adds 16 on/off lines over the Nucleo's two-wire I2C link. Its outputs drive ONLY the convenience loads: relay channels 1 to 4 and 6 to 12 (the switched loads, the charging relay on channel 8 and bilge pump 2's manual-on on channel 12) and the steering power-cut coil. Four of its pins are slow inputs: Water Witch 1 and 2, the release bus readback and the steering drive's fault line. All 16 pins are used; none is spare. Every release channel and the release enable relay (channel 5) stay on native Nucleo pins, so the expander is never in the release chain. If it or the I2C bus dies its outputs float, the same as a dead node (sec 11): the loads drop out, the steering power-cut relay closes so steering keeps its feed, and the charging relay drops out so charging carries on. Pull-down resistors under the terminal screws make a reset read as OFF. Pin map (settled 2026-09-26; A2 to the horn 2026-10-08): 21 of the 22 native pins used, A7 the one spare. It needs one bench step first, before burn-in: remove the four tiny solder links that join A4 and A5 to D4 and D5 (for Arduino-style I2C) and A2 and A7 to the USB serial link. That frees A2, A4, A5 and A7 and costs only the USB serial console (programming still works; logs go over CAN). Check the solder link numbers against the board manual, UM2397, when it arrives.
| Pin | Use |
|---|---|
| D0, D1 | Wind in and wind out: the post's Calypso CMI1018 (USART1) |
| D2, D10 | CAN TX and CAN RX |
| D4, D5 | I2C data and clock: the expander, the BME280 and the humidity board |
| D6 | Rudder encoder A (timer input) |
| D9 | Steering PWM (TIM1, 20 kHz) |
| D13 | Steering direction. It is the boot pin: at the bench, meter it at power-up and confirm nothing pulls it high |
| A0 | Steering current sense |
| A1 | Ewellix position wiper |
| D11 | 1-wire chain 1 (the battery probes: the cell-temperature trip) |
| D12 | 1-wire chain 2 (air) |
| D3 | Relay channel 5 coil (the release enable) |
| D7, D8, A3, A6, A4, A5 | Release gates for channels 1, 2, 3, 4, 5 and 6 |
| A2 | Horn: relay channel 13's coil (sec 11) |
| A7 | Spare |
The release gates avoid D12 and D5 (pulled high at reset) and D13 (the boot pin); each MOSFET board's own pull-down holds its gate off until the firmware starts. The release lines and channel 5 never move off native pins. No soldering on the boat. Every board is a screw-terminal module, pre-mounted, landing on DIN terminals (the list: Electronics and Wiring Plan sec 4.1). One-time bench soldering is allowed, off the boat, before burn-in and before anything is installed (Mike, 2026-09-27): terminal blocks soldered onto a board, or small solder links removed, as for the steering drive (Electronics and Wiring Plan sec 5.6) and the Nucleo above. After that every wire lands under a screw or a latched connector. Beside relay 8 lands the one run from the five charger remote on/off cables, which meet on the main bus board, tied together (decided 2026-09-26); the 1k 1 W resistor sits at the main bus board too (sec 11). Not in the box: burn-wire short relays, branch current meters, a VE.Direct reader board, a gas sensor breakout (deleted, do not re-propose). Every logic line between the Nucleo and the relay, MOSFET and opto boards lands under a screw terminal or a latched connector. No push-on jumper pins anywhere in release logic. Board preparation, in order: a week of powered burn-in, all the node boards as one lot; a heat cycling check; conformal coat; stake it in its carrier (after burn-in); seal it in its box. Steering command path. Pi A sends the rudder target over the control bus to the central node. The node drives the Ewellix CAHB-22 through the Pololu G2 24v13 drive (hardware and wiring: Electronics and Wiring Plan sec 5.6): it sets the direction, ramps the drive up, holds it, and stops at the target reading. The move limits and the stall-stop numbers are in the System Settings Sheet sec 3. PWM at 0 brakes the motor, and the self-locking screw holds the rudder with no power; a dead Nucleo leaves PWM at 0. After any stall stop the node sets PWM to 0 and opens the power-cut relay: a failed drive usually fails stuck on, and that relay is then the only stop. The power-cut relay in the feed, closed when unpowered, is the second cut; cutting a jammed or runaway actuator takes a live node. Rudder angle is read two ways and cross-checked: the actuator's own feedback (about 3,000 ohm over the stroke, 10 ohm per mm; measure it on the bench and scale the input to the measured span) and encoder A on the stock (an AS5048A magnetic encoder; the node reads its PWM output, a pulse-width signal, on a timer pin, not its SPI port; it comes in over the aft sensor trunk). Disagreement flags a sheared linkage or an encoder fault for shore. Jam logic (commanded effort with no encoder motion, or motor current at stall) cuts drive power within seconds: a stalled actuator burns its windings in 10 to 25 minutes. No J1939 and no CAN on the steering. Boat 1 steering has no backup, accepted on purpose. Ewellix publishes no rod-travel life, so rod travel sets no move budget; the deadband and minimum interval are in the System Settings Sheet sec 3. BOAT 2 ONLY: the backup is an aux rudder, not a second actuator, on release channels 7 to 9 (sec 10.3). Boat 2 keeps one pull-pin, on its primary actuator, so a dead actuator can let the rudder trail for the aux rudder. The Thomson Electrak primary and the aux-rudder Electrak take position commands over native J1939, straight from Pi A, on the main bank; the Electrak's own overload cut-out is the backstop. The backup VESC's phase-transfer contactor takes signed commands only, its coil pulsed by the central node from the main bank. No FSESC box on either boat. The wing's command path (since the slip ring). The wing's program runs on Pi A. The trim actuator (Thomson Electrak MD, CNO control option, MD24A025-0050CNO) stays on its own short CAN link through the slip ring, kept off the boat's two buses, and speaks J1939 (settings: System Settings Sheet sec 8). The MD sleeps under 2 mA, so its 24 V feed is not switched. The wing-tip wind meter (Calypso ULP CMI1017, RS-485) and the wing compass (WitMotion HWT905, RS-485, on its own pair) come down to the hull as well; the post's CMI1018 (UART/I2C) runs to the central node (pin map above).
One solenoid channel, repeated per channel on both release nodes, all screw-terminal modules, all landings on DIN rail screw terminals (the Erayco UK2.5B and MELIFE kits on hand). The power path:
Cord puller reverse, for shore tests (channels 3 to 6). Each puller's motor pair runs through its own reversing relay in the central node, which swaps the motor's two wires; all four coils run from relay module 2 channel 11, "reverse" (sec 11; the relays and their wiring: Electronics and Wiring Plan sec 4.1). A shore test is: reverse on, fire puller N as usual, reverse off. The release lock does not change: the enable relay (channel 5) and the puller's own channel must both be on, and "reverse" alone moves nothing. Normal release is forward only; the software refuses "reverse" unless a shore test command is active. No puller exercise at sea: the shore test is the only test (Mike, 2026-09-26). No extra wires aft. Cord puller run time and fire guard. Each pull runs a set time, then stops, with no stall sensing (System Settings Sheet sec 5). A slow-blow fuse per puller, between its MOSFET module and its reversing relay, is the fire guard (Electronics and Wiring Plan sec 4.1). DELETED on both nodes (do not re-propose): normally closed short relays across the burn wires and their daily continuity check (the release bus is dead until the enable relay closes, and each channel also needs its MOSFET on; nothing checks the burn wires between launch and use); confirmation microswitches and relay-state readback (the node knows what it commanded, and the speed drop and the aft camera confirm that the drogue is out). Release sequence (ENABLE and RELEASE, never ARM and FIRE: fire is a bad word on a boat):
One release bus at a time: the central node's channel first, then Release Node B's only if the first does not confirm, never both at once (what the small-branch fuse sees: Electronics and Wiring Plan sec 7.6). Channels across the two release nodes (Electronics and Wiring Plan sec 4.3 owns the full map):
| Node | Channels |
|---|---|
| Release Node B (stern) | The drogue horn-B channels only (sec 12) |
| Central node | The drogue horn-A channels, and the four release actuators for the Jordan series drogue (JSD, the long line of small cloth cones streamed from the stern in a storm). Channels 7-9 are reserved for Boat 2's aux rudder, unwired on Boat 1; the pull-pin channels 10-15 and the trim-decouple channel 16 are retired |
Open: the burn wire gauge, 28 or 30 AWG nichrome; the wet bake-off at 24 V settles it (Transom Drogue Box Guide sec 9). Open, BOAT 2 ONLY: the pull-pin solenoid's stroke and force; the pin geometry at fit-out sets them.
This section owns the switched-load channel map. The Electronics and Wiring Plan, the Box Wiring Sheets, Node Boxes Explained, the lid labels and the firmware follow this table. Print the map on the lid. The central node has two 24 V 8-channel relay modules. The first carries six switched loads (channels 1, 2, 3, 4, 6, 7), the release enable relay (channel 5) and the charging relay (channel 8); none of its channels is spare. The second (about $15; probably one of the two HiLetgo modules bought 2026-08-16 and 2026-08-18, so check the shelf before buying) runs off the MCP23017 expander: channels 9 to 12 in use, channel 13 the horn on a native Nucleo pin, and 14 to 16 with no driver, because those four expander pins now read slow inputs (sec 10.2). The expander drives every coil except channel 5's, which runs from a native Nucleo pin so the expander is never in the release chain (sec 10.2), and channel 13's (the horn, on pin A2). Timers and times for these channels: System Settings Sheet secs 1 and 5.
| # | Channel | Note |
|---|---|---|
| 1 | Starlink power | Normally open |
| 2 | Science payload group | Normally open. Switches the wet lab's own 15 A line from the bus bar (Electronics and Wiring Plan sec 7.6), not the node's feed. On only around each wet lab burst, so the wet lab draws nothing between bursts (timing: System Settings Sheet sec 5); the wet lab box (Electronics and Wiring Plan A18) switches each part inside that window. Open: confirm podlink does not already own that switching; settle it before the channel is wired |
| 3 | Side and stern nav lights | Normally closed contact, so it fails ON (below). On at the computed sunset, off at sunrise, and on in daytime too while fog mode is on (below) |
| 4 | em-trak AIS transponder | On demand, through its normally closed contact so it fails ON. The dAISy receiver is always on, unswitched |
| 5 | Release enable relay | Powers the release bus that feeds the drogue release actuators (sec 10.3). Coil on a native Nucleo pin, never the expander. Normally open: dead, the release bus is dead |
| 6 | Bilge pump 1 manual-on | Normally open. Joins pump 1's + terminal after its Water Witch, no diode |
| 7 | The two all-round motoring whites | Normally open contact, through a 30 A relay in the aft junction. On only when it is dark (channel 3's sunset-to-sunrise rule) AND the motor is running; in fog mode, whenever the motor runs (below) |
| 8 | Charging relay | Normally closed contact in the chargers' on-signal path. Coil on only by shore command: all five chargers stop (below) |
| 9 | Camera port UV-C | Second module. Normally closed contact, so it fails ON (below) |
| 10 | Orion-Tr float charger (emergency bank) | Second module. Normally open contact, so it fails OFF (below) |
| 11 | Cord puller reverse (test only) | Second module. Normally open contact: the coils of the four DPDT reversing relays (sec 10.3). On only under a shore test command; forward is the release direction |
| 12 | Bilge pump 2 manual-on | Second module, on the expander's last free pin, so all 16 expander pins are used. Normally open. Joins pump 2's + terminal after its Water Witch, no diode; runs the monthly 5-second dry exercise (Build Guide sec 3.6) (Mike, 2026-09-28) |
| 13 | Horn | Second module, driven from the Nucleo's native pin A2, not the expander (Mike, 2026-10-08). Normally open: a dead node leaves the horn silent. The contact feeds the horn's 24 V through a 5 A inline fuse. Sounded by signed shore command, and by fog mode: one long blast of 4 to 6 seconds every 2 minutes (below) |
| 14 to 16 | No driver | Second module. The expander pins of 13 to 16 read the two Water Witch states, the release bus readback and the steering drive's fault line (sec 10.2) |
The TSW202 network switch is on no channel. It is always on from the house branch: dropping it would blind Starlink and cut the Pi-to-Pi Ethernet. Not fitted: a white strobe. Do not re-propose: it is not a legal navigation light, and a flashing white light is a distress signal under US inland rules. Channel 3 fails ON. The node powers the coil by day to hold the lights OFF and drops it at the computed sunset (from GPS position and the date), so a dead or resetting node leaves the boat lit. Write the sunset rule to fail toward light (a missing GPS fix or a bad clock lights them), and run it in the node, not only the Pi, since the node holds its last command when the Pi dies. Not fitted: a dusk-to-dawn photocell (it draws power around the clock). No current sense on the light circuit; the cameras show the lights. The lights themselves: Electronics and Wiring Plan sec 6.4. A dead Pi never leaves the boat dark. If the node hears nothing from either Pi for the set time (the Pi-silence timer, System Settings Sheet sec 1, the same timer as the charging stop below), it turns the nav lights on by itself and keeps them on until a Pi is back. Channel 7 does NOT fail on. The motoring whites stay on the normally open contact: an all-round white shown while not motoring is the wrong picture. They come on only when it is dark and the motor is running, not whenever the motor runs (Mike, 2026-10-08): the rules ask for lights only from sunset to sunrise and in poor visibility; daytime whites cost only about 1.5 W. Fog mode (Mike, 2026-10-08), from parts already aboard. It turns on by itself when fog is likely: the air close to its dew point at very high humidity, worked out from the HygroVUE 10's temperature and humidity, or the air's dew point above the sea temperature from the wet lab (sea fog); the thresholds are in the System Settings Sheet sec 5. In daytime the forward camera confirms it by checking whether the horizon has faded; it cannot at night, since the camera IR LEDs are off. Near shore, the shore watch can also act on National Weather Service dense fog advisories. Shore can force fog mode on or off at any time by signed command. While fog mode is on, the side and stern lights (channel 3) come on in daytime too (the collision rules, Rule 20(c), ask for lights in poor visibility), the motoring whites come on whenever the motor runs, and the horn (channel 13) sounds one long blast every 2 minutes (the collision rules, Rule 35), under 2 W while fog mode is on. A false alarm only costs a little power. The logic is in the software repo. Channel 8, the charging relay. Charging is on by default in hardware. The relay's normally closed contact sits in the chargers' on-signal path: the thin battery positive from the 1k resistor at the main bus board, through the contact, to the yellow wires of the five Victron remote on/off cables. The wiring and its bench test: Battery Box Building Guide Part 8 ("Wire the charge-stop relay").
Camera port UV-C is on channel 9, switched in software on a timer (System Settings Sheet sec 5), not by a hardware timer, through the normally closed contact, so a dead node leaves the UV on nonstop and the window stays clear. Orion-Tr is on channel 10: the node runs it through its remote on/off terminal on a daily timer, and through the monthly emergency-battery test's recharge (the times and the idle draw saved: System Settings Sheet sec 1). Normally open contact, so a dead node leaves it off and the emergency bank's own panels and SmartSolar 75/10 keep the bank charged (sec 5.2; why not normally closed: Design Decisions R-8.1E). Like channel 9's, its wire leaves the node box through its own gland (Electronics and Wiring Plan A3). No current telemetry on these channels. The INA226 on the Pi A box feed is the only branch meter left. Both Iridium modem feeds are hard wired always on, never switched, so no distribution failure can cost the voice. Pi rail switching belongs to the watchdogs, never this board. The switched-load runs out to the aft junction: Electronics and Wiring Plan A3 entry 16. Nothing that keeps the boat alive runs through the relay boards except the nav lights, which fail ON. Node dead, or its feed lost: the relays drop out. Loads on normally open contacts go OFF, the Orion-Tr and the puller reverse among them; the nav lights, the em-trak and the camera port UV-C stay ON (normally closed); the release bus stays dead; the charging relay drops out, so the boat keeps charging. Propulsion, steering, the battery watchman, the bilge pumps (on their own Water Witch switches; only their two manual-ons are on this board), the network switch and Iridium never run through this board.
Every release path is three things in series: the command source, a separate enable relay powered only during a valid enabled window that times out by itself, and the release driver. Two independent faults are required to release. Release circuits are dead when idle: enable relay open, every MOSFET off (sec 10.3). Two nodes carry the release channels. Most sit on the central node (sec 10). Release Node B sits at the stern on the house bus with its own fuse and carries only the drogue-lid horn-B channels, under the same enable-then-release two-step lock, with no opto board and no short relays (Electronics and Wiring Plan A15). The four cord pullers have one path only, through the central node, and run backward only under a shore test command (sec 10.3).
| Channel | Who may release it | Why |
|---|---|---|
| JSD deploy (the nichrome coils part the lid strap) | Signed ground command only. No path the boat can take by itself | Deployment is forecast driven with hours to days of lead, and the two always-on Iridium chains keep the link |
| JSD release (the cord puller pulls the T8 trigger line) | Signed ground command only. Never by the boat alone | It cannot be undone, and there is no time pressure. It fails toward held: the puller's self-locking worm motor cannot be back-driven, so power loss leaves the drogue attached |
| BOAT 2 ONLY: the primary actuator's one pull-pin, and the aux rudder release (neither on Boat 1) | Same enable-then-release two-step lock | Aux rudder deployment is never automatic (Routing Guide 11.4) |
The boat does not shut down for lightning (Mike, 2026-10-02). Nothing is turned off; the boat keeps sailing, steering and logging through a storm. Not fitted: a go-dark mode. Do not re-propose it. Why:
Feathering stays with wind speed: the wing's program and the wing tab spring link handle it. What protects the boat from lightning is the copper path, the two isolated compute islands and the surge parts (Lightning Ground System Guide). The research: Reference/reports/USV lightning risk and protection.md.
The sensor: an AS3935 Franklin sensor on a Playing With Fusion SEN-39003 board (bought 2026-10-06), alone in a tiny plastic box within 3 ft of the Pi B box, touching no metal (the aluminum boxes would shield the signal it listens for). The board takes 2.4 to 5.5 V on power and signals, so it runs at 5 V from WD-B's own D36V6F5 rail, with no level shifter (Mike, 2026-10-14). It comes tuned from the factory: this board's value is 88 pF (System Settings Sheet sec 10). The lead: Belden 6504FE (22 AWG, six cores, foil shield and drain wire), under 3 ft, five cores used: red 5 V, black ground, blue IRQ to WD-B pin D2 (Claude's pick; no WD-B pin map exists yet), green SCL to A5, white SDA to A4; the brown core cut back flush inside the jacket at both ends. A PG7 gland fits it at each box. The drain wire lands at the WD-B end only, on the island's floating negative; at the detector it is cut back and taped. I2C address 0x03. The board's own wiring (the screw block on its pin row, the two jumpers that pick I2C, the nylon mounting): Electronics and Wiring Plan A12; drawing Lightning_Detector_Wiring_2026-10-14. WD-B never cuts Pi B's power for a strike. Range about 25 miles (40 km), with a distance estimate per stroke. WD-B keeps the strike log and puts the full data on the bus. What it is for: logging storms, setting the satellite check to every 15 minutes while strikes are heard (sec 13.4), and starting the post-strike audit after a very close strike (sec 13.2). Strike strength is not used: the detector's energy number has no physical meaning by its maker's datasheet, and a strong far strike reads like a weak near one. Keep it as far from the VESC as the lead allows: motor controllers are the classic source of false triggers. Bench-check the false-trigger rate with VESC A running before fixing the spot; if it is bad, move it forward on a small node with an isolated transceiver.
After a very close strike is logged, the POST_STRIKE audit walks every device, every surge protector that still reports, and the radio-link quality numbers, and reports a protection status to ground (Routing Guide sec 11.8). The meter checks at the dock: Lightning Ground System Guide sec 8.4.
The only new hardware is the detector mini-box. No coax safing relays: the two Iridium chains rely on their quarter-wave stub arrestors alone, with no sacrificial part, and the post-strike check is a meter reading of each stub against its first-setup record (Lightning Ground System Guide sec 7.1). Each stub ties its coax shield to the lightning copper, so each Iridium box makes its 5 V with an isolated 24 to 5 V converter (its 5 V side has no wire connection to its 24 V side; Mike, 2026-10-02): the tie stops at the Iridium side, and neither the main negative (Iridium A) nor the emergency island's floating negative (Iridium B) is tied to the copper a second time. The converter: a Mean Well DDR-15G-5 in each box (Electronics and Wiring Plan A11). Not fitted: a plain potted 24 to 5 V converter (the Klnuoxj class) in either Iridium box: it joins its 5 V minus to its 24 V minus, which would tie the main negative or the floating negative to the copper through the stub. Do not re-propose. On the data side, each modem's USB cable reaches its Pi through a USB isolator (secs 4.1 and 4.2): the Mean Well breaks the path to the battery, the isolator the path to the Pi. Starlink and the cameras are expendable behind their Ethernet surge protectors (Lightning Ground System Guide sec 7.4), AIS is entirely below deck, and the two GPS chains back each other up. Never touched by any mode: bilge pumps, Water Witch paths and the battery watchman. No gas sensors (deleted, do not re-propose); the cell-temperature trip on the pack probes is the battery trip.
Two weather satellites map every lightning flash and post the data free: NOAA's GOES lightning mapper (GLM; the Americas and the oceans beside them, under 20 seconds old, on AWS) and EUMETSAT's MTG Lightning Imager (Europe, Africa, most of the Atlantic and part of the Indian Ocean, from the EUMETSAT Data Store). They give real strike positions, about 5 miles accurate. Software only; no new hardware.
Size and power. A GLM file covers 20 seconds and runs about 150 to 210 KB, so 2 minutes is about 1.2 MB and 10 minutes about 6 MB; under 1 GB a month. A Starlink window is about 3.3 Wh; the 15-minute checks run only while strikes are heard, mostly in the tropics, where the panels make far more than the boat uses (Build Guide sec 4.6). The Starlink plan is picked at launch: check its data allowance against this. Where it does not work: no free map covers the west Pacific, Australia and the east Indian Ocean (roughly 70 E to 170 E, worked out from where the two satellites sit), or wherever Starlink is off or down. There the boat's own detector is the only lightning watch.
Rule zero: no relays. Then four rules:
| Term | Plain meaning |
|---|---|
| TVS | A fast clamp diode that catches a small spike and holds the voltage down |
| MOV | A ceramic block that dumps a large surge, and wears out doing it |
| SPD | Any surge protection device, used as the general word |
| Quarter-wave stub | A shaped piece of coax that shorts lightning frequencies to ground while passing the radio signal. Nothing in it can wear out |
The boat does not shut down for lightning (sec 13). These rules, the wiring rules and the copper path are what protect it. Per-conductor detail and the on-hand parts: Lightning Ground System Guide sec 7.
Iridium A and B coax: quarter-wave stub arrestors, Fairview FMSP1066 or PolyPhaser TSX class; nothing in them can fail. What is on hand: Lightning Guide sec 7.1. AIS: no arrestor and no changeover relay: nothing AIS is above deck. The em-trak transmit whip stands under the panel-free deck patch, the receive antenna is short and below deck, and the receivers are cheap. GPS coax, both chains: no arrestor and no bias tee. A plain stub is impossible, because the receivers feed DC up the coax to their active antennas. Not fitted: DC-pass gas tubes. Do not re-propose: one that fails shorted silently kills its antenna while still looking healthy. The protection instead: two independent chains (different receivers, antennas and boxes), the ferrite at the wall, and a cheap receiver cross-checked against Starlink's own position fix (Lightning Guide sec 7.3). Starlink and cameras: an Ethernet surge protector on each of the five cables that run on deck or near it, on a small board on a bulkhead near the deck where those cables first come inside (Electronics and Wiring Plan A6a), so a surge on one cable does not take the switch (Lightning Ground System Guide sec 7.4). Starlink and the cameras themselves stay expendable. CAN, both buses: TVS on the HATs. At the central node, Release Node B and wet lab box drops, a Taidacent RS-12V/2S DIN-rail surge module (all three on hand; the wet lab box's is one of the two bought 2026-10-07), not SM712 chips. Iridium USB links: a USB isolator (DSD TECH SH-G01A) at the Pi end of each modem's USB cable, Iridium A's on a hub port and Iridium B's at Pi B (secs 4.1 and 4.2). On data lines isolation beats clamping, and it keeps the islands apart. VESC UART and kill pair: unprotected, by choice (Mike, 2026-09-07). Battery bus surge modules: two Boartechs 2-pole DC modules, 48-70 V class (Amazon B0DFBPBW1R, $16.17 each, bought 2026-09-27 in place of the 12 to 24 V class first picked), in place of the graded MOV ladder of four modules stepped at about 35, 40, 45 and 56 V DC, all above the 29 V charge ceiling, which no US seller offers (decided 2026-09-27, on the latest real information). Each sits on its own 25 A blade fuse at the + solar bar, between the solar bars, and its ground terminal is jumpered to its own minus, so the bus keeps its one bond wire (the mounting: Electronics and Wiring Plan sec 7.6). Meter each at every refit and at the post-strike check (no status contacts; the central node gland they once used, A3 entry 20, is not drilled). Their working voltage sits far above the 27.6 V absorption and the VESC's 28.0 V maximum, so they never conduct in charging; read the label on arrival (System Settings Sheet sec 4). The listing gives a 1.8 kV protection level, so the TVS diodes at the box inputs do the fine clamping. Checked 2026-10-07: those diodes are safe with these modules. The battery holds the bus far below where the modules start, so each diode only meets what its own feed run picks up (Lightning Ground System Guide sec 7.6; detail: Reference, Surge Diode Check 2026-10-07). The TVS part is the 1.5KE33CA: about a dozen at the hull box inputs, plus whatever the wing box takes (Electronics and Wiring Plan A17), against the 20 on hand. Count the bag first; a second 20-pack ($5.89 last time) is Left to Buy item 64, bought only if the count is short. Solar strings: PV-type surge modules, Type 2, with a heat cutoff (the Phoenix Contact or Citel class), plain, no remote signal, one per charge controller, on a short DIN rail just below it on the main bus board, across its PV + and PV - (Reference/Solar_Surge_Wiring_2026-10-12). The wires that share a module terminal (two on DC+, three on DC-: panel minus, the short wire to the controller and the ground-screw jumper) are twisted together bare under the terminal's clamp plate, no ferrule (Mike, 2026-10-12); retighten once a week or two after wiring. The units: five NanQue Type 2 40 kA DIN modules, the 250 V DC option of Amazon B0DZ267ZV4, bought 2026-10-06 ($84.10 the five), and kept; the five bought 2026-09-27 are Boat 2's (Mike, 2026-10-07). They stay off at the strings' 75 V cold maximum. The listing's 100 V option would clamp lower, but in a close strike either option lets far more through than the controllers can take, so a swap would buy little. Check each has a heat cutoff on arrival. No contacts land on the central node. Behind them the MPPTs stay fused per branch, half sacrificial by design. Backup island: its own TVS at the Pi B DC-DC, at WD-B's D36V6F5, and on the detector lead, all referenced to the island's floating negative, never to the ground network. The detector lead is a 5 V line, so its TVS is a 5 V part, across red and black inside the Pi B box. Open: which 5 V TVS part; picking one settles it. Remote driver boxes: TVS on each power input. Release lines need nothing extra: they are dead until their node's enable relay closes (sec 10.3). The wing box (since the slip ring): every wire from the slip ring gets a fuse or a surge part at the wing box's entries, so a strike on the mast stops there and does not reach the other boxes (decided 2026-09-26). The CAN pair to the trim actuator and the two RS-485 pairs, to the tip wind meter and the wing compass, get a Taidacent RS-12V/2S surge module each where they enter the hull, and the 24 V pair its 5 A fuse and a 1.5KE33CA (Wing Power System sec 8.7). The three data pairs then end at isolated readers in the USB hub box (sec 10.2). The pod's surge diodes and fuses went with the pod. The 1.5KE33CA conducts from 31.4 V.
Ranked levers that add no parts that can fail:
Glands, connectors, cable rules and greases are wiring; their home is the Electronics and Wiring Plan sec 7 (gland sizes and drills 7.1, grouped entries 7.2, the disconnect rule 7.3, everything else and the greases 7.4). The two rules this plan leans on: every box entry is a nylon IP68 gland, and no M12 receptacle sits on any box.
Open questions for this guide are in Pending Amendments v10.4.