Nobotic USV, CD25D conversion, companion note to the Electronics and Wiring Plan

Node Boxes Explained v10.3

How the boxes around the boat work: what a node is, how the Pi talks to it, what an opto and a 1-wire chain are, and how the release channels fire a burn wire without ever letting go by accident. We build three boxes ourselves, the central node, Release Node B and the wet lab box; every other box is bought, with its own small computer inside. Sources: the Electronics and Wiring Plan, the Compute Bay Housing and Wiring Plan sec 9 to 11, the Build Guide sec 8.2A.

CAN data (blue)release path, live only when enabled (red)safe / idle state (green)field wiring, 24 V or sensor leads (brown)

1. The big picture

The Pi is the brain. It never touches a burn wire, a pump or a motor itself. It sends short messages down one wire, the CAN bus, and the boxes do the work where the work is. We build three of them. The central node is a Nucleo computer board (about $20) beside the battery and compute area. It listens on the bus, reads its sensors, fires its release channels and switches the convenience loads. Release Node B is a small second release node at the stern, on the house bus, fed on its own 30 A breaker. It carries the horn-B burn-wire channel of each drogue lid behind the same enable-then-release interlock, so one dead box can no longer take out every release on the boat. The wet lab box (A18) is a small node on the house bus, on the same Nucleo board, that switches each wet lab part one at a time and reads its sensors; its contents are in the Electronics and Wiring Plan, A18. The rest are bought controllers with their own brains: the VESC (motor controller; Boat 1 sails with one, and VESC B with its phase-transfer contactor is BOAT 2 ONLY) and the em-trak transponder. The rudder actuator, the Ewellix CAHB-22, has no brain and no bus; the central node drives it. If the bus goes quiet, every box drops to a safe state on its own: the Ewellix's self-locking screw holds the rudder, the release buses stay dead (no enable, no gate), the VESC ramps to zero. No safety default waits on a message arriving (Build Guide 5.1).

Pi A (brain) CAN HAT, both buses CAN bus: one twisted pair + 12 V, every box hears every message Central nodereleases, monitoring, loads Release Node Bstern releases, house bus FSESC boxnot fitted, either boat Ewellix CAHB-22rudder: no bus, central node drives it Aft junctionterminals (no node) Native gearVESC A, em-trak actuators, burn wires,probes, relays drogue lids (horn B) BLDC motor, encoder B rudder arm (self-locking) bilge, post;trunk to central
Figure 1. One wire (the CAN bus) carries every command and every report. The team-built central node sits beside the battery, busbars and compute area; the team-built Release Node B sits at the stern. The FSESC box is drawn only to show it is not fitted on either boat. The aft junction is a plain terminal box with no brain; its sensors ride a trunk to the central node (one trunk is likely enough; count at fit-out). The network switch, which feeds the cameras, is not on the CAN bus: it sits in its own aluminum box beside the aft junction box, glued to the hull for cooling (Electronics and Wiring Plan A6a).

How a message is addressed

Every box on the bus has a number (its address), and every message has a number that says what it is. We pick them once and print them on the lids. Example: central node = address 40; message "enable" = 40-01; message "release channel N" = 40-02 with N inside it; its status report back = 40-81. Release channels are numbered too: 1 and 2 = the drogue lids (horn A on the central node, horn B on Release Node B), 3 to 6 = the four release actuators for the Jordan series drogue, a long line of small cloth cones streamed from the stern in a storm (central node), 7 to 9 = reserved for the Boat 2 aux rudder (unwired), 10 to 15 = RETIRED (pull-pins), 16 = RETIRED (trim decouple). Release Node B has its own address and its own release channels; the Electronics and Wiring Plan sec 4.3 OWNS the release-channel map. Bought gear keeps its own numbering and the Pi speaks it (the VESC in its own frames, the em-trak in NMEA 2000; on Boat 2 the Electrak in J1939). The battery watchman (SBMS0), the small board that watches the battery, is on no bus and controls nothing. Its readings reach Pi A over USB, through the Electrodacus WiFi/USB add-on board and the USB hub box beside Pi A, and that stream is required, because the battery percentage comes from it. Charging is handled by the central node's relay channel 8 (section 5).

2. Inside the team-built nodes

The three team-built boxes share one skeleton. The central node is the full build: sensor breakouts (humidity, and a BME280 barometer moved inside from the deleted post weather pod; no gas sensor breakout: deleted, do not re-propose), one 8-channel opto board, two 8-channel relay modules, the MOSFET boards, the four cord puller reversing relays (section 3), the Ewellix reversing drive and its steering power-cut relay, and one MCP23017 I2C expander, a 16-output chip on a screw-terminal board, about $6. The expander drives only the convenience outputs: the six switched-load relays (Compute Bay Plan sec 11 owns the map), the charging relay on channel 8, the steering power-cut coil and the second relay module's channels 9 to 12; its last four pins read slow inputs (Compute Bay Plan sec 10.2). Pull-down resistors under its terminal screws make a reset read as OFF. If the expander or the I2C line dies, its outputs float: the loads drop out (same as a dead node, already accepted), the charging relay drops out so charging carries on, and the steering power-cut relay falls back to CLOSED, so steering keeps its feed. Every release channel and the release enable relay (channel 5 of the first relay module) stay on the Nucleo's own pins, so the expander is never in the release chain. The pin count is in the Compute Bay plan, sec 10.2. The central node, Release Node B and the wet lab box each get a NUCLEO-G431KB (the wet lab box's is to buy), which fits the screw-terminal carrier; the G431RB stays on the workbench for testing. Release Node B is the short build: the Nucleo on its screw-terminal carrier, the CAN transceiver, an enable relay and its MOSFET pairs, fed through its own breaker. Any later node would look the same (for example a small forward node for the lightning detector, if the AS3935 false-triggers next to Pi B). Nothing is soldered: the Nucleo plugs into a screw-terminal carrier, every board has screw terminals, the DIN blocks take ferrules, and the 5 V converter has wire leads.

plastic IP67 box, glued to a pad, cables in through glands at the bottom 24 V feed ingland 15 A autoreset breaker DIN terminal rail24 V, ground, every fieldwire lands here first 24 V to 5 V converterpotted, fixed, wire leads Nucleo-G431KBon a screw-terminal carrierruns the node's program CAN transceiverthe bus interface to the teeoutside the box Opto input board8 things in (switches,probes, contacts)8 clean yes/no out field Output boardsrelay module (24 V coils) orMOSFET switch moduleNucleo pin says on/off, board switches 24 Vloads via MCP23017 expander pins to solenoid,motor, relay coil 24 V for the loads Sensor breakoutshumidity, BME280, 1-wire probes
Figure 2. The skeleton of the central node; Release Node B is the same pattern with fewer boards (no opto board at all). Field wiring lands on DIN terminals; the Nucleo only ever sees 3.3 V, through the opto board for inputs and through relay or MOSFET boards for outputs. The MCP23017 expander (not drawn) sits on the I2C line and drives the convenience outputs: six switched loads, the charging relay, the steering power-cut coil and the second relay module's channels 9 to 12. The release MOSFETs and the enable relay stay on Nucleo pins. The 24 V that actually moves things never goes through the Nucleo. Not drawn: the central node's release bus has its own 12/2 feeder on a 30 A breaker at the bar, and Release Node B's feed is its own 12/2 on a 30 A breaker (Electronics and Wiring Plan sec 7.6).

What an opto is

An optocoupler, or opto, is a tiny LED and a light sensor sealed in one chip. Current from the field side lights the LED; the light switches on the sensor on the computer side. No metal joins the two sides, only light crosses. So the 3.3 V computer can read a 24 V switch out on the rudder, and the 24 V or a lightning surge on that long lead has no way in. The 8-channel opto board is eight of these with screw terminals: eight field wires in, eight clean yes/no signals to the Nucleo.

field side (24 V, long wires) computer side (3.3 V, Nucleo) LED switch closes → current sensor clean yes/no to a pin only light crosses this gap no wire between the sides: surges cannot cross
Figure 3. An optocoupler in one picture. The board does this eight times over, with screw terminals.

What a 1-wire chain is

DS18B20 temperature probes are called "1-wire" because many probes share one data wire, plus power and ground. Each probe has a serial number burned in at the factory. The Nucleo asks them one at a time, a few times a minute, and puts the temperatures on the CAN bus. A Nucleo can run several chains, one per pin: the central node has chain 1 for the six battery probes and chain 2 for compartment air (no motor jacket or bearing probes: deleted, because the motor's own heat sensor feeds the VESC, which backs off by itself). One 4.7 k resistor across data and 3.3 V, held under a terminal screw, is the only extra part.

Nucleo3.3 V, GND, pin D11 3.3 V data GND 4.7 k probe 28-A1… probe 28-7C… probe 28-F0…
Figure 4. Three probes on one chain (chain 1, on pin D11; chain 2 is on D12). Add probes by tapping the same three wires; the serial numbers tell them apart.

3. The release channels, step by step

These channels do things you cannot take back (drop a drogue lid, cut a burn wire), so they are built like a two-key vault: two separate conditions must both be true before anything moves, and with no power at all nothing can move.

Release Node B carries the horn-B channel of each drogue lid; the central node carries the rest. Channels 10 to 15, the old steering pull-pins, are retired. Each node has its own enable relay and release bus, and the two channels of each drogue box run to their nodes by separate routes (Transom Drogue Box Guide), so one dead box or one flooded trunk cannot take out every release.

The four cord pullers (channels 3 to 6) are motors, and nobody can crawl aft in the dark to test them. So each runs through its own reversing relay in the central node: a 24 V DPDT relay (two switches that move together; MY2NJ type, in a screw-terminal DIN socket, no soldering) that swaps the motor's two wires. All four coils sit on relay channel 11, "reverse". A shore test is reverse on, fire puller N as usual, reverse off. The lock does not change: channel 5 and the puller's own MOSFET must both be on, reverse alone moves nothing, and the software refuses reverse unless a shore test command is active. Each puller's 1N5408 flyback diode sits at its relay's input (MOSFET drain to the release bus), not across the motor, so it works both ways. No extra wires aft.

24 V infuse ENABLE relayopen when idle RELEASE BUS: dead until enabled, times out by itself Nucleo closes it on ENABLE(relay module channel 5) Channel type A: solenoid (T8 actuators) solenoidor actuator flyback diode MOSFETon = RELEASE N ground Nucleo pin Channel type B: burn wire (drogue lids) burn wirenichrome coils SHORT relayDELETEDtwo conditions remain MOSFETon = RELEASE N ground Nucleo pin Opto input board (readbacks DELETED) • confirm switches: all DELETED (drogue lid, pull-pins, JSD, bridle) - speed drop + aft camera • daily resistance test: DELETED • relay-state readback: DELETED Nucleoreports on CAN to Pi A"channel 3 released" (commanded)
Figure 5. Two kinds of channel on one release bus. Idle: enable relay open, release bus dead, MOSFETs off, so nothing can release even if a wire chafes to 24 V. Enabled: the bus is live for a timed window. Release N: the node turns on that channel's MOSFET. DELETED on both nodes (do not re-propose): the burn-wire short relays, the confirm switches, the daily resistance test and the relay readbacks. The node reports what it commanded, and a dropped drogue is confirmed by speed drop and the aft camera. On the central node the enable relay is channel 5 of the relay module; Release Node B has its own. Drawn before v10.0. Where it differs from the text, the text is right: channels 3 to 6 are the cord puller motors, not solenoids, each through a reversing relay whose coil is on relay channel 11, and the flyback diode sits at the relay's input, not across the motor.
StateEnable relay (condition 1)Release busBurn-wire short relay (DELETED)MOSFETs (condition 2)Can anything release?
No power at allopendeadnoneoffNo
Idle, poweredopendeadnoneoffNo, even with a shorted wire
Enabled (signed command)closed, timer runninglivenoneoffNot yet
Release channel NclosedlivenoneN on for set timeOnly channel N
Timer expiresopensdeadnoneoffNo

The parts, and why each is there

PartJob in one lineWhy not skip it
Fuse on the feedProtects the wire into the boxNormal practice; a short in the box must not take the branch down
Enable relay (central node: relay module channel 5)Cuts all release power until the Pi says ENABLE, then only for a timed windowThe first of the two separate faults needed to release. A relay is a real air gap; software cannot leak through it
MOSFET switch per channelThe Nucleo's RELEASE command turns one channel on for a set timeThe second separate fault. It carries the solenoid or burn-wire current the Nucleo pin cannot
Flyback diode (1N5408) per cord puller, at its reversing relay's inputAbsorbs the motor's kick when switched off, in either directionWithout it the kick kills the MOSFET. The burn wires need none
Reversing relay per cord puller (four, MY2NJ type, coils on relay channel 11)Swaps the motor's two wires so shore can run the puller backward for a testWithout it a fired puller cannot be reset without crawling aft; the relay adds no wire aft and does not touch the two-condition lock
Normally-closed short relay across each burn wire - DELETEDWas: shorted when idle so no stray current could heat the wireDropped on both nodes: the release bus is dead until the enable relay closes, and each channel also needs its MOSFET on. Two conditions are enough
Opto input board (central node only, one board)Reads the two Water Witch pump states (24 V present = pump running) without connecting 24 V to the computerLong field leads catch surges; with only light between the two sides, they cannot reach the Nucleo. Release Node B has no opto board
Nucleo on the CAN busDoes the sequence, keeps the timer, reports every stepPuts the thinking next to the loads and away from the Pi; if the bus goes quiet it does nothing, which is safe

4. Steering: the central node drives the rudder actuator

Boat 1 has one steering actuator, the Ewellix CAHB-22: a 24 V DC actuator with a self-locking screw and its own position feedback. A reversing DC drive in the central node runs it (a Pololu G2 24v13 motor driver), with current sensing to catch a jam. It is not a bus device. Not fitted on either boat: the FSESC box, its backup rudder motor and a second actuator. BOAT 2 ONLY: the steering backup is an aux rudder, and the primary actuator keeps one pull-pin so a dead actuator can let the rudder trail.

A power-cut relay in the central node sits in the Ewellix feed, ahead of the reversing drive. It is the Pi's way to drop a jammed or runaway unit. That relay rests closed with no power on its coil, so a dead central node leaves the actuator fed: cutting actuator power takes a live node and a command. BOAT 2 ONLY: the primary rudder is a Thomson Electrak, a J1939 device that takes position commands from Pi A on the control bus, holds on its internal brake, and reports position and overload, behind the same power-cut relay. Encoder A, the angle sensor on the rudder stock magnet, lands in the aft junction and rides the sensor trunk to the central node.

Pi Arudder-angle command CAN FSESCdriver + position loop 3 phase wires BLDC motorbackup rudder phase-short relayno power = wires joined = brake AS5048A encodertrue rudder angle encoder B on the FSESC's own port own 24 V feed, own fuse
Figure 6. The FSESC box, drawn before v10.0. Where it differs from the text, the text is right: no FSESC box is fitted on either boat. As drawn: rudder-angle commands go over CAN from Pi A to the FSESC, which drives the motor and reads its own encoder; at rest the relay joins the motor wires, so a dead box still damps the rudder.

5. Monitoring in the central node: volts, amps, temps, and the switched loads

An INA226 is a $6 chip that measures volts and amps on one wire, by watching the tiny voltage across a shunt (a precise low-value resistor the current flows through). The branch current meters are DELETED (do not re-propose): the bilge pump and Starlink/PoE INA226 modules and their I2C chain out of the central node are gone (a running pump is known from the Water Witch 24 V output). The only INA226 left on the boat is the feed monitor inside the Pi A box.

The central node reads: the two Water Witch pump states through its one opto board; the battery (six probes) and compartment DS18B20 chains; the humidity breakout and the BME280 barometer on the I2C chain; rudder encoder A on its PWM output; and the post wind sensor (Calypso CMI1018) on a UART. Also deleted: the high-water sensor (for good), the shaft seal probe, the shaft rpm sensor, the motor jacket and bearing probes, the nav light current sense, and the surge-device status contacts (check the bus board's surge modules and the PV surge devices with a meter at each refit and after a strike, not under way). No solar data reaches this box: the five MPPT VE.Direct ports carry the remote on/off cables, and Bluetooth is OFF on the MPPTs (too much watt-hour draw), so there is no per-string data at all.

Its first relay module carries eight channels, none spare (Compute Bay Plan sec 11 owns the map): 1 Starlink power, 2 the science group, 3 the side and stern nav lights, 4 the AIS transponder, 5 the release enable relay, 6 bilge pump 1's manual-on, 7 the motoring whites coil, 8 the charging relay. A second 8-channel module (probably on the shelf already: two were bought in August) runs off the expander: 9 the underwater camera port's UV-C lights, 10 the Orion-Tr charger that tops up the emergency battery (90 minutes a day, and as long as the monthly emergency-battery test's recharge needs, about 14 hours, an estimate; through its remote on/off terminal), 11 the cord puller reverse (the coils of the four reversing relays, shore tests only, section 3), and 12 bilge pump 2's manual-on. Channel 13 is the horn, driven from a native Nucleo pin. Channels 14 to 16 have no driver, and the expander pins of 13 to 16 read slow inputs, so all 16 expander pins are used. The expander drives every coil but channels 5 and 13, which stay on Nucleo pins (section 2). The TSW202 network switch is on no channel: it is always on. Four channels fail ON, through normally closed contacts, so a dead node or expander leaves them on: the nav lights (the boat stays lit), the AIS transponder (the boat stays visible to shipping), the charging path (the boat keeps charging) and the camera port UV-C (it runs nonstop instead of 5 minutes on and 10 off, so the window stays clear). The Orion-Tr on channel 10 is normally open, so a dead node leaves it off; the emergency battery's own two 50 W panels keep it charged. Channel 11 is normally open too: dead, the pullers run forward only. Channels 9 and 10 each leave the box through their own gland. The other load channels fail OFF, which is why nothing survival-critical runs through them. The release enable relay fails open, so the release bus stays dead. A dead Pi never leaves the boat dark either: on the same timer as the charging stop (below), if the node hears nothing from either Pi for the set time, it turns the nav lights on and keeps them on until a Pi is back.

Charging. It is on by default in hardware. Battery positive runs through a 1k 1 W resistor at the main bus board's positive bar, ahead of the branch breakers, then on its own thin wire, with no fuse, to this box and through relay channel 8's normally closed contact to five Victron VE.Direct remote on/off cables (ASS030550320), one per solar charger. The 1k limits a short anywhere on that wire to about 29 mA. The five cables meet on their joining strip on the main bus board, and one run comes from there to beside relay 8 in this box. Power the coil and all five chargers stop; drop it and charging resumes. Only shore can stop charging, through the Pi, and the stop expires after 5 days unless renewed. If the node hears nothing from either Pi for a set time while it holds a stop, it drops channel 8 by itself, so a dead Pi never holds the stop either. The Pi can never enable charging, and the watchman switches nothing. Not fitted: the watchman's charge pause on EXTIO4 and the node's 10-minute charge-continue bypass (withdrawn, do not re-propose).

24 V branch, e.g. Starlink shunt to the load INA226branch meters DELETED I2C Nucleovolts + amps → CAN to Pi A
Figure 7. One branch measured, kept for reference. The branch meters on the central node are DELETED; the pattern lives on only as the feed monitor in the Pi A box, which reports to the Pi, not to this Nucleo.

6. Why this arrangement

In Design Decisions and Rationale, and the change history (Archive, v9.7e set).

7. Options considered

In Design Decisions and Rationale, and the change history (Archive, v9.7e set).

8. Numbers you will see on lids and in the docs

ThingNumber means
Node address (e.g. 40)Which box a CAN message is for or from (40 = the central node; bought devices keep their own)
Message ID (e.g. 40-02)What the message is (enable, release, status, temperature report)
Release channel (1 to 6 live; 7 to 9 reserved for Boat 2, 10 to 16 retired)Which solenoid or burn wire, printed at both ends of its cable
Probe serial (28-xx…)Which DS18B20 a temperature came from; read once at the bench, then written into the config
I2C addressWhich device on the central node's I2C chain: humidity breakout, BME280 barometer, MCP23017 expander (set by its address jumpers). The INA226 branch meters and their 0x40 to 0x4F range are DELETED
Opto board input 1 to 8 (central node only)Which field contact; only the two Water Witch states are wired. Label the DIN block the same
Relay channel 1 to 16 (central node, two modules)Which switched output: 5 is the release enable relay, 8 the charging relay, 9 the camera port UV-C, 10 the Orion-Tr, 11 the cord puller reverse and 12 bilge pump 2's manual-on; 13 is the horn, on a native Nucleo pin; 14 to 16 have no driver; the rest are loads (Compute Bay Plan sec 11 owns the map). Print the map on the lid

Companion explainer to the Electronics and Wiring Plan; kept with the v10.3 set but not a numbered member. Decision records are in Design Decisions and Rationale; the change history is in Archive (v9.7e set).