[clbears]

Cascadia dashboards for TEL, MDI, DRD and STR

Rebuilds the four board dashboards for Cascadia, fixes cascadia.dbc where it disagrees with firmware, and adds a scenario simulator for testing dashboards without the car. The Brightside dashboards are unchanged.

Tested with simulated data and with real Cascadia data from the 2026-10-24 bench integration test (see Testing).

Layout (same on all four boards)

• Fault banner across the top, full width so it is noticed out of the corner of your eye:
  • Blank when everything is normal.
  • Red "ACTIVE FAULT: …" while a firmware fault bit is set. It names the bits.
  • Orange "WARNING: …" when something is wrong that isn't a fault bit: link lost, board rebooted in the last minute, GPS with fewer than 4 satellites or no 3D fix, CAN bus load ≥ 70 %, motor RTD not reading, motor in the 95–100 °C caution band, DRD cyclic-data timeouts, DRD drive state INVALID.
  • Faults and warnings together show red, faults first.
• Health row: link freshness (seconds since the board's last frame, red after 3 s), uptime, MCU Vdd and temperature (red at the firmware limits), and the board's key state. Tiles stay unfilled unless something is abnormal.
• Fault strip: one row per fault bit. Grey OK, red FAULT while set, then yellow CLEARED for the rest of the time range, so a fault that came and went isn't missed.
• Grouped data rows: trends with units and the alarm band drawn in (at most 4 traces each), and state timelines with a word for every state.
• Raw / diagnostic row (collapsed): every signal the board sent, with its latest value and arrival time, for bench work and bug reports.
• Titles follow docs/GRAFANA.md. Fixed uids cascadia-prod-{tel,mdi,drd,str}, and a link bar to switch between the Cascadia boards.

TEL	MDI	DRD	STR

More screenshots: GPS warning, motor overtemp, raw diagnostic row, simulator

TEL, GPS lost (warning, not a fault)

MDI, motor past 100 °C (fault)

DRD, raw diagnostic row expanded

Simulator control panel (tools/cascadia_sim)

DBC changes (dbc/cascadia.dbc), checked against firmware_v4 origin/main ecd8735

Message	Change	Firmware evidence
TEL 0x764	TELMCUHealth (MCU_Voltage mV, MCU_Temperature °C) replaces CPULoad; no firmware sends a CPU load	tel/app/diagnostics.c:74-93
TEL 0x751	+ MCUOvertemp, MCUUndervoltage (bits 5-6)	tel/app/diagnostics.h
MDI 0x501	+ MDIRTDFault, MDIRTDCommError (bits 3-4)	mdi/app/diagnostic.h:14-23
MDI 0x502	DLC 5→6, + RTDFaultFlags (byte 5) and a value table	mdi/app/diagnostic.c:96-131
DRD 0x402	new DRDMCUHealth	drd/app/diagnostic.c:142-157
DRD 0x403	+ DRDMCUOverTemp, DRDMCUUnderVoltage (bits 38-39)	drd/app/diagnostic.h:20-32
STR 0x591 / 0x592	new STRMCUHealth, STRFlags (bit 0 = undervoltage, bit 1 = overtemp; brightside.dbc has them reversed)	str/app/diagnostic_app.{c,h}

scripts/check_dbc.py passes. A hand-packed frame for each changed message decodes to the value the firmware encodes.

Panels from the Brightside boards that were dropped or renamed

• TEL GPSLatitude/Longitude/Fix/SatCount are now GpsLatitude, GpsLongitude, GpsFixType and GpsSatsUsed.
• DRD SpeedUnitKPH is now MotorCommand.SpeedUnits.
• MDI MCUVoltage/MCUTemperature/MDIOverVoltage are dropped: MDI has no MCU-health code, and 0x503 carries the commit hash.
• STR STRWatchdogReset is dropped: nothing sends 0x582.

Simulator (tools/cascadia_sim, separate commits, can be dropped)

Realistic, adjustable TEL/MDI/DRD/STR data for testing dashboards without the car. Frames use firmware IDs, rates and fault thresholds, are encoded and decoded with cascadia.dbc, and are written with sunlink's InfluxDB layout.

python tools/cascadia_sim/cascadia_sim.py                       # browser control panel on localhost:8090
python tools/cascadia_sim/cascadia_sim.py --headless --scenario motor_overheat --duration 120

• Control panel: 13 one-click scenarios (GPS lost, motor overheat, RTD failure, STR brownout, link down, …), toggles and sliders per board, and "Reset everything to normal", which also clears the simulated fault history.
• Tagging: every simulated point is tagged source=cascadia_sim, and reset deletes only those points. Real logs imported into the same local bucket are never touched. Dashboards ignore the tag.
• Safety: refuses to write anywhere but localhost unless --allow-remote is passed.
• How it differs from the existing CAN simulators:
  • firmware_v4/tools/Simulate_CAN_Bus (also in sunlite) replays fixed raw payloads from YAML onto a real CAN bus (socketcan). It is for bus timing, load and startup tests on hardware.
  • drd/test/test_scripts/simulate_mc.py fakes the motor controller for DRD bench tests.
  • This one needs no CAN hardware. It produces changing, realistic signal values with firmware fault logic straight into InfluxDB, so dashboards can be tested and demoed scenario by scenario. The tools complement each other.

Design basis (user interview, 2026-10-31)

• Who: pit crew during races; embedded on the bench.
• Situation: mostly live race and test drives (last 5 min, 1 s refresh).
• Decisions made from these boards: call the driver or stop the car; reboot or swap a board; file a bug for embedded.
• Doing at the same time: radio, other screens, typing/debugging, so attention is often elsewhere.
• Must be noticed peripherally: active faults and warnings, on the banner.
• Past faults: yellow CLEARED on their fault-strip row.
• Screens: a mix. The first screen (banner, health, fault strip) fits a ~15" laptop at 1440×900; detail sits lower.
• Threshold sources:
  • Firmware: MCU undervoltage/overtemp (TEL/DRD 2000 mV / 125 °C, STR 2400 mV / 105 °C), motor overtemp 100 °C set / 95 °C clear.
  • Designer default, confirmed by the user: link lost after 3 s; GPS < 4 satellites / no 3D fix; CAN bus load 70 % / 90 %; GPS accuracy 5 m / 20 m; PDOP 2.5 / 6; "rebooted" for 60 s after boot.

The layout follows situation-awareness and high-performance HMI principles (Endsley 1995, PAS High Performance HMI, Wickens 2008, FAA HF-STD-001B).

Testing

• Local stack: grafana-oss 9.5.2 + influxdb 2.7.1, fed by the simulator.
• DBC lint: every panel field exists for its board in cascadia.dbc (0 errors, 0 warnings on all four dashboards).
• Live query check: all 61 panel queries return data. Each scenario was run and its banner text, tiles and fault rows checked.
• Real Cascadia data, 2026-10-24 bench integration test (current firmware apart from the Oct 31 commits; decoded at the time with the old cascadia.dbc). The data is in CAN_prod on the Sunlink laptop used for that test (ckalitin-victus, 19:39–23:47 UTC). On a 30-minute slice (603k points), every panel whose signal was decoded shows real data, and the banners catch the real bench problems from that day:
  • DRD's cyclic-data timeouts (the motor controller only sent request frames).
  • A latched MDI watchdog reset.
  • TEL GPSReadFail toggling.
• Expected empty panels on that data:
  • MCU health (TEL/DRD/STR), STRFlags, RTDFaultFlags and SpeedUnits weren't in the old DBC, so they weren't decoded.
  • GPS sent no frames all day.
  • IMU sent only 302 frames in 1.5 h (IMU over CAN was fixed in firmware on Oct 31).
• DBC fix confirmed on real frames: the old DBC stored TEL 0x764 as CPULoad, reading about 2e-39. Re-decoded with this PR's layout, the same 28 frames give MCU_Voltage 3419–3467 mV and MCU_Temperature 21–25 °C. The slightly high voltage fits the uncalibrated TEL ADC noted below.
• Also tested: a 30-minute bench log from 2026-08-27 (team InfluxDB, CAN_log).
• Still to see on hardware: GPS, IMU and the new MCU-health/fault signals, once sunlink runs on the new cascadia.dbc.

🤖 Generated with Claude Code