The iteration O 3-tick ramp ended with dcCurController.SetRef(0) before
transitioning to STOP. Param::Change only fires when idcspnt or idclim
parameter values change — a normal STOP → OFF → WAITSTART → ENABLE →
ACTIVATE → RUN → EVSEACTIVATE restart cycle never touches those, so
nothing restored the PI reference. On the next entry to EVSEACTIVATE
the controller targeted 0 A and the charger sipped ~0.1 kW.
Bench-confirmed 2026-05-21 on iteration O firmware: after lowering the
ChargeLimit_pct to drop into COMPLETE and then raising it back, charging
restarted at 0.1 kW instead of ramping to 9 kW.
Fix: on the final ramp tick, restore SetRef to MIN(idcspnt, idclim)
instead of zeroing it. The GPIO clears (acpres_out, evseact_out) still
cut AC enable so the output current decays through the filter caps;
the reference value is irrelevant during STOP but matters for the next
EVSEACTIVATE entry.
The ramp-restore block in the vcucmd != Complete branch (charger.cpp:431
in iteration O) only handles the mid-ramp reversal case (counter ≠ 0);
it didn't help when the ramp completed cleanly. This commit closes that
gap.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
When vcucmd transitions to COMPLETE while in EVSEACTIVATE the previous
behavior was to clear acpres_out / evseact_out and transition to STOP
in a single 100 ms tick — cutting the OBC AC enable immediately. The
VCU's HV contactor opens at roughly the same moment; under active 9 kW
charging (~23 A) this hard disconnect transient floods the Tesla BMS
state-change CAN path and starves the VCU's task scheduler past the
IWDG window. Bench-confirmed 2026-05-21 as the trigger for the
iteration N limit-lower 0x002A reset loop.
Add a 3-tick (300 ms) ramp inside EVSEACTIVATE before the STOP
transition: stash the original DC-current reference, scale it 2/3 →
1/3 → 0 across the remaining ticks, and only then clear the AC enable
GPIOs and go STOP. The PI controller's natural response walks the AC
current limit down so output current decays smoothly rather than
hard-stopping.
If vcucmd reverts to Charging mid-ramp (user raised limit again),
restore the original reference cleanly before the existing branches
take over.
The VCU also defers its own HV-contactor open by 1500 ms (see VCU
iteration O commit 5781f47). Belt-and-suspenders: either fix alone
substantially reduces the disconnect transient; together they make
RUN → STOP graceful on any controller that interoperates with this
charger.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Promote 0x212 byte 3 from S11's binary vcustop to the full Dilong-style
OBC_ControlCMD vocabulary:
0 = Charging — walk to / hold at EVSEACTIVATE, produce current
1 = Stopped — sit idle in OFF/WAITSTART/ENABLE, no current
2 = Complete — drop to STOP (cycle terminated by VCU)
VCU is now authoritative for the charge-cycle lifecycle. The charger
keeps hardware-level wisdom (EVSE pilot via CheckStartCondition,
plug-out safety via CheckUnplugged, module-fault detection via
CheckChargerFaults) but obeys vcucmd for "should I be charging?".
FSM changes:
- ENABLE→ACTIVATE now gated on vcucmd==Charging (was unconditional).
ENABLE on vcucmd==Complete drops to STOP; vcucmd==Stopped holds
HV-armed waiting for next Charging command.
- EVSEACTIVATE self-decided exits (CheckVoltage, CheckTimeout) are
gone. Those were band-aids for the absence of an authoritative VCU
stop signal. vcucmd==Stopped now parks back in ENABLE (HV held);
vcucmd==Complete drops to STOP. CheckUnplugged + CheckChargerFaults
still own hardware-safety exits.
- STOP→OFF now also triggers on vcucmd != Complete (was unplug-only).
This is the fix for the "raise ChargeLimit mid-COMPLETE doesn't
restart charging" wedge — the VCU's vcucmd flip from Complete to
Stopped on a limit-raise lets the natural OFF→WAITSTART→ENABLE→
ACTIVATE→EVSEACTIVATE flow resume charging without an unplug cycle.
Param rename vcustop→vcucmd. External tooling that reads/writes the
S11 vcustop param via the openinverter web UI will see the new vcucmd
param with 3-state semantics — single-user bench, acceptable.
VERSTR bumped to S12.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Bastian de Byl