Catching the Islanding Fault That Hides Behind a Matched Load

An anti-islanding fallback that works at idle proves almost nothing about a bidirectional charger exporting at rated power into a local load that absorbs exactly what it produces. Under that condition the passive voltage and frequency trips see no excursion. The only thing between the island and a line crew is the active detection loop in the vehicle's inverter. The qualification question is whether that loop keeps working when the part that coordinates it, the charging session, fails during export.

Where redundancy quietly collapses

In an AC bidirectional architecture the vehicle carries the inverter and the wallbox carries the contactors, the grid code profile and the session, so anti-islanding depends on two parties that only coordinate over a communication link. When that link drops, most stacks fall back to a hold state: they keep the last power command, wait for the session timeout, then ramp down. The defect that keeps turning up is that hold also freezes the active perturbation, because the frequency shift injection was implemented as part of the commanded setpoint rather than as an autonomous loop in the inverter firmware. At light export the island collapses on its own, passive protection trips and the bug never shows. At matched rated export the island is stable, the injection is frozen, and nothing trips until the session timeout finally opens the contactors.

The budget nobody sized for this

Grid interconnection rules such as IEEE 1547 and UL 1741 allow 2 s to cease to energize an unintentional island, and a session timeout sized for charging convenience rather than protection can use up most of that budget before the fallback even starts to ramp.

The test that exposes it

Build the standard resonant RLC island and tune it so real and reactive power through the utility breaker are both near zero with the vehicle at rated export and the quality factor set where the standard requires. Then do the one thing the type test never does: cut the session link at a controlled instant, and sweep the delay between losing communication and opening the utility breaker from well before to well after the timeout. Record contactor state, inverter output current and island frequency on a common time base, and measure cease to energize time at every point of the sweep. The worst point is usually where communication loss comes slightly less than one session timeout before the island forms, because the stack is already in hold when the island appears. My estimate is that this window is a few hundred milliseconds wide, narrow enough that a test with a fixed offset will step right over it. Repeat the sweep after a thermal soak at full export and retune the load first, because derating changes the actual output current and detunes the match, which hides the fault again from the other side.

Deciding what counts as qualified

The rule I apply is blunt: a fallback path is not qualified until it has been triggered at the operating point where the primary protection is weakest, and for anti-islanding that point is matched load at rated export. In practice every degraded mode in the FMEA, whether a lost session, a lost voltage sense channel or a contactor welded on one pole, gets its own row in the matched island sweep instead of a separate functional test at nominal conditions. The cost is real, since each sweep point needs a retune and the RLC bench is the scarcest asset in the lab. The cheaper fix sits upstream: move the active detection loop fully inside the inverter so it never depends on the session, and the sweep shrinks to a confirmation run. Whether the vehicle supplier accepts that architectural change before SOP is the next decision, and it gets settled in negotiation, not on the test bench.

Back to AutomotiveV2G · Domains

← Back to AutomotiveV2G
Engineering & Architecture Brief

Catching the Islanding Fault That Hides Behind a Matched Load

An anti-islanding fallback that works at idle proves almost nothing about a bidirectional charger exporting at rated power into a local load that absorbs exactly what it produces. Under that condition the passive voltage and frequency trips see no excursion. The only thing between the island and a line crew is the active detection loop in the vehicle's inverter. The qualification question is whether that loop keeps working when the part that coordinates it, the charging session, fails during export.

Where redundancy quietly collapses

In an AC bidirectional architecture the vehicle carries the inverter and the wallbox carries the contactors, the grid code profile and the session, so anti-islanding depends on two parties that only coordinate over a communication link. When that link drops, most stacks fall back to a hold state: they keep the last power command, wait for the session timeout, then ramp down. The defect that keeps turning up is that hold also freezes the active perturbation, because the frequency shift injection was implemented as part of the commanded setpoint rather than as an autonomous loop in the inverter firmware. At light export the island collapses on its own, passive protection trips and the bug never shows. At matched rated export the island is stable, the injection is frozen, and nothing trips until the session timeout finally opens the contactors.

The budget nobody sized for this

Grid interconnection rules such as IEEE 1547 and UL 1741 allow 2 s to cease to energize an unintentional island, and a session timeout sized for charging convenience rather than protection can use up most of that budget before the fallback even starts to ramp.

The test that exposes it

Build the standard resonant RLC island and tune it so real and reactive power through the utility breaker are both near zero with the vehicle at rated export and the quality factor set where the standard requires. Then do the one thing the type test never does: cut the session link at a controlled instant, and sweep the delay between losing communication and opening the utility breaker from well before to well after the timeout. Record contactor state, inverter output current and island frequency on a common time base, and measure cease to energize time at every point of the sweep. The worst point is usually where communication loss comes slightly less than one session timeout before the island forms, because the stack is already in hold when the island appears. My estimate is that this window is a few hundred milliseconds wide, narrow enough that a test with a fixed offset will step right over it. Repeat the sweep after a thermal soak at full export and retune the load first, because derating changes the actual output current and detunes the match, which hides the fault again from the other side.

Deciding what counts as qualified

The rule I apply is blunt: a fallback path is not qualified until it has been triggered at the operating point where the primary protection is weakest, and for anti-islanding that point is matched load at rated export. In practice every degraded mode in the FMEA, whether a lost session, a lost voltage sense channel or a contactor welded on one pole, gets its own row in the matched island sweep instead of a separate functional test at nominal conditions. The cost is real, since each sweep point needs a retune and the RLC bench is the scarcest asset in the lab. The cheaper fix sits upstream: move the active detection loop fully inside the inverter so it never depends on the session, and the sweep shrinks to a confirmation run. Whether the vehicle supplier accepts that architectural change before SOP is the next decision, and it gets settled in negotiation, not on the test bench.

Related Architecture Specifications (AUTOMOTIVESDV): selfdrivingtwin.com · vehicle2infrastructure.com · vehiclenodes.com

More Technical Articles on automotivev2g.com

Implementing IEEE 1547 Islanding Protection in Vehicle Inverters →Wear-Free Reactive Power Support: Monetizing V2G Without Cycle Degradation →