No-Break Power Transfer (NBPT)

The previous article closed on a single iron rule: the generators are never electrically coupled — except on the ground, during a No-Break Power Transfer. This article is that exception, unpacked. NBPT is what lets a ground source change — APU generator handing over to external power, or an engine generator coming on line at start — happen so cleanly that the cabin reading lights do not even flicker. It is the most distributed function in the whole electrical chapter: the GAPCU sets the beat (it generates the frequency reference), the GCU matches the beat (it drives its generator's speed and phase onto that reference), and the ECMU closes the contactors against a stopwatch (it parallels the two sources for a controlled instant). Articles 03, 04 and 06 each described one corner of this; here the corners are assembled into one running story.

By the end you should be able to answer five things: which source pairs have NBPT and which do not (and how to read the FCOM matrix); what a successful NBPT does on a millisecond timescale; the full set of numeric conditions for the synchronisation window, and why the external-power channel does not check phase; the mechanism behind the "up to 15 s for the APU GEN with GPU" line, and what a time-out does; and the behaviour after NBPT fails or is suppressed, including why the suppression latches.

1. Where NBPT sits — the one controlled exception to "never parallel"

FCOM defines the function in four clauses that, between them, name every actor in this article:

"This function avoids busbar power interruption during supply source transfer on ground in normal configuration. It is inhibited in flight. The ECMU controls the simultaneous connection of the two sources for a short time. To achieve this, both sources are synchronized on a frequency reference signal sent by the GPCU or GAPCU."

Per FCOM DSC-24-10-20-30. Unpack the four clauses:

1. Ground only. NBPT is inhibited in flight — the air/ground discriminant (from the LGCIUs) gates the whole function off. In flight, every source transfer is a hard switch with a momentary break, which is acceptable because "electrical transients are acceptable for equipment" (the overview's opening premise — see Electrical Overview).
2. The ECMU is the executor. It is the unit that physically connects the two sources simultaneously for a short time — i.e. it commands a brief, controlled paralleling of two generators that the system otherwise forbids.
3. The GPCU/GAPCU sets the frequency reference. Something has to be the metronome both sources tune to; that is the GAPCU's Frequency Reference Unit (FRU).
4. Both sources synchronise on that reference. The transfer is "no break" only because, at the moment of paralleling, the two sources are at the same frequency and phase, so the circulating current between them is negligible.

The AMM gives the engineering one layer down:

"The NBPT function serves to perform a supply transfer between two electrical sources (IDGs, APU, EXT PWR) without busbar power supply cutoffs. On the basis of the data provided by the GCUs and the GPCU (GAPCU), the ECMUs manage the opening and closure of the line contactors (GLCs, APU GLCs, EPC) and of the transfer contactors (BTCs and SIC) in order to enable the paralleling of the two different sources during a short time without damaging the two sources."

Per AMM 24-29-00 §6.M(1). The phrase "during a short time without damaging the two sources" is the entire engineering tension of NBPT: paralleling two generators is dangerous if they are out of step, and dangerous if it persists — so NBPT is allowed only when they are in step, and only for an instant. The "for an instant" half is enforced by the same ECMU's Inadvertent Paralleling Trip (IPT), which opens contactors if any paralleling lasts beyond 80 ms (a second contactor at 130 ms) — per AMM 24-29-00 §6.N. NBPT and IPT do not conflict: NBPT is fast in and fast out, far inside the 80 ms line; IPT exists to kill paralleling that lingers. (The IPT itself is treated in ECMU and Contactor Management.)

2. The NBPT matrix — which source pairs are seamless

FCOM gives the crew a table and one instruction:

"As per design, the NBPT function is effective provided that the power supply sources connections/disconnections are done according to the below table. If the flight crew does not follow the sequences provided in this table, power transfers without NBPT can be observed."

Per FCOM DSC-24-10-30-20. Read literally: NBPT is not automatic on every source change — it only happens when the crew switches in the directions the table marks as NBPT. Switch the "wrong" way and you still get a transfer, but with a break.

The table is a 5 × 5 matrix. Rows are the source currently supplying the network; columns are the source being connected or disconnected. It is a three-state table, not two:

Legend:

• NBPT — NBPT is operative: the transfer is seamless.
• break — NBPT is not operative: the transfer still happens, but with a power break (a flicker).
• n/a — no electrical transfer is possible: this pairing never arises. The diagonal (a source "to itself", shown "—") is part of this state.

The AMM text pins the same logic down in words:

"The NBPT function is effective in the following configurations: ‐ transfer between the APU generator and any IDG, if the APU generator initially supplies the electrical network of the aircraft, ‐ transfer between an external power unit and any IDG, if this external power unit initially supplies the electrical network of the aircraft, ‐ transfer between an external power unit and the APU generator, if this external power unit initially supplies the electrical network of the aircraft, ‐ transfer between the APU generator and the external power unit A, if the APU generator initially supplies the electrical network of the aircraft."

Per AMM 24-29-00 §6.M(4). Every entry in that allowed list carries an "initially supplies" direction condition — and that single fact explains all three of the matrix's odd features:

1. Both IDG rows are entirely "break". An IDG that is supplying has no NBPT at all. The allowed list never names an IDG as the initial supplier — it is always the APU or external power that "initially supplies". This follows the priority order from Network Priority: the engine generator is the highest-priority source, so when an IDG is on the bus, connecting a lower-priority source produces no transfer (the IDG keeps the bus). NBPT for an IDG happens only the other way round — when the IDG comes on line (engine start) or goes off line (engine shutdown) while the APU or external power is the one supplying.

2. EXT A ↔ EXT B is asymmetric — and this is the trap. Read the figure cell by cell: EXT A row, EXT B column = n/a (when EXT A supplies, EXT B can never get onto the bus, so no transfer even exists), while EXT B row, EXT A column = break (a transfer that occurs, but with a break). The two directions are different states. It is wrong to flatten this into "the two externals swap with a break" — in one direction there is no transfer at all. The asymmetry traces directly to the per-bus priority strings the AMM lists alongside the occurrences:

   "1XP = IDG 1/APU GEN/EXT PWR B/EXT PWR A/IDG 2, 2XP = IDG 2/EXT PWR A/APU GEN/EXT PWR B/IDG 1."

   Per AMM 24-29-00 §6.M(4). On the left bus EXT B outranks EXT A; on the right bus EXT A outranks EXT B — so the two receptacles never sit at the same rank, and the "winner" depends on direction.

3. APU row, EXT B column = n/a. When the APU is supplying, EXT B can never get on the bus — it is the permanent bench-warmer of Network Priority. This dovetails exactly with the AMM allowed list, which names only "APU GEN ↔ external power A" and never B.

[!warning]- "Two IDGs on line, swap sources — surely that's seamless too?"

No. Both IDG rows in the matrix are "break"; an IDG that is supplying is never the initial supplier in any NBPT pairing. Seamless transfer only ever involves an IDG when the IDG itself is the one coming on line or going off line, while the APU or external power holds the bus. With two healthy IDGs each feeding their own half-network, there is no NBPT scenario at all — the question does not arise.

3. The actors — who sets the beat, who matches, who closes the contactor

NBPT is a three-role play, and each role lives in a different box:

GAPCU  (the metronome)
  ├─ FRU L  ──►  GCU 1 channel + APU channel  (left half)
  └─ FRU R  ──►  GCU 2 channel                (right half)
        │   RS 422 dedicated wiring carries the frequency reference;
        │   if the GAPCU sees a non-synchronous parallel is possible,
        │   it inhibits the reference  (latched)
        ▼
GCU / GAPCU  (the matcher + the referee)
        │   drives its generator's speed/phase onto the reference;
        │   all conditions met ──► sends a "synchronisation window"
        │   signal to the ECMU  ("I am ready to be handed over seamlessly")
        ▼
ECMU  (the executor)
        │   request + window ──► parallels (closes both source contactors)
        │   ──► quick replace ──► deletes the parallel
        │   IPT stopwatch supervises throughout  (80 / 130 ms)
        ▼
Result: busbar voltage with zero interruption

The FRU itself is the GAPCU's quiet centrepiece:

"The FRU is located in the GAPCU. The GAPCU generates two separate frequency references. The left reference (FRU L) for the left half of the aircraft (channels 1 and APU). The right reference (FRU R) for the right half of the aircraft (channel 2). The GAPCU transmits the reference signal (RS 422 signal) by separate wiring to each of the two GCU's."

Per AMM 24-29-00 §6.M(5)(c). Two references, one per half-network, sent over dedicated RS 422 wiring — this is why the GAPCU has to be alive for any NBPT to occur, and why a GAPCU fault can take seamless transfer away from both halves at once.

4. A successful NBPT, millisecond by millisecond

The AMM splits NBPT into a "starting" and a "shutdown" configuration, and both are the same trick in mirror image — parallel first, then replace and delete the parallel:

"A source is going to be connected to the busbars: ‐ first: a paralleling is performed between the effective busbar supply source and the source to be connected. ‐ Secondly: the source to be connected replaces quickly the previous one and supplies the busbars; paralleling is then deleted."

Per AMM 24-29-00 §6.M(2). Take the concrete "APU is supplying; engine 1 starts and GEN 1 comes on line" case (a connection):

1. Request. GCU 1 detects an NBPT condition and sends a request (the "D" signal) to its ECMU. The conditions that count as a request are spelled out below.
2. Match the beat. GCU 1 stops regulating to its own internal 400 Hz and instead drives its servo valve to bring GEN 1's frequency and phase onto FRU L — the same reference the currently-supplying APU is already tracking. (This is the "external beat" mode of the IDG speed loop from IDG.)
3. Open the window. When every condition is satisfied, GCU 1 raises the synchronisation window signal to the ECMU.
4. Parallel. Inside that window the ECMU closes GLC 1. For this controlled instant GEN 1 and the APU generator are paralleled — but at the same frequency and phase, so the circulating current is tiny.
5. Replace and split. The ECMU immediately opens the APU's contactor; the hand-over is complete and the parallel is deleted. The whole paralleling lasts far less than the IPT's 80 ms line, which is why NBPT and IPT never collide.

The shutdown case (an engine shut down, handing the bus back to the APU) is the exact mirror: parallel the incoming holder first, then drop the departing source.

The request signal — what makes a GCU/GAPCU decide an NBPT is wanted — has a closed set of triggers:

"The GCUs and GAPCU detect the NBPT condition (each GCU monitors its own channel) and generate the request (D) signals necessary to the appropriate ECMU to perform the NBPT function: ‐ for the GCUs and GAPCU (APU channel): switching of the Generator Control Switch (GCS), engine or APU starting, engine or APU shut down. ‐ for the GAPCU (External power channel): activation of any external power pushbutton switch."

Per AMM 24-29-00 §6.M(5)(a). So the triggers are: a generator control switch operation, an engine or APU start or shutdown, or an external-power pushbutton action — exactly the everyday ground events at which a source change is wanted.

5. The synchronisation window — five numeric gatekeepers

The "synchronisation window" is not a clock window; it is the AND of a set of numeric conditions that the GCU/GAPCU checks before it tells the ECMU "ready". The conditions differ between a generator channel and the external-power channel:

"A signal called synchronization window(s) is generated by the applicable GCU or GAPCU to the ECMU, in order to validate the NBPT request. … This signal depends on the following conditions: ‐ for the GCUs and GAPCU (APU channel): the aircraft is on the ground, the power ready (PR) is true, the difference between the generator frequency and the Frequency Reference Unit (FRU) is less than 0.5 Hz, the average of the three phase voltages at POR is 115 VAC plus or minus 5 VAC, the difference between the generator phase and the FRU phase must be: plus or minus 15 degrees to capture, plus or minus 30 degrees to hold. ‐ for the GPCU (GAPCU): the associated PR equation is true, the EXT PWR frequency is 400 HZ plus or minus 10 HZ, the three phase average is 115 VAC plus or minus 5 VAC."

Per AMM 24-29-00 §6.M(5)(b). Laid side by side:

The one asymmetry worth dwelling on is that the external-power channel does not check phase, and checks frequency as an absolute 400 ±10 Hz rather than as a difference from the FRU. The reason is structural: external power is the reference. A ground power unit cannot adjust its own speed to chase a beat — it has no speed loop — so the system makes the GPU the metronome and synchronises the generator onto it. Whatever can be tuned does the tuning; what cannot be tuned sets the standard.

[!warning]- Why two external sources can never seamlessly swap

Both external-power sources are "metronomes that cannot themselves be tuned". For NBPT you need one side that can adjust its speed/phase onto a reference, and a GPU has no such loop. So EXT A ↔ EXT B has no NBPT in either direction — but, as §2 showed, the matrix is not symmetric about this: EXT A → EXT B is n/a (no transfer occurs at all, A simply outranks B on the bus it holds), while EXT B → EXT A is a break transfer. A ground tow that swaps from cart A to cart B will always flicker — brief the cabin first.

6. Fifteen seconds versus milliseconds — two grades of difficulty

One source pair is conspicuously slow:

"Synchronization may take up to 15 s for the APU GEN with GPU, and some milliseconds in all other cases. If synchronization is not achieved within the allowed time, transfer is performed anyway (without simultaneous connection of two sources). This function has a backup in the GCU."

Per FCOM DSC-24-10-20-30. The AMM states the same time-out from the maintenance side and names the consequence outright:

"On ground during NBPT (APU generator with GPU), a maximum of 15 seconds is necessary for the ECMU to find the NBPT condition. After this time, the break transfer is forced. In all other conditions, only milliseconds are necessary."

Per AMM 24-23-00 §7.A. Two readings fall out.

Why is APU ↔ GPU the slow pair? (This part is an integrative synthesis, not a verbatim claim — the manuals give the 15 s figure but not its cause.) In every other pair, the side being matched is an IDG, whose hydromechanical speed loop chases a reference in milliseconds — that is its day job. The APU generator has no constant-speed drive; to change its frequency the GAPCU has to ramp the whole APU's speed through the engine control (the "ramp" channel of APU Generator and GAPCU), nudging a gas turbine's RPM to chase the GPU's frequency. A turbine's mechanical inertia makes that a seconds-scale action, hence the 15 s allowance. The GAPCU's own fault codes confirm the "ramp" vocabulary: one records "NBPT failed because ramp condition removed … The APU stopped ramping before the end of NBPT time window" (§7).

What does a time-out do? It cancels the seamless, not the transfer. "After this time, the break transfer is forced" — the hand-over still happens, the lights just flicker once. And per FCOM the function "has a backup in the GCU", so the loss of one path does not strand the bus.

7. Failure and inhibition — break before chaos

NBPT degrades in three tiers, each more conservative than the last.

Tier 1 — a window condition is not met. The seamless attempt is abandoned and a normal transfer with a break is performed:

"If one of the above mentioned conditions is not met … then the NBPT is not performed and the conventional transfer occurs with break."

Per AMM 24-29-00 §6.M(5)(c).

Tier 2 — the FRU fails or its reference goes out of limits. Each GCU falls back to its own internal 400 Hz and suppresses NBPT for its channel — without a trustworthy common reference there is no way to guarantee phase agreement:

"In case of FRU failure or FRU reference signal being out of limits, each GCU regulates the frequency to its own internal 400 Hz reference and inhibits the NBPT for its channel."

Per AMM 24-29-00 §6.M(5)(c).

Tier 3 — the GAPCU judges that a non-synchronous paralleling could occur. It then suppresses the frequency reference to the GCUs entirely, blocking all NBPT — and the suppression latches until the GAPCU is completely de-energised:

"If the GAPCU finds that its acquisitions can cause a NBPT with non-synchronous alternators, the computer inhibits frequency references sent to GCUs, which prevents any NBPT. In any case, the NBPT is inhibited definitively until the GAPCU power supply is completely cut off. This means that there is no AC and DC power source (external power(s), APU generator, IDG(s) and batteries) connected to the aircraft electrical network."

Per AMM 24-41-00 §6.A(1)(c). The design philosophy is explicit in that last clause: the inrush of a non-synchronous paralleling is far more damaging than one flicker, so the system would rather latch seamless transfer off — and stay latched until every source including the batteries is removed — than risk a bad parallel. Nothing short of fully de-powering the aeroplane clears it.

When a break transfer happens where an NBPT should have occurred and no network fault is present, maintenance can read a dedicated fault code in the BITE specific-data menu:

Per AMM 24-41-00 §6.A(1)(c). The crew-facing point is blunt: NBPT has no dedicated ECAM warning. The only symptom a pilot ever sees is a flicker at a source change. An occasional flicker is normal (any "break" cell in the matrix, or an APU↔GPU time-out). A flicker on every transfer is a maintenance lead — have engineering read the FC 889–893 family — not a dispatch issue: a "break" transfer is the lawful behaviour of every blank cell in the matrix.

8. Operational discipline — the table is written for the crew

The FCOM matrix is, in the end, an instruction to the crew: if you want seamless, switch in the directions the table marks NBPT. Translated into the ramp:

• Standard turnaround (the golden path). On taxi-in with the APU generator already on line, engine shutdown hands the bus to the APU seamlessly (APU row, IDG column = NBPT). Connect EXT A and select it on: APU → EXT A is seamless (APU row, EXT A column = NBPT). Departure runs the same path in reverse, all seamless. This is exactly the "Aircraft start" and "Aircraft return" case list the AMM enumerates under §6.M(4) — the cases the table was designed to cover.
• APU just started, taking over from the GPU, one flicker. APU ↔ GPU is the 15-second pair; a forced break after time-out is within design. Ignore an occasional one; a flicker every time means engineering should read the GAPCU codes.
• Swapping ground carts (A faults, change to B). Always a break (a blank/n.a. matrix cell). Brief the cabin first.
• Do not pull the APU and then reach for external power. That sequence is backwards — between the APU dropping and the GPU coming up the bus is dead. Confirm the GPU has taken the bus (seamless) first, then shut the APU down.
• Every transfer flickering, all day. Suspect the Tier-3 latch — the GAPCU may have logged a non-synchronous-paralleling risk earlier, and it stays latched until the aircraft is fully de-powered (batteries included). The fix is to report it and let maintenance read FC 889–893, not to keep retrying.

[!note]- Teaching analogy — the relay-race baton zone

NBPT is the changeover zone of a relay race. The zone exists only inside the stadium (ground only). The two runners must be shoulder to shoulder at the same speed and stride (frequency within 0.5 Hz, phase within ±15°) before the judge allows the baton to pass (the synchronisation window). The time they may run together is strictly capped (inside the IPT 80 ms line). The IDG is the seeded sprinter — while it leads, nobody else is allowed onto the track (its rows are all "break"). The APU is the heavyweight: it accelerates and decelerates slowly, so handing over to the GPU needs an extra 15 s of warm-up, and if the warm-up over-runs the race switches to "stop a step, then restart" (a forced break). And the head judge (the GAPCU) — the moment it suspects anyone is about to barge into the lane out of step — closes the changeover zone on the spot, and will not reopen it until the whole stadium is emptied (the aircraft fully de-powered).

Self-test

[!note]- Q1. Name the three odd features of the NBPT matrix, and the one rule that explains all three.

(1) Both IDG rows are entirely "break" — an IDG that is supplying is never an NBPT initial supplier; an IDG only gets seamless transfer when it comes on or off line while the APU or external power holds the bus. (2) EXT A ↔ EXT B is asymmetric — EXT A → EXT B is n/a (no transfer, A outranks B), EXT B → EXT A is a break; two metronomes have no NBPT in either direction, but the directions are different states. (3) APU row, EXT B column is n/a — with the APU supplying, EXT B never gets on the bus. The single rule behind all three is the AMM allowed list, where every NBPT pairing requires the APU or external power to "initially supply" — never the IDG, and only EXT A (not B) pairs with the APU. Legend is three-state: NBPT operative / blank = not operative (break) / grey = no transfer possible.

[!note]- Q2. Walk through the five steps of a successful NBPT.

Request (the GCU/GAPCU detects a GCS switch, an engine/APU start or shutdown, or an external-power pushbutton, and signals the ECMU) → match (the GCU abandons its internal reference and drives its servo valve onto the FRU) → window (all conditions met, it raises the synchronisation-window signal) → parallel (the ECMU closes the contactor inside the window, the two sources momentarily paralleled at the same frequency and phase, tiny circulating current) → replace and split (the ECMU drops the departing source; the parallel is deleted, far inside the IPT 80 ms line).

[!note]- Q3. List the five synchronisation-window conditions for a generator channel, and say why the external-power channel does not check phase.

On ground; PR true; frequency within 0.5 Hz of the FRU; three-phase average at POR 115 ±5 VAC; phase within ±15° to capture / ±30° to hold of the FRU. The external-power channel instead checks an absolute 400 ±10 Hz plus voltage, and does not check phase — because external power is the reference. A GPU has no speed loop to tune itself onto a beat, so the system makes it the metronome and synchronises the adjustable generator onto it.

[!note]- Q4. Explain the "up to 15 s for the APU GEN with GPU" line and what a time-out does.

The APU generator has no constant-speed drive; to change its frequency the GAPCU has to ramp the whole APU's RPM (through the engine control) to chase the GPU's frequency, and a gas turbine's mechanical inertia makes that a seconds-scale action — hence the 15 s allowance. Every other pair is matched by an IDG's hydromechanical loop in milliseconds. On time-out the transfer is forced as a break transfer — the hand-over still happens, only the "seamless" is given up — and the function has a backup in the GCU.

[!note]- Q5. What is the logic of the inhibition latch, and how does a pilot recognise and handle it?

If the GAPCU judges its inputs could cause a paralleling of non-synchronous alternators, it suppresses the frequency reference to the GCUs and blocks all NBPT — and the inhibition is definitive until the GAPCU is completely de-energised (no AC or DC source, batteries included, connected to the network). The philosophy is "break before chaos": a non-synchronous parallel's inrush is far worse than one flicker. The symptom is a flicker on every source change; the handling is to report it and have maintenance read FC 889–893, not to keep retrying.

Key takeaways

References

Per FCOM DSC-24-10-20-30 (NBPT definition, the "15 s vs milliseconds" line, forced transfer on time-out, GCU backup) and DSC-24-10-30-20 (the NBPT sequence table and the crew-discipline statement); AMM 24-29-00 D/O §6.M (the full mechanism — parallel-first sequence, the NBPT occurrences and per-bus source-priority strings, the request triggers, the synchronisation-window conditions, and the FRU L/R description) and §6.N (the IPT 80 / 130 ms staged trip); AMM 24-23-00 §7.A (the 15-second forced break transfer, maintenance side); AMM 24-41-00 §6.A(1)(c) (the GAPCU non-synchronous-paralleling inhibition latch and the FC 889–893 specific-data fault codes). The explanation of why the APU ↔ GPU pair is the slow one is an integrative synthesis from the APU's constant-speed-drive architecture and the GAPCU ramp-drive channel (the manuals give the 15 s figure, not its cause); the relay-race analogy is a teaching device. The actual duration of the paralleling instant (the specific value within the 80 ms line) is not given in the source and is not extrapolated.

Independent study material, not an Airbus publication. Refer to current operator FCOM, FCTM, and QRH for operational use.