What an Arc-Fault Device Responds To

Why this matters

An arc-fault device is the only thing in a building watching for a condition that produces no meaningful current, no heat at the breaker, and no measurement any ordinary tool will show you on a walk-through. It is also the device most often "repaired" by removing it, because its trips look random and its diagnostic story is unfamiliar. A tech who understands that it is a pattern recognizer rather than a current meter stops arguing with it and starts using it, because a trip from one of these is a report about the shape of the current, and the shape is the evidence.

What an arc is, in terms a meter can describe

An arc is a conducting plasma across a gap, re-established from the supply every half cycle. What makes it detectable is not how much current flows but how badly the current behaves.

• The waveform is not a clean sinusoid. After each zero crossing the arc has to re-ignite, so current sits near zero for a moment before it snaps back up. That dwell, sometimes called a shoulder, is the classic signature.
• Broadband high-frequency noise rides on top of the fundamental. The unstable plasma generates content far above line frequency.
• It is erratic cycle to cycle. A resistive load looks the same every cycle. An arc does not.
• The magnitude may be entirely unremarkable. This is the part that matters most in the field.

An arc-fault device samples the current and tests it against algorithms for those traits. It is not measuring amps against a threshold; it is deciding whether a pattern is present. Everything else about how it behaves follows from that.

Two kinds of arc, and only one of them could ever trip a breaker

A series arc is a break in the conducting path in line with the load: a loose terminal, a broken strand, a cracked splice, a nail through a cable that has partly severed a conductor. The load is still in the circuit, so the load still limits the current. A series arc on a circuit whose load draws a few amps produces a few amps. It cannot exceed what the load would draw normally, ever, under any circumstance.

A parallel arc is between conductors, or from a conductor to grounded metal. Here the arc's own impedance sets the current, and it can reach hundreds of amps. Even so it often fails to operate an overcurrent device, because it is intermittent: a magnetic element needs a multiple of rating, commonly around ten times for a standard molded-case breaker with the product curve as the authority, and a thermal element needs sustained current over minutes. An arc that strikes and self-extinguishes repeatedly delivers neither.

Both ends of that range point the same way. At the low end, a series arc is invisible to overcurrent protection by definition. At the high end, a parallel arc is invisible because of its duration rather than its size. The gap between them is precisely the population that starts fires, and it is what the device exists to close.

A combination-type device is listed to detect both series and parallel arcing under UL 1699, the listing standard it was tested to in the edition it was listed under, which binds through the listing itself and through the adopted NEC's requirement that the device be listed. The requirement to install one on specified dwelling-unit branch circuits sits in NEC Article 210 in the edition your authority having jurisdiction has adopted, and the list of circuits it covers has widened across editions, which is exactly why the edition clause is not decoration here.

The negative space: five things it will not save you from

Knowing what the device does not cover is what keeps a tech from over-trusting a panel full of them.

• A glowing high-resistance connection that is not arcing. A badly made joint can heat to ignition temperatures as a resistance rather than as an arc, drawing normal load current the whole time. No arc signature, no trip, real fire risk. Thermal inspection and torque, not electronics, is what finds those.
• An overload. A combination arc-fault breaker still contains conventional overcurrent elements, but the arc detection contributes nothing to that job. The two functions share a housing, not a mechanism.
• Personnel-level ground faults. Unless the device is specifically a dual-function type that also carries ground-fault circuit interrupter protection, milliamp-level leakage is a different device's question, covered by the sibling article separating ground faults from overloads.
• Anything outside its own branch circuit. It sees the current in its own conductors. A failing connection in a neighbouring circuit, in the panel feeders, or on the utility side is not in its field of view.
• Arcing it is designed to ignore. Switch contacts opening, relay contacts, thermostat contacts and brush motors all arc on purpose. The algorithm must reject those, and that deliberate rejection is where the hard field cases come from.

The look-alike population, and why "nuisance" is the wrong word for it

Loads that legitimately produce arc-like current signatures are a short and repeatable list: series-wound universal motors with commutator brushes, some phase-control dimming and speed-control loads, and certain switching supplies. Wiring conditions that produce them are shorter still, with a shared neutral serving two circuits in a way the device cannot resolve at the top of it.

When one of those trips a device, the device has usually recognised a real signature correctly. That is a different situation from a defective device, and it leads to a different repair: the load or the wiring arrangement is the subject, not the breaker. Calling it nuisance and moving on is how a genuine loose connection three rooms away survives a service visit.

Working on one of these without becoming the ignition source

Any work inside the panel or a device box is live work until proved otherwise. 29 CFR 1910.333(a)(1) requires de-energizing before working on or near exposed energized parts unless the employer can demonstrate that de-energizing introduces additional or increased hazards or is infeasible; isolation for panel and branch-circuit work runs under 29 CFR 1910.333(b)(2) in general industry or 29 CFR 1926.417 on construction, because 29 CFR 1910.147 excludes exposure to electrical hazards from work on conductors and equipment in electric utilization installations at (a)(1)(ii)(C); proving dead is live-dead-live per NFPA 70E-2021 120.5, and where live work is justified the boundary and arc-rated PPE come from the risk assessment at NFPA 70E-2021 130.5 and 130.7 in the edition your employer's program has adopted.

One hazard here is created by the diagnostic itself. Resetting an arc-fault device repeatedly to reproduce a trip re-energizes a circuit that may hold an active arcing fault, and an arcing fault is an ignition source next to building materials you cannot see. Inspect the accessible connections before the first reset, keep reset attempts to the minimum that answers the question, stay with the circuit while it is energized, and do not leave a circuit cycling unattended to "see if it happens again overnight." If a device will not hold at all, that is an answer, not a reason to keep trying.

Damaged devices are replaced, not restored. A receptacle or breaker with heat discolouration, melted insulation or a pitted terminal has taken damage that is not reversible by retightening.

Worked example: two populations of trip on one bedroom circuit

A 15 A dwelling branch circuit with a combination arc-fault breaker. Trips 2 to 6 times a week for a month. The homeowner has already been told twice that the breaker is faulty.

Sort the events before touching anything. Asking for times and circumstances rather than a story produces two clearly separate populations, and the separation is the finding:

• Most trips happen within seconds of a particular vacuum being switched on.
• Two trips happened overnight, with nothing on the circuit running.

Two populations means two causes, and the second one is the important one because it removes the load from the equation entirely.

Take the load measurement anyway. With everything on the circuit running, total draw is 5.4 A, which is 36 percent of the 15 A device rating. Fix that number in mind, because it bounds the whole series-arc question on this circuit: a series arc anywhere on it can produce at most that 5.4 A, since the load is still limiting the current. Against a 15 A thermal element, 5.4 A is not a signal in any sense. There is no version of this fault that a conventional breaker discovers.

Chase the overnight population first. With all loads unplugged the circuit still tripped once during a two-day observation, which implicates the wiring. De-energized, locked out and proved dead as above, every accessible device on the circuit comes apart. One receptacle has a conductor landed in a push-in back-wire terminal, and the terminal and the conductor insulation next to it are discoloured. The connection is remade by pigtailing to the screw terminals and torquing to the device manufacturer's marked value, and the discoloured receptacle is replaced rather than reused.

The vacuum population is separate and survives the repair at a much lower rate. That load is a brushed universal motor, which is in the look-alike list above, and its brush arcing is real arcing that the device is recognising correctly. The right next steps are testing the appliance itself, since a worn commutator or brush makes the signature markedly worse, and confirming the device is a currently listed model. What is not a right next step is defeating protection the adopted code requires on that circuit. Where a listed device and a listed appliance genuinely conflict, that is a question for the two manufacturers and, if the resolution involves changing anything about the installation, for the authority having jurisdiction. Whether any modification would void a listing is not something a technician settles on site.

The failure mode of getting this wrong: the two populations get averaged into "it trips sometimes," the vacuum explanation is accepted for all of it because it fits most of the events, the breaker is replaced or bypassed, and the discoloured back-wire terminal keeps arcing at 5.4 A behind a wall for another year. Nothing in that circuit will ever report it again.

What changes the approach: if every trip had correlated with one load and the circuit held for days with that load unplugged, the wiring teardown would not be the first move; the appliance would be. And if the device would not hold with the entire branch disconnected at the panel, the subject moves to the device or its own connections and the branch is exonerated in one test.

How to verify you got this right

• Use the test button for what it is. The device's own test button verifies its internal electronics and mechanism. It says nothing about the wiring, and a device that passes its self-test is not thereby cleared.
• Prove the load population and the wiring population separately. A circuit that holds for a full week with all loads disconnected has been shown clean in a way that no inspection can show it.
• Compare the trip current against the device rating and say the ratio out loud. If a series arc on this circuit can produce at most a third of the breaker's rating, then no story that depends on the breaker "finally noticing" the current is available to you.
• Recheck the repaired connection under load and warm. A remade joint that is right is at the same temperature as its neighbours after the circuit has run loaded for an hour, and a thermal comparison across identical terminals is more reliable than an absolute number.

References

• NEC Article 210, arc-fault circuit-interrupter protection provisions, in the edition your authority having jurisdiction has adopted
• UL 1699, the listing standard for arc-fault circuit interrupters, in the edition the installed device was listed under
• 29 CFR 1910.333(a)(1) and (b)(2); 29 CFR 1910.147(a)(1)(ii)(C); 29 CFR 1926.417 for construction work
• NFPA 70E-2021, 120.5, 130.5 and 130.7, as adopted by your employer's electrical safety program
• See related: What a Ground Fault Is, as Distinct From an Overload; Why Nuisance Tripping Is Usually a Real Signal