What a Fire Alarm Panel Is Actually Doing

Why this matters

A fire alarm panel spends essentially its entire service life not detecting fire. What it does instead, continuously, is watch itself: it pushes a small current down every circuit and measures what comes back, it watches both power sources, and it reports anything off-normal. Detection is the rare event; supervision is the job.

That inverts how the panel should be read. Its ordinary, meaningful output is trouble and supervisory information, and a panel showing a standing trouble is not a panel with a nuisance light on it. It is a partially unavailable system, and the part that is unavailable is the part that would have told you about everything else. It is also why the front of the panel is the wrong place to look: the lamp shows the present moment, and the event history shows how long the present moment has been going on. That duration is usually the finding.

Three signal classes, three different responses

NFPA 72 defines the classes, and it binds only in the edition the authority having jurisdiction has adopted and amended. The AHJ is a named role with legal authority over that adoption and its amendments, not a synonym for whoever performs the annual test.

Collapsing supervisory and trouble is the most common misreading, and the two point at different systems. A supervisory signal says the sprinkler system is compromised and the alarm system is working, since it just told you. A trouble says the alarm system is compromised and you know nothing about anything it supervises.

What supervision actually is

The panel maintains a small monitored current on each pathway. A device or terminating element sets the expected value, and the panel compares against it continuously. An open conductor changes the value one way and reports trouble. A short changes it the other way and, depending on circuit type, is either an alarm or a trouble. A ground fault, an unintended connection between a circuit conductor and earth, reports trouble because it degrades the panel's ability to distinguish the other conditions.

The consequence people miss: supervision is what makes every other supervised device meaningful. A valve tamper switch is only fast because the panel reports it in minutes. The control valve card gives that minutes-scale detection latency for an electrically supervised valve, and that figure assumes the panel and its transmission path are working. A standing trouble on the circuit carrying that tamper switch is precisely the condition that voids the assumption, without changing anything about the valve.

Pathway arrangement matters too. Some circuit classes are designed so a single open still leaves every device reporting and others are not, and which a building has is in the drawings rather than visible at the panel.

Power: primary and secondary, and the calculation nobody redoes

Two supplies. Primary is a dedicated, marked branch circuit. Secondary is a battery set sized to carry the system through a loss of primary power.

NFPA 72, in the edition the authority having jurisdiction has adopted and amended, sizes that set for 24 hours of standby plus 5 minutes in alarm for a protected premises fire alarm system, and for 24 hours plus 15 minutes for an in-building emergency voice and alarm communications system. The adopted edition owns which applies, and the two answers differ by enough to change a battery size. The arithmetic is simple, and it goes wrong because nobody redoes it after devices are added:

Required capacity = (standby current x 24 hours) + (alarm current x alarm period in hours), then multiplied by the safety margin NFPA 72 requires in the edition your authority having jurisdiction adopted and amended, commonly stated as 20 percent, so a factor of 1.2.

Every device added raises the standby current, and every notification appliance raises the alarm current. A building that has had three tenant fit-outs since the panel went in is running a calculation nobody has recomputed, on batteries nobody has resized. The annual load test is the only routine event that catches it, and only if the result is compared against a requirement rather than against last year.

Silence, acknowledge and reset are three different buttons

• Acknowledge tells the panel a human has seen it. The condition is unchanged.
• Silence stops the local sounder or the notification appliances. The condition is unchanged.
• Reset clears the latched state and re-interrogates the system. If the condition is still present it comes straight back.

Silence is the one that causes harm, because it removes the only continuous prompt the building had. A silenced trouble on a panel in a back-of-house room is indistinguishable, to a person who was not there when it started, from a panel that has always looked like that.

The case: a trouble that stood for 412 days

The signal. An annual alarm inspection at a distribution building. The panel's trouble indicator is lit and the sounder silent. The facilities lead says it has been like that a while and treats it as the panel's normal appearance.

The history, read first. The event log shows a ground fault first recorded 412 days earlier, silenced 6 minutes after it appeared, and re-reported at each subsequent power cycle. Nothing in the log shows it clearing. That number is the finding, not the ground fault itself.

Eliminating candidates on evidence. The technician isolates the fault by removing circuits at the panel one at a time, with the supervising station notified beforehand and the removals logged as an impairment of the alarm system, because lifting a circuit is exactly the act of removing protection.

• The fault persists with all field circuits removed: this would place it in the panel itself. It did not persist. Panel eliminated.
• The fault persists with only the notification appliance circuits connected: it did not. Notification circuits eliminated.
• The fault persists with only the initiating device circuit serving the loading dock connected: it did. Located.

What was true. Water in the backbox of a manual station on the dock wall. The maintenance record shows a roof drain clearing work order 415 days earlier, three days before the ground fault appeared. The fault was a consequence of an overflow already fixed, and the record of the fix is what dated it.

The second finding, from the battery calculation. With the panel open, the technician records the measured standby current at 0.85 A and the alarm current with notification appliances active at 4.2 A. Requirement: 0.85 A x 24 h = 20.4 Ah for standby, plus 4.2 A x (5/60) h = 0.35 Ah for the alarm period, giving 20.75 Ah, multiplied by the 1.2 aging factor stated above = 24.9 Ah. Installed: an 18 Ah set.

Working the installed set back the other way, at the same 1.2 factor: 18 Ah divided by 1.2 gives 15.0 Ah of design capacity, less the 0.35 Ah alarm reserve leaves 14.65 Ah for standby, which at 0.85 A is 17.2 hours rather than 24. That is a 28 percent shortfall in standby time. Direction check: a smaller battery gives less standby time, and the alarm reserve comes off the top because the alarm period has to be available at the end of the standby period, not at the start of it. Nobody had recomputed the load since two tenant fit-outs added devices.

Correction, printed. The 1.2 factor is a derating applied to the requirement, so it is a re-basing of the required capacity rather than an addition to it. It is applied once, at the end, to the sum of the two current-hour terms. Applying it to the standby term alone and then adding an underated alarm term would understate the requirement, and applying it twice by also comparing against a pre-derated battery rating would overstate it.

What the history told us that the front panel did not

The lamp said "there is a ground fault". The history said "this system has been partially degraded for 412 consecutive days, was silenced within 6 minutes of the first report, and nobody has entered anything against it since." Those are different findings and only the second one changes anybody's behaviour.

The 412 days also converts this from a repair into an impairment review. Under the impairment card's model, a standing trouble is a state: it had a start, it had no compensating measure, it had no owner, and it had no verified end. It also belongs to NFPA 72 in the adopted edition rather than to the water-based standard, which is the boundary the impairment card names.

Sibling-rule check, run against this case before it ships:

• The impairment card's five properties: applied to the 412-day trouble, which carries a start and fails the other four, and to the circuit removals during troubleshooting, which were opened as a state with the supervising station notified.
• The impairment card's rule that alarm impairment is NFPA 72 territory and not NFPA 25: applied above.
• The control valve card's minutes-scale detection latency for an electrically supervised valve: not contradicted, but conditioned. That latency assumes a working panel and transmission path, and this building's ground fault is the condition that voids it on the affected circuit.
• Count check, last mention first: three signal classes named and three used; three buttons named and three described; three candidates eliminated in the isolation sequence and three listed.
• No rounding runs in the flattering direction: 17.2 hours against a 24 hour requirement is reported as a 28 percent shortfall rather than as roughly three quarters of a day.

Hazard control for the work this case requires. Line voltage is present inside the enclosure at the primary supply terminals: de-energize the branch circuit and lock it out under 29 CFR 1910.333(b)(2), then prove dead with the live-dead-live sequence of NFPA 70E-2021, 120.5, before working in that section. 29 CFR 1910.147 is not the governing standard here, because it expressly excludes exposure to electrical hazards from work on electric utilization equipment. Batteries are a separate hazard from the panel: a sealed lead-acid set can deliver enough short-circuit current to vaporise a tool, so remove rings and watches, use insulated tools, and do not lay anything metal across the terminals. Do not disconnect the battery set while the system is running on secondary power, because that drops the system completely. Before lifting any circuit for troubleshooting, notify the supervising station and open an impairment, because removing a circuit removes exactly the protection the panel exists to provide, and a technician who lifts three circuits to chase a fault has taken the building down three times without saying so.

How to verify you got this right

Read the event history before you read the panel face, every visit. The first question is not what is wrong, it is since when.

Recompute the battery requirement whenever devices are added, and write the standby and alarm currents on the calculation so the next person can check it rather than repeat it. A load test that passes against no stated requirement has proven nothing.

Confirm supervisory and trouble are actually distinguished in the building's response plan. If the plan says "call the alarm company" for both, the valve tamper switch the control valve card counts on is not producing a faster response than an inspection would.

Close every trouble in a register, not in a memory. A trouble that clears by itself will come back, and the log is where its interval becomes visible.

References

• NFPA 72, National Fire Alarm and Signaling Code, in the edition the authority having jurisdiction has adopted and amended, for signal classes, pathway supervision, secondary power capacity and system impairment.
• 29 CFR 1910.333(b)(2) for work on energized electric utilization equipment, with NFPA 70E-2021, 120.5, for the proving sequence; 29 CFR 1910.147 does not apply to this exposure.
• Manufacturer documentation for the specific listed control unit, for the battery calculation form, the derating factor and the circuit isolation procedure.
• See related: the impairment card, the control valve card and the detector selection card in this category.