How to Work Out Why a Panel Keeps Going Into Trouble

Why this matters

A fire alarm panel that reports trouble every few days gets silenced every few days, and after about the third month nobody looks at the display at all. That is the actual hazard. The trouble itself usually means one circuit or one power source is compromised; the habituation it produces means the next signal, whatever class it is, lands on a building that has learned to ignore the panel. The method below converts an intermittent event into a number you can measure against a stated limit, which is the whole trick, because an intermittent fault you can only observe is a fault you can only guess at.

Step 1: Establish what is unprotected right now, and open the impairment

Find out what the trouble has taken away before you diagnose anything. A ground fault on an initiating device circuit means the panel may not report a fire on that circuit. That is a fire alarm system impairment under NFPA 72, in the edition your authority having jurisdiction has adopted and amended, not an NFPA 25 impairment, because NFPA 25 owns water based systems. The AHJ is a named role with authority to approve, commonly the fire marshal or the building official, and not a synonym for whoever inspects.

Open the impairment with a start time, a compensating measure and a named end, and understand that your own troubleshooting will widen it, because every branch you lift takes more devices out of service. The sibling card on supervisory signals owns the full procedure.

Skip this and you spend four hours making the system worse than you found it with no record of who was covering the gap.

Step 2: Print the history, do not read it on the display

Every listed fire alarm control unit keeps an event history with timestamps. Pull it to a file or print it. Reading it on a two line display is how people conclude "it does it all the time" when the log says eleven specific days. Extract four columns per event: timestamp, event class, point or circuit address, and whether it restored by itself. Self-restoring events are worth the most, because a fault that comes and goes is being modulated by something external, and external things have schedules.

Skip this and every later step is a guess dressed as a measurement.

Step 3: Sort into the three families before touching anything

1. Power supply troubles. AC loss, low battery, charger fault. About whether the panel can keep running.
2. Circuit integrity troubles. Open circuit, short, ground fault on an initiating device circuit, a signaling line circuit or a notification appliance circuit. About whether the panel can hear a device or drive an appliance.
3. Device troubles. A device not responding, a dirty chamber report, a device removed from its base, an address conflict. About one device.

Count each family before you form a theory. A panel with 30 events of which 25 are one address in family three is a device job; the same 30 spread across a circuit is a wiring job. These look identical on the display and identical to the customer.

Skip this and somebody replaces the device at the address on the display, the trouble returns, they replace it again, and two devices have been consumed on a wiring fault.

Step 4: Correlate the timestamps against something in the world

Lay the largest family's timestamps against three schedules: the building's occupancy and mechanical schedule, the weather record for that location, and any maintenance or construction activity in the log book. Water ingress correlates with rainfall, usually with a lag of hours rather than minutes. Thermal faults correlate with the mechanical schedule and outdoor temperature extremes. Induced noise correlates with a large motor starting. No correlation at all is a real answer too, pointing toward a marginal connection that moves with vibration.

Skip this and you disconnect circuits until the fault stops, which finds where it is without telling you what causes it, so it returns on the next branch.

Step 5: Localize by halving, with a meter, at the right test voltage

Isolate the suspect circuit at the panel with the power off and the batteries disconnected. Measure conductor to building ground resistance with a digital multimeter on its highest resistance range, each conductor separately, and record the number in ohms rather than writing "grounded." Then halve: lift the branches at their nearest junction one at a time and re-measure. The branch whose removal restores the reading to the meter's open circuit indication holds the fault. Repeat inside it.

Do not apply an insulation resistance tester to a circuit that has addressable devices or a panel connected to it. A test at 250 V or 500 V into a signaling line circuit destroys device electronics and the panel's line card, converting a wiring fault into a wiring fault plus 80 dead devices. If you need an insulation test, every device comes off first and the panel end is disconnected.

Two hazards attach to opening the panel, each with its own action. The primary supply inside the enclosure is line voltage, so open, lock and tag the dedicated branch circuit under 29 CFR 1910.333(b)(2) and verify absence of voltage at the terminals with a meter proved live, dead, live on a known source per NFPA 70E-2021, 120.5. The sealed battery set stays energized whatever the AC is doing, and a 12 amp hour cell will push hundreds of amps into a shorted tool, so lift the negative lead first with an insulated wrench, work one terminal at a time, and remove rings and a watch before your hands go in.

Skip the halving and you are back to swapping devices. Skip the test voltage rule and you replace the whole circuit.

Step 6: Turn the power family into a battery calculation

Family one is arithmetic. You need measured standby current, system alarm current, and the secondary supply period the adopted code requires. Measure standby current at the battery leads with the AC off and the system normal, and compute alarm current from the device schedule. Under NFPA 72 in the adopted edition, a protected premises fire alarm system commonly requires 24 hours of standby followed by 5 minutes of alarm, and an in-building emergency voice and alarm communications system commonly requires 24 hours followed by 15 minutes. The adopted edition owns which applies, and the two answers differ by enough to change a battery size.

Then check the date code. NFPA 72 in the adopted edition requires sealed lead acid batteries be replaced within 5 years of the date of manufacture, and a battery past that date closes a low battery investigation on its own.

Skip this and you fit a bigger battery because the old one seemed weak, which fixes nothing when the real load grew after somebody added an appliance circuit.

Step 7: Fix the cause, then prove it with the same measurement

The proof of a wiring fix is the resistance reading that found it, taken again and recorded. The proof of an environmental fix is time: a fault that correlated with rain has to survive rain. Close the impairment on the number, not on the word "repaired."

Worked example: 31 events in 90 days

A four story building, one panel, two signaling line circuits. The complaint is "it troubles constantly."

History, sorted into the three families. Ninety days of log, 31 trouble events. Family one, power: 6 AC loss and 3 low battery, 9 total. Family two, circuit integrity: 22 ground faults, every one on signaling line circuit 2. Family three, device: none, and that absence is itself a finding, because the two devices the previous contractor replaced were replaced on a wiring fault.

Family one, closed by arithmetic and a date code. All 3 low battery events occurred during one of the 6 AC loss events, so they are not independent, and the AC loss events matched the utility's posted outage notices. The battery date code was 6 years old against the adopted edition's 5 year replacement requirement, which closes them. The capacity calculation was run anyway, because appliances had been added since commissioning:

• Measured standby current 0.42 A over 24 hours: 0.42 x 24 = 10.08 amp hours.
• Alarm current from the device schedule 2.10 A over 5 minutes: 2.10 x (5 / 60) = 0.175 amp hours.
• Calculated total: 10.08 + 0.175 = 10.255 amp hours.
• The adopted edition's 20 percent margin: 10.255 x 1.20 = 12.31 amp hours required.
• Installed: two 12 amp hour cells in series for 24 V, so system capacity is 12 amp hours, not 24. Series adds voltage, not capacity.
• 12 against a required 12.31, short by 0.31 amp hours, 2.5 percent under. The size goes up at replacement, and the calculation goes in the file with the appliance schedule it was run against.

Correction, printed, and it is a re-basing rather than an addition. The 12 amp hour figure is published at a stated discharge rate, commonly a 20 hour rate, at a stated temperature, commonly 77 F. This system's standby draw of 0.42 A is a 12 / 0.42 = 28.6 hour rate, slower than the rating's basis, so delivered capacity there is at or slightly above the published figure. The alarm draw of 2.10 A is a 12 / 2.10 = 5.7 hour rate, faster than the basis, so delivered capacity is below it there, and that applies to only 0.175 of the 10.255 amp hours. The adopted edition's 20 percent margin exists to cover rate and temperature effects, so do not apply a second derate on top of it, which double counts the same allowance.

Family two, localized by halving. Circuit 2 carries 84 addressable devices across 6 branch runs. Against the site weather record, 22 of 22 ground fault events fell within 18 hours of measurable rainfall, and zero occurred during a 26 day dry stretch inside the window. The mechanism was named before a cover came off. With the panel de-energized and the batteries lifted per the actions in step 5, and the impairment open with a fire watch on the affected floors:

• Whole circuit 2, conductor to building ground: 41 kilohms. The panel's listed instructions state it reports a ground fault below 50 kilohms, so 41 is inside the reporting region and consistent with the log.
• Branches lifted one at a time. With run 4 lifted, the remainder read greater than 20 megohms.
• Run 4 alone read 41 kilohms, which closes against the whole circuit reading: 41 kilohms in parallel with more than 20 megohms is still 41 kilohms at the instrument's resolution.

Run 4 serves a below grade parking level and enters through a wall conduit with no seal at the exterior. Water was standing in the second junction box on the run. Conductors were replaced from the penetration to that box, the conduit was sealed at the exterior face, and the low point was rerouted above the penetration.

Proof, using the same measurement. After repair, run 4 alone read greater than 20 megohms and the whole circuit read greater than 20 megohms. Over the following 6 weeks the site recorded 4 measurable rain events and the panel logged zero ground faults. The rain events are what turned a repair into a confirmed repair.

Sibling rule check, printed. Against the supervisory signal card: every branch lift ran as a fire alarm impairment under NFPA 72 with a start, a fire watch and a stated end, and none was filed under NFPA 25, because no water based system was touched. Against the notification appliance card: the alarm current above includes the appliances added after commissioning, exactly the load that card warns gets omitted. Against the duct detector card: the building's duct detectors sit on circuit 1 and remain mapped as supervisory points rather than alarm points.

How to verify you got this right

The file should close with four things, and a job producing fewer than four did not finish: the sorted event count by family with the correlation you found, the resistance readings before and after with the branch identified, the battery calculation with the currents it used, and the impairment record with both timestamps. Closing paperwork that says "found ground fault, repaired" records that somebody was there.

References

• NFPA 72, National Fire Alarm and Signaling Code, in the edition your authority having jurisdiction has adopted and amended, which sets the secondary supply period, the capacity margin, the battery replacement interval and the impairment procedure
• Manufacturer listed installation instructions for the control unit, which own the ground fault reporting threshold and the permitted battery sizes
• 29 CFR 1910.164, fire detection systems, requiring prompt restoration to normal service after a test or alarm
• 29 CFR 1910.333(b)(2) and NFPA 70E-2021, 120.5, for the electrical exposure inside the control unit
• See related: What a Supervisory Signal Is and Why It Is Not an Alarm; The Detector That Alarmed Every Morning at the Same Time