The Detector That Alarmed Every Morning at the Same Time

Before anything else

An activation is a fire until somebody has proved it is not, and that proof is not available during the activation. So the building's plan runs every time: occupants leave or relocate by the route the plan names, nobody investigates on the way out, and the panel is not reset until the responding authority releases it. Diagnosis happens afterward, from the record.

If somebody has already decided to stop the nuisance by disabling the point, that is a fire alarm system impairment under NFPA 72, in the edition your authority having jurisdiction has adopted and amended, with a start time, a compensating measure and a named end. The AHJ is a named role with authority to approve, commonly the fire marshal or the building official, not a synonym for whoever inspects. A detector disabled without that record is a corridor with no detection and no one accountable for it, which is worse than the alarms.

The complaint and what the site had already done

Second floor corridor of a four story building. One address, L2-SD-14, a photoelectric spot smoke detector. The property manager's description was "it goes off every morning, we have replaced it twice."

Everything in this case was reconstructed after the fact, from three records that already existed on site: the panel's event history, the mechanical service file for the air handler serving that corridor, and the site weather record. Nothing here needed a live event to catch. That matters, because a fault that only appears at 05:45 is a fault most shops never see with their own eyes.

What the panel history alone ruled out

Six weeks of history, 42 calendar days, 30 of them weekdays.

• 14 alarm events, all at address L2-SD-14. No other address ever appeared.
• All 14 fell between 05:38 and 06:05. None outside that window, ever, in six weeks.
• Zero events on any of the 12 weekend days.
• Zero trouble events of any class on that circuit. No dirty chamber report, no device fault, no ground fault.

That is already most of the diagnosis. Take the classes in the sibling card on recurring troubles: family one is power, family two is circuit integrity, family three is device faults. Nothing here is a trouble at all, and family three in particular never appeared, which means the panel never once thought there was anything wrong with the detector. It thought there was smoke. Fourteen times, in a 27 minute window, only on days the building was scheduled to be occupied.

A photoelectric spot detector works by shining light into a chamber at an angle where the sensor cannot see it, and alarming when something in the chamber scatters that light onto the sensor. It does not identify combustion. Anything that scatters light does it: smoke, dust, and water droplets. That is not a defect in the technology, it is what the technology is.

The gaps in the record are part of the finding

Three earlier interventions were in the file, and each one is evidence.

The head was replaced on day 19 and again on day 31 of the six week window. Alarms continued after both. Two independent devices from stock behaving identically is strong evidence that the devices are fine and the environment is not.

On day 38 the sensitivity was moved to the least sensitive setting inside the listed range. Photoelectric spot detectors are commonly listed across a sensitivity range expressed in percent obscuration per foot, and the specific range belongs to the device's listing. Alarms continued after that too, which says the aerosol reaching the chamber was well above even the loosest setting the device is permitted to run at.

That third intervention is the serious one, and it is worth naming plainly. Desensitizing a detector because it keeps reporting is treating a correctly operating device as the fault. The corridor it protects is an egress path. The change reduced the device's response to a real fire, it was made with no impairment record, and it did not even work. Establish why a device operated before you treat the device as the problem. If the answer had been "the chamber is contaminated," the panel would have said so as a device trouble, and it never did.

What the file did not contain was any note connecting the alarms to the mechanical work. That connection was sitting eight days upstream.

The mechanical service file, read against the timestamps

The air handler serving that corridor runs an occupied schedule starting at 05:30, weekdays only. On weekends it stays in unoccupied mode. That alone explains the weekday pattern and the absence of any event outside a window that opens 8 minutes after the unit starts.

It does not explain why only 14 of 30 weekdays produced an alarm. So the 14 dates went against the site weather record, and all 14 fell on mornings following an overnight outdoor dewpoint above the range where the building's unoccupied space stays dry. The 16 quiet weekdays were drier mornings. Two correlations, one schedule and one moisture, and between them the whole pattern is accounted for.

And eight days before the first logged alarm, the file carries a fan sheave change. It was made in response to a complaint that the second floor corridor felt stuffy.

The mechanism, and the number that confirmed it

A cooling coil that is fully wetted will throw water droplets off its downstream face when the air velocity through it gets high enough to strip condensate off the fins faster than it can drain down them. That is carryover. The threshold is a coil face velocity, and for a plain fin coil without a moisture eliminator it is commonly cited in the neighbourhood of 500 to 550 feet per minute, with the coil manufacturer's own data owning the number for a specific fin design and spacing.

At the unit:

• Coil face area, measured: 20 square feet.
• Airflow by pitot traverse of the supply duct: 12,400 cfm.
• Face velocity: 12,400 / 20 = 620 feet per minute.

Correction, printed. The unit's nameplate airflow is 11,000 cfm, and 11,000 / 20 gives 550 fpm, which sits right at the commonly cited threshold and would have made this look marginal rather than clear. The nameplate figure is stated at rated external static with a clean filter and the original drive, so using it here is not a conservative choice, it is the wrong basis. The traverse re-bases the whole calculation onto the machine as it exists, and 620 fpm is the number every conclusion below rests on.

Fan law one says airflow varies directly with fan speed at constant system resistance and constant air density, so the sheave change that raised the fan speed raised the face velocity in the same proportion. Going from 550 to 620 fpm is a 12.7 percent increase, which is what a modest sheave change buys, and it moved the coil from just under the carryover threshold to above it.

The rest of the physical evidence lined up with that:

• The condensate trap was dry, so the pan was not draining against the fan's negative pressure and was holding water above the level of the coil's lower return bends. A coil sitting in standing water stays fully wetted from the first minute of operation.
• The detector was measured at 22 inches from the edge of the nearest supply diffuser. NFPA 72 in the adopted edition requires spot detectors not be located within 36 inches of a supply air diffuser. That is a second, independent finding: the detector sat inside the diffuser's jet, so it received the supply air stream undiluted rather than the corridor's mixed air.

The 8 to 35 minute delay is the last piece. Carryover does not start at fan start. It starts once the coil face is fully wetted and the pan is up to level, which on this unit took several minutes from a dry overnight condition, and it stopped once the space dried down enough that the coil cycled off.

Two hazards in that investigation, each with its own action. Opening the air handler exposes a rotating wheel and belt driven stored energy, so the unit was isolated at its disconnect and locked and tagged under 29 CFR 1910.147 before any panel came off, and the portion of the work inside the electrical enclosure forked to 29 CFR 1910.333(b)(2) with absence of voltage verified on a meter proved live, dead, live on a known source per NFPA 70E-2021, 120.5. The pan held standing water with visible biofilm: it was drained and wet cleaned with the unit off, and specifically not blown out with compressed air, which would have aerosolized it straight into an occupied supply duct.

The repair, and why it needed both halves

The coil side. Trap rebuilt and primed to the depth the unit's negative static requires, pan cleaned and its slope corrected to the drain, and the sheave returned to a ratio giving a measured 11,100 cfm, which is 555 fpm face velocity. That is still above the top of the 500 to 550 fpm band cited above rather than under it, so the coil remains in the carryover region on paper, and a moisture eliminator was quoted as the durable answer rather than as an upgrade, and the corridor airflow complaint that started all of this was reopened as its own job, because raising fan speed to fix a distribution problem is what created this one.

The detector side. L2-SD-14 relocated to 54 inches from the nearest diffuser edge, comfortably clear of the 36 inch requirement and out of the jet, and the sensitivity restored to the mid setting inside the listed range with the setting recorded.

Fixing only one half fails in a predictable way. Coil work alone leaves a detector inside a diffuser jet, which will alarm on the next aerosol anybody introduces into that duct, including a floor stripping crew two floors down. Relocating the detector alone leaves a coil throwing water into the supply duct, which shows up later as a stained ceiling grid, a corroded diffuser and wet insulation in the plenum, with no alarm to tell anybody it is happening.

Proof

Nine weeks after the repair, across 45 weekdays including 11 mornings with overnight dewpoints in the same range as the original 14 alarm days, the panel logged zero alarm events at that address and zero device troubles. The nine week window was chosen deliberately: it had to contain humid mornings, because a quiet dry spell proves nothing about a moisture driven fault.

Sibling rule check, printed. Against the duct detector card: the same air handler carries a duct smoke detector, it never activated during any of the 14 events, and it would not have sounded a notification appliance if it had, because it is mapped as a supervisory point under the adopted code routing. Its silence is consistent, not contradictory: it samples the return, and the carryover was on the supply side downstream of it. Against the supervisory signal card: the day 38 sensitivity change was an undocumented reduction in a required device's response with no impairment record, and it has been written up as such. Against the recurring trouble card: nothing in this case belonged to any of its three trouble families, which is precisely why replacing hardware twice produced nothing.

References

• NFPA 72, National Fire Alarm and Signaling Code, in the edition your authority having jurisdiction has adopted and amended, which sets spot detector clearance from supply air diffusers, sensitivity testing and the impairment procedure for alarm systems
• Manufacturer listed data for the detector, which owns its sensitivity range, and for the cooling coil, which owns its carryover face velocity
• 29 CFR 1910.147 for mechanical isolation and stored energy at the air handler, and 29 CFR 1910.333(b)(2) with NFPA 70E-2021, 120.5, for the electrical portion
• See related: How to Work Out Why a Panel Keeps Going Into Trouble; What a Duct Detector Is For and What It Does Not Cover