Why a Smoke Detector and a Heat Detector Fail Differently

Why this matters

Two detectors in the same building, both listed, both installed correctly, both quietly out of service. One told you about it for eleven months before it went silent. The other has never said anything and never will, and its silence today looks exactly like its silence when it was working.

That asymmetry is the point of this card. Smoke detection and heat detection do not just differ in sensitivity, they fail in opposite directions, and the direction determines which instrument finds the failure. Auditing a heat detector by reading the alarm history is reading a blank page and concluding the page is fine.

Two failure directions

Smoke detection fails toward the alarm, and then toward removal. It is sensitive to a signature that ordinary activity also produces, so its characteristic malfunction is announcing itself. People respond to a device that alarms without a fire, and the responses available to a non-specialist are all subtractive: cover it, bag it, unplug the base, ask for lower sensitivity, or have the address disabled at the panel. The device becomes unavailable through human action taken in response to information the device itself provided.

Heat detection fails toward late, and then toward never. It is deliberately insensitive, responding only when enough energy has arrived at the ceiling. A heat detector painted over, buried under new insulation, fitted with the wrong temperature rating, or removed with its base left capped produces exactly the same output as one working perfectly: nothing at all, for years.

Everything follows from this. The smoke detector's problem is that its information gets acted on badly. The heat detector's problem is that it produces no information at all until the day it matters.

Why smoke detection fails toward disablement

The nuisance sources are covered in the detector selection card and are not re-derived here. What matters for failure behaviour is the sequence, which is consistent:

1. A device alarms with no fire, more than once.
2. The occupants stop believing the system, a broader failure than the one device.
3. Somebody makes it stop, rarely by a decision anyone records.
4. The panel goes quiet, and quiet reads as fixed.

Step 4 is the trap. A quiet address after a noisy one is treated as an improvement and is at least as likely to be a covering, a disconnected base or a disabled point. Coatings deserve a mention because they arrive with a repaint hitting several devices at once: a coated smoke detector is a replacement item on the same reasoning the coated sprinkler card gives, since paint blocks the openings the signature travels through and cannot be verified as removed from the chamber.

Why heat detection fails toward silence

A heat detector has no self-generated event stream. It has two states, and one of them has never happened in most buildings.

The failure modes are all physical and all invisible from the panel: overspray or coating on the element, storage or insulation stacked against it, an ambient that has crept up under it, a rating that was wrong the day it was installed, and damage that leaves the device looking intact. None produce a trouble signal, because the panel supervises the circuit, not the element. The panel card owns that distinction: supervision proves the pathway, not the sensor's ability to sense.

That is why the adopted editions treat the two families with different maintenance logic. Smoke detector sensitivity is verified on a cycle that can be extended once readings prove stable: in the commonly adopted editions of NFPA 72, within a year of installation and every alternate year after, with an extension up to a maximum interval once two consecutive tests come back within the listed range. Heat detectors are addressed by functional testing plus laboratory testing of a sample after a long service interval, commonly beginning around fifteen years after installation and repeating thereafter. Both belong to NFPA 72 in the edition the authority having jurisdiction has adopted and amended, and the AHJ is a named role with legal authority over that adoption rather than a synonym for the inspector. Go to the adopted edition for the intervals rather than to memory, because these are numbers editions move.

Thermal lag, and why the rating is not the air temperature at operation

A fixed-temperature heat detector operates when its sensing element reaches its rated temperature. Heat only flows into that element while the surrounding air is hotter than the element, so at the moment of operation the air is always above the rating, by an amount that grows with how fast the fire is growing. That is thermal lag, a property of the device's mass and surface rather than a defect. Rate-compensated devices exist to shrink it.

Two consequences follow, running in opposite directions, which is why rating selection is a real decision rather than a default:

• Rating too close to the maximum expected ambient and the device nuisances on a hot day or a process cycle. The adopted editions require a margin above the maximum expected ambient at the device's location, and state it there.
• Rating too far above ambient and the fire has to get larger before the element gets there, pushing an already-late device later.

Anything added to the outside of the element behaves the way a coating on a sprinkler does: added mass and added insulation, both increasing lag. The coated sprinkler card owns that mechanism and the direction is identical here.

Rate of rise elements carry their own blind spot: a fire growing slowly enough never crosses the rate threshold, and the element self-restores after a transient, so a rate of rise device that responded to a non-fire event leaves no trace. This is why they are almost always combined with a fixed-temperature element in one housing.

The audit instrument follows the failure direction

One question, two spaces

One building, one question asked in both spaces: what would tell me this device is unavailable?

Space A: office corridor, photoelectric spot smoke detection. The panel history for address 1-14 shows 4 alarms in 11 months, a rate of 0.36 per month against that same 11 month base, followed by 9 months with none.

Test the silence against the rate rather than against an impression. At 0.36 alarms per month, the expected count over 9 months is 0.36 x 9 = 3.24. Observed: zero. Under a constant-rate assumption, zero events where 3.3 were expected has a probability of about 4 percent, which makes the silence worth investigating and does not make it proof; a genuine fix to a nuisance source produces the same history. The point is that the question got asked, and it got asked because the device had a history to compare against.

Investigation found a plastic bag taped over the device during a ceiling repair, still there, applied inside the quiet window. The nuisance record is what made the silence legible.

Space B: mechanical room, fixed-temperature spot heat detection, 6 devices. The panel history for all 6 addresses shows nothing, ever, which is the correct history for working heat detectors and for failed ones. So the history is set aside and the audit is physical.

Findings: 2 of the 6 carry overspray from the same repaint that produced the bag in Space A, which under the coating reasoning above makes them replacement items rather than cleaning candidates. A third is stamped at a 135 F rating above a boiler where the measured maximum ambient at the ceiling was 122 F, a margin of 13 F. Against a required margin of the order of 20 F above maximum expected ambient in the adopted editions, 13 F is short, and that device is a nuisance risk on the hottest process day rather than a late-response risk.

What the two spaces prove together. Same repaint, same contractor, same week. In Space A the failure left a readable signature in a record. In Space B the identical failure left nothing, and would have for the rest of the building's life. Three of six heat detectors in that room are defective, 50 percent of the devices in the space, and none would have been found by any amount of history reading.

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

• The detector selection card's direction: smoke detection responds earlier than heat detection to a smouldering fire and heat detection requires a larger fire. Used unchanged, and it is why the bagged smoke detector in Space A is the more urgent restoration despite being one device against three.
• The coated sprinkler card's mechanism, that an added coating increases thermal mass and insulation and therefore delays operation: used unchanged for the heat detector elements and for the coated smoke detectors, in the same direction.
• The panel card's rule that supervision proves the pathway rather than the sensor: applied, and the reason none of the three defective heat detectors produced a trouble signal.
• The impairment card's five properties: applied to the bagged detector, an unrecorded impairment with a start nobody logged, no compensating measure, no owner and no verified end.
• Count check, last mention first: 2 coated plus 1 wrongly rated is 3 of 6, stated as 50 percent, on a base of the 6 devices in that room rather than the building. The 0.36 per month rate uses an 11 month base and is applied over 9 months in the same unit.
• No rounding runs in the flattering direction: a 13 F margin against an order-of-20 F requirement is short, not close.

Hazard control for the work this audit requires. The mechanical room measured 122 F at the ceiling: limit time at that level, work in pairs, and follow a heat illness plan with water and rest breaks, because a technician on a ladder at head height there is in the hottest air in the building. Do overhead work from a properly set ladder or platform, with the selection routed to the ladder and lift cards. Before testing any device, notify the supervising station and open an impairment, since a zone placed in test is protection removed. Test heat detectors only with the manufacturer's listed heat applicator: a heat gun or open flame can drive the element past its design condition, damage it permanently, and ignite dust or insulation above the ceiling. Do not put a coated device back in service after removing a bag or cover; a coated detector is replaced, on the same reasoning as a coated sprinkler.

How to verify you got this right

For smoke detection, verify by history first and physically second. Pull the per-address event history for a full year and look for two shapes: an address with repeat alarms, which is a selection or placement problem, and an address that went from noisy to silent, the shape this card exists to teach.

For heat detection, verify physically and never by history. Count the devices against the drawings, look at each element from close range, and measure the ambient at the device on the hottest condition the space sees, not the day you visit.

For both, check the coating question building-wide after any repaint, in the same visit as the sprinkler check: one painter produces findings in both systems and one walk finds them.

Record what you inspected, not only what you found. On a device family whose working and failed states produce identical evidence, a dated physical inspection record is the only thing distinguishing verified good from nobody has looked in years.

References

• NFPA 72, National Fire Alarm and Signaling Code, in the edition the authority having jurisdiction has adopted and amended, for sensitivity testing intervals, heat detector testing, temperature rating margins above maximum expected ambient, and replacement criteria.
• Manufacturer listing and maintenance documentation for the detector, including the listed test method and applicator.
• See related: the detector selection card, the fire alarm panel card, the coated sprinkler card and the impairment card in this category.