What a Duct Detector Is For and What It Does Not Cover

Why this matters

A duct smoke detector is one of the few fire protection devices an HVAC tech touches routinely, and it is the one most often described wrongly on both sides of the trade. Mechanical contractors treat it as smoke detection for the space it serves. Alarm contractors treat it as somebody else's device because it hangs off a duct. It is neither. It exists to keep an air handler from taking smoke from one part of a building and delivering it to the rest, and it can only do that while air moves past it at a rate its listing covers. Get that wrong and you can pass every functional test on the schedule while the device would not have shut anything down on the day it mattered.

What the device is actually doing

A duct smoke detector is a spot sensing chamber mounted outside the duct, fed by two tubes through the duct wall. The inlet tube runs across the duct with a row of holes facing upstream; the exhaust tube is shorter and opens downstream. Moving air creates a pressure difference between the two, air flows through the housing, and a fraction of the airstream crosses the sensing chamber. The chamber is an ordinary photoelectric or ionization sensor. Nothing about it is special. The sampling arrangement is the whole device.

That single fact drives everything else. The detector does not sense the duct, it senses a small sample the duct hands it, and the sample only exists because there is a pressure differential across the two tubes. No differential, no sample, no detection. A duct detector on a stopped fan is not a slow detector, it is not a detector at all.

Airflow is the coverage, not floor area

An area smoke detector has a coverage figure in square feet. A duct detector does not. Its coverage is a velocity band printed on the device label and in the listed installation instructions. Duct detectors are commonly listed somewhere between roughly 300 and 4,000 feet per minute, and the number that governs is the one on the device you are holding.

Velocity in feet per minute equals airflow in cubic feet per minute divided by the duct's free area in square feet. Below the low end the pressure differential is too small to pull a representative sample and response time stretches out unpredictably; above the high end the sample crosses the chamber too fast for the sensor. Both ends fail quietly, and nothing on the panel tells you the device is outside its band.

This is where variable speed retrofits bite. A unit that was inside the band at commissioning gets a drive ten years later and runs at 30 percent of its old airflow through the shoulder seasons. Fan law one says airflow varies directly with fan speed, and that holds at constant system resistance and constant air density, so a 30 percent speed command is a 30 percent airflow as long as nobody also changed the ductwork or the filter bank.

Where it has to sit to get a real sample

The sample has to represent the whole duct cross section, which means the air has to be mixed. At a fan discharge, elbow, transition or damper it is not. Listed instructions state the straight run as a multiple of duct width, commonly on the order of six to ten duct widths downstream of any disturbance, and your device's instructions own the multiple.

        upstream disturbance detector housing
        (fan, elbow, damper)          outside the duct
             |                              |
   =========\|/=============================[]=========
   air  >>>  turbulent, unmixed mixed     ||
   flow >>>  do not sample here sample    ||
   =========================================[]=========
                                    inlet tube spans
                                    the duct, holes
                                    facing upstream
        |<----- six to ten duct widths ----->|

Two placement facts get missed. The inlet tube has to span most of the duct width, because a tube reaching a third of the way across samples a third of the duct and misses stratified smoke on the far side. And the holes face upstream while the exhaust faces downstream, so a tube installed backwards produces a detector far slower than its listing says. A backwards tube passes a magnet test and an aerosol test at the housing. It fails the only test that matters.

What signal it produces, and what it does not cover

Where a building has a fire alarm system, the model codes generally route duct detector activation to a supervisory signal at a constantly attended location rather than to the occupant notification appliances. The device says an air handler is moving smoke, which is a reason to shut that unit down and investigate, not by itself a reason to evacuate. The requirement lives in the fire code and building code your jurisdiction adopted and amended, and the adopted edition owns it. Where a routing question has to be resolved, it belongs to the authority having jurisdiction, a named role with authority to approve, commonly the fire marshal or building official, and not a synonym for whoever inspects. The sibling card on supervisory signals owns that distinction in full.

• It is not open area detection. It cannot see a fire in the room the duct serves, only smoke already drawn into the return. Required area coverage is a separate design with separate devices.
• It is not occupant notification. Under the routing above it sounds nothing and flashes nothing.
• It does not work with the fan off. Nights, weekends, unoccupied setback: the device is inert, and most building fires that get people are at night.
• It does not cover the outside air stream in economizer mode unless located and sized for it. A return duct detector on a unit at 100 percent outside air sees very little return air.
• It does not replace a required duct penetration protective. The detector shuts a fan down; a damper closes a hole. The sibling card on dampers covers that split.
• It does not tell you the fan actually stopped. Shutdown is a separate control path and a separate test.

Testing it means moving air

The sensitivity and response test at the housing confirms the sensing chamber works: introduce the listed test aerosol at the test port per the instructions and confirm the panel receives the signal. Use only the aerosol the listing names, because a non-listed aerosol leaves residue in the chamber and drifts its sensitivity, showing up months later as a device that will not reset.

The airflow test confirms the device is getting a sample. Measure the differential pressure across the sampling tubes with a manometer at the ports provided, with the unit running at the condition you are qualifying, against the range in the listed instructions. That test catches a backwards tube, a plugged tube, a tube cut short, and a unit whose airflow dropped after a coil change. Putting a code required initiating device out of service to run either test is an impairment with a start time, a compensating measure and an end time.

Two hazards attach to the work, each with its own action. Opening an air handler exposes a rotating wheel and belt or spring stored energy: isolate at the disconnect and lock and tag under 29 CFR 1910.147, and where you are inside an electrical enclosure the exposure forks to 29 CFR 1910.333(b)(2), so verify absence of voltage with a meter proved live, dead, live on a known source per NFPA 70E-2021, 120.5. Rooftop units put you near a roof edge: work inside a guarded perimeter or under a fall protection system before you open a panel, never while carrying one.

Worked example: the unit that passed every test and would not have shut down

A four story office building, one rooftop unit serving floors two and three. Supply duct at the detector is 30 in by 20 in, so 600 square inches, which is 4.17 square feet. Nameplate supply airflow is 8,000 cfm, and a drive added during a controls retrofit turns the unit down to 30 percent.

Raw figure, then the correction, printed. Nameplate 8,000 cfm is stated at rated external static with a clean filter, so it already contains an assumption about a system you no longer have. That is a re-basing, not an addition: a pitot traverse measured 6,900 cfm at full speed, and 6,900 is the number every velocity below comes from.

• Full speed: 6,900 / 4.17 = 1,655 fpm, inside a 300 to 4,000 fpm listing at 5.5 times the low limit. Pass.
• Minimum turndown, at constant system resistance and constant air density: 0.30 x 6,900 = 2,070 cfm, so 2,070 / 4.17 = 496 fpm, inside the band at 1.65 times the low limit. Pass, and worth recording, because that margin is what a future filter upgrade eats into.
• Housing aerosol test with the listed aerosol: panel received the point in 21 seconds. Pass.
• Fan shutdown on activation, verified at the unit. Pass.

Then the placement check. The detector sits 2 ft downstream of the fan discharge. Duct width is 30 in, which is 2.5 ft, so it is 2 / 2.5 = 0.8 duct widths downstream of a major disturbance, against listed instructions calling for six duct widths, or 15 ft. The device sits at about 13 percent of the required straight run.

Why the other four passes did not catch it. Every passing test above ran either at the housing or on the control path, and none of them asks whether the duct is delivering a representative sample. Two feet off a fan discharge the air is a rotating, unmixed jet, so pressure at the inlet holes varies hole to hole and moment to moment, and the housing receives whatever part of the jet happens to be loading those holes. Smoke entering on the far side of the wheel can pass without being sampled. The device is not broken and not miswired; it is in the wrong place, and the only test that would have shown it is the differential pressure reading against the listed range, never recorded on any of the six annual reports in the file.

What flips the answer. Where the duct has no room for six widths, some listings permit a shorter run behind a flow conditioner or an alternative location upstream on the return. Where none works, the resolution is a design change routed through the alarm contractor and the AHJ, not a field decision on the roof.

Sibling rule check, printed. Against the supervisory signal card: the panel maps this device as a supervisory point, matching the routing above, and pulling the head to test ran as a timed impairment with a start, a compensating measure and an end. Against the notification appliance card: no appliance was expected to sound on this point and none did, which is the correct outcome rather than a failure. Against the damper card: the same duct crosses a rated shaft wall two floors down, and because that is a fire-rated assembly the penetration's protection is a FIRE damper, heat-actuated, or a combination damper where the shaft is also a smoke barrier. The barrier type selects the device.

How to verify you got this right

Ask for four numbers and refuse a passing report without them: the measured airflow at the detector location and the date measured, the duct free area, the computed velocity against the listed band, and the measured differential pressure across the sampling tubes against the listed range. A report with a checkmark and none of those four recorded that somebody was on site, not that the device works.

References

• NFPA 72, National Fire Alarm and Signaling Code, in the edition your authority having jurisdiction has adopted and amended, which reaches the trades through the permit and the acceptance test
• The mechanical code and fire code adopted in your jurisdiction, which set when a duct detector is required and how its signal is routed
• Manufacturer listed installation instructions, which own the velocity band, the straight run multiple and the sampling tube span
• 29 CFR 1910.147 for mechanical isolation and stored energy, and 29 CFR 1910.333(b)(2) for the electrical work the same job creates
• See related: What a Supervisory Signal Is and Why It Is Not an Alarm; Why a Fire Damper and a Smoke Damper Are Different Devices