Why Occupancy Sensing Fails in Predictable Places

Why this matters

Occupancy sensing fails in the same handful of geometries in every building, and the standard field response makes it worse. A tech turns the sensitivity up, the drop-outs continue because sensitivity was never the constraint, and now the zone burns after-hours run time chasing a banner moving under a supply diffuser. The failures are geometric and they are predictable from the coverage drawing before anyone is annoyed. What follows is the geometry, then one office where the wrong hypothesis was tested three times before anybody looked at a plan.

Two technologies, two different blindnesses

Passive infrared (PIR) reads long-wave infrared through a segmented lens that carves the field into alternating zones. It fires when a warm body crosses a zone boundary. That means it needs three things at once: line of sight, motion ACROSS the zones rather than straight toward the sensor, and enough temperature difference between the body and its background. Ordinary window glass blocks the 8 to 14 micron band a PIR works in, so a PIR does not see through a glass partition even though you can see through it.

Ultrasonic floods the space at roughly 25 to 40 kHz and reads the Doppler shift off anything moving. Sound diffracts, so it works around a partition and detects small motion the PIR misses. It also cannot tell a person from moving air, and its range falls as humidity and temperature change the attenuation.

Dual-technology is not "both sensors on." The logic is asymmetric and the asymmetry is the whole product: initial turn-ON requires BOTH technologies to agree, and HOLD-ON requires only ONE. That gates out the false-on that ultrasonic alone produces while keeping the fine-motion hold that PIR alone cannot deliver. Confirm on the device that this is how it ships, because some products default to either-for-on.

The angular problem, and why sensitivity does not fix it

A PIR lens divides its field into zones by ANGLE. The linear width of a zone at the work surface therefore grows with slant range, so the smallest detectable motion gets bigger the further out you go.

Say the lens uses zones roughly 5 degrees wide, a figure you read off the manufacturer's coverage drawing rather than assume, and treat as a uniform-zone simplification because many lenses deliberately narrow their peripheral zones to fight exactly this. Mount the sensor on a 9 ft ceiling over people whose hands and torsos sit around 3.5 ft above the floor, a 5.5 ft drop. At 10 ft horizontal offset the slant range is 11.4 ft and a 5 degree zone spans about 12 in at that height. At 17 ft offset the slant range is 17.9 ft and the same 5 degrees spans about 19 in. In the corner at 19.2 ft the slant is 20.0 ft and the span is about 21 in.

Keyboard typing moves a hand 3 to 6 in laterally. It does not cross a 12 in boundary, let alone a 19 in one. This is why manufacturers publish TWO coverage patterns for the same head, a large walking-motion circle and a much smaller small-motion circle, and why the small-motion circle is drawn for a hand-and-forearm gesture rather than for typing. Raising sensitivity lowers the temperature-change threshold. It does not narrow a zone. It buys a little at the margin and it buys false-on everywhere.

Line of sight is the other hard gate, and no setting reaches past it:

   ceiling  ----*---------------------------------
                |\
                | \
                |  \
                |   \   line of sight
                |    \
                |     \       +----+  partition
                |      \      |    |
   work plane --+-------+-----+----+------+------
              under near blocked
              sensor desk desk

The predictable places

Time delay is a duty-cycle decision

The delay is not a comfort setting. It trades false-off against run hours, and with LED sources the old fluorescent lamp cycling penalty that used to argue for long delays is gone. The penalty moved rather than vanished: the inrush at each switch-on is a stress the driver absorbs, so on a zone switching many times an hour the switching count belongs in the driver selection, which the driver card owns. Do the arithmetic on the actual occupancy pattern rather than picking a number.

A restroom entered six times an hour, ten minutes apart, three minutes per visit, is occupied 18 of 60 minutes, 30 percent. At a 20 minute delay the timer never expires between visits and the zone runs 60 minutes an hour, 100 percent. At a 5 minute delay each visit holds the zone from entry to 5 minutes after exit, 8 minutes of each 10 minute cycle, 80 percent. Twenty points of run time for a delay change, and no drop-out risk because nobody in that space is motionless for 5 minutes.

Where people ARE motionless, the answer is not a longer delay. It is manual-on with automatic-off, which typically saves more than any delay tuning because it removes the on-events nobody needed.

The case: two people in a twelve-person office

Signal. Two of twelve people in a 30 ft by 24 ft open office report the lights dropping out on them, several times a day. The other ten never see it. Both are on the west side of the room.

Hypothesis one, sensitivity. The ceiling PIR was raised from its mid setting to its highest. Drop-outs continued at the same rate. The controller log then showed 3 new after-hours on-events per week between 11 p.m. and 4 a.m., which turned out to be a paper banner moving under a supply diffuser. Sensitivity was reset and the events stopped, which is itself the confirmation that the sensitivity change caused them.

Hypothesis two, a bad head. The sensor was swapped for a new one of the same model. Identical behaviour on the same two desks. A fault that reproduces on new hardware in the same two seats is not a fault in the hardware.

Hypothesis three, delay too short. Raised from 10 to 20 minutes. Drop-outs became less frequent and did not stop. Logged weekday run time went from 13.6 hours to 15.1 hours against a 9 hour occupied day. The workaround cost 1.5 hours of run time per weekday and did not close the complaint.

What the drawing showed. The head's published patterns at a 9 ft mount are a 40 ft diameter walking-motion circle and a 20 ft diameter small-motion circle. The sensor is at the room centre, so the far corners at 19.2 ft are inside the 20 ft walking radius, which is why nobody had a problem walking in. The two affected desks are at 16 ft and 17 ft from the sensor, both outside the 10 ft small-motion radius. Both people work keyboard-only, no phone, no paper. At 17 ft the zone width at their hand height is about 19 in against a typing hand travel of 3 to 6 in.

The fix. A dual-technology head added over the west side, initial-ON gated on both technologies agreeing and hold-ON on either, with the ultrasonic aimed away from the nearest diffuser. Delay returned to 10 minutes.

Confirmation over four weeks. Drop-out reports from those two desks: 0. After-hours on-events: 0. Logged weekday run time 13.9 hours, which is 1.2 hours below the 15.1 hours the 20 minute delay had produced and 0.3 hours above the 13.6 hours the same 10 minute delay logged before the change. That 0.3 hour rise is the price of actually detecting the two people, and it is the correct direction: the old figure was low because the zone was switching off on occupants.

How to verify you got this right

Print the figure next to the criterion. A check written as a declaration cannot fail.

• Every occupied position is inside the SMALL-motion pattern, not the walking pattern. Furthest desk 17 ft from the head against a 10 ft small-motion radius: fails on the original head. After adding the west head, furthest desk 8 ft from the nearest small-motion centre against 10 ft: passes.
• The initial-ON logic is AND, and it was read off the device, not assumed. Verified on the installed head: both technologies required for ON, either for hold.
• No sight line is blocked to any position the head is relied on for. 12 workstations checked, 0 with a partition or rack between the seat and a head.
• The delay was chosen from an occupancy pattern, not a habit. 10 minutes, logging 13.9 h against a 9 h occupied day, and the after-hours count that justifies it is 0 events in 4 weeks.
• The sensitivity setting is back where the false-on count is zero. Mid setting, 0 after-hours events in 4 weeks, against 3 per week at the highest setting.

Hazards this work creates

• Do not deliberately let a zone time out with a person still in a stair, a machine area or a corridor to prove the drop-out. Test the timeout with the space empty and the adjacent zones up, or watch the controller log instead of the ceiling.
• Never bypass, disable or place in test any sensor in an emergency or egress lighting path. Those luminaires answer to a listed device, not to a sensor, and the sibling card on emergency lighting circuits owns why.
• Reaching a ceiling head is work at height. Use a properly set ladder or a lift under 29 CFR 1910 Subpart D for general industry or 29 CFR 1926 Subpart X for construction, never a chair or a desk, and never a ceiling grid as a handhold.
• The head is low voltage; its power pack is not. Do not open a line-voltage power pack enclosure energized. De-energize, lock and tag under 29 CFR 1910.333(b)(2), and prove dead with a meter checked on a known source before and after (NFPA 70E-2021, 120.5, binding through your employer's electrical safety program or your contract).

References

• 29 CFR 1910 Subpart D (walking-working surfaces and ladders) for general industry; 29 CFR 1926 Subpart X for construction
• 29 CFR 1910.333(b)(2), electrical work on utilization equipment conductors
• NFPA 70E-2021, 120.5, as adopted through an employer electrical safety program or by contract
• Manufacturer coverage drawings, which publish separate walking-motion and small-motion patterns for the same head
• See related: What a Lighting Control System Is Being Asked To Do; What Daylight Harvesting Needs To Work