What Daylight Harvesting Needs To Work

Why this matters

A daylight control loop does not measure the thing it is being paid to deliver. It regulates a photosensor's own reading in the ceiling, and the deliverable is illuminance on a work plane several feet below that nobody is measuring after commissioning day. Everything the system is worth therefore rests on one quantity: the ratio between what the sensor sees and what the task plane gets, and whether that ratio holds still as the sun moves. It is a property of where the sensor is and how the room is shaped, not of the controller, so no amount of setpoint tuning fixes a sensor in the wrong place. That is why the two most common outcomes are a zone that never dims and a zone that dims when the room is already too dark, and why both of them log as normal operation.

Two architectures, two different things going wrong

Closed loop. The sensor sees the combined result, daylight plus the electric contribution it is controlling, reflected off the surfaces below, and drives to hold its own reading at a setpoint. Because the electric contribution is inside the measurement, the loop self-corrects for depreciation and dirt. Because whatever sits under it is also inside the measurement, it is fooled by a dark desk mat, a stack of boxes or a partition moved into its view.

Open loop. The sensor sees daylight only, aimed so the electric lighting cannot reach it, and the controller applies a stored relationship between what it reads and what the task plane needs. It cannot be fooled by what is under it, and it has no feedback at all, so the stored relationship drifts with sun angle, season, blind position and any change to the electric system.

Neither is better in general: closed loop suits an interior zone with stable surfaces, open loop suits a floor that changes constantly, such as a warehouse or a sales floor.

The setpoint is referenced to the maintained target, and that is where the margin lives

The design target is a maintained illuminance, which already contains its light loss factors. The light loss factor card owns how that product is built; take only the consequence, which is sharp and usually missed.

Set the loop to hold the maintained target measured at the task plane, not whatever the electric system happens to deliver today. On commissioning day the system is at its initial output, so a loop holding the maintained target sits below full output even with the blinds shut, and that gap is the loop's dimming headroom. The headroom is exactly the depreciation the design has not spent yet. At the end of the maintenance interval, with the light loss factor fully consumed, full output delivers the maintained target and the electric-only headroom is zero. That is the design working as written, and it is why the record has to carry the as-found full-output level and the date: without that pair, nobody can tell whether a zone that no longer dims at night has lost its sensor or simply spent its margin.

The placement test: one ratio, two genuinely different conditions

This is the load-bearing procedure of the whole subject and it takes about twenty minutes. With the electric lighting off, read the task plane and read the sensor under one daylight condition, then again under a materially different one, and divide task-plane illuminance by sensor reading each time. If the two ratios agree, the sensor is watching a part of the room that tracks the task plane. If they disagree by more than instrument spread, the sensor is in the wrong place and no controller setting rescues it.

The error handling matters because both ratios come from the same two instruments. A calibration error in the hand meter appears in both numerators and a scale error in the sensor appears in both denominators, so both divide out of the comparison between the ratios. What survives is repeatability, an independent random spread on each of the four readings, and independent spreads combine in quadrature.

Three placements fail this test every time: a sensor a direct sun patch crosses at some hour, a sensor straddling a daylit and a non-daylit zone, and a sensor inside the throw of a task light or a monitor. One condition no placement fixes: a control zone spanning a daylit strip and an interior area has to serve its worst point, so it holds at full and harvests nothing while the window end is overlit. Draw zones to the daylight, not to the furniture.

Deadband, delay and fade

Three parameters keep the loop from chasing weather, each answering a different question. Deadband is the band around setpoint inside which nothing is commanded, so it sets how small a change is worth acting on. Delay is how long the reading must stay outside that band before the loop acts, chosen against the length of the disturbances the site actually produces, because a transient shorter than the delay produces no change at all. Fade rate decides whether the occupant perceives the correction at all. A cloud crossing in under two minutes is a disturbance you want ignored, and the delay is the parameter that ignores it.

The commissioning record for one zone, filled in

A perimeter row, 6 luminaires, 0-10 V dimming on a linear curve, closed-loop ceiling photosensor 8 ft in from the glazing. Every illuminance figure is horizontal at 30 in above finished floor, taken as-found on one hand meter stated at plus or minus 5 percent of reading, with no light loss factor applied to any reading. Sensor readings are in the controller's own units.

Placement test as run. Electric lighting off for all three conditions.

Correction line, error basis and character. Comparing ratio B against ratio A, the hand meter's plus or minus 5 percent of reading and the sensor's scale factor both appear in both ratios and divide out. What remains is repeatability, taken as an independent random 2 percent on each of the 4 readings, combining in quadrature to 4.0 percent. B against A is (2.34 - 2.29) / 2.29 = 2.2 percent, inside 4.0 percent, so those two conditions agree and the placement passes on them.

Condition C is not instrument spread. 0.46 against 2.29 is 80 percent low, twenty times the allowed 4.0 percent. Trace what the loop does with it: the sensor reads 26 units against a 13-unit setpoint, so the loop drives the zone to minimum, where the measured electric contribution at the task plane is 2 fc. Total delivered is 12 + 2 = 14 fc against a 30 fc target, 16 fc short, at the hour of day when the space is still occupied. The controller log for that hour shows the loop at setpoint, because it is, on the only quantity it can see.

The correction is a placement change, not a setting change. The sensor was shielded and moved out of the low-sun patch. Re-run at the same late-afternoon condition: 13 fc task plane, 5.6 units, ratio 2.32 fc per unit, which is (2.32 - 2.29) / 2.29 = 1.3 percent from condition A, inside the 4.0 percent spread. Setpoint, deadband, delay and fade were not touched, because none of them was ever the problem.

Checking your own figures

• The maintained target is used once, as a re-basing. 30 / 0.72 = 41.7 fc predicted initial. No reading here is divided by 0.72 a second time.
• The commissioning reading is checked against the prediction with a bound. 41 fc against 41.7 fc is 1.6 percent low, inside plus or minus 5 percent of reading.
• The dimming figure carries its curve and control point. 73 percent of full at 7.2 V measured against 7.3 V predicted on the linear curve the submittal states.
• The sensor cross-check is arithmetic, not assertion. 0.73 x 18 = 13.1 units against 13 measured.
• The ratio spread is derived before any ratio is judged. 2 percent repeatability on each of 4 readings, in quadrature, gives 4.0 percent. A to B 2.2 percent, pass; A to C 80 percent, fail; A to relocated C 1.3 percent, pass.
• The shortfall is printed in the target's own units. 12 + 2 = 14 fc against 30 fc, 16 fc short.
• The headroom claim is arithmetic. 41 - 30 = 11 fc, 11 / 41 = 27 percent of full output, and at the end of the interval full output and the maintained target are the same figure, so headroom is zero by construction.
• Every illuminance figure carries plane, height and basis: horizontal at 30 in AFF, as-found, no light loss factor applied, electric lighting off for all three placement conditions.

Hazards this work creates

• Do not force a zone to minimum or off in an occupied space to take a low-end reading. Tell the occupants first, keep the adjacent zones at full, and do it out of hours in any area with stairs, moving equipment or machinery.
• Never override, calibrate or place in test any device in an emergency or egress lighting path. Those luminaires answer to a listed device, and the sibling card on emergency lighting circuits owns the arrangement.
• Reaching a ceiling sensor to shield, aim or relocate it is work at height, from a properly set ladder or lift under 29 CFR 1910 Subpart D for general industry or 29 CFR 1926 Subpart X for construction, never off furniture and never using the ceiling grid as a handhold.
• The sensor head is low voltage and its power pack is not. De-energize, lock and tag the branch circuit before opening a line-voltage power pack or a control enclosure, under 29 CFR 1910.333(b)(2) for electrical work on utilization equipment, 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).
• Taking a task-plane reading near glazing at the late-afternoon condition puts you looking toward a low sun. Position yourself so the meter reads the plane and your eyes are not on the source, and do not walk a grid backwards while watching a display.

References

• 29 CFR 1910.333(b)(2), electrical work on utilization equipment conductors; 29 CFR 1910 Subpart D and 29 CFR 1926 Subpart X for the height half
• NFPA 70E-2021, 120.5, as adopted through an employer electrical safety program or by contract
• Manufacturer photosensor coverage and aiming data and the controller's commissioning instructions, which own every device-specific value in this subject
• See related: What a Lighting Control System Is Being Asked To Do; What a Light Loss Factor Is Accounting For; Why Occupancy Sensing Fails in Predictable Places; How to Take a Light Level Reading Somebody Else Can Repeat