What Flicker Is and Why People Notice What They Cannot See

Why this matters

A light meter integrates. It hands you an average over a period far longer than one cycle of the supply, so a source modulating by a third of its output and a source holding steady read the same footcandles at the same point. That is why the complaint and the reading disagree in this subject more than in any other: the level went up, the space measures better than it ever has, and two people are getting headaches while a lathe chuck at one particular speed looks like it has stopped turning. Nothing in the meter reading is wrong. It is answering a question about quantity, and the complaint is about a waveform.

Before you diagnose anything near a machine

Never judge whether a rotating machine has stopped by how it looks. Under a modulated source a spinning shaft, chuck, blade or fan can appear stationary or appear to creep slowly backwards while it is running at full speed. Confirm a stop at the machine's own controls and its energy isolation, and where you are going to put a hand near it, isolate and lock out the mechanical and stored energy under 29 CFR 1910.147 before it goes anywhere near the moving part. Do not run a stroboscopic test with a guard removed or with anyone within reach of the machine.

The two numbers, and neither means anything without a frequency

Percent flicker, sometimes called modulation depth, is (maximum output minus minimum output) divided by (maximum plus minimum), times 100. It describes how deep the modulation is and says nothing about its shape.

Flicker index is the area of one cycle above the mean divided by the total area of that cycle, running from 0 to 1. It is shape-sensitive, so it separates a gentle ripple from a narrow spike with the same peaks.

Both are stated at a frequency, and a figure quoted without one is not a specification: thirty percent modulation at 120 Hz and thirty percent at 30 kHz are two different physical situations. Two newer descriptors also appear on some manufacturers' data and in some energy-program specifications, the short-term light modulation indicator (Pst-LM) and the stroboscopic visibility measure (SVM), each with its own test condition. IEEE Std 1789-2015 sets frequency-dependent modulation recommendations and binds only where your specification, contract or employer standard adopts it, in the edition named there. None of it is a code minimum on its own.

Where seeing stops and the effects continue

Direct, visible flicker fades out for most people in the low tens of hertz, and the periphery of the visual field is more sensitive than the centre, which is why a ceiling flickers in the corner of your eye and stops when you look at it. Above that threshold the modulation does not go away, it stops being seen as flicker and shows up as three other things: the stroboscopic effect, where a moving object is sampled rather than lit continuously so its apparent motion is wrong, which is the one with a machine-guarding consequence; the phantom array, where a small bright source smears into a row of beads as the eye saccades past it, noticeable well into the hundreds of hertz; and non-visual load, the headache and eyestrain reports that arrive without anyone saying the word flicker.

The practical rule: the absence of visible flicker is not evidence of low modulation. Anyone who says "I looked at it, it's fine" has tested the one band where the eye is least useful.

The freeze arithmetic

A rotating object appears frozen when the light's modulation frequency divided by the object's rotational frequency is a whole number, because every flash then catches it in the same orientation. Write the modulation frequency as f and the rotation as r revolutions per second: apparent standstill at f / r = 1, 2, 3, 4 and so on.

Full-wave rectified 60 Hz supply gives f = 120 Hz, and that lands exactly on the two commonest machine speeds in North American shops. A 2-pole motor at a synchronous 3,600 rpm is 60 rev/s and 120 / 60 = 2. A 4-pole motor at a synchronous 1,800 rpm is 30 rev/s and 120 / 30 = 4. Both are freeze conditions.

An induction motor under load runs below synchronous speed, and slip rises with load, so a loaded 4-pole machine sits under 30 rev/s rather than on it. That does not remove the effect, it converts a standstill into a slow apparent creep: the apparent rate is the actual rate minus the nearest freeze rate. The direction of the creep tells you which side of the freeze condition the machine is on, and a machine that appears to be turning slowly backwards is running slightly slower than a freeze rate, not backwards.

Where the modulation comes from

• The driver. A converter with little energy storage passes the supply ripple straight through to the output; one with adequate storage flattens it. That is why two drivers with identical output current behave completely differently on the same board. The driver card owns the rest of the part.
• The dimming method. Phase-cut control chops each half cycle, and what the driver does with a chopped input is its own behaviour. Pulse-width modulation imposes a carrier of its own, and a carrier of a few hundred hertz is invisible as flicker and fully effective as a stroboscopic source.
• The dim level. On most drivers percent flicker rises as output falls, so a measurement at full output does not describe the setting the space runs at. Measure at the level in use.

The case: a retrofit that measured better and worked worse

Signal. A machine shop replaced fluorescent high bays with LED high bays. Illuminance at the bench line went from 32 fc to 58 fc, horizontal at 36 in above finished floor, as-found on one meter, no light loss factor applied to either figure. Two second-shift operators then reported eye fatigue, and one reported that the lathe chuck "looks stopped" at a particular speed setting.

The first hypothesis, glare. A different distribution can put a bright source in the field of view, and the glare card owns that assessment. It was run: the luminance ratio at the operator positions had not materially changed, and the complaints tracked machine speed rather than head position. Glare does not switch on and off with a speed dial. Set aside.

The second hypothesis, too much light. 58 fc is not a fatigue level by itself in a machining space, and the complaints came from 2 operators out of a shift of 9, both at lathes. If level were the cause it would not select for the people running rotating work. Set aside.

The measurement that separated it. A phone camera in video mode showed banding across the new fixtures and none across the one remaining original fixture. That is a screening test, not a measurement: rolling-shutter banding detects presence, not magnitude, and its sensitivity depends on shutter speed. It justified the real one. Percent flicker at the bench under the new high bays: 38 percent at 120 Hz at full output. Under the surviving fluorescent fixture on an electronic ballast operating in the tens of kilohertz: under 5 percent at 120 Hz, which is why this shop had never had a stroboscopic complaint.

Applying the freeze arithmetic. The lathe's problem speed was its 1,800 rpm setting: 1,800 / 60 = 30 rev/s, and 120 / 30 = 4, a whole number, so the chuck is caught in the same orientation every flash. At the neighbouring 1,750 rpm setting it runs at 29.17 rev/s, and the nearest freeze rate is 30 rev/s, so the apparent rate is 29.17 - 30 = minus 0.83 rev/s, an apparent slow reverse rotation of 0.83 x 60 = 50 rpm. The operator's description of a chuck that "drifts backwards" was an accurate report of the physics.

The fix, and what it does not do. The 18 high bays over the machine bays were redistributed across the three phases, 6 per phase, so each machine position receives comparable flux from fixtures on all three. The three ripples then arrive evenly spaced across the 120 Hz cycle and their 120 Hz components largely cancel wherever all three reach. Measured composite modulation at the lathe operator position afterwards: 9 percent, with the residual at 360 Hz and above rather than at 120 Hz, because the 120 Hz and 240 Hz components of three phases spaced 120 degrees apart both sum to zero and 360 Hz is the first that adds in phase. The freeze condition is not abolished, because 240 / 30 = 8 is also a whole number; what changed is depth, and stroboscopic visibility scales with depth. A task luminaire at the lathe, selected on a stated percent flicker at the level it runs at, was added as the durable control.

Confirmation. Illuminance at the bench line after the change: 58 fc, horizontal at 36 in AFF, as-found, unchanged, because only phase assignment moved. Operator reports of an apparently stationary chuck over the following month: 0. That is the point of the whole exercise. The reading that certified the retrofit was identical before and after the fix that resolved the complaint, so the reading was never going to find it.

Checking your own figures

• Every modulation figure carries its frequency. 38 percent at 120 Hz, under 5 percent at 120 Hz, 9 percent composite with residual at 360 Hz and above (360 Hz because the 120 Hz and 240 Hz components of three phases spaced 120 degrees apart both cancel). No bare percentage appears.
• Freeze condition recomputed. 1,800 rpm = 30 rev/s, 120 / 30 = 4, freeze. 3,600 rpm = 60 rev/s, 120 / 60 = 2, freeze.
• Apparent rate printed with its sign. 29.17 - 30 = minus 0.83 rev/s, 50 rpm apparent reverse, stated as slower than the freeze rate rather than as reverse rotation.
• The claim about the fix is bounded. 38 percent to 9 percent is a depth reduction; 360 / 30 = 12 is still a whole number, so no claim of elimination is made.
• Both illuminance figures carry plane, height and basis. 32 fc and 58 fc horizontal at 36 in AFF, as-found, no light loss factor applied; the 58 fc appears twice at the same value because nothing changed it.
• Complaint counts checked against the shift. 2 operators of 9, both at rotating work, which is the selection that killed the level hypothesis.

Hazards this work creates

• Isolate and lock out the machine's mechanical and stored energy under 29 CFR 1910.147 before any measurement is taken within reach of a moving part, and keep every guard in place for any test run with the machine turning.
• Redistributing fixtures across phases is panel work. De-energize, lock and tag 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). Re-check phase loading after the change so you have not created an imbalance.
• Do not dim, switch or otherwise alter lighting over an occupied machining area to run a test. Stop the machines first, or test out of shift with the adjacent zones at full.
• Reaching a high bay is work at height, from a properly set lift or ladder under 29 CFR 1910 Subpart D for general industry or 29 CFR 1926 Subpart X for construction, with the area below cleared of people and work in progress.

References

• 29 CFR 1910.147, control of hazardous energy for mechanical and stored energy isolation; 29 CFR 1910.333(b)(2) for electrical work on utilization equipment conductors, which 1910.147 expressly excludes
• 29 CFR 1910 Subpart D for general industry walking-working surfaces and ladders; 29 CFR 1926 Subpart X for construction
• NFPA 70E-2021, 120.5, as adopted through an employer electrical safety program or by contract
• IEEE Std 1789-2015, recommended practice for modulating current in high-brightness LEDs, binding only where a specification, contract or employer standard adopts it in the edition named
• See related: What a Driver Is and Why It Fails First; What Dimming Compatibility Actually Depends On; What Glare Is and the Two Kinds That Matter