Pulsed LED overdrive in machine vision: how bright is safe

The part is moving, the image is two stops dark, and the temptation is to turn the light up and leave it there. Pulsed overdrive is the better move: fire the LED at many times its continuous current for just the camera's exposure. Done within the light's limits, the flash is far brighter than the continuous rating while the LED dissipates less average heat than the same light at rated continuous output. Done outside them, the light dies on the spot. This page gives the rules that decide which side you land on: the peak-current table, the duty-cycle math behind it, the fastest safe repetition rate for your pulse width, and the controller settings that hold all three at once. The worked example runs a 700 mA light on a 2 Hz station; substitute your own light's table row and line rate and the same three lines of arithmetic decide it.

What does pulsed LED overdrive do to a light?

Overdrive runs the LED at many times its continuous current — typically 10× — for a strobe lasting tens to hundreds of microseconds, so the part receives a flash roughly an order of magnitude brighter while the LED's average heat stays below its continuous rating.

Electrically nothing exotic happens: the controller's constant-current channel is raised from its continuous setting, say 700 mA, to a peak such as 7 A, and the camera exposure is gated to the pulse instead of timed independently. Brightness rises roughly with current, because output scales with the carriers pushed through the chip each second. The gain is sublinear — the chip droops as it saturates — so plan on about half the multiplier: 10× the current is worth roughly 5× the light, and the remainder is margin.

The short pulse replaces a long exposure, and that is where the image wins come from. Motion blur is capped at part speed times pulse width, ambient light has almost no time to integrate, and a 100 µs flash at a few hertz is invisible to anyone watching the cell — the basis of the hidden strobe technique. Extra light also buys aperture: 10× the photons is about 3.3 stops, enough to move an f/4 station toward f/13 for depth of field, which the depth-of-field calculator turns into numbers for your lens and sensor.

How much extra current can a strobe light take?

Use the light's peak-current table, whose typical shape is 10× the continuous rating at 1% duty cycle, and treat the table's numbers — not the multiplier alone — as the binding limit.

Two ceilings decide the table. The first is instantaneous: bond wires and the chip carry a maximum current density, so no duty cycle rescues 20 A through a light whose peak row ends at 10 A. The second is thermal, and it is the subject of the next section. Published tables trade one against the other; illustrative rows for a typical machine vision light, to be replaced by your light's printed table, look like this:

Small, high-luminance spot lights are the natural overdrive loads — short thermal path, small lit field, and a trigger input built for exactly this pulse pattern:

Why does duty cycle decide survival rather than flash brightness?

Junction temperature follows average power, not peak power: for pulses far shorter than the chip's thermal time constant of roughly a millisecond, the junction barely warms within one pulse, so a 10× flash at 1% duty heats the LED less than continuous running at one-tenth of its rated current.

For widths of tens of microseconds to about a millisecond, junction temperature tracks the average power dissipated, not the peak inside a pulse. The chip is too small a thermal mass to heat much within 200 µs; heat accumulates pulse by pulse toward the steady state that the mean value sets. Average current is the handiest yardstick:

D is dimensionless, t_pulse is the strobe width in seconds, f_strobe is the repetition rate in hertz and D_max is the table's duty ceiling. At the 10×/1% row the average current is 10% of the continuous rating, so an overdriven light runs cooler than the same light at full continuous output. The width cap exists because the within-pulse temperature ride-up grows as pulses stretch toward the millisecond scale; that is why this arithmetic is the sanity check and the printed rows are the rule. Under its own table, pulsing is thermally gentler than continuous running at full output.

How fast can you strobe at a given pulse width?

Divide the allowed duty cycle by the pulse width: at a 10×/1% limit a 200 µs flash may repeat at 0.01 / 0.0002 s = 50 Hz, a 100 µs flash at 100 Hz, and a 20 µs flash at 500 Hz.

After the example the decision is mechanical: compute the duty at your worst-case rate, compare it with the row, and shorten the pulse until both the duty and the blur budget fit. LED strobe duty cycle: matching lights and controllers walks the same arithmetic from the controller side, and strobe overdrive, duty cycle and timing covers the trigger timing around it.

What must the lighting controller enforce?

The controller has to hold three settings at once: pulse current at the table's peak, pulse width at the table's maximum for that current, and the duty cycle at the table's limit — which is why strobing is a controller job and not a supply job.

A constant-current channel closes the loop on amperes, so the peak stays set while the chip warms; a constant-voltage channel switches a light that carries its own strobe driver, and that driver's rated pulse input then governs the peak. Either way, the controller should hold peak current, pulse width and duty cycle together, and the better units refuse a width-and-rate combination that breaks the row instead of letting you discover it optically. If you drive an LED stage from your own board, a pulsed driver wastes far less power than a series resistor doing the same job — the difference is measured in this NE555 pulsed LED driver versus resistor efficiency analysis.

For current that holds steady pulse to pulse — regulation, settling, droop — read LED overdrive and current control. Multiple channels also open sequencing: four overdriven pulses microseconds apart, one per channel, give a single camera a full photometric-stereo set, as multi-channel light controller sequencing lays out; the broader case for a controller in the cabinet is why use a machine vision lighting controller. The lighting-controller finder separates pulse-rated models with its strobe feature filter.

Compact single-camera stations need one clean strobing channel:

High-output lights need constant-current channels with power in hand:

Lights that accept both a 24 V voltage-mode drive and current-mode strobing suit dual-output channels:

When is pulsed overdrive the wrong answer?

Leave overdrive out whenever duty climbs toward the table — continuous inspection, video-rate previews, line scan at speed — or when a continuous light at rated current already gives you a clean histogram, because a flash budget spent where it is not needed buys nothing.

Continuous operation is the first disqualifier: a 20 ms exposure at 30 fps already spends 60% duty, far above any overdrive row, so dim a continuous light in a closed loop instead — closed-loop lighting control holds brightness against LED aging, dust and line drift.

Line scan is the second, because the same D = t_pulse × f applies per line and line rates are brutal: a 20 µs line exposure at a 20 kHz line rate is 40% duty, which is why a fast line is usually a continuous-power problem — line scan light power: how much a fast line really needs works the numbers.

Third, when the process needs steady illumination rather than flashes — live video, an operator preview, an inspection that integrates over many frames — the enemies change to mains flicker banding, the black bars from lighting flicker failure, and ambient drift, which ambient light variation in pick and place vision addresses with shielding and exposure discipline.

What do you check before switching a light to overdrive?

Before the first overdriven pulse, four numbers must agree on paper: the table's peak current, its pulse-width cap, its duty-cycle ceiling, and your worst-case trigger rate — compute the duty at that rate and only then power the channel.

• The light's peak-current table: peak current, maximum pulse width and maximum duty cycle for the row you plan to use. Every other check on this list is arithmetic against those three numbers.
• The controller channel's mode and range: constant-current output that reaches the peak, or voltage switching behind a light whose driver is rated for the pulse.
• Worst-case duty: D = t_pulse × f_strobe at the fastest rate the cell will ever see, not the rate at commissioning.
• Average current: I_avg = I_peak × D against the continuous rating — at the 10×/1% row this is 10% of the rating, which leaves generous thermal margin at short widths.
• Optical rise time: an overdriven pulse still takes a few microseconds to reach its flat top, so keep the camera exposure inside the plateau; measure what you get with LED strobe optical rise time, and watch pulse-start droop with LED strobe current droop diagnosis.
• Ambient temperature: peak rows are quoted near room ambient, and a light inside a hot cabinet earns a lower row — apply the maker's derating curve before reusing your duty math.

The whole decision in one line: take the 10×/1% row as the default, compute D at your worst-case rate, shorten the pulse until duty and motion blur both fit, and let a strobing controller hold the three numbers for you. To pair a light and controller with these ratings in one pass, start at the light configurator or the controller configurator; the light finder and the lighting topic page carry the rest of the range. If a combination needs a quote, the contact page is the fastest route.