LED overdrive and current control: stable machine vision lighting

A machine vision image is only as good as the light that made it, and an LED's brightness follows exactly one variable: the current through the diode. Lights are usually specified by voltage, but voltage is a poor handle on brightness — a small change in rail voltage produces a much larger change in light output. That gap is why measured dimensions wander between shifts, why thresholds get retuned, and why a brighter strobe is the cheapest way to reject factory ambient light. This guide covers how LED overdrive works, how far a strobed light can be pushed (about ten times its continuous rating under a duty-cycle cap), the arithmetic that keeps the pulse safe, and what to check on a light controller before you order one.

The decision: regulate the current, not the voltage

An LED emits light in proportion to the current flowing through it, while its forward voltage barely moves across the useful range. That is why two identical lights on the same 24 V rail can differ visibly: cable length, connector resistance and junction temperature all nudge the current, and brightness follows. Voltage drive also sits the diode on the steep part of its curve, so a small rail change buys a large current change.

So the decision is simple. Any light feeding a measurement, a match or a pass/fail call runs from a light controller that regulates LED current and supports strobing. Constant-voltage drive stays acceptable for indicator-style go/no-go stations, where a few percent of drift changes nothing. Everything else here follows from that one choice.

What overdrive buys, and where the ceiling sits

Overdrive means pulsing the LED at a current above its continuous rating. It is permissible at all only because the light is off for most of each cycle, so the pulse's heat drains away before the next one arrives. The usual strobe ceiling is about ten times the continuous rating, and it always comes as a set: a current multiple, a duty-cycle cap and a maximum pulse width, stated in the light's strobe rating.

The intensity gain roughly tracks the current ratio — 5x the current targets 5x the intensity — but LED efficiency droops as pulse current climbs, so the achieved gain sits below the nameplate multiple. The current-droop diagnosis shows how to measure the real gain on your own light. One limit matters above all: only lights with a published pulse rating may be overdriven. A light sold for continuous operation has no stated headroom, and its ceiling is its continuous rating.

The payoff is ambient rejection. A strobe 10x brighter lets you cut exposure 10x; ambient light contributes in proportion to exposure time, so the background falls by the same factor while the strobed image stays bright. The measurement procedure is in strobed lighting ambient rejection, and hidden strobe operation keeps the flashes invisible to operators.

The same arithmetic covers every form factor — ring lights included:

A ring that clears a 48 mm inner diameter seats around the lens, and its pulse is sized with the same duty-cycle formula used below.

The duty-cycle math

A light overdriven k times may be on for at most 1/k of the time if average power is to stay at the continuous rating: 10x means 10%, 5x means 20%, 2x means 50%. The overdrive multiple, duty cycle and pulse timing form one package; the companion guide on overdrive, duty cycle and strobe timing works through the same coupling from the timing side.

Worked example. Take a 24 V bar light rated 0.5 A continuous:

• Continuous power: 24 V × 0.5 A = 12 W.
• A 10x pulse: I_pulse = 10 × 0.5 A = 5 A, so peak power is 24 V × 5 A = 120 W.
• Duty-cycle cap: D_max = 12 W ÷ 120 W = 10%.
• With 1 ms pulses, t_off must be at least 9 ms, the period is 10 ms or more, and the light manages 100 strobes per second.
• A line at 200 frames per second cannot use that setting. Drop to 5x: D_max = 20%, period ≥ 1 ms ÷ 0.20 = 5 ms, and 200 strobes per second fit. Or keep 10x and shorten the pulse to 0.5 ms: period ≥ 0.5 ms ÷ 0.10 = 5 ms again, and the shorter exposure cuts motion blur at 100 mm/s from 0.1 mm to 0.05 mm.

Two conclusions follow. The controller channel must deliver the peak, not the average: that channel faces 120 W for a millisecond even though the light draws 12 W on average. And the frame rate you need constrains overdrive depth before any optical argument does. How pulse width and exposure interact is covered step by step in how strobe pulse width freezes motion blur.

Why 10% of brightness is 0.5% of the measurement

Edge and calliper measurements react to light level. The working rule: a 10% change in light level shifts a simple calliper measurement by about 0.5%. On a 100.0 mm dimension that is 0.5 mm of apparent movement — enough to fail a ±0.3 mm tolerance with a perfectly good part. On a 25 mm bore it is 0.125 mm.

Brightness drifts for three reasons. A continuously-on light warms up, and output sags over the first tens of minutes as the junction heats. LEDs age, losing output across months of continuous running. And mains-powered ambient beats against the exposure as flicker; the diagnosis paths are in image brightness changes frame to frame and camera flicker under artificial lighting.

Strobing removes the first two causes: the LED is off between images, so the junction stays near ambient, and on-time shrinks by the duty factor, so ageing accumulates at one tenth the rate at 10% duty. Whatever drift remains is what a closed loop is for — closed-loop lighting control trims the drive until the image, not an assumption, holds level. As a benchmark, one published metrology study measured run-to-run flux stability of a controlled source at about 0.4% (repeats of 0.385%, 0.399% and 0.478%); that is the order of stability a regulated, strobed chain should approach.

Current-regulated against voltage-fed drive

Sizing a controller means counting three things: peak power per channel, strobe support and channels for the lights you fire independently. Three outputs from the catalog span the range from a small single-light station to overdriven pairs:

Values as printed in each product's data sheet; open a part for its drawing and full specifications.

The LS-CTL-0054 carries 240 W per channel. A 120 W pulse like the worked example's fits with a 2x margin, which is the margin to keep, because pulse ratings assume a cool junction. The shop finder for light controllers filters the catalog by channels, control type, output type and strobe feature.

Trigger to pulse: budget the timing

A strobe controller fires from a trigger input, opto-isolated so PLC, sensor and robot outputs can drive it directly; wiring one from robot I/O is walked through in trigger strobe lighting from robot IO. Two delays stack between the trigger edge and full light output: the controller's trigger-to-light delay and the LED's optical rise time. The rise time of an LED light is microseconds, and why microseconds of light delay blur your image shows what those microseconds cost, while measuring optical rise time shows how to separate rise time from controller delay on the bench.

Treat the total as a fixed offset: measure it once, then trigger that much early. The arithmetic is unforgiving on moving lines. Suppose trigger-to-light jitter is ±0.1 ms and the part travels 100 mm/s: position scatters by 100 mm/s × 0.0001 s = ±0.01 mm in the image, a tenth of the 0.1 mm budget the light chain is working to hold. When the camera fires but the image comes back black, pulse and exposure missed each other; the timing fix for a black strobed image is the first check. If the camera is set so the trigger pulse width defines exposure, then trigger pulse width sets exposure time. Hardware triggering is also the tool for holding apparent brightness steady across a shift, as this guide to stabilizing lighting changes with triggers sets out.

Checks before you order

1. Output regulation. Constant-current output for any light doing measurement work; constant-voltage only for indicator duty.
2. Peak power per channel. At least the pulse power, not the average: the worked example's channel must supply 120 W for a millisecond.
3. Strobe and trigger. An opto-isolated trigger input your PLC or robot can drive, with pulse widths down to your shortest exposure.
4. Channels. One per light you fire independently; a sequential multi-light station needs one channel per light.
5. Readback. Per-channel current and voltage, plus fault states such as disconnected or short-circuited lights, so a dead light is a reported error instead of a mystery batch.
6. Wiring. Connector and current rating on the light side; adapter connectors and current ratings covers what changes when pulse current rises.

When overdrive is the wrong tool

High frame rates push duty cycle past the cap first. Past roughly 200 frames per second at millisecond pulses, the honest alternative is more light, not more current: a physically larger light run continuously within its rating. One option of that size class:

Run continuously within rating, it trades peak intensity for freedom from duty-cycle arithmetic. Overdriving a light that has no published pulse rating is the second dead end, because there is no headroom to spend. Third, characterise multi-colour lights at the working pulse before committing: colour balance can shift with pulse current. And when a pulse shrinks toward the LED's optical rise time, most of the energy lands during the ramp and usable light falls, so size pulses comfortably above it.

Where to start

Choose light geometry from the surface and feature first — the light geometry selection guide maps surface types to geometries — then size the pulse with the duty-cycle formula, then pick the controller whose peak channel power covers the pulse. The lights finder filters bar, ring and the other forms by size, colour and angle, the lighting guides hub collects the companion articles, and the configurator opens at the controller step:

If the arithmetic above lands between two options, the contact page puts the application in front of the people who configure these stations every day.