LED lifetime L70: why strobe mode makes lights last longer

A machine-vision light datasheet promises a lifetime in hours, and the number looks simple until a strobed application has to plan around it. The figure is an L70 rating: how many hours the LEDs run before output has fallen to 70% of the starting value, measured in continuous operation at a stated ambient temperature. Strobe mode changes both sides of that ledger, cutting junction heat and the number of hours the chips are actually energized. This guide gives you the arithmetic: read any L70 rating correctly, estimate what strobing adds for your cadence, and decide where LED lifetime belongs on your risk list.

What an L70 rating actually promises

LEDs rarely fail the way filament lamps do. Nothing burns out; the chip fades. Industry convention calls an LED source end of life when it has lost 30% of its light, and that point is written L70: the operating hours at which output has decayed to 70% of the initial value. A light rated L70 at 50,000 hours still works at hour 50,001 — it is about 30% dimmer.

For an inspection, that fade is a defect mode in its own right. At L70 you need 1.43× the exposure time (that is 1/0.7) at the same f-number, roughly half a stop, or you accept 30% less signal. With white LEDs the phosphor ages too, so colour temperature drifts and a colour inspection can fall out of tolerance before the brightness margin is gone; the white LED colour temperature guide covers how to specify against that. Treat the rating as a brightness budget: design the tooling to pass at L70 output, not at day-one output.

What the rating is based on

The hours come from component-level aging tests: LEDs run at a fixed drive current and case temperature for thousands of hours, the decay curve is fitted, and the time to 70% is extrapolated from it. The printed number holds for the printed conditions. A typical machine-vision rating reads 50,000 hours of continuous operation at up to 40 °C ambient air, with the light bolted to a metal surface that acts as the heatsink.

Every condition is a lever. Drive the chip harder and the hours fall. Bolt the light to thin painted sheet instead of bare metal and the junction runs hotter than the test did. Park it in 55 °C panel air instead of 40 °C and, by the anchor in the next section, roughly 40% of the hours are gone before you switch anything on. Ambient temperature is the condition most often ignored on a running machine.

Junction temperature is the aging clock

Degradation inside the chip — electromigration in the metallisation, darkening of the encapsulant, wear of the phosphor — accelerates with junction temperature, the temperature at the die itself rather than at the case. Published component data makes the sensitivity concrete: one widely cited pairing gives about 50,000 hours to L70 at a 65 °C junction against about 25,000 hours at 85 °C. Twenty degrees hotter, half the life. That pairing describes one LED family rather than a law of nature, but it is a sound anchor for estimates:

Every 20 °C you remove from the junction roughly doubles the hours; every 20 °C you add roughly halves them.

The duty-cycle arithmetic

A strobed light is off for most of every cycle, and three quantities follow from the duty cycle D:

Here t_on is the flash time and T the trigger-to-trigger time, both in seconds; P_rated is the continuous electrical rating in watts; H_run is the hours the machine runs per year and H_on the hours per year in which the LED is energized. Rating hours are hours of energized operation, so H_on is what consumes them. Because average heat input scales with D, the first-order estimate of average junction temperature is:

where Tj_cont is the junction temperature behind the continuous rating. Flash per part instead of running constant light, and both entries in the ledger — hours consumed and heat per hour — fall in proportion to D.

Worked example: a two-shift backlight line

The arithmetic is easiest to trust on a real part. The LS-BKL-0865 is a 400 x 300 mm high-power backlight rated 92.2 W at 24 V.

The 125-year figure is an estimate stacked on an anchor, not a promise. The useful reading is simpler: at any realistic strobe duty cycle, L70 leaves the failure list. On a strobed line, the parts that age on wall-clock time are the cable at the moving axis, the connector pins and the controller's power stage, and they give out decades before the chip fades measurably. The same light run continuously on the same schedule is what produces the 12.5-year figure, and that is the number a non-stop line has to plan around.

Overdrive: brighter flashes, the same ledger

Strobe mode also permits peak drive current well above the continuous rating, which is how a 20 ms flash freezes motion that continuous light smears. The thermal ledger does not care what the current is called: average heat is still D × peak power, and the peak junction temperature adds its own stress. Two hardware limits apply before lifetime ever enters the argument — the controller's duty-cycle and pulse-width ceilings, covered in the LED strobe duty cycle guide, and the optical rise time that decides how much of each pulse is usable light, covered in the rise time measurement guide. The overdrive, duty cycle and timing guide walks the timing set-up end to end.

Strobing carries a second benefit unrelated to aging: with the sensor exposed only while the flash is on, ambient light stops contaminating the image, and the strobed ambient rejection guide shows how to measure that margin. Strobe-capable parts span both the backlights and bar lights ranges, and strobe output is a filterable feature across the light controllers category.

When the light cannot be strobed

Two cases resist the whole idea. The first is hardware: some lights are specified for constant-on operation only, with no driver and no trigger input. If you want an infrared source that simply runs whenever the station has power, the LS-BKL-0416 is built for exactly that: a 136.4 x 136.4 mm, 850 nm backlight needing no driver, specified for constant-on operation only. Its rating hours are its whole story, and the production schedule has to fit inside them.

The second is physics: line-scan and full-frame web inspection expose every millisecond the web moves, so D = 1 by construction and strobing buys nothing thermally. There the levers are thermal, not electrical. Bolt the light to bare metal, keep it out of the hot corner of the panel, and dim through a regulated controller rather than a supply rail — the closed-loop lighting control guide shows why regulation beats an open supply as parts age. If brightness is short, look at the lens before the light: opening from f/4 to f/2 multiplies the light reaching the sensor by four with no extra watt anywhere, a trade the M12 lens light gathering guide works through. And if the dimming method is PWM, pick a frequency the exposure integrates over cleanly, because banding in images shares its signatures with mains-light flicker — the camera flicker guide lists the diagnostics.

Ultraviolet: the short-lived exception

Short-wavelength chips age faster under their own light. Industry tech notes commonly rate 405 nm parts around 10,000 hours of continuous operation against 50,000 hours for visible-light models — about a fifth of the life. Strobe duty cycle stretches UV life by the same arithmetic as any other colour, so a UV light flashed 10% of the time consumes its rating at a tenth of the calendar pace. Ultraviolet options in the shop appear at 365, 395 and 405 nm.

Checks before you order

• Read the rating line, not the headline: hours, operating mode, maximum ambient and mounting surface. The conditions are the rating.
• Compute the worst-case duty cycle and annual energized hours for your cadence. A few hundred H_on hours a year moves LED aging off the risk list; H_on near H_run keeps it on.
• Measure air temperature at the light's position at the end of a shift, not at the room thermostat. Panel air at 55 °C costs roughly 40% of the rated hours on its own.
• Match the control input to the plan: a constant-on light cannot join a strobe sequence, and the controller finder filters strobe output directly.
• Mount for heat: bare metal under the base, through-bolted, no insulating gasket unless the datasheet calls for one.
• For ultraviolet parts, divide the hours by five and order the guarding with the light.

The decision, case by case

Triggered part inspection. Strobe, and stop budgeting for LED fade. At 20% duty on a two-shift schedule, L70 sits past 60 years of calendar life on the rating alone, and the thermal anchor stretches it further. Pick a strobe-capable controller and size the flash for the motion. One current option is the LS-CTL-0070, a four-channel digital control box rated 200 W with an RS-232 port and strobe among its features.

Continuous web or 24/7 operation. D = 1 is physics, so plan to the rating: 50,000 h ÷ 8,760 h/year ≈ 5.7 years to L70. Mount to bare metal, hold the panel temperature down, dim through a regulated controller as the light ages, and leave a 30% brightness margin in the tooling so hour 50,000 still passes.

Constant-on-only lights. The rating is the whole plan. If the schedule will not fit inside the hours, change the part rather than the arithmetic.

Ultraviolet. Divide by five, strobe where the inspection allows, and buy the shielding with the light.

To choose light and controller together, start at the lighting configurator or the light finder; the lighting topic page collects the companion guides.