What Lumen Depreciation and Lifetime Ratings Mean

Why this matters

A lifetime figure on a lighting product is not a measurement of how long the product lasts. It is a projection, made from a test of one component, run at a drive current and a temperature that may or may not match the product you are holding, extrapolated past the data by a rule with a hard limit. Every one of those qualifications is dropped when the number reaches a proposal, and what arrives on the customer's desk is a round hour count that reads like a warranty. Then the space goes dim at year six of a stated hundred thousand hours and somebody has to explain the difference between a projection and a promise, usually after the fact.

The definition had to change when the source stopped failing

A lamp-based system had a rated life that meant something plain: the point at which half a tested population had failed. It was a failure statistic, it described going dark, and a maintenance program could be built directly on it, because the event it predicted was self-announcing.

A solid-state array mostly does not go dark. It fades. So the industry replaced a time-to-failure statistic with a flux threshold, and that is what an L-number is: the point at which output has fallen to a stated fraction of initial. L70 is the point at which 70 percent remains. L90 is the point at which 90 percent remains. They are different criteria and cannot be compared to each other at all. A shorter L90 hour count is a stricter claim, not a worse product, and a sheet that sets one against the other has compared two different questions.

There is a second letter that usually goes missing. B carries the fraction of the population that has fallen below the L-threshold by that time. L70/B50 says half the units are below 70 percent at that hour; L70/B10 says one in ten is. A bare "L70" is a median claim by default. If the number is going into a contract, ask which.

What is measured, and what is projected

Two separate documents sit behind any hour count, and only one of them contains data.

The measurement is a long-duration flux maintenance test of the LED package, not of the luminaire. Packages are run at stated drive currents and stated case temperatures and their output is logged over thousands of hours. That is real data about a real component.

The projection takes the later portion of that data, fits a decay curve, and extrapolates it forward. The method that governs this caps what may be reported at six times the tested duration. Ten thousand hours of data supports a reported figure up to 60,000 hours and no further. A claim past that limit is an extrapolation that the method does not authorize, and it should be labelled as one on the page. The cut-sheet card uses this cap as a screening test on a sales document; this card is where the number underneath it comes from.

The gap between those two documents is where the confusion lives. The measurement is of a package. The thing you are buying is a luminaire, which runs its packages at its drive current inside its thermal design. The manufacturer bridges that gap by measuring the in-situ case temperature in the finished product and interpolating the package curves to that condition. If that interpolation was not done, the published curve belongs to a different object.

That in-situ temperature is a controlled laboratory measurement, not something a field tech should go get. Opening an energized luminaire at height to land a thermocouple on a board is energized work at height for a number the manufacturer already has. If your ambient is in question, ask for the derating curve instead.

The two conditions the number holds constant

A projected decay rate is stated at a fixed drive current and a fixed in-situ case temperature, and those two conditions travel with it exactly the way a torque spec travels with its lubrication condition. Raise either and the decay accelerates; the published curve does not follow you there. This is why the same LED package carries several different published curves and why a luminaire that drives it harder to advertise a bigger lumen figure is spending lifetime to do it.

It is also why a hot mounting location moves the answer without changing the product. A luminaire tested at a 25 C bench ambient and installed under an uninsulated roof deck is running its packages above the case temperature the curve was interpolated to, and the curve is now optimistic by an amount only the derating data can tell you.

The gate: four things a lifetime figure must state

A lifetime figure is usable in a design only when all four are on the page or obtainable from the report:

1. The criterion. Which L-value, and ideally which B-value.
2. The tested duration behind it, so the reporting cap can be checked.
3. The drive current and the in-situ case temperature the projection was interpolated to.
4. The reported hours inside six times the tested duration.

Miss any one and the figure informs a question to the manufacturer. It does not carry a decision.

Two claims through the gate

Same job, two products.

Claim A, as printed: L70 greater than 100,000 hours. Package test duration 10,000 hours. No drive current stated, no case temperature stated.

Claim B, as printed: L90 at 54,000 hours, at 1,050 mA and 85 C in-situ case temperature, projected from 9,000 hours of package data.

On the sheet A looks close to twice as good. Run the gate.

A is struck. Its hours sit 1.67 times past what its own test duration supports, and with no drive current or case temperature there is nothing to interpolate to this luminaire's condition. It is not a smaller number than it claims. It is not a number.

B stands, on a stricter criterion, at a defensible duration, with both held-constant conditions printed. The ranking on the sheet reversed, and it reversed because of the qualifications rather than the magnitudes.

Turning the survivor into the number a design needs

What a design actually wants is not an hour count. It is the depreciation term inside its light loss factor at your maintenance interval, and the light loss factor card owns how that term joins the others.

The space runs 14 hours a day, 6 days a week, 52 weeks a year: 14 x 312 = 4,368 operating hours per year. Claim B's 54,000 hours is therefore 54,000 / 4,368 = 12.4 years, which is far past the shop's 6-year cleaning and replacement cycle. So the interval that matters is 6 x 4,368 = 26,208 hours, and the question is what fraction of flux survives to there.

Read it off the manufacturer's projection report, which gives flux at any hour. If you have to interpolate between the two points you have, know what you are doing. A straight line from 100 percent at 0 hours to 90 percent at 54,000 hours puts 26,208 hours at 100 - 10 x (26,208 / 54,000) = 95.1 percent. The real curve is a decay, which lies below its own chord, so linear interpolation always reads high. The exponential decay that passes through the same two points gives 95.0 percent at 26,208 hours, so here the error is 0.1 percentage point and the shortcut is defensible.

State the condition on that, because it does not travel. The curvature error scales with how much total decay the interval spans. On an L70 claim, where the endpoint is a 30 percent loss instead of a 10 percent one, the same shortcut at the halfway hour gives 85.0 percent by straight line against 83.7 percent by exponential, an error of 1.3 percentage points, about ten times larger and still in the flattering direction. Use the chord on a shallow L90 span. Do not use it on an L70 span; go get the report.

So the depreciation term entering the design is 0.95 at 26,208 hours, at 1,050 mA and 85 C in-situ, and both of those conditions go in the design notes next to the 0.95. If the luminaire ends up somewhere hotter, the number moves and the note is what tells the next person why.

What actually ends a luminaire's life

The flux projection answers one question and the product has several ways to stop being acceptable, whichever comes first:

• The driver. In most assemblies this is the shortest-lived component in the box, and the driver card owns why and what to do about it.
• Optical materials. Lens and reflector yellowing or hazing that cleaning does not reverse, which shows up as a light loss factor term rather than as a lifetime term.
• Color shift. Output can remain well above L90 while chromaticity has moved far enough that a partial replacement no longer matches the units beside it. That is a service life limit in any space where appearance is the product.
• Enclosure integrity. Gaskets and seals in a wet or dusty location fail on their own schedule, and a breached enclosure ends the unit regardless of what its packages are doing.
• Availability. A luminaire whose replacement module or driver is no longer made has reached end of service life at the moment the last spare goes on a truck.

Checking your own figures

• The cap arithmetic is printed both ways. A: 10,000 h x 6 = 60,000 h cap against a 100,000 h claim, ratio 1.67, struck. B: 9,000 h x 6 = 54,000 h cap against a 54,000 h claim, ratio 1.00, stands.
• L70 and L90 are never compared. A's 100,000 h and B's 54,000 h are each run against their own cap and against nothing else.
• Operating hours recomputed, not assumed. 14 h/day x 6 days x 52 weeks = 4,368 h/yr; 6 years x 4,368 = 26,208 h; 54,000 / 4,368 = 12.4 years.
• The interpolation error is printed with its span. L90 span: 95.1 percent by chord against 95.0 percent by exponential, error 0.1 point. L70 span at the halfway hour: 85.0 against 83.7, error 1.3 points. Both run high, and the second is stated as the case where the shortcut is refused.
• Held-constant conditions carried onto the derived value. The 0.95 depreciation term is written as 0.95 at 1,050 mA and 85 C in-situ, the same pair the claim was stated at.
• Four gate items named, four rows in the table for each claim, plus the cap check. No item is dropped on either product.
• Nothing here derives a light loss factor. The 0.95 is one term handed to that calculation, not the calculation.

References

• IES LM-80, the LED package lumen maintenance test method, and IES TM-21, the projection method including the limit on reported projection duration, in the editions the manufacturer's report names, which bind through that report rather than as regulation
• IES LM-79, the electrical and photometric measurement of the finished luminaire, which is the source of its initial output and input power
• Manufacturer projection report, in-situ case temperature measurement and ambient derating curve for the exact ordering code, which own every value in this subject
• See related: How to Read a Luminaire Cut Sheet; What a Light Loss Factor Is Accounting For; What a Driver Is and Why It Fails First