Why an LED Fails Differently From What It Replaced

Why this matters

A lamp told you when it was finished. It went dark, somebody called, you carried the right lamp on the truck, and the whole maintenance program rested on that one self-announcing event. A solid-state luminaire mostly does not do that. It has five ways of becoming unacceptable, and of those five, one arrives at the service desk as an accurate lighting ticket, one arrives as a ticket that gets closed wrong, two arrive described as something other than lighting, and one never arrives at all. That is why a floor can sit ten percent under its own design criterion with a ticket log that never once mentions the light level.

What the old event was, and what a program could hang on it

A lamp had a rated life that was a failure statistic: the hour at which half a tested population had gone dark. That single property carried an entire maintenance model. You could group relamp on a calendar, because the event predicted was a population median and the calendar was a defensible proxy for it. You could spot relamp on complaint, because the failure was binary and visible from the doorway. You could stock by lamp type and wattage, because the replaceable unit was a consumable any tech could swap from a ladder. The lamp-plus-ballast fork was two parts and you diagnosed it by substitution.

None of that survived the change of source, because the criterion changed from a time-to-failure statistic to a flux threshold. The sibling card on lumen depreciation and lifetime ratings owns what an L-number is and what the projection behind it may claim. Take from it only the consequence: the hour count on the sheet predicts a fade, and a fade generates no phone call.

The five presentations, and where each one lands

Two are worth sitting with. The partial outage is the one your ticket system actively mishandles: a tech looks up, sees light, closes the call, and the record now holds positive evidence that the fixture is fine. The flicker case leaves the lighting category entirely, so it is invisible to anyone reading a lighting log. The flicker card owns the measurement and the mechanism; what belongs here is that it does not arrive labelled.

The replaceable unit moved, and it moved at purchase

In a lamp system the field-replaceable unit was chosen by the maintenance department, in the aisle, off a shelf. In a solid-state luminaire it is whatever the manufacturer decided, and that decision was locked in when somebody signed a submittal: driver and board both replaceable (a failure is a component job), driver replaceable with the board integral (a board failure takes the whole luminaire), or a sealed assembly where every failure does. Those are three different labor-hour profiles per event and three different spares plans.

Spares are therefore held by ordering code rather than by lamp type, and an ordering code can be discontinued. A luminaire whose driver and board are both out of production has reached the end of its service life the day the last spare leaves the shelf, whatever its projection report says.

What comes off the list, and what has to go on it

Strike these three. Group relamp on a calendar keyed to rated life, because the event that calendar predicted does not occur, so the calendar now replaces working hardware on a schedule derived from a statistic that no longer exists. Complaint as the primary detector, which reliably catches one of the five presentations and unreliably catches a second. Lamp inventory by type and wattage, replaced by drivers and boards held by ordering code with a live check on whether the code is still made.

Add these three:

• A measured illuminance survey on a fixed grid at a fixed interval, same points, same plane and height, same meter, recorded with all three. The how-to on taking a repeatable light level reading owns the method; a survey is now the only detector for the presentation that never generates a call.
• Cleaning as a scheduled task with a date on it, not something done when a fixture looks dirty. The light loss factor card owns how dirt enters the design; a design that assumed a cleaning interval and did not get one is running a term it never budgeted.
• A rule that closes a partial-outage ticket only on a count of lit boards or modules, not on "fixture is on."

Worked example: the survey four years after a retrofit

An open-plan floor, 24 luminaires, 24 measurement points on a fixed grid. Every illuminance figure below is horizontal at 30 in above finished floor, taken as-found on one meter stated at plus or minus 5 percent of reading, and none of them has any light loss factor applied to it.

What the design said. Maintained 30 fc horizontal at 30 in AFF at the end of a 6-year cleaning and replacement interval, on a light loss factor of 0.73 read from the design notes. A maintained figure already contains its loss factors, so the initial condition predicted was 30 / 0.73 = 41.1 fc.

What the ticket log said. Eight tickets in four years. Six were fixture-dark calls closed by a driver replacement, 6 of 8, 75 percent of all lighting work on this floor. Two were "the new one does not match" after those replacements. Zero mentioned light level.

What the survey found. Grid average 27 fc. Grid minimum 14 fc at 2 of the 24 points, both directly under one luminaire. Grid maximum 36 fc.

Correction line, and the trap it avoids. The 27 fc is an as-found measurement and it already contains every loss the space has actually suffered, so it goes straight against the maintained criterion, 30 fc, with no factor applied to either. Note the one asymmetry: the reading is at year 4 of a 6-year interval and the criterion is written for year 6, so it has two years of loss still to take. That makes the comparison conservative against the installation rather than against you, which is the safe direction to be wrong in. Dividing 27 by the 0.73 to get 37 fc and calling that "initial equivalent, still fine" applies the light loss factor a second time to a number that already contains it. It is the commonest arithmetic error in this subject, and it runs in the flattering direction, which is why it survives.

Correction line, meter tolerance. Plus or minus 5 percent of reading is a percent-of-reading bound on a single absolute measurement, not a spread to be split. The band on 27 fc is 25.7 to 28.4 fc, entirely below 30 fc, so the deficit is real at the instrument's own worst case.

Correction line, depreciation. The projection report gives flux maintenance of 0.97 at the 11,440 operating hours this floor has run (11 h/day x 5 days x 52 weeks = 2,860 h/yr, x 4 years), so depreciation accounts for 3 percentage points of loss. It cannot explain a floor sitting 10 percent under a criterion computed to absorb six years of it.

What was taking it. Two of the 24 luminaires had one of their two boards dark, 2 boards of 48, 4 percent of installed flux, and both sat over the two 14 fc grid points. Both had been ticketed earlier and closed as operating. No lens had been cleaned in four years against a design that assumed a 6-year cleaning interval, and the residual gap after depreciation and the dark boards is where that sits.

What the log could and could not have told you. One condition was reported and closed wrong: the 2 partial outages. One was never reported at all: the floor-wide deficit, 27 fc against 30 fc, 10 percent under the criterion with two years of the interval still to run. No complaint-driven program finds the second, on any budget, because there is no complaint to drive it.

The uniformity question belongs to the card on why uniformity matters more than average level; note only that the two worst points on this grid sit under a fixture the ticket system had recorded as working.

What changes this

• A 24/7 space. Past roughly 8,000 operating hours a year the depreciation term stops being a rounding error and the survey interval shortens with it.
• A vertical criterion. In a warehouse aisle or at a parts counter the deliverable is vertical illuminance at a stated height, and a horizontal grid at 30 in AFF will not detect its loss at all.

Checking your own figures

• Maintained criterion against as-found reading, same basis. 27 fc as-found against 30 fc maintained. No light loss factor applied to the 27; the 0.73 is used once, to derive the 41.1 fc initial prediction, and never again.
• Tolerance stated as a bound with its basis. Plus or minus 5 percent of reading on 27 fc gives 25.7 to 28.4 fc, and 28.4 is still under 30.
• Deficit arithmetic printed. 30 - 27 = 3 fc, 3 / 30 = 10 percent under the maintained criterion, at year 4 of a 6-year interval.
• Depreciation printed against the deficit it is asked to explain. 0.97 at 11,440 h is 3 points of loss against a 10 percent deficit, so it is not the cause.
• Ticket shares recomputed. 6 driver calls of 8 tickets = 75 percent; 2 partial outages of 24 luminaires; 2 dark boards of 48 = 4 percent of installed flux.
• Five presentations, five table rows, routing counted. 1 accurate lighting ticket (driver), 1 mis-closed one (partial outage), 2 filed under another category (flicker, color shift), 1 never reported (fade). 1 + 1 + 2 + 1 = 5.
• No lifetime projection is derived here. The 0.97 is a value read from the report the depreciation card explains.

Hazards this work creates

• Do not open an energized luminaire at height to count lit boards. Count from the floor, or work with the branch circuit de-energized, locked and tagged under 29 CFR 1910.333(b)(2) for electrical work on utilization equipment, proved 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, not on its own).
• Reaching any ceiling luminaire is work at height, on a properly set ladder or lift under 29 CFR 1910 Subpart D for general industry or 29 CFR 1926 Subpart X for construction, never off a desk and never using the ceiling grid as a handhold.
• Walking a fixed grid in a warehouse aisle puts a person on foot in a powered-equipment lane. Clear the aisle with the operator or survey out of shift, in the high-visibility apparel the site requires.
• Cleaning a lens puts liquid near a fitting. De-energize and lock out the circuit first, and use the agent the manufacturer's instructions name, because a solvent that crazes a lens turns a recoverable loss into a permanent one.

References

• 29 CFR 1910.333(b)(2), electrical work on utilization equipment conductors, which 29 CFR 1910.147 expressly excludes from its scope
• 29 CFR 1910 Subpart D for general industry ladders and walking-working surfaces; 29 CFR 1926 Subpart X for construction
• NFPA 70E-2021, 120.5, as adopted through an employer electrical safety program or by contract
• Manufacturer projection report and ordering-code spares availability, which own every product-specific value in this subject
• See related: What Lumen Depreciation and Lifetime Ratings Mean; What a Driver Is and Why It Fails First; What a Light Loss Factor Is Accounting For; Why Uniformity Matters More Than Average Level; How to Take a Light Level Reading Somebody Else Can Repeat