Brand Logo
Control Notes

How to Test an LED Driver: A $4,300 Lesson From a Failed Chandelier Batch

2026-10-27 · Adrian Flores

I've been handling LED component and fixture orders for going on nine years. In that stretch I've personally made—and documented—seven significant mistakes, totaling roughly $12,000 in wasted budget. I now maintain our team's qualification checklist, mostly to stop other people from making my mistakes. This is the story of the one that forced the rewrite.

Here's how I used to test an LED driver: wire it to the chip, dial the bench supply to the nominal forward voltage, and watch the light come on. Five minutes. Pass or fail. Move on.

If that sounds like your process, I get it. I ran it for almost four years. And most of the time it works—the fixture lights up, the current lands in range, nobody complains. Until the one order where it doesn't, and the driver that "passed" takes a chandelier batch down with it.

That was me in September 2022. A hanging chandelier order, 48 fixtures for a hotel renovation, dimmable to 5%, CRI 90+. The chips were fine. The chips were always fine. The test was the problem.

The "Five-Minute Test" Problem

The issue isn't that the bench test is wrong. It's that it's testing the easiest version of the driver's job. Bench supply, room temperature, steady load, no enclosure, no dimmer. The other 95% of what that driver actually has to survive is invisible to the test.

Three specific things I got wrong, in roughly the order I learned them.

The datasheet gives you a range, not a number

Every LED datasheet lists a typical forward voltage. The Nichia 519A we were buying for the chandelier work, for example—the datasheet gives you a typical Vf and a distribution window. What I didn't account for is that the driver is designed around a window, and each reel of chips sits somewhere slightly different inside it.

So I'd test one sample from a batch, see a clean result, and sign off on the whole lot. Then production would pull a different reel from the same order and get a Vf sitting a few percent higher. The driver was technically in spec. It was also right at the edge of the operating range I'd never actually checked, and at low dim levels in a warm chandelier housing, that edge turned into visible flicker.

From the outside, the driver "worked." The reality was that it was working at the margin, and the margin was the one thing I never measured.

The driver and the chip aren't separate purchases

This one cost me the most money. I used to treat the driver and the LED as two line items on the same PO, each tested on its own. That's fine for cheap fixed-output stuff. It is a disaster for anything where dimming, thermal regulation, or high CRI actually matters to the customer.

What most people don't realize is that dimming behavior is a property of the pair, not the chip and not the driver. The driver's dimming curve, the chip's thermal coefficient, the enclosure's ability to shed heat—change any one of those and the assembly that "passed" on the bench can start flickering in the field. I didn't learn that from a textbook. I learned it from a hotel's commissioning walk-through.

Specialty applications don't inherit general lighting test protocols

The latest Nichia LED news cycle tends to be about headlights, UV, and laser applications, and for good reason—that's where the engineering attention is. But here's the trap: a chip that wins on a general illumination datasheet can behave like a completely different component once it's in a narrow-beam, high-ambient fixture.

The stage lighting crews working on spotlight LA projects know this better than most general lighting people. A spotlight at full tilt inside a hot truss is a different thermal world than a hanging chandelier. Same chip, totally different driver tuning, completely different failure modes.

Same thing on the automotive side. Nichia's laser headlight work and their UV LED line get a lot of press, and the test requirements for those programs are not the general-illumination playbook with a different label on it. If you assume the flow carries over just because the vendor is the same, you're going to have a bad quarter. Those programs need their own safety envelope, their own thermal cycling, their own acceptance criteria.

What the Mistake Actually Cost

The chandelier order shipped in what looked like good shape. On the wall in our shop, everything was clean. Inside the actual fixture housing, with 48 chips pushing heat into a small canopy, it wasn't.

About 15% of the batch showed visible flicker at low dim. Three units went into thermal shutdown during a six-hour continuous burn-in. The hotel's commissioning was two weeks out, and the customer's lighting designer had already signed off on warm-white renderings that our first article wasn't matching in the real housing.

We reworked the driver spec, re-qualified against the new pairing, and shipped replacements. By the time it was done, the number landed around $4,300 in replacement drivers and rework, plus three weeks of schedule slip and an uncomfortable call with the GC. The chip was never the problem. I had just never tested the system as a system.

I'd also skipped the burn-in step because I thought "we've never had a problem before." That was the one time it mattered. A lesson learned the hard way.

My Current LED Driver Test Checklist

Here's what I actually run now. It's not elegant. It's just longer than five minutes.

  1. Test three samples per reel, not one. At least one at cold start, one at thermal steady-state, and one after a few hours of running in the enclosure.
  2. Measure flicker at 100%, 50%, 10%, and minimum dim. Use a meter that reports TLA and SVM, not your eyes. Per NEMA 77-2017, the visible flicker thresholds are lower than most people assume.
  3. Run a four-hour burn-in inside the actual fixture housing before approval. Not on the bench. In the housing. With the driver mounted the way it'll ship.
  4. Check the Vf distribution across the reel and confirm the driver's operating window covers the high and low end, not just the typical.
  5. If CRI is part of the spec, measure it in the final assembly. CIE 13.3-1995 is the reference method, and real-fixture measurements can differ meaningfully from the chip-level number on the datasheet.
  6. Document the matched driver/chip combination by part number. Treat a chip substitution as a new test event, not as a drop-in replacement.

The long version lives on a one-page form that takes about 90 minutes per batch. Annoying? Yes. Also the reason we haven't had a driver-related field failure since Q2 2023, which is a streak I'd like to keep.

The Cost Math Nobody Runs

Look, I'm not going to pretend the five-minute test is always wrong. For prototypes and non-critical dimming, it's fine. The part I want to press on is what happens next time somebody on your team compares two drivers by unit price.

The $2.40 driver versus the $3.10 driver. Obviously the cheaper one wins the spreadsheet, right? The $2.40 driver cost me $4,300 in rework. The $3.10 driver would have cost $336 more up front and exactly zero in rework on this order. That's the whole argument in one line, and it only holds if you're counting the returns, the schedule, and the customer confidence—not just the line item.

I didn't believe the checklist people until I'd skipped the burn-in and paid for it. The odds caught up with me. They always do.

Adrian Flores

Adrian Flores

Adrian Flores is an architectural and decorative lighting analyst specializing in pendant lights, chandeliers, wall lights, floor lamps, table lamps, and track lighting. He applies IEC 60598-1 luminaire requirements and photometric data to examine beam angle, shielding, surface temperature, mounting geometry, dimming compatibility, and glare. He writes practical guides for designers and buyers balancing visual character with usable illumination, safe installation, maintenance access, and total project cost.