How to Audit an LED Power Supply Factory in 48 Hours

Factory size is the least useful thing you will be shown. A 12,000 m² plant with eight SMT lines tells you about capacity, not about quality. The two areas that actually predict whether your first 5,000 units arrive good are the burn-in room and the potting station — and neither one makes a good photo for the sales deck.

Here is how to spend 48 hours, and the 23 things to check.

Day Structure

If the factory will not let you into the burn-in room unescorted, or says “it’s not running today,” treat that as the answer.

Section 1 — Incoming and SMT (6 checks)

1. Capacitor brand on the shelf. Walk to the component store and read the reels. 105°C / 10,000 h rated, named brand. Unbranded or 85°C parts: stop here.
2. Reel date codes. Anything over 18 months old on an electrolytic reel has been sitting and has degraded.
3. Moisture-sensitive device (MSD) storage. ICs and MOSFETs should be in a dry cabinet or sealed with desiccant and a humidity indicator card. Open bins on a shelf = latent field failures.
4. SMT line age and placement accuracy. Ask for the model and the stated accuracy; anything still placing 0402 reliably is fine for power supplies, which mostly use 0805 and larger.
5. Solder paste storage. Refrigerated at 5–10°C, with a log of removal and thaw times. Paste left out overnight is a real and common problem.
6. AOI after reflow. Is there automated optical inspection, and does it gate the line or just log data?

Section 2 — Assembly and Potting (6 checks)

1. Torque drivers on terminal screws. Loose terminals are a top-three field failure. Are the drivers set to a specified torque and calibrated?
2. Potting vacuum level. Should be around -0.09 MPa. Lower, and you get residual bubbles.
3. Degassing time. Under about 10 minutes at that vacuum, bubbles remain. Bubbles become micro-channels under thermal cycling, and water arrives 18 months later — with no visible defect on the outside.
4. Potting compound batch and type. Ask for the datasheet and the thermal conductivity. Below 1.0 W/m·K, the unit will run hot.
5. Curing rack discipline. Units should cure undisturbed for the full specified time. A rack of units being moved at half cure is a real thing and it wrecks the seal.
6. Gland torque check. If cables enter through glands, is there a pull test?

Section 3 — Burn-In and Test (7 checks) — The Decisive Section

1. Burn-in coverage: 100% or sampling? This is the watershed. Full-load burn-in costs roughly 3–5% of unit cost in electricity, rack space and labour. Factories that skip it save that and ship you the infant-mortality curve.
2. Burn-in duration and temperature. Look for ≥4 hours at 45°C under load. Two hours at room temperature is theatre.
3. Burn-in failure rate log. Should be below 0.3%. Ask to see the last three months. A factory that does not log this number does not measure it.
4. Load type during burn-in. Resistive dummy loads are fine; electronic loads that cycle are better. Neither is acceptable if the load is below rated output.
5. Hi-pot (dielectric withstand) testing. Typically 3,000 V AC for 1 minute between primary and secondary, per IEC 61347. Ask what voltage and what duration, and whether it is 100% or AQL.
6. Final ATE test coverage. A Chroma 8000 or equivalent should measure output voltage, current, ripple, efficiency and OVP/OCP trip points. Ask to see a printout with serial numbers.
7. Immersion or spray testing. For IP67, is it 100% or batch? A 30-minute immersion on every unit is expensive; most factories do AQL. Know which one you are buying.

Section 4 — Systems and Traceability (4 checks)

1. Serial number traceability. Give them a random serial from finished stock and ask which capacitor reel and which SMT shift it came from. If they can answer within an hour, the system works.
2. Certificate validity. Get the certificate number and check it on the issuing body’s site — not on the PDF the salesperson emailed. Expired and scope-limited certificates are common.
3. RMA process and rate. Ask for the 12-month RMA rate as a percentage. A factory that has never measured it is telling you something.
4. Engineering change control. Ask to see the last ECN. Was the customer notified? Was it re-tested?

The Three Questions That Expose a Trading Company

1. “Show me the SMT line running my product, right now.” A trading company will take you to a “partner factory” — and usually cannot name which line your order would run on.
2. “What is the burn-in failure rate for the last three months?” Traders do not have this number. Factories that do burn-in always do.
3. “Open a unit from finished stock and tell me the capacitor brand.” A factory knows its BOM instantly. A trader will need to ask someone.

None of these is hostile. All three are normal questions from a serious buyer, and a real factory answers them in under five minutes.

What to Do With What You See

Score it while it is fresh. Write down three numbers before you leave the building: burn-in coverage %, burn-in failure rate, capacitor brand and rating. Those three predict your first-year experience better than everything else combined. For a weighted scoring sheet, use our 23-check supplier scorecard.

Cannot travel? Do it on video. We run live factory walkthroughs covering all 23 points — burn-in room included, unedited, at a time you choose. → Book a video audit · See the plant