Inside a Waterproof Power Supply Line: 11 QC Gates

100% burn-in is the dividing line in this industry. Full-load burn-in costs roughly 3–5% of unit cost in electricity, rack space and labour. Factories that do it charge for it; factories that do not can quote 3–5% lower and still look identical on a datasheet. That 3–5% is precisely what determines whether your first 90 days are quiet or full of RMA emails.

Here is every gate a waterproof unit should pass, in order, and which three get quietly skipped.

The Line, End to End

The 11 Gates

Gate 1 — Incoming IQC (AQL 0.65) Components sampled on arrival. Critical: capacitors are checked for brand, temperature rating and date code. Anything beyond 18 months on an electrolytic reel is rejected. Receiving inspection catches about 0.8% of incoming lots.

Gate 2 — Solder paste inspection (SPI) Paste volume measured on every board before placement. Insufficient paste on a high-current joint is a latent field failure that no downstream test reliably catches.

Gate 3 — Post-placement AOI Automated optical inspection after the pick-and-place, before reflow. Catches missing and misoriented parts while they are still cheap to fix — roughly 40× cheaper than catching the same defect at final test.

Gate 4 — Post-reflow AOI + X-ray (sample) AOI on every board. X-ray抽样 on BGA and hidden-joint packages. Solder void rate should be under 25% on thermal pads.

Gate 5 — DIP and hand-assembly check Through-hole components, transformer mounting, terminal torque. Torque drivers set and calibrated; a loose terminal is a top-three field failure and a genuine fire risk at 16.7 A.

Gate 6 — Potting, in-process Vacuum level around -0.09 MPa, degassing ≥10 minutes. Compound batch and thermal conductivity logged (we use filled compound at 1.5 W/m·K or better on units above 150 W).

Gate 7 — Cure check Full specified cure time, undisturbed. Rushed cure is the defect you cannot see: the housing looks perfect and the seal fails in month 18.

Gate 8 — Hi-pot (dielectric withstand) 3,000 V AC for 1 minute, primary to secondary, per IEC 61347-1. Leakage current trip typically set at 10 mA. This is a safety gate, not a quality gate — a unit that fails it can kill someone.

Gate 9 — Burn-in: 100% at 45°C for 4 hours, under load The watershed. Logged failure rate should sit below 0.3%. Units that fail here are exactly the ones that would have failed in your customer’s first month.

Gate 10 — ATE final test A Chroma 8000 or equivalent measures output voltage, load regulation, ripple, efficiency, OVP/OCP trip points and — on dimmable models — the dimming curve. Results logged against the serial number. Ask for the printout for your batch; a factory running real ATE can produce it in minutes.

Gate 11 — Ingress verification IP65: spray test at 12.5 L/min, 30 kPa, 3 m, 3 min. IP67: 30 minutes at 1 m. Most factories run this on an AQL sample, not 100%, because full immersion testing adds real time and requires drying. Know which one you are buying — and if you need 100%, say so in the RFQ and expect to pay for it.

The Three Gates That Get Skipped

1. Burn-in (Gate 9). The 3–5% cost saving is the entire motive. Symptoms in the field: failures clustered in the first 30–90 days, then a long quiet period. If your RMA curve looks like that, you bought a supply chain without burn-in.

2. Degassing during potting (Gate 6). Here is why it matters and why it is invisible. Residual bubbles in the compound do not cause immediate failure. Under three years of thermal cycling, the bubbles expand and contract and eventually interconnect into micro-channels — and water follows them to the PCB. The unit looks perfect from outside. There is no leak path you can find. This is the single most common cause of “IP67 unit failed after 18 months, no visible water entry.”

Verify it by asking for the vacuum gauge reading and the degassing timer setting, not by asking whether they vacuum.

3. Hi-pot at 100% (Gate 8). Some factories run hi-pot on an AQL sample. That is a genuine safety exposure, not just a quality compromise. Ask what percentage and what the rejection protocol is.

Reading the Factory’s Own Numbers

Ask for four figures, and judge the factory by whether it has them:

A factory that cannot produce all four numbers is not managing its process, whatever the tour looked like.

What to Ask on a Walkthrough

Five questions, in this order:

1. “Show me the burn-in room and the log for the last three months.”
2. “What is the vacuum reading on the potting station right now?”
3. “What percentage of units get hi-pot, and at what voltage and duration?”
4. “Pull a serial number from finished stock and trace it to the capacitor lot.”
5. “What was your 12-month RMA rate, as a percentage?”

All five are reasonable. All five are answered in minutes by a factory doing the work. For the full 48-hour version, see How to Audit an LED Power Supply Factory; for a weighted scorecard, use 23 checks.

See our line live. We run unedited video walkthroughs through all 11 gates — including the burn-in log and the potting vacuum gauge. → Book a video audit · About the plant