I once designed a product for Avaya, a 1U x 19" rack with an open-framed OEM power supply. Yields were good at first. Later, HIPOT (high-potential) testing started to fail on random units at a 10 percent rate, which was much higher than I expected, since the supplier tested before shipping.
The product supplied power for Avaya's VOIP phones in a distribution box with 10 watts for each device inline on RJ45 ports which ran at T1 or 1.544Mbps rates. I was head of design at C-MAC in Winnipeg almost 10 years ago. I was responsible for test process and fault isolation.
I chose two suppliers for the 180W 1U open-frame power supply units (PSUs). We had a contract to supply Avaya with 10,000 units in the first year. Then the PSUs started failing in HIPOT testing. After many emails and phone calls to the supplier, it refused to believe its product could have such a high failure rate. I found out it had recently transferred the production from California to Mexico. I started to investigate the root cause of the failures.
Not only did the 1.5kV DC HIPOT test fail, but many units became nonfunctional after the test. Some passed initially, then failed later in functional testing. To figure out what could cause such failures, I decided to current limit the HIPOT tester with a high-voltage series resistor. Even though it was designed for a 2mA current limit, the capacitance in the tester was enough to create high-current impulse if low-impedance arcing occurred. Normally, it gives a slow-rising output to 1.5kV for one second and then shuts off.
That prevented PSU failures from catastrophic failure of arc discharge, but 1.5 mA overcurrent still occurred on 10 percent of the units.
Then I examined the root cause of the failures and discovered poor process methods in assembly with arcing. I highlighted all the areas of the PSU where insulation gaps had to increase and insufficient silicone or PU material had been applied.
The general manager of the company supplying the PSUs didn’t believe his design team in California and his production team in Mexico could not detect the same problems in their facilities. The supplier blamed us for breaking the product. I couldn’t believe it either, so I was determined to figure out why there was such a major discrepancy in yield and catastrophic failure rates.
I concluded that they were testing it wrong. When I asked how they performed the test with DC high-voltage input, it was OK. But when I asked how they terminated the outputs, they explained they did it with outputs loaded but floating, whereas we tested with DC output grounded open circuit.
Eureka! I demanded they test the same way we were testing, and they started to see the same problems. My reasoning was that the increased insulation on the floating secondary side was enough to make their test pass and our test fail. A gap increase of only 1mm or 2mm made it pass.
The PSUs were not being produced with enough insulating paste on exposed passive device leads. Details... details... tsk, tsk. They never did apologize, but I'm sure they learned their lesson. Also, the polarity of the HIPOT test might affect failure modes, since the passive parts were connected to active parts.
Normally, neutral and line (hot) are shorted together, and the high-voltage input is applied between them and the ground wire. But if the secondary is also floating, any arcing between primary (AC) and secondary (DC) could still pass if the DC insulation breakdown were enough. But a positive DC HIPOT arcing to ground would put a reverse spike on the DC output by means of positive pulse going from ground to DC-negative output.
The odds of a high-potential spike on the power line are high in Florida, where we tested, and they can be as much as the meter 6kV suppressor. That is usually between hot and neutral and less likely between hot and neutral and ground, unless there is a ground fault, since neutral is grounded at the main transformer. HIPOT is intended to be a leakage to ground test. The noise input caps are chosen for 60Hz at the rated voltage to ground to prevent an unsafe leakage current, but it is tested with an impulse for line ingress on an ungrounded unit. If the DC output is grounded externally in an application, for shielding or whatever reason, it must be tested in this manner with DC-negative output grounded for HIPOT.
Don't forget to current limit your probes just above the test failure threshold to prevent capacitor discharge on short. How many PSU HIPOT escapes have you seen?
This entry was submitted by Anthony Stewart and edited by Rob Spiegel.
Anthony started life as a professional engineer after his EE graduation in 1975 at the University of Manitoba in Winnipeg. He contributed 30 years in R&D industries such as aerospace, telemetry, SCADA, telecom, disk drive, nuclear instrumentation, and contract manufacturing.
Tell us your experience in solving a knotty engineering problem. Send to Rob Spiegel for Sherlock Ohms.