Burn-in, burn-in: dc inferno
By Paul Breed, NetBurner Inc - December 15, 2009
The company I worked for had just finished a rush project to deliver a batch of ruggedized power converters to an integrator that was combining them with ruggedized computers and shipping them out for military use in the first Persian Gulf War, Operation Desert Storm. These power converters would take any standard voltage from 100 to 300V at a frequency of 40 to 400 Hz and a voltage of 28V dc and deliver clean, regulated, 117V, 60-Hz ac with 10 minutes of UPS (uninterruptible-power-supply)-like battery backup at 1500W. We had a commercial product that met this specification, but it was lacking in the ruggedness and environmental specifications that this customer required. We underwent an arduous effort to upgrade the mechanicals and thermal design to meet this customer’s requirement. We accomplished this task after about three months of 80-hour weeks.
The most difficult test to pass was the 24-hour test at full load and 85°C. We passed that test on the first attempt and tested 10% of our production run to ensure that we complied. Everything passed, and we shipped all 200 power converters to the integrator. I then took a well-deserved week off.
The customer took all 200 units, attached them to systems, and started a 48-hour burn-in at 70°C. At 36 hours, 100% of the units failed catastrophically. My employer quickly recalled me from my week off and put me on a plane to find out what happened. I found that the primary power bridge in every unit had blown out. There was almost nothing left in the power section that had not melted. I sheepishly returned home with no idea of what had gone wrong. We took our engineering unit and ran it at temperatures as high as 70°C; it, too, failed at 36 hours. There was nothing in the unit that did not reach thermal equilibrium in less than two hours, so how could the unit tell the difference between 24 hours and 36?
It took us two weeks to hunt down the culprit: The output-power stage in our system operated in a patented resonant circuit that used current feedback to keep the output transformer balanced and to fold back to protect from overloading. This current used a digital optocoupler to send pulses back from the power side to the control side. It was one of the parts we had upgraded; it offered 125°C operation and 3500V-ac-rms isolation. We were using it at only 300V dc.
We had assumed that if it was good for 3500V ac, it should be good for 300V dc. With an ac application, no charge migrates across the isolation barrier. Some charge might migrate back and forth, but it changes direction with each ac cycle. When you put a large dc bias across the device, the leakage migrates in only one direction. This leakage increases with temperature. Over time at elevated temperature, charge built up on the receiver side of the device until there was enough charge to affect its operation. It stopped receiving pulses, and the control circuit lost its feedback, driving the output transformer into saturation and emitting smoke.
It turns out that optocouplers for long-term dc use have a screen inside that shunts all of the drifting charge to ground. In short, alternating current and direct current are not interchangeable. It would have been nice if the optocoupler’s data sheet had specified “not for dc use.”