Fire Inside a Chiller Variable Frequency Drive, Texas (2019)

In 2019 we were asked to find the cause of a fire inside the variable frequency drive of a large water chiller at an institutional central plant in Texas. The fire had happened two years earlier. Client details are confidential.

A drive that was gone

The drive that burned had been replaced within two weeks, and it was no longer available to examine. What remained was the record around it: photographs and video taken while it was still burning, the construction punch lists, the protective device documents, and the event history the drive controller had stored before it failed. Then a second source arrived. While our review was under way, the identical replacement drive failed as well, and we examined it on site.

Reading the event log

The controller saved a record whenever something out of the ordinary happened, with voltages, currents, run hours, start counts and several internal temperatures. One record showed the voltage on a single input phase collapsing, which matched a cable that had shorted to its conduit fitting above the drive. By then the two internal air temperature sensors were already reading far above anything logged in the weeks before. The fire was burning before that cable failed, so the cable failure was a result of the fire, not its start.

The same log counted one more start and about 11 hours of run time in the 16 and a half hours before the failure. Statements that the chiller had not been running could not be reconciled with that record. Tracing the drive schematic also showed that its input capacitor bank is energized whenever the drive has input power and its main breaker is closed, whether or not the motor is turning.

What the replacement drive showed

In the second drive the damage was in the corresponding capacitor bank. Bolted connections in electrical equipment are usually marked with a paint line after they are torqued, so that any later movement shows. On this drive several marks were broken, on the terminals of the failed capacitor and the one below it, and on main connections below the drive breaker. A poor connection at a capacitor can heat over time until the capacitor fails. The video of the first fire fit the same sequence: the capacitors at the rear were burning while the cables and the electrical tape in the drop area below the conduits were still intact.

Breaker settings and arc flash

Large circuit breakers ship without settings, and someone has to set them from a coordination study. Here the study had not been completed when the plant went into service. We built a model in ETAP^® to show what that meant. With the feeder breaker left at its maximum settings, a fault on the line side of the drive breaker would take about 17 seconds to clear, with an incident energy of about 712 cal/cm². With that breaker set from a study, the same fault would clear in about 0.05 seconds at about 2.5 cal/cm². For the ground fault that actually occurred, the settings made no difference: in the model the feeder and the building main both trip in under 50 milliseconds, which fits the loss of power to the whole plant.

Ruling out surges

The surge protective device for the plant showed a healthy status and a running count of surge events, and lightning records showed no strikes within 5 miles in the two days around the fire. A second chiller on the same service had come through every one of those events, and no other sensitive equipment in the building had failed. Surge damage was an unlikely explanation.

Our opinion

We concluded that the most probable cause of the original fire was a failure in the capacitor bank inside the drive, which then damaged the incoming cables until one shorted to its fitting. The exact failure mode inside the original drive could not be determined without the drive itself. Losses like this one are the core of our electrical failure analysis work, and the breaker question is what our protection coordination studies settle before equipment goes into service.