July 12, 2012

Over the last 10 or so years, the entire data center industry has become much more aware of the problems with switching medium-voltage (MV) distribution transformers with vacuum breakers, especially after some very high-profile failures of dry-type transformers in data centers around Manhattan and New Jersey, although there have been dozens of other similar failures around the country.

A number of Institute of Electrical and Electronics Engineers (IEEE) committees have performed some solid research, papers have been published, and new IEEE guidelines proposed and issued in IEEE C57.142. There is finally a good understanding of the full nature of the problems, and the steps then can be taken to mitigate. There have been some excellent papers by David Shipp of Eaton Corporation and a number of presentations made by him to various IEEE conferences, as well as some excellent papers and presentations by Phil Hopkinson, owner of HVOLT.

I’ve closely followed David’s and Phil’s works (and IEEE’s work) on this topic since their very beginning, because of my 30+ years of keen interest in this problem. The recommended solution always seems to be the installation of RC snubbers connected directly to, or very close by, the transformer primary windings.

Can snubbers be effective, and produce the desired result? Emphatically, “yes.”

But I think this is the wrong question to be asking, and it begs the more fundamental question: “Why install a data center substation transformer in the first place that requires a snubber in order to survive for a normal lifetime?”

Even the most ardent proponents of use of snubbers generally add a post-script of advice, saying something like “RC snubbers - while useful and often necessary - will likely become points of failure at some time, when the capacitors eventually fail. A systems study should be performed to determine whether a snubber is actually required at any particular transformer. If a snubber is not required, then don’t install one - what’s not there can’t fail.”

I’ve read a number of such system studies performed by study engineers for various projects. They generally end with a conclusion and recommendation that goes something like this:  

Transformers T15 and T16 do not require use of snubbers. Transformers T1 and T2 do require snubbers, because of the very short lengths of the primary feeder cables. Transformers T3 through T14 are ‘borderline’. Prudence suggests that it might be wise to install snubbers on these twelve transformers.

After reading a recommendation like this one from the study engineer, most data center owners and their consulting engineers wouldn’t be able to sleep at night until after they had made a decision like “Why don’t we just install snubbers on every transformer?” (And, who could argue with the logic of that conclusion?)