Comparing MV Generator Power Plant Architectures for Data Centers

April 2, 2018 Patrick Donovan

Schneider Electric Reliability Engineers Publish Study – Finds Simple, Redundant Architectures vs. Redundant & Concurrently Maintainable, More Reliable

Cloud and colocation data centers with power capacities of 5 MW and greater tend to connect to the grid at medium voltage (MV) and even at the high voltage (HV) grid transmission level. In a 2N configuration, the generator power plant represents about 30% of the total CAPEX investment in the vast electrical distribution network that make up these massive compute facilities. Not only are they expensive, but they take up a significant amount of real estate, particularly when you include the fuel storage tanks and all the related switchgear. Furthermore, in our experience, generator systems are one of the leading causes of data center outages. So, there is much to be gained by optimizing and simplifying the design of the generator power plant architecture. Although not part of the study, this simplification, it is reasonable to assume, would also simplify operations, decrease maintenance, and perhaps speed up the initial deployment since switchgear and genset engines tend to be long lead time items.

An Experts’ Analysis on MV Generator Power Plant Architecture

technicians discussing over laptop in colocation facility

With so much at stake, a global team of Schneider Electric’s best power distribution and reliability engineers came together to compare and evaluate various MV generator power plant architectures in the context of large data centers and critical facilities. The team’s expertise spanned high, medium, and low voltage domains, as well as protection/control systems and reliability analysis. This cross-domain expertise, I believe, is a key differentiator for Schneider Electric. There are very few people in the world with intimate experience and knowledge of connecting data centers to the HV transmission grid, for example. The team spent a few months evaluating and comparing the cost and reliability of various power architectures. Some of the study’s results fit with common sense while other results were quite surprising.

Since this was done in the context of large data centers, the architectures were developed and grouped by their level of redundancy based loosely on the Uptime Institute’s Tier standard. This led to five different generator power plant architectures being analyzed in the study: A Tier 2, a “double-fed” Tier 3, an “open loop” Tier 3, a “double-fed automatic” Tier 4, and a “closed loop” Tier 4 type architecture. The principal way to optimize and simplify a traditional 2N genset power plant architecture is to move to a simpler N+1 (or N+2) design. This optimization is discussed in detail. But, given the critical function of the generators during a utility outage, who is going to be willing to do that? Well, for starters, with these large sites, there are usually redundant MV distribution feeders or the site is connected at the HV transmission level…the grid at HV is very reliable. This raises the question as to whether 2N is worth it. Is it worth the extra CAPEX and OPEX cost? Is it worth the complexity? Reliability of the grid aside, this study surprisingly showed that a Tier 2 type generator power plant architecture was not only less expensive from a CAPEX perspective, but it was also more reliable with more inherent availability than the 2N fed Tier 3 generator power plant architectures.

Looking Objectively Finds the True Answer

Now, how could this be? The team used a failure mode and effect analysis (FMEA) combined with a fault tree analysis for multiple contingency analysis. It considered equipment failure rates, all equipment failure modes, all time elements (failure detection times, spare part delivery, etc.), as well as all scheduled maintenance operations and periodic testing that would be typical. A 2N data center with a Tier 2-type generator power plant turned out to have a very high level of reliability and (we think) should be considered as sufficient, particularly if the other parts of the overall system are well considered in the design (i.e., the LV equipment, mechanical system design, other common mode failures, and human factors as governed by an Operations & Maintenance program). The Tier 3-based generator power plant architectures did not provide increased reliability because they involve more distribution equipment (more equipment meaning more potential failures) and they do not avoid the issue of an undetected failure on the generator system (i.e., something that shows up only when the generator system is energized) potentially preventing it from serving critical loads in standby mode. These reasons are explained in more detail, of course, in the study.

The analysis further shows that the best way to improve reliability beyond that of the fully redundant architectures with a Tier 2 generator power plant is to move to a fully redundant architecture with a Tier 4 closed loop-type generator power plant architecture with differential protection. The whole Tier 2 vs. Tier 3 thing sounds very counter-intuitive, I know…it certainly did for me. But, I encourage you to read the study in full for yourself. The white paper is very detailed and open in terms of its methodology and assumptions. Thoughts and questions are welcomed; the authors of the study are happy to discuss it with you in context of your next generator power plant project. Comment below!

The post Comparing MV Generator Power Plant Architectures for Data Centers appeared first on Schneider Electric Blog.

Previous Article
(Webinar) Attracting Investors: what colocation providers need to know
(Webinar) Attracting Investors: what colocation providers need to know

With $18 billion in M&A activity in 2017, the colocation industry is riding the bubble of rapid growth. The...

Next Article
Investment Banker Paints Rosy Picture of Colocation Market in 2018 
Investment Banker Paints Rosy Picture of Colocation Market in 2018 

We’ve all been hearing about how robust the market for data center space is, but a presentation by an inves...