Data center electrical commissioning and power QA field guide

Data center electrical commissioning, and what power QA actually proves

Data center electrical commissioning is the structured, witnessed process that proves the power chain will carry critical load before any IT load is on it. Critical load is the IT load, the servers and storage the building exists to keep alive. Everything upstream of it, the utility service, the transformers, the switchgear, the UPS, the generators, the batteries and BESS, the busway, the PDUs and RPPs, and the grounding, is there to deliver clean power to that load without a single unplanned interruption.

Power QA is the electrical half of that work. It is not a final inspection and it is not a punch walk. It is a series of tests and demonstrations, each one witnessed and signed, that builds confidence stage by stage: the gear is the gear that was specified, it was installed and terminated correctly, it powers up, it does what the sequence of operations says, and the whole plant rides through a utility failure at design load without dropping the critical bus.

The thing that separates a commissioned plant from an installed one is proof under realistic conditions. An installer can hand over gear that looks finished and passes a visual. Commissioning is what catches the breaker with factory default settings still loaded, the generator that starts but cannot accept block load, the UPS that transfers to bypass but never transfers back, and the transfer scheme that works on paper and stalls when you actually pull the utility. Most data center outages traced back to construction are not bad equipment. They are commissioning that nobody finished.

The commissioning levels: L1 through L5

Most data center programs organize commissioning into levels, commonly numbered Level 1 through Level 5, that move from the factory floor to the fully integrated plant. The levels build on each other. You do not start functional testing until installation checks are signed off, and you do not run the integrated test until every system has passed its own functional test. Skipping forward is how deficiencies hide until the worst possible moment.

The exact boundary between levels shifts from one commissioning program to the next. Some split static installation checks and pre-functional energization into separate levels, some fold them together, and some add a Level 0 for design review and commissioning planning and a Level 6 for post-occupancy and seasonal testing. Do not assume the numbers mean the same thing across two jobs. The project commissioning plan is the authority for what each level includes and who witnesses it.

The keystone is Level 5, the integrated systems test. Everything before it proves a piece. Level 5 proves the pieces work together under a simulated failure, and it is the level that schedule pressure attacks first because it needs the whole plant, full load banks, and a block of uninterrupted time.

Who owns commissioning: CxA, EOR, contractor, and owner

The commissioning authority, the CxA, runs the process. On a data center the CxA is usually a third party hired by the owner, independent of the installing contractor, because the whole point is a check that is not marking its own homework. The CxA writes or owns the commissioning plan, builds the test scripts, schedules and witnesses the tests, keeps the issues log, and assembles the turnover package. The CxA does not perform the electrical tests with their own hands as a rule. They witness and accept them.

The acceptance testing itself, the NETA work, is performed by a separate independent test agency or qualified technicians. NETA accreditation exists precisely so the testing party is independent of the installer. The engineer of record, the EOR, owns the design intent and the coordination and arc-flash studies, and the EOR or the equipment manufacturer answers when a functional test does not match the sequence of operations. The installing electrical contractor builds it, supports energization, and corrects deficiencies.

The owner sits at the top and inherits everything at the end. The commissioning record becomes the owner's permanent baseline, the document a future maintenance contractor or the operations team measures against for the life of the building. When the roles blur, when the installer witnesses its own work or the CxA reports to the GC instead of the owner, the independence that makes commissioning worth anything quietly disappears.

Two documents anchor the whole effort, and the commissioning agent traces everything back to them. The Owner's Project Requirements, the OPR, is the owner's plain-language statement of what the facility must do, including the redundancy target, the temperature limits, and the ride-through expectation. The Basis of Design, the BOD, is the engineer's record of how the design meets that OPR. Under ASHRAE Guideline 0 the verification chain runs OPR to BOD to design to installation to test, so a deficiency is judged not against opinion but against whether the built plant still satisfies the OPR.

The power chain and what power QA checks at each stage

Power QA follows the power, from the utility service down to the receptacle at the rack. Each component in the chain has its own set of checks, and each one has to pass before the load downstream of it can be trusted. The table below maps the chain a power QA scope walks through. The depth at each one depends on the voltage class, the topology, and the tier the facility is chasing.

Two patterns repeat all the way down. First, every connection that carries current gets a resistance check, because a loose or high-resistance joint is where heat and failure concentrate. Second, every active component gets a functional demonstration, not just a static test, because a breaker that megs fine can still fail to trip on the settings it was given, and a generator that starts can still fail to take load.

Acceptance testing under NETA ATS

The electrical acceptance tests on a data center are performed to the NETA Acceptance Testing Specifications, ANSI/NETA ATS, the current edition being ATS-2023, so confirm the latest against NETA. NETA ATS is the field reference for what tests to run on each class of electrical power equipment before initial energization and final acceptance, and what the results should look like. It is the standard the independent test agency works to, and the test reports it produces become the de-energized baseline for the life of the gear.

The acceptance test battery is not a single test. It is a set, and which subset applies depends on the equipment. The table lists the ones a power QA scope leans on most. The specific test voltages, durations, and pass criteria come from the NETA tables and the manufacturer's published data, so confirm the value against the standard and the nameplate rather than carrying a number in your head. NETA ATS gives the framework; the equipment listing gives the limit.

The reason these matter beyond a checkbox is that they are the baseline. The first time anyone megs a cable or ductors a bus joint is during acceptance, when the gear is new and clean. Every maintenance test for the next twenty years, run to NFPA 70B and the IEEE 3007 series for maintenance, compares back to that acceptance number to spot the trend before it becomes a failure. Skip the acceptance baseline and you have taken away the reference that future maintenance depends on.

What is the integrated systems test?

The integrated systems test, the IST, is the full-plant test that simulates a real utility failure and proves every system reacts together to keep the critical load up. It is also called the pull-the-plant test or the black building test, because the cleanest version starts by actually opening the utility and watching the whole facility ride through on its own resources. It is the keystone of data center commissioning, and it is the level a Tier certification witness cares most about.

A real IST loads the plant with load banks to a defined percentage of design, then drops the utility. The UPS and BESS carry the load through the gap. The generators sense the loss, start, build voltage and frequency, parallel, and take block load within the time the sequence allows. The transfer scheme moves the load to generator power without dropping the critical bus. Then you restore the utility and watch the plant transfer back, also without a drop. Along the way you fail individual components on purpose, a generator, a UPS module, a feeder, to prove the redundancy is real and not just drawn on the one-line.

The reason it is the keystone is that nothing else exercises the seams. Every system can pass its own functional test and the plant can still fail at the IST, because the failure lives in the handoff: the timer that was set wrong, the load-shed step that sheds the wrong bus, the generator that accepts load fine alone but trips when the second one tries to parallel. The IST is the only test that puts the whole sequence under a clock and a load at the same time.

It is also the test that schedule pressure goes after first, because it needs the entire plant complete, megawatts of load banks staged, fuel, and a long uninterrupted window. When a project is behind, the IST is where corners get cut: shortened, run at partial load, or quietly broken into pieces that never get strung together. A plant that never had a complete IST has never actually been proven. That is the one to refuse to sign.

The integrated systems test is where redundancy stops being a single-line claim and gets proven. A facility built N has no spare, so it is never tested by failing a unit; N+1 carries one extra module or path beyond the load, so the IST drops one unit and confirms the rest hold; 2N runs two fully independent systems at no more than half load each, so the test kills an entire side and watches the other carry the block; 2(N+1) adds a spare within each of those sides. The script fails exactly the unit each topology promises it can lose, then proves the IT load never saw it.

The role of load bank testing

Load banks are how you put real load on the plant before the real load exists. At the IST and at generator and UPS functional tests, there are no servers yet, so portable resistive or resistive-reactive load banks stand in for the IT load. They let you prove the generators carry rated kW, the UPS holds its runtime, the cooling rejects the heat, and the whole chain behaves at design load instead of at the trickle a half-built building actually draws.

Sizing and staging the load banks is part of the test plan, not an afterthought. You need enough capacity to load the plant to the percentage the acceptance criteria call for, the cabling and connection points to land it safely, and a way to step the load to test block-load acceptance and load-shed sequences. Resistive load banks test real power. Reactive load banks add the lagging power factor that proves the generator and the alternator behave under a realistic, not purely resistive, load.

The acceptance criteria for these runs, generator load-acceptance steps, run duration at load, temperature stability, are spelled out separately. For the full breakdown of how load bank results are judged pass or fail, see the load bank test acceptance criteria guide.

The role of switchgear receiving

Commissioning does not start at energization. It starts when the switchgear lands on the truck. Receiving inspection, the Level 2 step, is where you catch shipping damage, missing components, wrong configuration, and moisture intrusion while you still have the time and the standing to make the manufacturer fix it. Once the gear is set, bused, and the schedule has moved on, a problem found at energization is ten times the disruption it would have been at the dock.

The receiving check verifies the gear against the approved submittal, confirms the impact and tilt indicators on the crate did not trip, looks for cracked bushings and bent bus, confirms space heaters were energized or the desiccant is intact, and documents the as-received condition with photos before anyone signs the delivery ticket. A megger reading at receiving on MV gear catches insulation that got wet in transit before it gets buried behind connections.

The detail of what to look at, crate to bus, lives in the switchgear receiving inspection checklist guide. Treat it as the first commissioning record, because a damage claim is far easier to win with dated receiving photos than with a memory of how the crate looked.

UPS, static transfer switches, and battery QA

The UPS is the no-break heart of the plant, so its commissioning is about transfers and runtime, not just startup. Functional testing proves the UPS carries load on normal, transfers to internal and maintenance bypass and back without dropping the critical bus, and rides through a source loss long enough for the generators to come up. The transfer back to inverter is the step that gets skipped and the one that bites, because a UPS that goes to bypass and stays there has lost its protection silently.

Static transfer switches downstream get their own transfer-timing tests, proving they move the load between two sources fast enough that the equipment never sees an interruption, and that the preferred-source logic returns correctly. The numbers are in milliseconds, and the test gear has to be fast enough to actually capture the gap.

The battery is the part everyone trusts and few prove. A battery that floats fine can still fail to deliver capacity under discharge. Acceptance includes a capacity or discharge test that confirms the battery delivers its rated runtime at load, plus impedance or ohmic baselines on each cell or jar so the maintenance team can trend degradation later. A UPS with a tired battery passes every test except the one that matters, the one where the utility actually fails. For lithium and other chemistries in a BESS configuration, the protection, EPO, and thermal and fire interlocks get proven alongside the electrical performance.

Generators, paralleling switchgear, and NFPA 110

Standby and prime generators are commissioned against the clock. NFPA 110, the standard for emergency and standby power systems, sets the framework for how the engine-generator set and its transfer equipment are installed, tested, and timed, including the start-and-transfer time the system is classified to. NFPA 110 also classifies the system by the Class, how many hours it must run at rated load, and the Type, how fast it must pick up the load. Confirm the Class, Type, and Level required against the project documents and the adopted edition, because they drive the acceptance criteria.

Functional testing proves the genset starts on a signal, builds voltage and frequency, accepts block load without stalling or excessive dip, and runs at load with stable temperature and no alarms. Block-load acceptance is the test that finds the underrated set: the generator idles fine and then sags or trips when a large step of load lands all at once.

On a multi-generator plant, the paralleling switchgear is its own commissioning effort. You prove the sets synchronize, share load proportionally, shed and add load in the right order, and execute the sequence on utility loss without two sets fighting each other. The failure mode here is subtle and dangerous: each generator passes alone, and the plant trips when they try to parallel under load because the sharing or sync settings were never proven together. That only shows up when you run them as a system, which is why it folds into the IST.

BESS commissioning in the power chain

Battery energy storage is showing up in data center power chains as ride-through, peak support, and in some designs as a replacement for the traditional UPS battery or even the standby generator. Commissioning a BESS pulls in everything a UPS battery needs plus the power conversion system, the inverter or PCS that moves energy in and out, and a heavier set of protection and safety interlocks because of the energy density and chemistry.

Power QA on a BESS proves the state of charge and capacity, the PCS function and its transfer and ride-through behavior, the protection and emergency power off, and the thermal management and fire-detection interlocks that have to act before a cell event becomes a fire. The electrical performance and the safety system are tested together, because a BESS that delivers power but does not isolate on a fault or trip its thermal interlock is not commissioned, it is dangerous.

BESS commissioning also has more punch-list churn than older gear, because the controls, firmware, and protection coordination are newer and less settled. The deficiency tracking and punch process for a BESS is heavy enough that it benefits from a dedicated workflow; the besscxpunch tool exists to run that BESS commissioning punch list so findings do not get lost between the integrator, the EOR, and the CxA.

Grounding, bonding, signal reference, and surge protection QA

Grounding is the QA that gets rushed because it is invisible when it works and catastrophic when it does not. An open or high-resistance ground means fault current has no low-impedance path back to the source, so the protective device may not see the fault and clear it. Power QA verifies the grounding electrode system meets the design ground-resistance value, that bonding is continuous across the equipment, and that the connections are torqued and not just landed.

Data centers add a layer of signal reference and equipment grounding that ordinary buildings do not, to keep ground potential differences from corrupting the low-voltage signals and the sensitive electronics on the floor. The IEEE grounding guidance, including IEEE 142, the Green Book, for grounding of industrial and commercial power systems, is the reference for how the system is designed and what good practice looks like. Verify the bonding to the gear, the raised floor structure, and the signal reference grid against the design.

Surge protection rounds it out. SPDs only do their job if they are installed with short, straight leads, because lead length adds let-through voltage, and if their status indication is checked so a failed SPD does not sit dead and unnoticed. The lightning and transient protection scheme, including any SPD coordination across service entrance, distribution, and branch, gets verified as installed against the design, not assumed.

Are the arc-flash and coordination study settings actually loaded?

The coordination study and the arc-flash study are not paperwork that lives in a binder. They produce the actual settings that go into every protective relay and trip unit in the plant, and the single most common power QA failure is gear running on factory defaults instead of the study values. Power QA verifies, device by device, that the as-left settings in the gear match the approved coordination study. A breaker set to its default instead of its coordinated value will either trip the whole bus on a downstream fault or fail to trip at all.

The coordination study, built on IEEE practice for protection and coordination, sets the time-current curves so the device closest to a fault clears it first and the rest of the plant stays up. That selectivity is the whole reason a fault on one rack does not take down a data hall. It only works if the settings in the relays match the study, which is why the as-left settings sheet is one of the most important turnover documents in the set.

The arc-flash study, performed to IEEE 1584 for arc-flash hazard calculations, sets the incident energy and the labels that tell a worker what protection to wear at each piece of gear. The settings that affect clearing time also affect incident energy, so a relay set wrong is both a coordination problem and a safety problem. Power QA confirms the labels are installed and the settings behind them are the studied ones. Verify the gear, not the binder.

Sequence of operations verification

The sequence of operations, the SOO, is the script the plant follows automatically: what happens on a utility loss, in what order the generators start and parallel, how the load transfers and sheds, and how the plant recovers when the utility returns. Functional and integrated testing exist to prove the plant actually does what the SOO says, step by step, under real timing.

Verifying the SOO is tedious and it is where the real defects live. You walk each branch of the logic: normal operation, single utility loss, generator failure during the event, UPS module failure, a feeder fault, and the return to normal. Each branch has an expected outcome and a timing window, and you record what the plant actually did against what the SOO said it should do. A controls scheme that handles the clean case and mishandles the second concurrent failure is exactly what the SOO verification is built to catch.

Write the test script straight from the SOO with explicit pass or fail criteria for each step, and have the witness sign each step, not just the page. A SOO that was demonstrated step by step and signed is worth something to the operations team. A SOO that was watched once in a hurry and checked off as a block is not.

How long does data center commissioning take?

Data center electrical commissioning runs in parallel with construction over months, not as a single event at the end. Factory tests happen as gear is built, receiving happens at each delivery, installation checks track the install, and functional testing follows energization room by room. The integrated systems test is the concentrated push at the end, and on a large facility the IST alone can take one to several weeks of staged runs, depending on the number of systems and failure scenarios in the script.

The honest answer is that commissioning takes as long as the plant needs to be proven, and the variable that actually moves the schedule is how much rework the testing uncovers. A clean plant flows. A plant full of deficiencies turns each test into find, fix, re-test, which is where the time goes.

The dangerous version of this question is the one asked to compress it. When construction runs late, commissioning is the last activity before turnover, so it absorbs the slip. The fix is to protect the commissioning window in the master schedule from the start and to front-load the levels that can run early, so the only thing left at the end is the integrated test that genuinely needs the whole plant. Commissioning that gets squeezed into the gap left over is commissioning that gets shortened, and the IST is the first thing to go.

The commissioning record set and turnover package

The commissioning record is the deliverable. The tests prove the plant in the moment; the record is what the owner keeps. It is the commissioning plan, the factory test reports, the NETA acceptance test reports, the as-left relay and trip settings, the signed functional test scripts, the IST script and results, the issues log with every deficiency tracked to closure, and the final turnover or closeout package that bundles it for the owner.

Two parts of the record carry the most weight later. The NETA acceptance reports are the de-energized baseline every future maintenance test trends against, so they have to be complete and filed, not summarized. The as-left settings sheets are the proof that the coordination and arc-flash studies were actually loaded, and they are the first thing a future engineer needs when the plant gets modified. A record set missing those two is missing its spine.

The deficiency log deserves its own discipline. Every finding gets logged, assigned, tracked, and closed with evidence that the fix was re-tested, not just reported done. A log that shows open items closed on paper, with no re-test, is the document that tells you the deficiencies are still in the building. Keeping that log clean across a long project with many parties is exactly the kind of field tracking the tradeos workflow is built to carry, so findings do not fall through the gaps between the installer, the test agency, and the CxA.

What do I do if a deficiency is found during the IST?

When the integrated test turns up a deficiency, you log it, you stop and decide whether the test can continue, and you do not close it until the fix has been re-tested under the same conditions that found it. The instinct under schedule pressure is to note it, keep running, and circle back. For a deficiency that affects the rest of the test path, that is how you get a test result that proves nothing, because the plant you finished testing is not the plant you started with.

Triage by severity. A finding that does not affect the sequence under test gets logged and the test continues. A finding that breaks the path, a generator that will not parallel, a transfer that drops the bus, stops that branch until it is corrected, because everything downstream of it is now untrustworthy. The CxA makes that call, not the contractor trying to keep the clock running.

The classic failures all share one root: a deficiency that got closed without proof. Settings noted to be fixed but never re-verified in the gear. A failed transfer marked resolved on the strength of a controls change nobody re-ran. An IST that hit a wall at hour three, got broken into pieces to save the night, and never got strung back together into one continuous run. The discipline is boring and it is the whole job: log it, fix it, re-test the thing you changed, sign it. A deficiency closed on paper is a deficiency still in the building, waiting for the night the utility actually fails.

Commissioning and Uptime Institute Tier certification

For facilities chasing an Uptime Institute Tier certification, commissioning is not a parallel nice-to-have, it is how the constructed facility gets certified. Uptime separates Tier Certification of Design Documents, which reviews the design before construction, from Tier Certification of Constructed Facility, which verifies the built plant through on-site inspection and witnessed testing. The constructed-facility certification is, in practice, a witnessed demonstration that the plant meets its claimed Tier under fault and maintenance scenarios.

That means the integrated systems test and the redundancy demonstrations do double duty. The same runs that prove the SOO for commissioning are the runs an Uptime witness watches to confirm the topology delivers the concurrent maintainability or fault tolerance the Tier claims. A well-run commissioning program lines its IST scenarios up with the Tier demonstration so the plant is proven once, thoroughly, rather than tested twice.

Thorough integrated testing across electrical, mechanical, and controls is the strongest predictor of certification going smoothly. The certifications fail or stall on the same things commissioning catches: a redundancy that is drawn on the one-line but does not hold up when you actually fail the component.

Field checklist

What to document

The record is judged by whether someone two years from now can reconstruct what was proven and compare today's readings to it. Capture the plan and the level definitions, the test reports at every level, the as-left settings, the witnessed scripts, the IST results, and the deficiency log with closure evidence. The table below is the spine of a power QA turnover set.

Common mistakes

• Compressing commissioning to recover schedule slip, so the IST gets shortened, run at partial load, or skipped.
• Energizing before NETA acceptance tests are complete, losing the de-energized baseline for good.
• Loading factory or default relay settings instead of the approved coordination study values.
• Closing deficiencies on paper without re-testing the change that was supposed to fix them.
• Running each system's functional test but never stringing them into one continuous integrated test.
• Testing on a healthy utility and never proving the plant on a real simulated outage at load.
• Treating the arc-flash and coordination studies as a binder instead of verifying the settings in the gear.
• Skipping the battery capacity or discharge test, so a tired battery passes everything except the real event.
• Letting the CxA report to the GC instead of the owner, which erases the independence of the check.
• Turning over a record set with missing test reports, leaving the owner no baseline to maintain against.

Standards and references

The acceptance testing follows the NETA Acceptance Testing Specifications, ANSI/NETA ATS, current edition ATS-2023, which is the field reference for what to test on each class of power equipment before energization and what the results should show. Ongoing maintenance after turnover follows NFPA 70B, the standard for electrical equipment maintenance, and the IEEE 3007 series for the operation, maintenance, and safety of industrial and commercial power systems, both of which trend against the NETA acceptance baseline.

The protection work draws on IEEE practice: the coordination study sets the protective device curves, and the arc-flash study follows IEEE 1584 for arc-flash hazard calculations. Switchgear and circuit breakers are governed by the IEEE C37 series for their ratings and test methods, and grounding design references IEEE 142, the Green Book, for grounding of industrial and commercial power systems. Generators and standby power follow NFPA 110, the standard for emergency and standby power systems, including its Class, Type, and Level classification and the start-and-transfer timing. Worker safety during energized testing follows NFPA 70E.

The installation itself is built to the NEC, NFPA 70, as adopted and amended by the jurisdiction. The commissioning process framework comes from the ASHRAE commissioning guidance, commonly ASHRAE Guideline 0 for the commissioning process, and the Building Commissioning Association best practices, while the Uptime Institute Tier standards drive the certification testing where a Tier is claimed. Edition numbers and clause references change between cycles, so confirm the specific edition and any local amendments against the project documents before citing a standard on a submittal. The commissioning plan and the contract documents control scope; the standards give the framework.

Units, terms, and acronyms

Data center power QA carries its own vocabulary, and the same idea reads differently across a commissioning plan, a NETA report, and a Tier document. The terms below are the ones that travel across the whole power chain.

Cx / CxA

Commissioning, and the commissioning authority or agent who runs and witnesses the process

IST

Integrated systems test, the full-plant test simulating utility failure at load, also called pull-the-plant or black building test

FAT

Factory acceptance test, the witnessed test of equipment at the manufacturer before shipment, the Level 1 step

NETA ATS

Acceptance Testing Specifications, the standard for field acceptance tests on electrical power equipment

SOO

Sequence of operations, the automatic logic the plant follows on utility loss, transfer, and recovery

Critical load

The IT load the facility exists to protect, the load the power chain must never drop

BESS / PCS

Battery energy storage system and its power conversion system, the inverter that moves energy in and out

STS

Static transfer switch, which moves load between two sources fast enough to avoid an interruption

EOR

Engineer of record, who owns the design intent and the coordination and arc-flash studies

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What voltage drop is acceptable?
• What load rating does a data center raised floor need?
• What do I check when switchgear is delivered?
• How long is a generator load bank test?
• What is commissioning in construction?