Cooling commissioning checklist: the test at each level, the number that passes it, and the reading the operations team keeps

The chiller carries a white tag. Its Level 5 test was signed off on a Thursday in July, and the plant has run on utility power since. At 04:10 the utility drops for forty seconds, the generators take the load, and when the transfer switch returns to utility the chiller trips on the dip and restarts eleven minutes later. Nobody on shift can find a reading from July that says what the inlets did the last time this happened.

A cooling commissioning checklist is the ordered set of tests a data center's cooling plant passes before it carries live load: the factory witness test, site acceptance, start-up and pre-functional checks, functional performance tests of each system, and the integrated systems test of the whole plant under load banks. Integrated systems testing, Level 5, interrupts utility power and removes redundant chillers, pumps and CRAH units to prove the environment holds through every failure the design claims to survive. Each test on this list carries the number that passes it and the reading recorded to prove it, because that reading is the first baseline the operations team will ever have.

• Five levels, five tags. Red for the factory test, yellow for delivery, green for start-up, blue for the functional performance test and white for the integrated systems test. A white tag on a unit whose blue-tag failure response was never fixed is the commonest way a commissioned plant fails.
• The numbers exist and nobody on page one quotes them. AHRI 550/590 sets the chiller test stand's tolerances, ASHRAE TC 9.9 sets the 18 to 27 °C band and the 5 °C-in-any-15-minutes limit an inlet may not exceed, and Uptime's Tier Standard says what Continuous Cooling has to show through a utility transition.
• The test that fails is the second half of the blackout. Commissioning agents report more findings on the return to utility power than on the transfer to generator, and the retransfer dip is what trips a chiller's controls.
• The record is the point. The IST is the only time the plant's failure response is measured under load with instruments in place, and the readings it produces are the baseline every later repair is verified against. A PDF in a handover binder is not where they should live.

What is a cooling commissioning checklist, and what is integrated systems testing?

Commissioning is the process that verifies a building and its systems meet the Owner's Project Requirements, and ASHRAE writes the process down twice. Guideline 0 describes a commissioning process capable of verifying that the building and its systems meet the OPR from predesign through warranty expiration, and its 2020 addendum retitled it for new buildings and new systems. Standard 202 sits beneath it as the floor, written as mandatory requirements a code official can adopt rather than suggestions, and Guideline 1.1-2025 adds the HVAC&R technical detail that augments Guideline 0.

Data centers run that process in five numbered levels, and the trade marks each level with a coloured tag on the equipment. The levels are the same on every commissioning agent's page, which is the consensus this checklist is built on.

Level 1, red tag

Factory witness test. The equipment is run on the manufacturer's test stand against its selection, before it ships.

Level 2, yellow tag

Site acceptance. Delivered undamaged, placed to the drawings, nameplate matching the design, ready to energise.

Level 3, green tag

Start-up and pre-functional. Flushed, pressure-tested, wired, calibrated and started by the installer and the OEM technician.

Level 4, blue tag

Functional performance test. Each system run through its sequence of operations and its own failure modes, alone.

Level 5, white tag

Integrated systems test. The whole plant under load banks, in every mode, through injected failures, with every other system watching.

Level 6, no tag

Closeout. Issues closed, systems manual and commissioning report issued, operators trained, the record handed over.

Integrated systems testing is the fifth level, and it is a different kind of test from the four before it. A functional performance test asks whether a system does what its own sequence says. The integrated systems test asks what every system does when one of them fails, and its common form is the blackout test, in which utility power is cut to the whole site with the load banks running and the plant is watched all the way back to normal.

Uptime Institute puts the level in its own words when it describes what a facility should have finished before certification: Level 5 Commissioning, Integrated System Testing, with the testing having exercised all of the components as though the data center were operational. The same article gives the view it is written against, that commissioning is turning everything on to see if it works.

Why does a chiller that passed its factory test still fail the integrated systems test?

Because the factory test proves the chiller can make cold water at four load points, and nothing else. Trane's description of its own AHRI-approved test stand says the factory results serve as a benchmark during the commissioning process, with AHRI 550/590 defining the temperatures and flow rates for 100, 75, 50 and 25 percent load, flow held within 5 percent, leaving evaporator and entering condenser water within 0.5 °F of target, and four sets of readings taken at five-minute intervals and averaged. That is a precise answer to a narrow question.

The integrated systems test asks a wider one. Alex Mathers of GMC Commissioning writes that an IST is not solely, or even mainly, an electrical test, that the most common findings are mechanical, and that he has seen more issues once normal power is restored than going on to generator. His list is the list every cooling plant should expect: BMS field controllers that lose memory and their stored flow coefficients, equipment that does not restart, HVAC equipment that takes thirty seconds to five minutes to come back, and server-room CRAC units that do not come back at all.

A chiller that passed on the test stand has proved it can make cold water. It has not proved it can come back.

The physics of the gap is documented. Schneider's White Paper 179 notes that CRAH fans, chilled-water pumps and chillers are typically not connected to UPSs and may not even be connected to backup generators, that without airflow the rate of temperature rise could easily be 5 °C a minute or more, and that a chiller's controls ride through only a quarter of a cycle before demanding a restart. The retransfer from generator back to utility is a dip of 100 milliseconds to a second in Schneider's account and 15 to 150 milliseconds in Mathers's, and either is long enough.

The schedule is the other reason. Uptime's certification team writes that when construction slips, the commissioning just gets reduced in scope to meet that same schedule, and describes a site where the demonstration found cooling tubes clogged with dirt and silt that a full Level 3 flush would have caught. The white tag is only as good as the four tags under it.

What is on the cooling commissioning checklist, level by level?

The table is the list, assembled from the five-level consensus, ASHRAE's guidelines, Uptime's demonstrations and the OEM test-stand documents. Two columns are added to what any page-one list carries: the number that passes the test, from the document that sets it, and the reading that is recorded to prove it.

Two things follow from the fourth column. ASHRAE's 5 °C figure is a change within a window, not a rate: note f to Table 2.1 says 20 °C in an hour and no more than 5 °C in any 15-minute period, measured as the maximum minus the minimum inlet within the window, so the trend has to be recorded at a resolution that can show it. And the allowable band is not a reliability statement, the same card says, but a statement that the IT manufacturers test their equipment to function inside it, which is why the recommended band is the target and the allowable is the limit an injected failure may touch.

The factory row deserves one more sentence. Trane's document says rapid restart after a power loss can be proven on the test stand for mission-critical applications, and a plant that specifies a quick-start chiller and never asks for that test has bought a number it cannot check until Level 5.

Which failures does the integrated systems test inject, and what passes each one?

The failure set comes from what Uptime's consultants ask an operations team to demonstrate and from what commissioning agents report finding. Uptime's list is short and exact: interrupt the incoming utility supply to see that the UPS carries the load until the generators come on line and that the cooling system maintains the critical load during the transition, then remove redundant capacity components to show that N generators carry the load or that the redundant chilled-water distribution loop can be taken out while still sufficiently cooling the data center.

How long the plant has in the first row is not this page's question. Which cooling system gives you the most time after its own worst failure is ranked and modelled in its own guide, with the stored water the chiller restart needs. What the IST adds is a measurement of that window for this plant, and the measurement is the point.

One published study shows what the measured window looks like. Cho, Park and Jeong modelled a 20 MW plant under six fault conditions as part of its commissioning verification and found that at 17 °C chilled water and 24 °C CRAH supply the inlets stayed inside 18 to 27 °C for approximately 320 seconds after the cooling outage, while at 18 °C and 25 °C a rapid rise began, so the backup had to be active within 300 seconds. Cold-aisle containment and warm setpoints, the paper notes, raise the chance of the rapid rise, which is the trade-off every efficiency decision makes against the window.

The alarm row is where the rationalised list gets its first real exercise. An IST that trips fifty alarms from one injected failure has found the flood before the plant is live, and the fields a rationalised alarm has to carry are what the log is checked against.

Which readings does the test produce, and where do they go after handover?

Standard 202's deliverables are the OPR, the basis of design, the commissioning plan, the issues log, the systems manual, the training plan and the final report. Guideline 0's scope adds requirements for acceptance and documentation at each phase and for the initial establishment of an ongoing commissioning process, which is the clause that says the record is meant to outlive the project.

The last edge is the one this page exists for. When the readings arrive in operations as a report, the first repair on a CRAH in row 14 has no baseline but the day's trend, and the next utility loss is judged against memory. When they arrive as values against the unit and the row, the repair has a pre-work baseline that predates the fault, and the outage has a measured window to be compared with.

Is NFPA 4 integrated testing the same test?

No, and the confusion is built into the search results. Half of page one for the phrase is fire-system testing, and the AI answer is software testing, so a project that needs the data-center Level 5 test and the fire test will hear the same three words for both.

Level 5 integrated systems test

The data-center commissioning test on this page: the whole mechanical, electrical and controls plant under load banks, through injected failures, against the OPR. Witnessed by the commissioning authority, with Uptime's consultants watching where a Tier certification follows.

NFPA 4 integrated testing

The fire and life safety test. NFPA 4, the Standard for Integrated Fire Protection and Life Safety System Testing, applies when two or more fire protection or life safety systems are tied together, runs as a complete sequential end-to-end test under an integrated testing agent, and repeats on a plan the authority having jurisdiction accepts.

Software system integration testing

The AI Overview's meaning: testing that software modules work together. Not a facility test.

The two facility tests meet at the containment. A drop-away aisle roof that has to release on smoke detection ties the fire alarm to the HVAC containment, which is the kind of interconnection NFPA 4 exists to test, and the cooling plant's Level 5 test has to know the roof will open, because an open roof changes where the air goes. Both tests need the same load banks in the same week, and the schedule usually has room for one.

How to write the cooling IST script this week

A script is a list of failures, the reading each one is judged on, and the number that reading has to stay inside. Most scripts have the first and neither of the others, which is how a test passes on the day and proves nothing later. The order below is the order that keeps the load banks and the operators safe.

• Pull the pass criteria out of the OPR first. If the OPR says the plant holds the environment and gives no band, write the band now: recommended 18 to 27 °C as the target, the class allowable as the limit, 5 °C in any 15 minutes as the change limit, and get the owner to sign it before anyone opens a breaker.
• Confirm the four tags below the white one. Every Level 3 and Level 4 item signed, the issues log down to items the owner has accepted in writing, and every sensor the script will read calibrated with its offset applied.
• Size and place the load banks to the design load at rack granularity. Uptime's tell for an under-prepared site is a 1,200 kW hall with two large banks. Avtron's guidance is to place resistive load banks where heat loads will ultimately be present, network them for one control and one data capture, and use them to create the hot spot deliberately.
• Take the baseline at design load before any failure: every inlet, every supply and return, every flow, every setpoint, with the load banks stable for long enough that nothing is still moving.
• Inject the failures in the order of the table above, one at a time, and return to baseline between each. The return is a test too, and it is where the reversed valve and the controller that forgot its coefficients show up.
• Run the blackout last and run both halves. Hold on generator long enough for every cooling component to have restarted and settled, then retransfer to utility and watch the plant through the dip as closely as through the loss.
• Record the readings against the asset, not the day. Time to generator, time each unit returned, peak inlet, minutes above recommended, and which units tripped on the retransfer, each against the unit or the row it belongs to.
• Close the issues log before the white tag, and hand over the values, not the report. The report is the proof the test happened. The values are what the operations team will use.

The eighth step is the one that decides whether the July test is findable in November. It costs nothing on the day and it is the difference between a baseline and a binder.

What do the BMS, DCIM, CMMS and the issues log already hold?

Each system holds one column of the readings table, and none holds the join between the injected failure, the reading it produced and the unit it belongs to. The gap is not that the systems fail to communicate. It is that the test happened before most of them had an owner.

The join is cheap once the readings are values. The CRAC and CRAH maintenance checklist that starts at handover carries the reading that proves each PM item, and the first value in every one of those rows already exists, taken under load with calibrated instruments, in the Level 4 sheet. Annual repeats of the IST, which the load-bank trade recommends, add a row to the same record rather than a binder to the same shelf.

Who starts the chiller, who pulls the breaker, and who signs?

The commissioning authority plans, witnesses and documents, and does not operate the plant. The OEM's start-up technician starts the chiller at Level 3 and is the one qualified to interpret its controller through the restart at Level 5. The controls contractor owns the sequences and is in the room for every injected failure, because half the findings are theirs.

The operations team performs the demonstrations. Uptime's consultants observe the operations team as it performs them, and the sites that pass first time are the ones that wrote MOPs and SOPs for each demonstration before the visit. That puts the person opening the breaker and the person removing the loop on the owner's side of the table, which is where they will be at 04:10 in November, and which credential that person holds is the same question then as now.

Uptime names the failure mode too: contractors and vendors who depart before operations arrives, and staff unfamiliar with the systems they have inherited. A Level 6 training row that lists every mode tested at Level 5 is the only defence, and it is in the checklist above for that reason.

What this cannot do

A checklist assembled from ASHRAE's guidelines, Uptime's demonstrations and the OEM test-stand documents is not your plant's OPR. The bands are ASHRAE's, the tolerances are AHRI's as Trane states them, the demonstrations are Uptime's, and none of them knows your load, your class or the restart time your chiller was bought with. The 320-second window is one modelled plant's result at its own setpoints and is here as the shape of a measurement, not a rule.

Nothing here opens a breaker, starts a chiller, steps a load bank or changes a setpoint. A decision layer that holds the Level 5 readings against each unit can say that the standby chiller took eleven minutes in July and fourteen tonight, and that row 14's inlets went 3 °C higher than the test showed, and it does not restart, stage, isolate or rebalance anything. A named person on the operations team runs every step of the test, the commissioning authority signs it, and the layer records what was read and whether it agreed with the last time.

Answered