CRAC and CRAH preventive maintenance checklist: every item by interval, the unit state it needs, and the point that proves it

The semi-annual visit on CRAC-04 is booked for Tuesday. The work order lists twenty-two items copied from the manufacturer's checklist, the technician has a Type II card, and the hall is running at N+1 with CRAC-07 already flagged for a compressor fault. Which of the twenty-two items need CRAC-04 off, and whether the hall can spare it on Tuesday, is not written anywhere on the order.

A CRAC and CRAH preventive maintenance checklist runs on a monthly, quarterly, semi-annual and annual cycle set by the unit's manufacturer, with ASHRAE Standard 180 as the published floor. Filters, alarms and noise are monthly, fans, humidifier and the refrigerant or chilled-water readings are quarterly to semi-annual, coil cleaning and the electrical panel are semi-annual, and subcooling, thermography and belt replacement are annual. In a data hall every item also carries two things the manufacturer's list leaves out: whether the unit must be out of rotation, which needs the standby confirmed first, and the point on the unit's controller whose before-and-after reading proves the item was done.

• The intervals are the manufacturer's minimums. Vertiv's guides call their monthly and semi-annual schedules minimum intervals, and ASHRAE 180 calls its Section 5 frequencies the minimum required for compliance.
• Roughly a third of the items need the unit off. Those are the ones a live hall has to gate behind a confirmed standby, because two units out in one window is how a PM becomes an incident.
• Since 1 January 2026 a DX unit on an HFC refrigerant with 15 pounds or more of charge is under 40 CFR 84.106, with a 10 percent leak-rate threshold and a 30-day repair clock. The old 50-pound figure belongs to R-22 units.
• Most items can be proven from points the unit already reports: filter differential pressure, supply and return air temperature, fan and compressor amps, chilled-water delta-T, condensate pump status and the alarm log. The work order should close on those readings, not on the tick.

What is a CRAC and CRAH preventive maintenance checklist?

It is the list of inspections, measurements and cleaning tasks a computer room air conditioner or air handler receives on a fixed cycle, so that it fails less often and so that a failure is caught while the standby can still carry the load. A CRAC is the direct-expansion unit with a compressor and a refrigerant circuit inside it, and a CRAH is the chilled-water unit with a fan, a coil and a control valve fed by the chiller plant. The two lists share the air side and differ on the cooling side.

Four documents set the intervals, and they do not agree with each other. The manufacturer's installer and user guide is the binding one, ASHRAE 180 is the reference the sector cites, the EPA's refrigerant rules fix the leak-repair clock on a DX unit, and the site's own method of procedure (MOP) decides how a unit is taken out and put back. The table below is what each of the opened documents actually says.

The words that matter in those guides are the ones about minimums. Vertiv's DS guide tells the reader to follow the monthly and semi-annual maintenance schedules as minimum intervals that may need to be more frequent on site conditions, and ASHRAE's 2018 errata restates clause 4.3.2.2 as the Section 5 frequencies being the minimum required for compliance. A site that runs the manufacturer's list once a year is below both.

Why is a data-hall PM checklist different from a commercial HVAC one?

The commercial checklist assumes the building can lose the unit for the afternoon. A data hall cannot, so every item that needs the unit off is also a redundancy decision, and the standby unit that covers it has to be on the same list rather than a lighter one. Schneider's white paper puts it as redundant cooling designs that allow for concurrent maintenance, where the IT load is protected while maintenance is being performed.

The envelope is also narrower than comfort cooling. ASHRAE's thermal guidelines recommend 18 to 27 °C at the IT inlet and limit the change to 20 °C in an hour and no more than 5 °C in any 15-minute period, so the pull-down after a unit comes back is itself a thing the checklist has to watch. There is no off-season to schedule around, either, because the load does not drop in winter.

On a commercial roof the checklist is a list of tasks. In a data hall it is a list of tasks, each with a question in front of it: can the hall spare this unit right now?

Standard 180 itself points the other way for buildings. Its 2012 text gives the example of an old chiller kept as a backup whose inspection may be adjusted to reflect fewer operating hours, which is exactly the relief a data hall cannot take, because the standby CRAC is the unit that carries the hall the moment the duty unit is taken out. The data-center results agree on this and call it the standby rule, and the reason is the ride-through the hall has when a unit is out and the standby does not start.

What is on the checklist, monthly to annual, and what proves each item?

The table below is the list, assembled from the Vertiv DS, PDX, PCW and CRV guides, the Schneider InRow manuals and ASHRAE 180's Table 5-2, with the interval each source gives. Two columns are added to what any of those documents carries: the state the unit has to be in, and the point whose reading proves the item afterwards.

The CRV guide is worth reading for one sentence beyond its table. Its maintenance section says that all tasks and time periods listed are the manufacturer's regulations and must be documented in an inspection report, which is the manufacturer saying that the tick is not the record and the reading is. The proof column above is what an inspection report should contain.

Johnson Controls' 2026 CRAH guide shows how much of that proof is already wired. Its controller point list carries discharge and return air temperature, coil flow, a filter differential-pressure transducer, condensate pump status and a status contact per supply fan on a BACnet bus, and Schneider's InRow units keep run-hours counters for the fans, the air filter and the condensate pump that the manual says to reset only after the maintenance is done.

Which items need the unit out of rotation, and what has to be true first?

Six of the seventeen items put hands inside the fan section, the coil section or the electrical panel, and those need the unit off. Vertiv is explicit that the controller does not isolate power from the unit even in Unit Off mode and that the only way to be sure there is no voltage inside is to open a remote disconnect, with PPE per NFPA 70E before the control enclosure is opened. Schneider's manuals say the same in a DANGER block: turn off all power supplying the equipment before working on or inside it.

In a live hall that is a redundancy decision before it is an electrical one. The Uptime Institute's own explainer defines a Tier III site as one that requires no shutdowns for equipment replacement and maintenance, because every component needed to support the IT load can be shut down and maintained without impacting it, and a Tier II site as one with select maintenance opportunities. The gate below is what that sentence means for one CRAC on a Tuesday.

• Confirm the standby is ready, not just present: in automatic, no active alarm, no manual override, and load-tested since its last PM. A standby in alarm is a single point of failure that the BMS is already reporting.
• Confirm the hall's load against N with this unit out and any other unit already down. Two units out in one window is the case the data-center results warn about, and CRAC-07 on a compressor fault already counts as one.
• Take a baseline before the unit is touched: supply and return air, filter differential pressure, fan and compressor amps, delta-T, and the rack inlet temperatures in the rows the unit serves.
• Isolate under the site's written energy control procedure. 29 CFR 1910.147 requires a documented procedure per machine, applied by an authorized employee, and the lock comes off only by the person who put it on.
• Do the off-state items in one visit, so the unit comes out of rotation once and not three times.
• Return the unit to automatic, put the standby back in its rotation position, and watch the pull-down against the 5 °C in 15 minutes limit at the rack inlets.
• Close the work order when the post-visit readings match the proof column, not when the last box is ticked.

The filter row is the one worth arguing about. 1910.147 exempts minor servicing activities that are routine, repetitive and integral to the use of the equipment when alternative measures give effective protection, and many sites treat a filter change on a running unit that way, because the filter track is on the air side and nowhere near the panel. The coil clean and the panel are not that, whatever the visit schedule says.

Who does the off-state items is a separate question from whether the hall can spare the unit. The refrigerant readings need an EPA Type II technician, the panel needs an OSHA qualified person, and the lock needs the authorized employee on that unit's procedure, and which of those the person on shift actually holds is a record, not a job title.

What does the refrigerant side of a DX CRAC require since January 2026?

The refrigerant rows changed at the start of this year and the data-center checklists have not caught up. The EPA's Emissions Reduction and Reclamation rule put appliances charged with HFC refrigerant, which is what R-410A is, under leak-repair requirements at 15 pounds and up, and the agency's own fact sheet dates it: beginning January 1, 2026, owners, operators and certified technicians must comply. The 50-pound threshold that still appears on the top data-center result is the older rule for ozone-depleting refrigerants, and it applies to an R-22 unit, not an R-410A one.

What the new section asks for is a set of records, and each one is a checklist row. The list below is the EPA's own summary of 84.106, and it is the reason the leak check in the table above is annual while the unit is over threshold and not merely when the technician is on site.

Leak-rate calculation

Each time refrigerant is added, using the annualising or the rolling-average method, so every top-up on the work order carries a rate against the full charge.

Repair and two tests

Repair within 30 days, an initial verification test in that window, and a follow-up test within 10 days of the initial test or of the unit reaching normal operating characteristics.

Leak inspections

Once a year for a comfort-cooling or other appliance that has exceeded the rate, until a year of calculations shows it has not.

Retrofit or retirement plan

Within 30 days of deciding not to repair, of failing to act, or of a unit still leaking above the rate after repair and tests, with the work done inside a year.

Chronic leaks

A unit that leaks 125 percent or more of its full charge in a calendar year is reported to the EPA by March 1 of the next year.

Records

Kept for three years, on paper or electronically.

Two things the rule does not change. The definition of comfort cooling in 40 CFR 84.102 speaks of occupied facilities, and whether a data hall's CRAC is comfort cooling or an other appliance makes no difference to the threshold, because both carry 10 percent. And the certification rule is the one it always was: 40 CFR 82.161 requires anyone who could reasonably be expected to violate the integrity of the circuit to hold the card, Type II for the high-pressure appliance a CRAC is.

What do the unit controller, BMS, DCIM and CMMS already hold about a PM?

Each of them holds one column of the checklist table, and none of them holds the join between the item, the state the hall is in, and the reading before and after. The gap is not that the systems fail to communicate. It is that the PM record lives in the one system that never sees the points.

Standard 180 quietly allows the join. Its 2012 text says inspection may include data provided by sensors or a building management system, and that if unacceptable condition indicators are found on two successive inspections the owner shall investigate, while three successive acceptable inspections allow the frequency to be reduced and documented. A site that closes each PM on a reading has the two-and-three history for free; a site that closes on a tick has nothing to reduce or investigate from.

How to turn the checklist into a work order that closes on evidence this week

The table above becomes a site document in one afternoon per unit type, and it changes what a closed work order means. The steps are in the order a facilities lead would actually do them.

• Copy the manufacturer's checklist for each unit model you run, keep its intervals, and add the two columns: unit state and the point that proves it. Where the model's guide has no interval, use the CRV or Schneider band for that item and note the source.
• Mark the off-state items and group them into one visit per unit, so each unit comes out of rotation once per cycle.
• Write the redundancy gate as three lines at the top of every off-state work order: standby ready in automatic, load against N with this unit out, and any other unit already down.
• List the baseline points to capture before the unit is touched, and the same points after return, with the window you will watch them for.
• For every DX unit, write the full charge on the work order from the nameplate or the methods 84.102 lists, and add the leak-rate calculation as a mandatory field on any top-up.
• Name the credential each row needs: Type II for the refrigerant rows, a qualified person for the panel, the authorized employee for the lock.
• Give the standby unit the same document, on the same cycle, and log its load test as an item.
• Close the work order only when the post-visit readings sit inside the proof column, and hold it open when they do not.

What closes a PM item is a reading that agrees with the tick. The filter change shows as a drop in differential pressure and a counter reset, the coil clean shows as discharge temperature back at the same load, and the leak repair shows as two verification tests on file. A visit that leaves none of those behind was a visit, not a maintenance record.

What this cannot do

A checklist assembled from four manufacturers' guides is not your unit's guide. The intervals here are the ones those documents give, the ASHRAE rows are the 2012 edition's text with the 2018 minimum rule quoted from ASHRAE's errata, and the refrigerant rule is the regulation as it stands on the eCFR this month. Your model, your refrigerant and your site's reading of the lockout exemption decide the rest.

Nothing here takes control of a unit or a hall. A decision layer that reads the points can say that the standby is in alarm and the order should not release, or that the filter counter reset without the differential pressure moving, and it does not start, stop, isolate or reset anything. A named human takes the unit out, a named human applies the lock, a named human does the work and returns it, and the layer records what was read before and after and whether the two agreed.

Answered