UPS Battery Minutes Do Not Equal Thermal Ride-Through Seconds

Five minutes [TARGET] of battery holds the uninterruptible bus. One cubic metre of secondary fluid, allowed one kelvin, buys 0.33 seconds [MODELED] at 12 MW. Those are not two readings of one clock. At the instant utility supply is lost, the non-protected heat-rejection equipment stops while the secondary loop, held up by the UPS, keeps delivering roughly 12,600 kW into a primary loop that is no longer rejecting anything [MODELED]. Survival then depends on stored heat capacity. The result is an order-of-magnitude statement, not a design value. No ride-through time is fixed here.

Natural Thermal Inertia Here Is Seconds, Not Minutes

One cubic metre of the secondary fluid absorbs 3,968 kJ per kelvin of allowed rise. Against 12 MW that is a third of a second. Minutes of ride-through would therefore need hundreds of cubic metres across ten or more kelvin, while the estimated secondary-side fill volume of the whole site is 25 to 45 cubic metres with an uncertainty band of plus or minus 40 percent [MODELED]. The direction of the answer is settled before any detailed calculation: natural thermal inertia on this plant is a seconds-scale quantity.

Worked across the plausible range, secondary inventory alone gives 25 seconds at the pessimistic corner of 25 cubic metres and 3 K of allowed supply-temperature rise, and 121 seconds at the optimistic corner of 45 cubic metres and 8 K [MODELED]. That band straddles typical generator transfer times, which is precisely why the question cannot be settled by judgement.

Two details inside the arithmetic are easy to lose. The denominator has to include secondary pump heat, because the pumps keep running at full power during the event — holding flow is itself adding heat. And the levels do not add. Primary and secondary inventories share one temperature criterion chain, so a loop temperature rise transfers to the supply temperature with almost no attenuation, and adding the separate ride-through times together overstates the result.

Ride-Through Depends on Which Pumps Keep Running

Compose the levels properly and three cases appear. With every pump held up, roughly 35 cubic metres of secondary and 80 cubic metres of primary inventory plus exchanger metal give about 187 seconds at 5 K of allowed rise [SENSITIVITY]. Drop the primary pumps and the primary inventory leaves the coupled mass entirely: about 59 seconds [SENSITIVITY]. Stop the secondary pumps as well and the case collapses to the cold plate, where the few tenths of a litre inside it give 5 to 15 seconds [SENSITIVITY].

Those three cases differ by factors of roughly three and six, and the discriminating variable is not how much liquid is stored — it is which pumps are still running. Flow retention outranks temperature retention: with flow maintained, a supply temperature a few kelvin high raises junction temperature by a few kelvin; with flow at zero, a pipe full of cold fluid is irrelevant to a cold plate that holds a fraction of a litre.

That ordering has a cost consequence that runs against intuition. Placing the primary circulating pumps on protected supply adds about 202 kW to protected load and about 7 kW to facility load, so it does not move PUE [MODELED]. It takes the composed ride-through from about 59 seconds to about 187 seconds. Buying the same 128 seconds as stored cooling capacity needs roughly 40 to 65 cubic metres of tank, with the footprint, structural load, insulation, water treatment, stratification control and freeze protection that follow [SENSITIVITY].

A Five-Minute Store Would Be 95 to 190 m³ on the Primary

Aligning a thermal store to the electrical five-minute floor is 95 to 190 m³ [MODELED]. Sixty seconds is 19 to 38 m³ [MODELED]. Time and volume are strictly linear. Each extra ten seconds of target is another 4 to 6 m³ of tank. If a store is bought, the heat-rejection basis puts it on the primary, not the secondary: three isolated secondary zones would need three tanks, and a large secondary volume fights the cleanliness the glycol loop is written to.

The interface this creates is a timing interface with the electrical design, not a capacity one. What the cooling side needs is not the generator start time but the moment the rejection fan motors are re-energised, which is detection plus start plus the complete staged-loading sequence. Staged loading exists to protect the generator from voltage and frequency dips, and it usually places large mechanical loads in the last step — a correct electrical decision that lengthens the cooling outage. The loading step assigned to rejection fans and primary pumps therefore cannot be set by one discipline alone.

Fail-Closed Valves Turn a Rejection Loss Into a Flow Loss

Valve and actuator failure position must be specified as hold-position rather than fail-closed, because a fail-closed valve turns a recoverable loss of rejection into an unrecoverable loss of flow. The controls basis records that requirement as an inherited cooling constraint, and records the opposite error — expecting the management system to put the valves back — as a banned default.

The UPS room itself sits outside all of this. Its heat is rejected by equipment that is not on protected supply, and the path from utility loss to UPS over-temperature does not pass through the cooling system at all, so no amount of cooling redundancy touches it.

Limits and open items

Modelled: the scale rule, the inventory bands, the 95 to 190 m³ five-minute tank, and the three composed cases. Every temperature-rise allowance in them is an example value chosen to show sensitivity, and the metal thermal capacity of pipework has been set to zero, which is conservative. [MODELED]

Sensitivity only: the 5 to 15 second cold-plate figure and the 187 / 59 / 10 second composition. [SENSITIVITY]

Open: the IT supplier's time curve from loss of flow to throttling, which is the only source for the cold-plate limit; the maximum allowable secondary supply temperature; the primary-side fluid and its inventory; the six electrical timing quantities behind the effective transfer time, none of which has been obtained; and whether the owner accepts capacity reduction under a combined worst-case weather and maintenance condition, which is the same question as how much stored cooling to buy. [HOLD]

No ride-through time is established, no storage volume is selected, no protected-load change is approved, and no Engineer of Record has reviewed any of this.

A Larger Battery Makes the Primary Heat Up Faster

The counter-intuitive part is the direction of the battery argument. If the effective transfer time already exceeds the thermal ride-through, a larger battery does not help — it keeps the secondary pumps at full power for longer while nothing is rejecting heat, so the primary loop reaches its limit sooner. The cooling side's request to the electrical side is therefore not more minutes. It is a shorter, instrumented transfer, and a written answer to one question: at which loading step do the rejection fans come back?

© 2026 K&K Data Service Inc. All rights reserved. Reproduction or republication is permitted only with clear attribution to K&K Data Service Inc. and a working hyperlink to the canonical URL of this article. Excerpts must preserve the technical context, maturity labels, assumptions, and limitations. No excerpt may imply project approval, field validation, certification, or endorsement that the original article does not state.

Source: K&K Data Service Inc., “UPS Battery Minutes Do Not Equal Thermal Ride-Through Seconds,” https://www.kkdatasvc.com/lab/hydraulics-and-transients/five-minutes-of-battery-seconds-of-thermal-margin/.