Guaranteed efficiency in a specification: ITU-T L.1320 gives you no tolerance band

The specification asks for an efficiency of no less than 95%. The supplier’s test report declares 95.8. At acceptance you repeat the measurement following the very recommendation named in the contract, and you read 94.6. It is not that one of you got it wrong: you can both be right, because that text never says how far apart two measurements of the same unit may legitimately fall. The acceptance band is not there. The certificate, though, is yours to sign.

What L.1320 actually defines

Recommendation ITU-T L.1320, Energy efficiency metrics and measurement for power and cooling equipment for telecommunications and data centres, was approved on 22 March 2014 by Study Group 5 under the AAP procedure. First and only edition, listed as In force, with no amendment and no corrigendum. Twenty-four pages, of which the normative body is a single clause, clause 5.

The level is stated in the scope: “The methodologies defined in this Recommendation are applied at single equipment level.” This is not room-level performance, which is measured with other indicators in the L.1300 series. Two families: power feeding (UPS units, DC/AC inverters, AC/DC rectifiers, DC/DC converters, photovoltaic, wind, hydrogen fuel cells) and cooling (air conditioners, outdoor air cooling, heat exchangers).

The formulas are elementary, and that is a virtue. For power feeding, efficiency is always η = Po/Pi, active output power over active input power, with Po becoming V × I for the rectifier.

For cooling it changes, and the change is worth noting. For the air conditioner efficiency is QT/Pi, where QT is the total cooling capacity, sensible plus latent. For outdoor air cooling and for the heat exchanger it is QS/Pi, the sensible part alone. Under two headings of the same clause you are not measuring the same quantity: whoever compares the figures has to say first which formula was applied.

Eight requirements, and all of them ask you to declare

The tally of modal verbs is only a counterpoint here, and it runs the opposite way to the suspicion: shall occurs nine times, but once in the cover note that explains what the word means, so the real requirements number eight, against twenty should and eleven may. The document does prescribe. The question is what.

Three clauses hold five of them, the ones a specification can genuinely work with. Clause 5.1.1: “Environmental conditions (such as temperature, humidity and air pressure), electrical conditions (such as input DC voltage, input AC voltage and frequency) and metrology requirements shall be reported when measuring energy efficiency.” Clause 5.1.3: “As such, parameters used for static and dynamic load conditions shall be documented”, followed by “The load range shall be consistent with the operational range of the application targeted loads.” And clause 5.1.3.1, the only one that points to an external test standard: “Measurement conditions for UPS shall be considered according to [IEC 62040-3]” and “The Pi and Po calculation shall be conducted in accordance with [IEC 62040-3].”

The remaining three ask the same for cooling (5.2.1, with external static pressure added to the list), for the crest factor of the UPS load (5.1.3.1), and for the photovoltaic test environment, which has to comply with IEC 61215 and IEC 61646 (5.1.3.3).

Not one of the eight says where a value has to land. They say report it, document it, keep it consistent. That is a discipline for the test certificate, not a threshold.

The band that is missing, and the words nearby

Across the twenty-four pages the word uncertainty never appears. Neither does reproducibility, nor confidence, nor error. The neighbouring words are there, and where they sit is what matters.

Repeatability appears exactly once, in clause 5.1.2, and not as a requirement but as a rationale: “For test repeatability and consistency, load conditions and test configurations are fixed and static.” Tolerance appears three times. One is in the body, clause 5.1.3, and it is a should attached to the load rather than to the result: “the load ranges and tolerance during testing should be restricted and documented” — restricted and documented, without saying within what. The other two, and all six occurrences of accuracy, sit in Appendices I and II.

Both appendices carry the same warning printed at the head: “This appendix does not form an integral part of this Recommendation.” They are written in the past tense, as an account of tests already performed. And ISO/IEC 17025 sits in the bibliography, among the informative references.

The recommendation tells you how to calculate efficiency and what to write in the certificate; it does not give you the number against which to decide whether the unit passes. In a dispute, that number is one you had to write yourself, in advance.

The load points are an example, not a rule

The same goes for the test conditions. The load points 25%, 50%, 75% and 100% appear exactly once in the whole document: in Table I.1 of Appendix I, as an example on a 100 kVA UPS. And that table demonstrates the problem better than any argument: the same unit yields 95.78% at 50% linear load and 93.02% at 25% with a non-linear load at crest factor 3. Almost three points of spread, every one of them compliant with L.1320.

The appendices are also where you find the test temperature (25 °C ± 3 °C for power feeding, with the unit settled for at least three hours), humidity between 30% and 75%, pressure between 81,200 and 102,000 Pa, supply within ±1% on voltage and frequency, at least one hundred evenly spaced readings per test cycle, instruments calibrated within ±1% by a national metrology institute, and for cooling 24 °C ±1 dry bulb in the room against 35 °C ±1 outdoors. Excellent figures. Not one of them mandatory.

The body concedes as much, in a note to clause 5.1.2 that contrasts a test run at room temperature with a facility operated at 35 °C: “Therefore test conditions must be carefully set.” Appendix II is blunter still: “If the test conditions are different, the result will be completely different.”

What it does not cover, and what to write

This is not a superseded text: it is in force and its formulas hold. But it dates from March 2014, and among the three cooling categories it lists there is no liquid cooling, neither direct-to-chip nor immersion; there is no partial load profile for modular UPS units; and by its own declared scope it does not reach room level. Read it for what it is: a way of calculating a ratio, not an acceptance protocol — the same distinction that separates a redundancy calculation from a redundancy test.

In a specification, then, the line efficiency measured in accordance with ITU-T L.1320 has to be completed, and the items are few: the formula chosen, total or sensible; the load points and the load type, with the crest factor; test temperature and humidity with their tolerances and the settling time; test duration and number of readings; the class and calibration certificate of the instruments; and the band within which the measured value is accepted. The first five can be copied out of the appendices, provided you write them as your own clause and not as an obligation imposed by the recommendation. The last one it will not give you under any circumstances.

The point

A guaranteed efficiency with no acceptance band is not a guarantee: it is a sentence that defends itself in a dispute. In the compliance specifications we deliver, that line arrives complete — method, load points, conditions, instruments and band — and the same line becomes the running order of the certificate signed at acceptance, with measurements and documentation handed over together with the works. Before the test, not after. On the other axis, efficiency is not a commissioning figure: it is a curve that shifts with load and with the season. Rectifier readings, chiller consumption, supply and return temperatures, UPS alarms and the as-built record of the room come together into one map of the plant and the network, where an AI recognises the drift before it becomes an outage and our own team intervenes, with no subcontracting. Where an AI is needed on that data, it runs inside the client’s perimeter, on-premise or in a dedicated cloud with data centres in Italy, with CSIDIA, the other company in the group.

Do you have to sign off a test certificate on a guaranteed efficiency, or are you writing that clause right now? Talk to an engineer: the site survey is free of charge, and an acceptance band is written before the test, not after.