Data centre noise assessments that hold up at night.

Standby generators are the loudest items on most data centre sites and the least often heard. In a grid failure they run for as long as they must; the rest of the time they run for testing. Councils generally treat those two situations differently, and a report that assesses one without saying what it has done about the other leaves the officer to fill the gap.

For testing, the assessment needs the schedule the operator actually intends: how often each set runs, for how long, at what load, whether sets are tested singly or together, and at what time of day. Testing one set at a time on weekday daytimes is a very different proposition from a load-bank test of every set at once. We assess the schedule that will really be used, and where it does not work at the nearest homes we say which part of it to change.

That schedule often becomes the planning condition, so it is worth proposing one the operations team can keep. A testing window that looks tidy in a report and cannot be met in practice turns into a breach later. Generator noise itself comes from the exhaust, the radiator fan and the engine casing, and each needs its own treatment, but which way the sets face and where they sit is settled on the layout drawing long before the silencer schedule.

Transformer hum sits at 100 Hz, twice the 50 Hz mains frequency, with harmonics at 200 Hz, 300 Hz and above. Fans produce tones at their blade passing frequency, which moves as the fan speed changes. BS 4142 adds a correction of up to 6 dB for tonality when the rating level is built, so on these schemes the tonal judgement deserves as much care as the background survey.

We make that judgement from frequency analysis, of the plant itself or of comparable installed plant, rather than from a manufacturer's single A-weighted figure, which says nothing about tones. Plant that is audibly tonal but carries no correction in the report is where a consultee or an objector will look first, and the omission rarely stands up.

Low-frequency tones also change the mitigation. A barrier that does well against a fan's broadband noise does much less at 100 Hz, so a transformer is controlled mainly by distance, by specifying a lower-noise unit and, where that is not enough, by a full enclosure or close walls around it. That is a reason to settle the substation's position early.

Data centres go where grid capacity is, so they arrive in groups around the same substations. When another scheme near the same homes is consented or in the system, the council will want the combined effect, and the list of schemes to include is best agreed with the authority at the outset rather than chosen at the end.

The trap is a creeping background. If each new scheme is assessed against a background that already includes the last one, the area gets louder one acceptable result at a time. We report the scheme's own BS 4142 result. Alongside it we report the combined rating level of this scheme and the others agreed with the council, against a background that excludes them: measured before they ran, or derived where it can no longer be measured. The officer sees both without rebuilding the picture from several separate reports.

Where a larger campus is screened as EIA development and noise is scoped in, it becomes a chapter of the Environmental Statement, covering construction and operation, cumulative effects and significance. Our EIA noise page sets out how we write those chapters. The survey, the source model and the cumulative work described here carry straight into it, so nothing is measured twice.

Good data centre acoustics starts on the layout drawing, and the cheapest mitigation comes first. Layout comes first: where the plant yard sits, which way the dry coolers and generator discharges face, and whether the building itself screens the nearest homes. Plant selection is second: low-noise fan variants, larger fans turning more slowly and transformers specified for a lower sound power level cost less than treating a noisy choice afterwards. Then the hardware: attenuators on generator exhausts and air paths, acoustic louvres, enclosures and barriers, each stated as a performance the installed plant can be checked against.

Consent for these schemes usually carries an operational noise condition: a rating level limit at the nearest homes, a commissioning survey once the site is running, a noise management plan, or all three. The difficulty is that a well-designed site may sit close to the background at the homes, where its own contribution cannot be measured cleanly. We draft the condition with that in mind, naming a measurement position nearer the plant where the site's sound can be isolated and the level at the homes calculated from it, so compliance can be demonstrated rather than argued about.

The engineer who scopes the survey runs it, writes the report and answers the council's questions in writing. Where the quote says the Approval Guarantee applies and you adopt the approach the report recommends, a written technical objection from the council to the methodology, calculations or conclusions is met by revising the report at no additional cost until it is resolved, and if revision cannot resolve it the fee for that work is refunded. It guarantees the work, not the permission.