Electrical design

Cable sizing to IEC 60502-1 and IEC 60364 in a data center

Four constraints size a feeder, and the largest of the four wins. On a data center route the one that usually wins is not the one people compute first.

The four constraints

A conductor is acceptable only when it satisfies all four. Sizing means computing each and taking the largest.

Start from the design current, not the connected load

The design current is what the circuit is expected to carry in service, after diversity, and before any derating. For a three-phase feeder:

Data center loads are unusual in two ways that matter here. They are close to constant, so diversity is far lower than in a commercial building. And the UPS rectifier front end and DC plant are non-linear, so the current is not the clean sinusoid the simple calculation assumes.

Derating: the factors stack, and on a data center route they stack hard

The tabulated current capacity in IEC 60364-5-52 assumes reference conditions. Real routes are not reference conditions, and the correction factors multiply.

• Grouping. The dominant factor. Nine circuits bunched on one tray can pull capacity to around 70% of the single-circuit value. Data center containment is congested by definition: A and B routes, every PDU feeder, and the mechanical circuits often sharing the same containment.
• Ambient temperature. Tabulated at 30 °C in air. A ceiling void above a hot aisle, or a route through a plant room, is not at 30 °C. Above a containment roof it can be 45 °C or more.
• Installation method. Free air on a perforated tray, in trunking, buried, in a duct — each has its own table. The method assumed on paper is often not the method the contractor uses.
• Soil thermal resistivity for buried sections between buildings, which on a campus is frequently taken from a default rather than a site test.

Voltage drop, and why it governs more often than expected

Long horizontal runs are normal in a large hall — a feeder from the electrical room to a remote PDU can be well over a hundred metres. The limits usually applied are around 3% for lighting and 5% for power, taken from the origin of the installation, and IEC 60364-5-52 Annex G gives the customary values.

The point people miss is that voltage drop is proportional to both current and length, so it scales with the very things a data center maximises: high sustained current, long routes. On any feeder past about 80 metres, compute the drop before sizing on current capacity — it will often be the constraint that wins, and it is cheaper to discover it now than after the cable is ordered.

Fault withstand near the source

Close to a transformer or a large UPS the prospective fault current is high and the conductor has to survive it for as long as the protection takes to clear. The adiabatic check is the standard one:

This is the constraint that most often governs the earth conductor rather than the phases, and it is the one that quietly fails when a protective device is changed late in the project without the cable schedule being revisited.

Where cable schedules go wrong in practice

1. Sized on connected load rather than design current. Oversized, expensive, and harder to install — wrong in the other direction, but still wrong.
2. Grouping factor omitted because the route was not known at the time of sizing, and never revisited when it was.
3. Ambient taken as 30 °C for a route above a hot aisle containment.
4. Voltage drop computed at nominal voltage rather than at the lowest expected supply voltage.
5. The schedule and the single line drift apart. The diagram is revised, a board is moved, a feeder is re-terminated — and the schedule is a separate spreadsheet that nobody updated.

The last one causes the most trouble on site, and it is purely a process failure. A schedule generated from the diagram cannot disagree with it.