RCDs for Data Centers and IT Equipment: High Leakage

Why do standard RCDs misbehave in data centers and IT equipment rooms? Every switch-mode power supply, EMI filter, and VFD-fed cooling unit in a server room bleeds a small steady-state leakage current to earth through its filter capacitors, commonly 1-3.5 mA per unit under IEC 62368-1 limits, and that current adds up across a rack or a PDU. Populate one final circuit with forty servers and the cumulative leakage alone can sit at 40-140 mA, past a 30 mA personal-protection threshold before any actual fault exists, which is why a single blanket RCD across an IT distribution board trips on load, not on a fault. This article covers cumulative leakage sizing, earthing system choice for server rooms, RCD type and sensitivity selection for racks and UPS output, selective discrimination across PDU tiers, and where leakage monitoring replaces hard tripping on dual-fed critical loads.

Where the Leakage Current Actually Comes From

Three sources dominate. First, the EMI/RFI filter on every switch-mode PSU has Y-capacitors bridging line to earth by design, each contributing a fixed leakage figure the manufacturer publishes in the datasheet. Second, three-phase UPS rectifiers and VFD-driven CRAC or CRAH fan motors add pulsating or, on some topologies, smooth DC-like residual content on top of the AC leakage. Third, long runs of shielded data and power cable in a raised floor add distributed capacitive leakage that scales with cable length, not equipment count.

What we see in the field: PDU circuits sized for 30 mA out of habit trip within days of populating a rack, not because of an insulation fault but because nobody added up the per-unit datasheet leakage figures before commissioning.

Sizing the RCD Around Cumulative Leakage

Vendor guidance for avoiding nuisance tripping is to keep IΔ_total under roughly a third of the rated IΔn, since capacitor aging and ambient humidity push published leakage figures upward over the equipment's service life. A rack drawing 60 mA of steady leakage has no headroom left on a 30 mA RCD and will trip the first humid week or the first PSU added to the rack.

TN-S vs IT Earthing for Server Rooms

On a TT system, high earth-loop impedance means overcurrent devices alone cannot clear an earth fault fast enough, so an RCD is essential. Most data centers run TN-S from the sub-distribution board down specifically to keep the leakage path clean and avoid combined PEN conductor noise reaching sensitive IT loads; the MCB or MCCB clears line-to-earth faults, but 30 mA RCDs are still required on socket circuits and maintenance outlets for additional protection. Some critical process areas use IT earthing instead, where the first insulation fault is monitored rather than tripped, and RCDs are applied only on sub-circuits downstream. For the full breakdown of which system needs what, see RCDs across TT, TN and IT earthing systems.

Breaker-level terminology also matters at this stage of design: whether the rack PDU uses a separate RCCB plus MCB or a combined RCBO changes both panel width and how faults are isolated per circuit. That distinction is covered in the MCB, RCCB, RCBO and RCD terminology comparison.

Matching RCD Type to the Load: AC, A, F or B

Type AC is unsuitable for IT loads outright — pulsating DC from a rectifier bridge can blind its detection coil, and most codes exclude it from electronic-load circuits entirely. Type A is the working default for standard server PSUs and single-phase UPS units. Type F adds coverage for single-phase VFD-fed loads, relevant to variable-speed CRAC fan controllers. Type B is required where three-phase VFD-driven chillers or DC-coupled battery/UPS systems can produce smooth DC residual current that a Type A device will not detect at all.

Siemens covers this range with the 5SV Type A/F RCCB and the 5SM3 Type B device for DC fault currents; ABB's F200 series extends to an F200 B variant; Schneider's Acti9 iID range includes B-rated units for the same reason. Browse the current stock across brands in the residual current devices collection. Type selection by waveform, including where SI (super-immunized) devices fit, is covered in RCD types AC, A, F, B and SI explained.

Personal Protection vs Fire Protection Sensitivity Tiers

Socket outlets and maintenance access points in a server room need 30 mA for additional protection against direct contact, matching the standard human-safety threshold used in most codes. Non-socket IT distribution feeding fixed racks does not need that low a threshold and benefits from sitting higher, at 100-300 mA, precisely because it removes personal-protection headroom pressure from a circuit that is carrying 40-140 mA of normal leakage anyway. The higher band still exists for a reason: leakage current at fire-protection levels can ignite insulation before it reaches a level that is dangerous to touch, so the circuit is never left unprotected, just protected at the sensitivity appropriate to its actual contact risk.

Sensitivity selection across the full 10 mA to 500 mA range, with the reasoning for each step, is in RCD sensitivity from 10 mA to 300 mA explained.

Selective Discrimination Across PDU and UPS Tiers

A data center's power train is layered: utility feed, UPS, main distribution board, PDU, rack. Put an RCD at more than one of those layers without coordination and an upstream device trips for a fault the downstream device should have cleared alone, taking down far more load than the fault warrants. The fix is time-graded S-type (selective) devices upstream, rated at roughly double the downstream IΔn and given a short intentional delay, so the local device always clears first. For feeders above the practical range of a fixed RCCB, a residual current relay with a separate toroid — the Siemens 5SM2 is a common choice at this tier — measures leakage on the busway or large feeder without being built into the breaker itself.

Leakage Monitoring Instead of Tripping on Dual-Fed Critical Loads

This depends on whether the load actually has redundant A+B feeds. Where it does, an uncoordinated trip on one cord is often worse than a slow, monitored leak, so some data centers run continuous insulation or leakage monitoring on the critical PDU output instead of a hard-tripping RCD, alarming staff to investigate rather than dropping the load. That is not a substitute for RCD protection upstream or on any socket and maintenance circuit — it is an added layer specifically for single points that cannot tolerate an unplanned trip. Recurring nuisance trips on non-critical circuits usually trace back to the cumulative leakage math above, not a real fault; the troubleshooting sequence for isolating which circuit is responsible is in why an RCD keeps tripping, causes and fixes.

Commissioning still matters regardless of architecture. The front test button on any device proves the trip mechanism only; a proper commissioning check uses an RCD tester to measure actual trip time at IΔn and at 5x IΔn, which is the only way to confirm a data center RCD will clear within the IEC 61008 time limit under real load, not just that it can trip at all.

Building the Rack-Level Protection Scheme

Start from the load, not the panel schedule. Add up published leakage per PSU and per VFD stage for the actual equipment going into a row, pick IΔn with margin for aging, choose Type A as the default and step up to F or B only where the load genuinely produces that waveform, and grade sensitivity so sockets get 30 mA while fixed PDU feeders sit higher. None of that requires exotic hardware — it requires doing the addition before the rack ships, not after the third nuisance trip. For a broader walkthrough of the full selection sequence, see the RCD protection guide, and for RCBO options where a combined device suits the PDU better than separate RCCB plus MCB, check the RCBOs collection.

Conclusion

Data center leakage is arithmetic, not fault-finding. Every PSU, filter, and VFD stage on the load list has a published leakage figure, and once that sum is compared against IΔn before commissioning, most of the nuisance-trip and under-protection failures in this article stop happening. Set sensitivity by contact risk, type by waveform, and grading by feeder tier, and the RCD scheme holds up as racks are added rather than degrading with every server delivered.