Power systems · protection and arc flash
Fault, cleared
A fault at any board of a 2N data-centre power train we modelled: which breaker clears it, and how far the arc-flash boundary reaches on utility supply and on generator.
Overview
We modelled one side of a reference 2N data-centre power train: a 10 kV intake, a 2.5 MVA transformer, a 2.5 MVA standby generator, the main LV switchboard, a distribution board (board B) and a sub-board. We put a fault at each board. Short-circuit currents follow the IEC 60909-0:2016 method. Every protective device on the path runs a generic characteristic with its tolerance band: long-time, short-time and instantaneous for the LV breakers, standard-inverse for the MV relays. Incident energy is estimated following the IEEE 1584-2018 model; it is not a PPE study.
On generator, the fault current does not stay put. It decays the way an alternator’s does, from 23.4 kA towards the roughly three times rated current its excitation can hold. So the trip units see a falling current, and we step through the fault half a millisecond at a time: trip progress, arcing current and incident energy are all integrated as the current falls. At board B the generator’s fault current is half the utility’s, and its arcing current (15.7 kA at most) stays below the feeder breaker’s instantaneous trip, so the breaker waits out its 0.3 s short-time delay: 320 ms instead of 50 ms. The incident energy at the working distance rises from 2.2 to 4.7 cal/cm², and the boundary moves out from 0.89 m to 1.44 m.
The result depends on where the instantaneous setting sits, and we say so. With it set higher, or switched off, the utility case also clears on short-time and becomes the worse one. At the main board, utility is the worse case anyway. A maintenance setting brings board B back to 50 ms and 0.54 m on generator. While it is on, though, a fault at the sub-board can trip the feeder before the sub-board’s own breaker clears: that is the price of the faster trip.
What we built
- A short-circuit model of the power train after the IEC 60909-0:2016 method, in pandapower and in our own impedance code. The two agree to within 1e-13 kA, and pandapower reproduces the IEC TR 60909-4 reference network to within 0.1 A.
- An alternator decrement model (subtransient, transient, then current held by excitation support), compared with the IEC 60909-0 breaking-current factor μ.
- Our own trip units: generic long-time, short-time and instantaneous elements for LV air and moulded-case breakers, and IEC standard-inverse MV relays. Each integrates its progress over a time-varying current, with fast and slow tolerance edges, and pairs are checked for selectivity on bolted and on arcing current.
- Our own implementation of the IEEE 1584-2018 model, integrated over the decaying arcing current for both the full and the reduced arcing current. We checked it against the published Annex D worked examples and 105,120 cases from the IEEE calculator, and cross-checked it against the open-source arcflash library.
- A time-dilated 3D room and an interactive view: pick a board, a supply mode and the maintenance setting, replay the fault, and open its time–current bands.
Tech stack
Python (pandapower, NumPy) for the short-circuit, protection and incident-energy models; three.js for the room; Canvas 2D for the interactive view.
Numbers
| Number | How we measured it |
|---|---|
| 45.7 kA → 23.4 kA | Initial bolted fault current at board B, on utility supply and on generator, after the IEC 60909-0:2016 method (c = 1.10). The generator current falls to 11.8 kA by the time the breaker clears. |
| 50 ms → 320 ms | Clearing time at board B, from our trip-unit model integrated over the arcing current, using the slow edge of each tolerance band. |
| 2.2 → 4.7 cal/cm² | Incident energy at 610 mm from board B, following the IEEE 1584-2018 model, integrated in 0.5 ms steps. VCB electrode configuration, 32 mm gap, 508 mm enclosure. An estimate, not a PPE study. |
| 0.89 m → 1.44 m | Arc-flash boundary (1.2 cal/cm²) at board B, utility vs generator, from the same integration. |
| 0.54 m | Boundary at board B on generator with the maintenance setting on (feeder trips instantly above 3.2 kA, 50 ms). |
| 20 ms vs 25 ms | With the maintenance setting on, the feeder breaker’s fastest trip vs the sub-board breaker’s slowest clearing, for a fault at the sub-board. The feeder can trip first, so selectivity is lost. |
| 19.2 kA | The feeder’s instantaneous setting (12 × 1600 A, ±10 %). The headline holds while it sits between the generator’s arcing current (at most 15.7 kA) and the utility’s (at least 22.6 kA). In our sensitivity sweep, setting it to 15 × In or off reverses the result. So does a weaker grid (100 MVA) or an 80 m feeder, which pull the utility’s arcing current below the setting. |
| 105,120 cases | Cases from the IEEE 1584-2018 calculator spreadsheet (as collected in the open-source arcflash repository), all reproduced by our implementation to within 0.0003 %. |
| 0.1 A | Largest deviation of pandapower’s Ik'' from the IEC TR 60909-4 reference values (10 buses), using the reference values in pandapower’s own test file. |
| 15–23 % below μ | Our alternator decrement at 20–250 ms, compared with IEC 60909-0’s breaking-current factor μ for the same machine. μ is a conservative envelope; with time constants fitted to it instead, the generator case at board B rises to 5.7 cal/cm². |
| 22 integrations in about 3 s | 11 energised fault states (board × supply × maintenance setting), each integrated for the full and the reduced arcing current at 0.5 ms steps, plus selectivity checks and the two short-circuit runs: 2.2–3.5 s per full run on an Apple M1 Pro laptop shared with other jobs. |
Stills
Try it live
An interactive view that runs in your browser. Nothing loads until you open it.
