OCP 800 V DC fault protection: a 2026 ratings and acceptance checklist

What does acceptance of an OCP-based 800 V DC system require?

Start with five linked records: a fault-contribution matrix, device-duty register, ground-fault case register, precharge state diagram and vendor evidence checklist. Tie each to the installed one-line revision, equipment revision, operating mode and acceptance disposition. Accept when evidence covers the installed duty; hold when a source path, rating, required test or isolation function is missing.

OCP's DCF LVDC white paper v1.1.0 is the engineering basis used here, with protection in chapter 7 and operating sequences in chapter 10. Reconcile its recommendations with the adopted installation code, certification conditions and project requirements. NEMA's August 2020 NEC adoption record distinguishes adoption from effective dates: enter the permit's NEC edition and local amendments.

Product evidence establishes qualified applications; the study establishes duty; factory/site acceptance testing (FAT/SAT) establishes delivered behavior. This guide belongs to the data center field-guide hub.

Which voltage must the protection actually interrupt?

Record the highest applicable voltage at each device, including fault-state voltage to earth. An 800 V two-wire circuit and a bipolar ±400 V circuit both have 800 V between rails. Under OCP LVDC v1.1.0 chapter 6, a solid midpoint reference establishes nominal ±400 V to earth, while an unearthed system's healthy rail can approach the full 800 V to earth after a hard first earth fault.

Map nominal voltage, highest sustained voltage, temporary overvoltage, impulse and rail-to-earth stress to the supplied device's operational and insulation ratings. Obtain the approved pole connection, polarity and current-direction drawings. Adding individual pole voltages without the manufacturer's approved arrangement does not establish an interruption rating.

OCP v1.1.0 §5.7 identifies 840 V as the Current/OS U5 protection anchor, but describes proposed voltage-time curves and ongoing alignment concerning operation above 840 V and short-term withstand to 1,080 V. Enter the approved voltage-time envelope and its source in the register. OCP voltage-band discussion

For an IT earthing arrangement, meaning isolated from earth or connected through impedance, distinct from information technology, OCP §7.2.4 calls for individual-pole full-system-voltage breaking capability where operation continues after a first earth fault. Keep that case separate from a series-pole rail-to-rail rating. Use the rack DC architecture guide for voltage/current comparisons.

Fault-contribution matrix: who feeds which fault?

Trace each source to collector-bus, feeder and rack-input faults, then identify every device crossed by that current. The utility-to-rack power chain establishes interface boundaries; the matrix below adds capacitor and reverse-current paths. Its directions are engineering interpretations of the stated topology, based on OCP §7.2.1 and Schneider Electric White Paper 219, version 1.

Inventory capacitance, equivalent series resistance (ESR), equivalent series inductance (ESL), cable resistance/inductance, converter control delays and failure behavior. Separate capacitor discharge, controlled converter current and sustained storage contribution. A storage converter has capacitors on both ports; their voltages and paths determine which energy reaches the fault.

Build cases for ties open and closed, batteries charging and discharging, module loss and maintenance modes. Include maximum close-in interruption duty and minimum detectable current: remote or resistive faults, unavailable storage and the current-limited tail after capacitors discharge. A large initial peak does not prove the remaining fault reaches pickup.

Sum source currents on a common time axis at the device terminals. Noncoincident peaks cannot simply be added, and a capacitor downstream of a breaker can discharge locally without its current crossing that breaker. Preserve fault inception, pickup, trip command and actual interruption as separate timestamps.

Why can current-limited converters still produce a severe capacitor fault?

OCP's LVDC white paper v1.1.0 models an approximately 92 kA initial fault peak in its 800 V DC reference system, showing why converter current limiting does not remove capacitor-discharge duty. The model uses a 4 MW AC service, four 100 kW server branches and converter-connected 600 V batteries; initial fault-current rise is about 1 kA/µs. OCP chapter 7, pages 113 to 116

In OCP's LVDC v1.1.0 reference simulation, a healthy feeder falls below 90% of nominal voltage about 0.26 ms after a server fault, so selective clearing and load ride-through must be assessed together. The same simulation falls below 50% at about 0.56 ms. Establish the installed load's voltage-time envelope separately.

Stored joules, peak amperes and the current-squared time integral describe different stresses. OCP's sustained-current narrative and Figure 7.6 caption disagree, giving about 85 kA and 41 kA respectively; neither is used here as a selection rating.

Anvilfield estimates use OCP Table 7.1 inputs. Divide voltage by inductance for the initial branch ramp; the complete network determines the faulted-server peak.

Device-duty register: what must each supplier prove?

Give every protective and switching function an ID that matches the one-line. Record normal and reverse current, source mode, grounding, environmental derating, voltage envelope, fault waveform, clearing interval, isolation and backup.

OCP LVDC v1.1.0 §7.2.5 and ABB's November 2025 DC application paper distinguish device interruption from the duty of the surrounding equipment. A breaker rating does not independently qualify terminals, connectors, bus bracing or the assembly's short-circuit current rating (SCCR). Use the available fault current and SCCR guide and interrupting-rating glossary for definitions.

Record operational voltage Ue, insulation voltage Ui and impulse withstand Uimp separately. For applicable IEC devices, distinguish ultimate/service breaking capacities Icu/Ics, making capacity and short-time withstand with its stated duration; retain UL interrupting ratings and assembly SCCR under their own qualification basis. IEC 60947-2:2024 and IEC 60947-10:2026 define product scopes, while the selected manufacturer's documentation establishes the supplied model's values.

Schneider Electric's Electrical Installation Guide distinguishes switching from isolation under IEC 60947-3. Specify a switch-disconnector for load interruption; an isolation-only disconnector needs an interlock preventing opening under load. Neither function establishes short-circuit interruption without the corresponding rating.

For capacitor-fed duty, retain peak, rate of rise, pulse duration, I²t, reverse current and recovery voltage. For a converter-limited tail, retain detection and clearing evidence at the minimum relevant fault current. Match the qualified DC time constant or transient envelope, utilization duty, cooling, ambient temperature and altitude to the installed conditions.

Anvilfield installed-device record, one per controlling case: ID | one-line revision | model/serial/firmware | source mode | grounding/poles | voltage | load/derating | forward/reverse waveform file | peak/di-dt/I²t/tail | interruption/making/withstand/isolation | clearing limit | backup | certification/report revision | as-left settings | capture ID | pass/hold/reason | approver/date. Unsubstantiated duty receives a hold disposition.

Which protection technology fits the actual waveform?

Select protection by its qualified waveform and total response. OCP v1.1.0 pages 125 to 127 describe mechanical clearing from about 1 ms for small miniature breakers to a few tens of milliseconds for power breakers, semiconductor switching in microseconds and semiconductor hybrids in a few hundred microseconds.

IEC 60947-10:2026 covers semiconductor and semiconductor hybrid circuit breakers up to 1,500 V DC and includes mechanical isolation contacts in series with their semiconductor switching paths. Electronic turnoff and mechanical isolation must each appear in the functional record. IEC 60947-10 public scope

A fuse needs sufficient fault energy to melt and a qualified interruption envelope after melting. Compare pre-arcing and total-clearing I²t, minimum breaking current, repetitive inrush and the complete pulse-plus-tail waveform. Minimum breaking current is the lower boundary of the specified interruption capability, rather than a generic pickup setting. An unqualified low-current tail requires complementary protection or a changed device/source combination.

Mersen's automotive fuse examples show how application evidence can conflict; they do not establish data center branch-circuit approval. Its MEV100C10-S item page states 20 kA in its description, while the linked 2025 MEV100 datasheet gives 50 kA in the table. Require written, model-specific application evidence resolving the Mersen and ABB conflicts below. Include spare fuses, post-fault inspection, auxiliary power, cooling and firmware support in the lifecycle scope.

How do you prove selectivity without losing healthy racks?

Compare detection, decision, interruption and recovery with the healthy load's voltage-time tolerance and local hold-up. Selective coordination means limiting the outage through protective-device selection and operation; coordination study records need the DC source and waveform cases established here.

Healthy rack capacitors and distributed storage can feed a remote fault in reverse through a healthy branch. Under OCP §7.2.2, reverse blocking, fast protection, local energy storage and independent feeders provide different continuity strategies. Test the strategy actually installed at normal and light load, with specified storage modes, a failed blocking element and an unavailable redundant path.

Retain synchronized faulted-device current, healthy-feeder reverse current, healthy-bus voltage, trip command, interruption indication and recovery. Identify the upstream action if the branch fails to clear, including converter/storage blocking, exposed equipment, total clearing interval and resulting outage boundary.

Distinguish an electrical fault from converter instability using current, voltage and converter status on the same time axis. OCP §§10.6.9 and 10.9 include impedance and small-signal stability checks; verify permitted tie and load-admission combinations.

IEEE 1584-2018 covers three-phase AC calculations from 208 V to 15 kV and excludes DC calculations. Record a separate DC method, enclosure, working distance, time-dependent sources and clearing assumptions, as discussed in Schneider White Paper 219. The arc-flash study and labels guide supports record organization; retain the separate calculation basis for this bus. IEEE 1584-2018 catalog

Ground-fault cases: what changes with the grounding topology?

Separate protective earth (PE) continuity from the DC system's intentional reference to earth. The equipment grounding conductor provides a protective path; ground resistance and bonding tests do not establish DC first-fault detection. Identify which current-carrying conductors each disconnect opens while retaining protective continuity.

OCP v1.1.0 chapter 6 describes the rail-voltage shift after a first fault and the possible two-feeder PE loop after a second fault on the opposite rail. Continued first-fault operation needs an engineered monitoring and response state. OCP §6.4 identifies cumulative filter Y-capacitance as a touch-current and scaling constraint: inventory capacitance to earth across connected racks and converters. An insulation monitoring device (IMD) alarm does not limit that initial discharge; test both fault polarities and the approved response to monitor failure.

Bender's June 2026 iso685-x-B manual lists a general DC monitoring range up to 1,000 V but limits its stated UL-application range to 600 V, so the general voltage rating alone does not qualify an 800 V UL application. The manual's adjustable response range is 1 kΩ to 10 MΩ; its typical 4 s response is profile-dependent at RF equal to half a 10 kΩ setting and 1 µF system capacitance. Bender manual revision 12, page 93

Its factory alarms are 40 kΩ and 10 kΩ; profile-dependent capacitance capability reaches 1,000 µF. Enter the selected settings, delay, profile, measured capacitance and observed response; defaults do not establish project acceptance. OCP page 96 advises one actively measuring IMD per galvanically connected IT system; coupled systems need manufacturer coordination.

Continuous monitoring of insulation resistance and an offline insulation resistance test answer different questions. OCP §7.2.4 identifies unresolved IEC/NEC differences for high-impedance arrangements; the project's adopted-code interpretation and approved pole drawing must resolve the installation. Follow connected electronics' insulation-test restrictions.

Voltages are ideal nominal circuit relationships derived from OCP chapter 6, not allowable touch voltages. Test the actual alarm/trip action, loss of reference and monitoring failure.

Precharge state diagram: what must happen before the main path closes?

Close the main path only when measured voltage differential and actual device feedback satisfy the approved sequence of operations (SOO). TI's March 2026 TIDA-050082 design guide specifies active buck precharge of 2 mF from zero to 800 V within 400 ms. Its passive RC discussion reaches about 99.3% after five time constants; neither figure establishes a universal closing permissive.

Prove topology, work clearance, healthy monitoring, valid voltage sensing, source envelope and precharge thermal capacity before starting. Keep load admission blocked throughout precharge.

Use the approved model-specific closing differential, timeout, current limit, reset and thermal retry settings. Test an open or bypassed precharge path, welded main contact, stuck auxiliary indication, sensor disagreement, timeout, auxiliary-power loss and command-channel loss. A failed ramp prevents Run, and failed isolation or discharge enters LatchedHold without automatic retry.

Confirm that the precharge branch is bypassed or removed as designed before load admission. A failed bypass must not leave its resistor carrying rack load. On power or communications loss, verify the specified hardware fail state; a software state change alone does not establish isolation.

Anvilfield state diagram, based on OCP chapter 10 and TI TIDUFH5:

Blocked
  | work clear, permissives proven
  v
Ready -- permissive lost --> Blocked
  | source valid, load admission blocked
  v
Precharging -- timeout / abnormal ramp / feedback error --> FaultLatched
  | acceptable closing differential + valid feedback               |
  v                                                               | block sources
Run -- sensor / control shutdown fault ---------------------> FaultLatched
  | stop / protection                                             |
  v                                                               v
Isolating <--- controlled abort from Precharging --------------------+
  | isolation confirmed, backfeed blocked
  |---- isolation timeout / welded device ----> LatchedHold
  v
Discharging
  |---- decay failure / invalid sensing / rebound ----> LatchedHold
  | required decay, no rebound measured
  v
MeasuredDischarged -- authorized restart gates --> Blocked
  | lockout/tagout + independent absence-of-voltage verification
  v
WorkRelease -- work released, return-to-service checks --> Blocked

LatchedHold: no automatic retry; authorized fault recovery required.

Discharge evidence: when is the bus ready for physical work?

Release physical work after isolation, lockout/tagout (LOTO) and independent absence-of-voltage verification under the approved procedure. OCP v1.1.0 §10.8 calls for monitored decay, rebound detection, discharge-device temperature supervision and fault lockout.

OSHA 29 CFR 1910.333(b)(2)(ii)(B)-(C) requires disconnection from every electrical source and release of hazardous stored electrical energy. Capacitors must be discharged; high-capacitance elements must be short-circuited and grounded where stored energy could endanger personnel, using the approved procedure and treating handled equipment as energized. For circuits over 600 V nominal, §1910.333(b)(2)(iv)(B) requires checking the test equipment immediately before and after verification. Qualified personnel implement these requirements through the approved isolation procedure. OSHA 1910.333, 2025 CFR edition

Enter the applicable product safety standard, manufacturer's discharge threshold and maximum time in the acceptance record. OCP §9.5.3.4 refers those criteria to applicable safety standards. Keep product accessible-part compliance, completion of monitored decay and authorization for physical work as separate decisions.

Capture rail-to-rail and both rail-to-PE voltages at identified test points alongside source-disconnect feedback, backfeed isolation, discharge command and temperature. Test at the approved maximum initial voltage/capacitance and limiting temperature/control-power state. Failed discharge paths, invalid sensors, power loss and rebound must retain lockout; timer expiry cannot replace a measured state.

Worked example: an 800 V, 2 mF DC link

Using Texas Instruments' March 2026 test inputs of 800 V, 2 mF and a 20 kΩ discharge path, Anvilfield estimates 640 J of initial stored energy and about 111 seconds to reach a 50 V calculation checkpoint. TI forms its bank from four parallel 500 µF, 2 kV capacitors. TI TIDUFH5, sections 2 and 3

Anvilfield estimates below use half the capacitance times voltage squared for energy and charge divided by ramp time for average capacitor current. Multiply resistance by capacitance for the time constant, then multiply by the natural logarithm of the initial/final voltage ratio for discharge time. Passive precharge assumes 40 Ω, an uncharged bank, fixed source, disabled load and inactive discharge.

Method: E = CV²/2; Iavg = CΔV/t; τ = RC; td = RC ln(V0/Vf); Vc = Vs(1-exp(-t/RC)); P0 = V²/R. Discharge assumes isolated source/backfeed and a connected, fixed resistance.

Anvilfield selects a 120 s target to the 50 V calculation checkpoint, with 840 V, +20% capacitance and +5% resistance as tolerance bounds. Those bounds require about 142 s and leave about 78 V at 120 s: hold the 20 kΩ design.

Changing to a 15 kΩ, ±5% resistor reduces the bounding decay time to about 107 s. Specify at least the calculated 847 J energy duty and about 50 W initial power at the -5% resistance bound, then qualify temperature and repeat duty. Record the installed decay and failure tests; physical work still requires isolation and independent absence-of-voltage verification.

Vendor evidence checklist: what must FAT, SAT and turnover retain?

OCP v1.1.0 §10.9 calls for studies, simulation where practical, FAT/SAT and periodic proof testing. Anvilfield's acceptance approach separates laboratory interruption qualification from installed functional tests: approved injection or simulation proves only the sensors, logic and outputs exercised. Record each test boundary and require qualification reports for full-power interruption.

Assign model-specific reports to the original equipment manufacturer (OEM), study/settings reconciliation to the design engineer, test execution to the test agency, witnessing/deficiencies to the commissioning authority, and operational acceptance to the owner. The authority having jurisdiction addresses installation-code approval where required.

Retain raw timestamped captures, signed pass/hold scripts, one-line revision, serial numbers, firmware, as-left settings and instrument/calibration records. Connect the package to electrical commissioning and power QA and the BESS commissioning punch list wherever storage crosses the acceptance boundary.

For a 10 µs event, Anvilfield's planning criterion of ten samples within the event requires at least 1 Msample/s per channel. OCP §10.10.1 discusses 10 to 25 µs semiconductor-breaker clearing and MHz-level sampling.

Use sensors that capture the transient and sustained DC. PEM's Rogowski probe guidance confirms that these probes cannot measure DC; pair transient capture with a qualified DC measurement channel. Actual rise time controls bandwidth and sample rate; document probe common-mode/DC ratings, transient range, saturation, channel timing uncertainty, trigger/pretrigger windows and processing settings. A clipped or missing waveform is incomplete evidence.

Checklist synthesis: OCP LVDC v1.1.0 chapters 7 and 10, IEC product scopes, ABB, Schneider, TI, Bender and OSHA. Pass only against the approved installed-system criteria.

What scope, staffing and retest interval should the contract carry?

Anvilfield estimates 32 labor-hours for a bounded document review and functional-test package, plus eight labor-hours for one retest. The allocation below assumes ready documents/settings and one accessible test scope; vendor review cycles, outage access and laboratory availability determine elapsed calendar time.

Scope study/model reconciliation, qualification review, FAT witnessing, site scripts, recorder/probe rental, load-bank interfaces, travel and outage access separately. New multi-vendor models and full-power laboratory qualification need their own scope.

Estimators pursuing protection studies and acceptance testing get early signals of public electrical upgrades from budgets and capital plans through PorchRocket's Bid Control.

Method: add the stated engineering and technician hours; the base plus one retest totals 40 labor-hours. These are Anvilfield planning inputs dated November 5, 2026.

Test before acceptance and repeat affected checks after source, battery, capacitance, grounding, firmware, setting or protection changes under OCP §10.9. Record maintenance/proof-test intervals from equipment instructions and the site maintenance program, including the basis and next due date. Coordinate the wider package using the UPS and STS hold points.

Sources checked November 5, 2026

1. S1. Open Compute Project, DCF Low Voltage Direct Current Power Distribution, v1.1.0, 2026-09-02.
2. S2. IEC 60947-2:2024, edition 6.0, publication 2024-09-18.
3. S3. IEC 60947-3:2020+AMD1:2025 CSV, edition 4.1, publication 2025-04-25.
4. S4. IEC 60947-10:2026, edition 1.0, publication 2026-05-12.
5. S5. ABB, Protection Devices for Direct Current Applications, 9AKK108470A9606, revision B URL, 11/2025.
6. S6. Schneider Electric, DC Arc Flash Analysis: A Practical Study on 800 VDC AI Data Centers, White Paper 219, version 1.
7. S7. Mersen, EVpack-fuse MEV100, DS-DCBPMEV100-23-0524-EN, ©2025.
8. S8. Texas Instruments, TIDA-050082 High-Voltage, Positive Rail Active Precharge Reference Design, TIDUFH5, March 2026.
9. S9. Mersen, MEV100C10-S product page, undated current page.
10. S10. Bender, ISOMETER iso685(W)-x-B manual, iso685-x-B_D00177_12_M_XXEN/06.2026.
11. S11. OSHA, 29 CFR 1910.333, current web text and official 2025 CFR PDF.
12. S12. IEEE, IEEE 1584-2018 catalog. Publisher scope.
13. S13. UL Standards & Engagement, official UL 489I catalog listing.
14. S14. NEMA, Implementation of the National Electrical Code, August 1, 2020; adoption mechanism reference.
15. S15. Power Electronic Measurements (PEM), Rogowski probe Troubleshooting Guide, accessed November 5, 2026.
16. S16. Schneider Electric, Electrical Installation Guide, Elementary switching devices, accessed November 5, 2026.

Update log, November 5, 2026: Clarified disconnect duties, first-fault capacitance and test boundaries; completed the discharge redesign and revised staffing scope.

How to cite this page: Anvilfield, OCP 800 V DC fault protection: a 2026 ratings and acceptance checklist, 2026, https://anvilfield.com/field-guides/datacenter/data-center-800v-dc-fault-protection-2026/.

Related tools

FAQ

People also ask

• Why are data centers moving to DC power for AI racks?
• What is available fault current?
• How long does data center electrical commissioning take?