Why do a UPS and generator hunt? Voltage and frequency field checks

Why do a UPS and generator hunt together?

Hunting is repeated variation or repeated source rejection. A single voltage dip followed by a damped recovery is a transient response; a sequence that keeps rebuilding after apparent recovery needs a stability investigation. Electrical frequency variation and harmonic distortion of the voltage waveform are separate measurements.

Start with four branches: engine and governor response, alternator and automatic voltage regulator (AVR) response, UPS rectifier controls, and switching or filter actions. Cummins Power Topic 6014 (2007), pp. 2 and 7 through 11, explains why equipment that operates separately can interact when connected. Its SCR rectifier cases describe that converter technology; STAMFORD AGN025 Issue B distinguishes converter characteristics from inverter characteristics. Identify the installed rectifier generation before applying either discussion. [S1; S2]

Record UPS topology, normal versus ECO or bypass operation, module count, and the actual generator-mode contact state. In normal double-conversion operation, batteries can support the DC bus while the inverter continues supplying the load; rectifier rejection alone does not prove an output transfer. Record line-interactive transfers separately. Recover the original [generator acceptance evidence](/field-guides/datacenter/generator-acceptance-loadbank-turnover/) for comparison; use the [GPU transient test plan](/field-guides/datacenter/gpu-power-transient-test-plan/) when designing new workload disturbances.

Which control interaction matches the symptom?

Voltage movement with steady electrical frequency prioritizes excitation, sensing and reactive loading; frequency movement with real load prioritizes engine response and load application. These signatures select measurements, not a failed component.

A generator can respond correctly to a load that keeps changing. Cummins PT6014 documents a 200 kW generator and 125 kVA UPS interaction, but those ratings establish neither a sizing ratio nor a remedy for another installation. Distinguish input acceptance, inverter synchronization, bypass eligibility and actual transfer; a bypass alarm alone does not establish protected-output loss. [S1]

Locate the departure before tuning. If the generator bus stays steady while one UPS-input phase departs, compare the intervening feeder, transformer, contactor, connections and sensing path. Mark cooling starts and ATS contact changes on the same timeline.

What evidence should you preserve before a reset?

Export UPS histories, generator events and waveforms, ATS or switchgear events, EPMS trends and relevant BMS cooling events before cyclic logs overwrite them. Save model, serial number, firmware, AVR/governor configuration, rectifier and charger limits, qualification delays, ramp settings, bypass windows and filter commands. Preserve the original configuration files alongside the exports.

Use the [EPMS metering checks](/field-guides/datacenter/epms-power-monitoring-metering/) to validate channels and the [incident-management procedure](/field-guides/datacenter/data-center-incident-management-outage-response/) for evidence custody and operational handoff. The timeline below follows a rejection/restart branch; use the measured order for other mechanisms.

Each worksheet row needs the original timestamp, corrected relative time, timing uncertainty, value and unit, phase/channel, instrument or controller ID, file ID and setting revision. Keep the first reported alarm separate from the first electrical departure.

How do you establish which event came first?

Compare a common observable edge at all recorders and measure residual offset before and after the run. Include timezone, synchronization method, clock drift, scan or polling delay, averaging-window alignment and sensor/acquisition delay in the timing record. A shared UTC label does not establish event order.

Under Anvilfield’s timing method, event order remains unresolved when the separation is no greater than the sum of the two records’ total timing-error bounds. Preserve native waveforms, frequency-estimator settings, reject reasons and the corrected plot together.

Where and how should voltage and frequency be recorded?

Record both ends of the source path and the protected output. One recorder can represent electrically common points only when their common connection is established; separate rectifier and bypass feeders require distinct source records. Record all phases, true kW, kVA, signed kVAr, true and displacement power factor, frequency, voltage distortion, current harmonics and installed battery telemetry.

How should recorders be connected safely?

For US general-industry work, OSHA 29 CFR 1910.333(a), (b) and (c), July 2025 edition, requires deenergizing exposed live parts before work unless the employer establishes an allowed exception, isolation from all electrical sources, control of hazardous stored energy, lockout/tagout and qualified verification before treating parts as deenergized. Energized work requires qualified personnel and protective work practices. Identify utility, generator, bypass and battery feeds in the connection plan. [S12]

Fluke’s 1773/1775/1777 Safety Information, PN 5125091, Rev. 1 (October 2021), requires approved accessories with suitable category, voltage and current ratings; the lowest-rated component limits the assembly. It prohibits Ethernet or USB accessories where exposed hazardous live parts are accessible. Plan safe recorder connection/removal, closed-panel recording and a permitted timing/data connection before switching. [S13]

Fluke’s January 2022 specifications for the 1773/1775/1777 distinguish GPS timing accuracy below 1 ms absolute to UTC from NTP timing typically below 0.1 s, depending on network latency. Clock accuracy does not establish phase-angle comparability between recorders. Its frequency results use 10/12-cycle windows, about 200 ms at nominal 50/60 Hz; a precise timestamp does not make these instantaneous frequency measurements. [S7]

The same Fluke specification provides 80 kS/s voltage/current acquisition, DC-to-30 kHz PQ bandwidth, one-cycle voltage/current RMS refreshed each half-cycle, and 200 ms power/AUX extrema. Its triggered records retain 10 seconds of RMS and 10/12 cycles of waveform, about 0.2 seconds at the respective 50/60 Hz mains frequencies. The 1775/1777 high-speed transient function records voltage, not equivalent high-speed current. Verify exports and retrigger dead time before mobilizing. [S7]

What capture window should the field plan specify?

Anvilfield’s starting capture plan retains 30 seconds of pre-event trends and at least 120 seconds afterward at 1-second intervals with interval extrema, plus separate waveforms at each initiating edge, rejection and restart. Arm the trend session before switching. Extend it through every qualification, start, ramp and recharge delay; use verified continuous or repeated-event acquisition when short built-in waveform records leave gaps.

For subsecond hunting, acquire voltage, current and derived power fast enough to resolve each swing; the trend supplies context. Half-cycle RMS refresh calculates to 8.33 ms at 60 Hz and 10 ms at 50 Hz by taking half the reciprocal of line frequency. Select raw sampling, bandwidth and pretrigger duration for the suspected notch or switching event, and calculate Hz/s over a stated interval from valid frequency estimates.

As an Anvilfield diagnostic check, compare an apparent frequency jump with engine-speed telemetry and the raw voltage: notches, missing cycles or loss of tracking can corrupt the estimator. Do not calculate a slew rate across invalid samples or compare differently averaged frequency channels as though they were simultaneous. [S7]

Deliver one aligned plot with panels for generator and UPS-input per-phase RMS voltage, valid frequency, input kW and signed kVAr, UPS current/mode/reject reason, bypass permissive and protected-output voltage. Mark departure, rejection, recovery, requalification and restart with actual limits and timing uncertainty. Each panel needs readable units and ranges. IEC 61000-4-30:2025, Edition 4.0, corrected July 2026, defines measurement methods; its public scope supplies neither a UPS acceptance band nor a universal waveform sample rate. [S6]

Which voltage and frequency limits govern the diagnosis?

Use the installed UPS manual and exported settings, generator performance requirements and project operating criteria together. Record pickup/dropout delays and firmware revisions with each limit. A generic steady-state band cannot replace a model’s qualification logic.

Schneider Electric’s Galaxy VX technical specifications 990-5451L-001 separate rectifier input, bypass and output tracking limits for the 1500 kW UPS with 1500 kW I/O cabinet; use the 480 V column below. [S9]

The sizing advice in Schneider’s small-UPS FAQ has a different product scope from the Cummins and STAMFORD compatibility guidance. Resolve the installed converter characteristics and source response in the generator sizing study; a multiplier cannot diagnose an existing oscillation.

Why does bypass remain unavailable when the rectifier is online?

The inverter can remain DC-fed while bypass synchronization is unavailable. Plot bypass voltage, frequency, phase/sync status and configured tracking separately, using a separate source ID where supplied separately. Define whether bypass is required continuously or in designated recovery states under the approved [UPS/STS commissioning hold points](/field-guides/datacenter/ups-sts-commissioning-hold-points/). Widening a window changes transfer behavior; it does not prove that source instability was corrected. [S1, p. 11]

How does AVR and rectifier interaction cause hunting?

Distorted source voltage can affect regulator sensing and rectifier operation, while constant-power behavior increases current as voltage falls. Overlay bus voltage, UPS input current/kW, excitation command or available field-current telemetry, frequency and filter status. Obtain excitation channels from controller exports or installed isolated transducers. [S1; S2]

Keep THDv and THDi separate, and retain absolute current and load state with distortion percentages. A higher THDi percentage at light load does not alone establish more disturbing current. Compare waveform notching as well as aggregate distortion; the [harmonics investigation guide](/field-guides/electrical/harmonics-power-quality-mitigation/) covers broader mitigation studies.

Cummins PT6014 (2007), p. 8, explains why voltage reduction need not unload a constant-power UPS: input current rises while real demand remains. Review the AVR’s underfrequency voltage roll-off with the measured current-limit behavior; a motor-oriented response can worsen this interaction. [S1]

Match the correction to the evidence: excitation/sensing review for voltage-led oscillation; ramp, charging or generator-mode coordination for load-led cycling; filter/capability review for light-load behavior; and load-share repair for per-set exchanges. If the accepted linear-load baseline also hunts, investigate fuel supply, actuator and sensing faults before attributing the problem to the UPS.

An OEM-controlled change needs a named parameter, allowed range, saved original value and rollback trigger. Change one relevant variable and rerun the same state. Cummins PT6014 (2007), pp. 9 and 10, warns that damping for UPS compatibility can worsen other-load response, so repeat affected motor-start and transfer checks. Permanent preload or filter changes require a load, losses and capability review. [S1]

Can rectifier walk-in or recharge make the generator hunt?

Cummins Power Topic 6014 (2007) describes a rectifier restart loop in which UPS loading depresses generator frequency, input rejection removes the load, and frequency recovery permits the UPS to reconnect. The repeated load cycle can sound like governor hunting even when the governor is responding to the commanded load. [S1, p. 11]

Record commanded and actual ramp, start delay, input-current limit, generator-mode input, charger command and battery current. Separate inverter load, recharge power and conversion losses. A delay postpones load; a walk-in rate changes its slope; a current limit caps its magnitude.

Verify the on-generator dry contact at the UPS and the setting it actually selects. TSEWG TP-19 (October 2, 2017), p. 20, calls for coordinating that signal with reduced battery charging. A correct charger setting is ineffective if the source-state signal is absent or incorrectly mapped. [S5]

Capture aggregate demand when several UPS modules restart together. Their individual ramps can overlap into a larger source demand change. Compare recharge inhibited and enabled only where supported and approved, preserving battery state of charge and temperature with each run; compare the measured ramp with the configured ramp instead of assuming the command was delivered.

Why does hunting depend on load level or generator lineup?

The Tri-Service Electrical Working Group’s TP-19 (October 2, 2017) identifies passive UPS input filters as a possible source of leading power factor below 40% loading, making the actual filter state part of a light-load stability investigation. Use that condition to select a comparison, not to impose a threshold on every active-front-end UPS. [S5]

True [power factor](/field-guides/glossary/power-factor/) is kW divided by kVA; displacement power factor describes the fundamental voltage/current phase relationship. Record the meter’s import/export and leading/lagging conventions, and verify current-probe polarity with a known load. A low PF magnitude alone cannot identify capacitive loading. Capture actual kW, signed kVAr and connected filters against the alternator’s own operating chart. [S3; S4]

For [parallel generators](/field-guides/datacenter/generator-paralleling-switchgear/), record each set’s kW and kVAr as well as the common bus: a stable total can conceal units exchanging load. Identify unit count, breaker states, droop or isochronous control, load-share status and islanded versus utility-parallel operation. STAMFORD AGN045 Issue B treats mains-parallel operation separately. Record whether the disturbance begins at islanding or reconnection and which control mode takes over at that transition. [S10]

Which load-state comparisons should you run?

Start with the event records and accepted generator baseline. Run controlled comparisons within the authorized [switching and recovery procedure](/field-guides/datacenter/data-center-mop-sop-eop-procedures/), holding source lineup, load composition and settings fixed except for the variable under investigation. Record actual kW/kVAr, temperatures and battery condition for each state.

Anvilfield’s starting observation plan holds each permitted steady state for 5 minutes after programmed transitions finish, extending through slower timers or thermal behavior. This is a diagnostic hold, not a maintenance-test duration. Compare failed and corrected traces at the same load, lineup, ramp and relevant charger/temperature conditions.

Add affected motor/cooling transitions and the actual multi-UPS restart sequence to the corrected-state run. Bypass comparisons require an approved operating plan that protects the load. Stop at the preapproved source, protected-load, equipment or battery-reserve limit, on loss of the required recovery path, or on a developing equipment alarm; retain the evidence and execute rollback. Mark unperformed states and the resulting operating restriction.

Worked examples: how do the numbers change the diagnosis?

The worked checks use a balanced 480 V, 60 Hz path. The rectifier AC demand already includes conversion losses; recharge is added separately. Anvilfield estimates the current and ramp results while holding real power and true PF fixed until current limiting or another operating state changes them.

Method: divide real input watts by the product of the square root of three, line-to-line volts and true PF: I = P / (√3 × VLL × PF). Divide applied kW by ramp seconds for average slope and compare currents before rounding for percentage change. Linear ramp shape is a planning input; measure the delivered ramp.

What does the leading-power-factor operating chart change?

Spare nameplate kW does not establish margin against leading reactive load. STAMFORD’s AGN004 Issue E compares usable alternator output in both power-factor directions in the published chart example below. [S3]

Method: multiply the base rating by the relevant per-unit chart ordinate: output quantity = 670 × chart factor. For the estimated reduction, subtract leading kVA from lagging kVA and divide by lagging kVA. The independently rounded chart readings remain tied to their PF direction. Apply the installed alternator’s chart to the measured filter/load state.

How do you scope, schedule and price the investigation?

Anvilfield estimates $7,640 and about 3 working days of active sequential preparation, fieldwork and reporting for the two-feeder scope below, including one 8-hour field window. One additional field repeat adds $5,120 before added analysis. The estimate uses selected November 2026 planning rates and allowances.

Method: sum the stated labor products and allowances. The 38 labor-hours include concurrent field roles; 6 preparation hours, 8 elapsed field hours and 8 reporting hours total 22 active hours, rounded up to 3 eight-hour workdays. OEM scheduling, access and battery recovery determine calendar duration. The stated rates and allowances are Anvilfield planning inputs.

The allowance provides one recorder at each of two independent three-phase voltage/current locations. Separate bypass/output buses and parallel sets need additional instruments or validated installed channels, all qualified for timing and export. Load-bank hire, temporary cables, fuel, permits, taxes, overnight premiums and repairs are separate; travel and recorder rental duration also change the allowance.

Assign generator OEM tuning, UPS OEM settings, qualified electrical installation/switching, owner operations and analyst/commissioning review explicitly. The owner controls the operating window and recovery plan; each OEM controls its proprietary tuning instructions. Put additional runs and evidence deliverables in the [commissioning subcontract and retest scope](/field-guides/datacenter/data-center-commissioning-subcontract-scope-retests/).

A repeatable stability-investigation service needs a defined capture package, controls attendance and a retest allowance for each site. Service managers building recurring generator investigation programs start with named accounts: PorchRocket’s Asset Search finds generator assets visible from above and returns the owner of record for every find.

What permits restoration to normal service?

Restore normal operation when the original troublesome load state and relevant initiating transition pass after the documented correction. Anvilfield’s starting method requires two consecutive successful repeats with the required battery reserve restored before each; increase repetition when the failure is intermittent. Required bypass and transfer functions must work under the operating design.

Save final settings, firmware, active lineup, charger/filter states and trace filenames. Restore test overrides and record why alarms were cleared. A temporarily disabled charger or filter cannot silently become the normal operating state.

Repeat affected stability checks after UPS control/firmware changes, AVR/governor replacement or tuning, filter or generator-lineup changes, major load additions and recurrence. Routine EPSS exercise intervals remain in the maintenance program.

Field checklist for the closeout record

1. Identify generator, alternator, AVR, governor, UPS and switchgear model/serial/firmware; attach the one-line.
2. Preserve original logs before resetting alarms or changing a setting.
3. Export as-found generator and UPS configuration, including all qualification, restart and ramp delays.
4. Identify rectifier, bypass and protected-output sources and their approved operating limits.
5. Record battery reserve, charge state and the approved rollback path; hold reruns without adequate reserve.
6. Approve the recorder connection/removal and energy-isolation plan; confirm ratings, frequency span, waveform export, probe configuration, phase labels, CT ratios/polarity and calibration status.
7. Document clock reference, residual offset, native scan/averaging intervals, communications latency and combined timing-error bounds.
8. Arm event waveforms and trends before the source transition; verify memory and trigger operation.
9. Capture generator/UPS V/I, source kW/kVAr, protected output and relevant controller commands together.
10. Mark each comparison state with lineup, load composition, filter state and recharge condition.
11. Match first electrical departure to the timeline; label events unresolved when timing uncertainty overlaps.
12. Match the correction to the measured mechanism, change one approved variable, retain its original value and repeat the failed state plus affected motor/transfer checks.
13. Complete restoration checks, including required bypass function and recovered battery reserve.
14. Deliver native files, annotated traces, final settings, restrictions and owner acceptance.

Closeout records the approved operating state, remaining restrictions, responsible operations contact and next review trigger, with both raw files and annotated traces. Review date: 2026-11-05. Sources checked with curl: 2026-11-06. Update log: Final article review; manufacturer limits and calculations rechecked, safe recorder connections and frequency-estimator checks added, repeated limits removed.

How to cite this page: Anvilfield, Why do a UPS and generator hunt? Voltage and frequency field checks, 2026, https://anvilfield.com/field-guides/datacenter/ups-generator-hunting-diagnosis/

Related tools

FAQ

People also ask

• How do you size a data center generator?
• What is an NFPA 110 acceptance test?
• What is a commissioning hold point on a UPS?
• How do you perform GPU power transient testing for UPS and generators?

Sources checked

• S1: Cummins, Power Topic 6014: Generator set and UPS compatibility 2007; printed pp. 2, 7 through 11. Compatibility mechanisms, rejection/restart and bypass.
• S2: STAMFORD | AvK, AGN025: Non Linear Loads Issue B. Converter effects and sizing-factor limits.
• S3: STAMFORD | AvK, AGN004: Operating Charts Issue E, 2024. Published 670 kVA operating-chart example.
• S4: STAMFORD | AvK, AGN087: Power Factor Issue C, 2024. PF direction and leading-load capability.
• S5: Tri-Service Electrical Working Group, TP-19: Static Uninterruptible Power Supply October 2, 2017; p. 20. Passive filters and generator-mode charging coordination.
• S6: IEC 61000-4-30:2025 2025, Edition 4.0; corrected July 2026. Public measurement-method scope.
• S7: Fluke, 1773/1775/1777 Product Specifications January 2022; pp. 3 through 5 and 8. Acquisition, retention, timing and frequency span.
• S8: Schneider Electric, The UPS won’t operate online when powered by generator FA158891, modified May 28, 2026. Back-UPS/Smart-UPS application only.
• S9: Schneider Electric, Galaxy VX 480 V technical specifications Rendered revision 990-5451L-001. 1500 kW UPS/1500 kW I/O, 480 V column; URL retains older E token.
• S10: STAMFORD | AvK, AGN045: Alternators for Parallel Operation Issue B. Mains-parallel operating context, source transitions and configuration.
• S11: STAMFORD | AvK, AGN113: Application General Guidance and Rating Adjustments Issue D, 2024; p. 9. Excitation-selection guidance.
• S12: OSHA, 29 CFR 1910.333: Selection and use of work practices July 1, 2025 annual edition, paragraphs (a), (b) and (c); checked against OSHA’s current public text. General-industry electrical isolation, stored energy, verification and energized-work rules.
• S13: Fluke, 1773/1775/1777 Safety Information PN 5125091, Rev. 1, October 2021, pp. 1 and 2. Accessory ratings and restrictions on Ethernet/USB around exposed hazardous live parts.