Energy Optimized

•

Software-Defined Electricity

Products

Software-Defined Electricity

A voltage sag the load rides through
A voltage sag the load rides through

Resilience: more margin between events, not fewer events

A short sag stops a line and costs the plant restarts and lost hours, long before a reliability index counts an outage. We do not change the sag count. We correct the current at the panel, so the plant’s own conductors keep more margin.

Resilience is a facility’s ability to keep running through a disturbance and to recover when one wins. Every part of that has a clock.

A generator picks up load in seconds, and an uninterruptible power supply (UPS) transfers in milliseconds. Most of what a plant loses falls between those clocks. Fluctuation means waste, damage and dollars.

Every layer answers on its own clockThe sag that shuts a line down falls between the clocks. The correction row is our own figure.Every layer answers on its own clock1 µs100 µs10 ms1 s1 min1 hCorrection responds in under 5 microsecondsA series sag corrector restores the missing voltage within 2 msThe ITI curve gives 20 ms of dropoutA 230 kV fault lasting 42 to 66 msSEMI F47 asks tools to ride through 50% for 200 msAn ISO 8528-5 class G3 generator recovers voltage within 4 sNFPA 110 Type 10 restores power within 10 sIEEE 1366 counts a sustained outage from 5 minutesAverage recovery after a sag shutdown, 51 minutesTime (log scale)Illustrative placement on a logarithmic time axis. Every value on it is printed and sourced in the body of this page. Thecorrection row is the company’s own figure and is marked in blue.Every layer answers on its own clock1 µs100 µs10 ms1 s1 min1 hCorrection responds in under 5 microsecondsA series sag corrector restores the missing voltage within 2 msThe ITI curve gives 20 ms of dropoutA 230 kV fault lasting 42 to 66 msSEMI F47 asks tools to ride through 50% for 200 msAn ISO 8528-5 class G3 generator recovers voltage within 4 sNFPA 110 Type 10 restores power within 10 sIEEE 1366 counts a sustained outage from 5 minutesAverage recovery after a sag shutdown, 51 minutesTime (log scale)Illustrative placement on a logarithmic time axis. Every value on it is printedand sourced in the body of this page. The correction row is the company’sown figure and is marked in blue.
The sag that shuts a line down falls between the clocks. The correction row is our own figure.

What resilience is

Event counts stay the same, and upstream conductors run cooler. The controller does not replace a UPS, a sag corrector, a transfer switch or a generator. It changes none of the outage or sag counts 17.

  • Production lines. Sags that fall between the clocks. A survey of 33 industrial plants found about eight sag or short-interruption shutdowns a month, each costing an average of US $7,365 3.
  • Transformers and feeders. Upstream conductors and switchgear. Source-side current falls to near the working current, so the transformer, feeder and switchgear run cooler 17.
  • Generators. Sites that run on their own alternator. The harmonic burden that heats the generator’s rotor windings moves to the controller, up to its rating 21.
  • If it stops. A controller removed or powered down. The electrical system continues to operate as before 17.

Under IEEE 1366, SAIFI counts the sustained outages an average customer sees in a year. SAIDI counts the minutes they last. MAIFI counts the momentary ones under five minutes 1. By that count, the average U.S. customer lost 7.8 hours of service in 2017 2.

A two-cycle sag that drops a production line for 51 minutes appears in no SAIDI figure, because by the utility’s definition the supply was never interrupted 2,3. IEEE 1159 files the same event as an instantaneous sag 4.

An illustrative sag, not a capture. On a 480 V, 500 kVA, 5% panel, 60 A of reactive current moves the bus half a percent of nominal. So the corrected readout rounds to 71% against 70% 5. The dashed line sits at 80% of nominal, where the ITI curve allows 10 s and SEMI F47 allows 1 s 6,7. Ride-through stays with stored energy.

What it costs

In the survey of 33 industrial plants, recovery from each shutdown averaged 51 minutes 3. A 2025 Lawrence Berkeley survey of nearly 4,000 non-residential customers priced a momentary interruption at US $609 and an 8-hour one at US $6,172 8. Sustained interruptions cost U.S. customers US $44 billion a year in 2015 dollars, by a 2018 Lawrence Berkeley estimate 23.

Sags arrive far more often than outages: EPRI found a distribution-fed site 15 times more likely to see a sag than an interruption 9. On 2024-07-10, a 230 kV fault lasting 42 to 66 ms dropped about 1,500 MW of data center load off the grid on its own protection 10. That is the domino the stack is supposed to stop. A facility with the full mitigation stack in place still loses production when a sag falls between the layers 3.

How the industry has dealt with it

Facilities layer a stack, and each layer is rated on its own clock. Two of the clocks are curves. Engineers specify equipment to the ITI curve, which gives 20 ms of dropout tolerance and half a second at 70% voltage 6. SEMI F47 asks tools to ride through 50% for 200 ms 7.

Two are machines that put the voltage back. A series sag corrector restores the missing voltage within 2 ms 11, and a UPS carries the load for the minutes a generator needs.

Two are the generator’s own rules. NFPA 110 Type 10 restores power within 10 seconds 12. An ISO 8528-5 class G3 generator recovers its voltage within 4 seconds of a load step 13.

Each layer is sized against the worst event and paid for whether or not it comes. Meanwhile the sources are turning into inverters, with less inertia and faster current limits than the generator the stack was built around 14. Nothing in the stack answers the current that made the bus soft.

Every sag test point sits below the controller’s bandEvery sag test point sits below the controller’s band.Every sag test point sits below the controller’s bandRemaining voltage (% of nominal)020406080100120Controller stated operating band, ±15% of nominal.Behavior outside the band is not stated.0.0010.010.1110100ITI (CBEMA) 2000SEMI F47-0706IEC 61000-4-11 Class 3ERCOT NOGRR282Duration (s, log scale)The lower envelope of the ITI (CBEMA) 2000 curve and the SEMI F47-0706 required immunity. IEC 61000-4-11 Class 3test points at 60 Hz. The two ERCOT NOGRR282 points are the lower edges of the bands read from the SPWG slides.The band is the controller’s stated operating band in the VectorQ technical overview. Drawn only from points read forthis series. ITI upper points are omitted because the application-note text does not state them.Every sag test point sits below the controller’s bandRemaining voltage (% of nominal)020406080100120Controller stated operating band,±15% of nominal. Behavior outsidethe band is not stated.0.0010.010.1110100ITI (CBEMA) 2000SEMI F47-0706IEC 61000-4-11Class 3ERCOT NOGRR282Duration (s, log scale)The lower envelope of the ITI (CBEMA) 2000 curve and the SEMI F47-0706required immunity. IEC 61000-4-11 Class 3 test points at 60 Hz. The twoERCOT NOGRR282 points are the lower edges of the bands read from theSPWG slides. The band is the controller’s stated operating band in theVectorQ technical overview. Drawn only from points read for this series. ITIupper points are omitted because the application-note text does not statethem.
The envelopes are the ITI lower curve and SEMI F47, with the IEC 61000-4-11 Class 3 and ERCOT ride-through points 6,7,15,16. The shaded band is the controller’s stated operating range 17.

How Software-Defined Electricity handles it

Say first what it does not do. It changes none of the counts: not SAIDI, not the sag count, not the transfer time 17. It changes the hours between events.

The stack acts on the voltage after the event. The controller acts on the current every cycle, before any event. Within its band, it supplies the loads’ reactive, harmonic and unbalanced currents locally, in under 5 microseconds. Upstream loss falls as the square of the current removed 17.

We assessed two server racks. Total current fell from 23.78 A to 22.60 A and from 15.86 A to 13.84 A. Reactive current at the fundamental fell from 7.27 A to 0.18 A 18.

Thermal margin upstream is resilience that a thermometer can measure. A transformer study found harmonic losses near half of the total at 40% to 50% current distortion 19. On a generator, the effect is larger, because the source is weaker. A generator set at 12% to 20% impedance has up to about four times the impedance of a 5% to 7% utility transformer 20.

The harmonic burden that heats the alternator’s damper cage moves to the controller, up to its rating 21. Heat that leaves the damper cage is life that stays in the alternator.

An illustrative trace, not a capture. A facility at 0.85 power factor pushes 18% more current through its transformer than the work needs 22. The corrected trace is the working current alone, and the margin is what is left in the winding.

Where it matters

Two Opportunity Pillars meet here. In Delivery, feeders built for one direction now carry power both ways. In Generation, an inverter has no rotor to absorb the first instant.

A defense facility runs on generators for weeks at a time and sizes them for the worst hour. A data center now carries ride-through obligations written into grid rules. A network coming back in a blackstart has no stiff grid to absorb inrush, resonance and unbalance. Each page says which share of the problem is the plant’s own.

Read more: the Delivery and Generation pillar pages.

What to measure

Log the bus with a Class A meter for a week. Read the sag count and the interruption count against the equipment’s own curve. Then read the transformer temperature and the feeder current before and after correction, on the same load. Expect the counts to stay the same and the margins to change. Have a third party log it, so the baseline is theirs and not ours.

Start by logging one switchboard for a week before sizing a generator for the worst hour.

Sources
  1. California Public Utilities Commission, Electric System Reliability Annual Reports, reproducing the IEEE 1366 definitions.
  2. U.S. Energy Information Administration, Today in Energy (2018-11-30): 7.8 hours of interruption per customer in 2017.
  3. Motoki et al., Energies 2021, 14(10), 2874. https://doi.org/10.3390/en14102874
  4. IEEE 1159-2019, Table 2.
  5. Rule review for this series: the illustrative 480 V, 500 kVA, 5% panel. USDA Bulletin 1724D-112, Eq. 4. First-party derivation.
  6. Information Technology Industry Council, ITI (CBEMA) Curve Application Note (2000).
  7. SEMI F47-0706 (Reapproved 0812), Table 1.
  8. Larsen et al., ICE Calculator 2.0, Lawrence Berkeley National Laboratory (2025-05-29), Table 4.6.
  9. EPRI, Wright, Power Quality and Reliability Seminar (2018), slides 19 and 21.
  10. NERC, Incident Review: Considering Simultaneous Voltage-Sensitive Load Reductions (2025-01-08).
  11. A leading maker’s series sag corrector product page.
  12. NFPA, An Overview of NFPA 110 (2023-01-23).
  13. ISO 8528-5:2018, Table 4, as reproduced in a leading generator maker’s Bulletin 5544421.
  14. Shi, Bai and Wang, Nature Communications 2026, 17, 8498. https://doi.org/10.1038/s41467-026-76692-0
  15. IEC 61000-4-11:2020, Table 1, Class 3.
  16. ERCOT, NOGRR282, Large Computational Load Ride-Through Requirements, effective 2026-08-01.
  17. 3DFS, VectorQ Technical Overview (current revision), p. 9 and Technical Specifications table. First-party.
  18. 3DFS, electrical network analysis of server racks at a data center, overview sheets. First-party.
  19. León-Martínez et al., Environments 2023, 10(10), 177, Section 3. https://doi.org/10.3390/environments10100177
  20. The Green Grid, White Paper 55, Table 3, p. 15.
  21. A leading maker’s generator application note AGN 025, Non Linear Loads, p. 6.
  22. 3DFS, Power Quality Impact White Paper, System Losses. First-party.
  23. Hamachi LaCommare, Eto, Dunn and Sohn, Energy 2018, 153, abstract.

Version 5.1