What is a voltage sag?

The mechanics

A fault somewhere on the grid — a tree contacting a line, an animal across a bushing, a car into a pole — draws enormous current until a protective device clears it, typically in three to six cycles. During those few cycles, voltage across a wide area collapses toward the fault. Every facility on that part of the network sees a brief, deep dip.

The facility never loses power. Lights may flicker. And a variable-frequency drive on a 400 hp extruder trips on DC bus undervoltage, a contactor drops out on a chilled water pump, a PLC resets mid-sequence, and a CNC machine faults with a part in it. The line stops, and recovery takes hours because the process has to be purged, re-sequenced and re-qualified.

Then it happens again the following week, and nobody can explain it, because there is no record.

Two properties define a sag: remaining voltage (expressed as a percentage of nominal, or as the depth of the dip) and duration. A "40% sag" is ambiguous unless the convention is stated — it may mean 40% remaining or 40% lost. State remaining voltage and remove the ambiguity.

What equipment is supposed to tolerate

Two curves define the expectation, and the gap between them is where most industrial trouble sits.

The ITIC curve (formerly CBEMA, from the Information Technology Industry Council) describes the voltage envelope within which information technology equipment should continue operating. It is widely used as a general reference for whether a disturbance should have caused a trip.

SEMI F47 is the harder specification and the more useful one for industry. It requires semiconductor process equipment to ride through:

Remaining voltage	Required ride-through duration
50% of nominal	200 milliseconds
70% of nominal	0.5 seconds
80% of nominal	1 second

The important consequence: most industrial control components do not meet SEMI F47 and were never required to. A standard electromechanical contactor typically drops out somewhere between 70% and 50% remaining voltage within a cycle or two. A drive's DC bus undervoltage trip depends on bus capacitance and loading. So the same 65%-for-83-milliseconds event can leave a properly specified machine running and stop the machine beside it, and the difference is a $40 coil, not a power problem.

Worked example

A fault on a distribution circuit 12 miles away is cleared by a recloser in 5 cycles — 83 milliseconds. Voltage at the plant's service entrance falls to 65% of nominal.

• Against SEMI F47: the requirement at 50% remaining is 200 ms of ride-through. An F47-compliant machine at 65% for 83 ms rides through comfortably.
• Against a standard contactor: at 65% remaining, a typical AC coil is at or below dropout. It opens, the motor stops, and the drive faults on the loss.
• The cost: the affected line draws 900 kW. At the field-survey figure of $6.72 per kilowatt interrupted, the event costs roughly $6,048 — before any scrap, and before the four hours of purge and re-qualification that a food or chemical process needs.
• Frequency matters more than depth. Four such events a year is $24,000. Eight a month, which is what the survey's participating companies reported, is a different conversation entirely.

(Per-kilowatt figure from a Brazilian field survey of 33 companies across selected sectors, Motoki et al., Energies 2021. Use it to size the problem, then replace it with the facility's own logged events.)

Why nothing in the facility recorded it

This is the part that surprises people, and it is a hard technical constraint rather than a product gap.

Detecting a sag requires continuous half-cycle RMS measurement — a voltage magnitude computed every half cycle, roughly 8 milliseconds at 60 Hz, without gaps. Anything that averages over a longer window loses the event completely:

• A 15-minute interval meter: the 83 ms event moves the interval average by about 0.01%. Invisible.
• A 10-second-reporting energy sensor: also invisible. It is not a matter of being slightly too slow; it is three orders of magnitude away.
• A BMS trend logging every minute: invisible.
• The utility's meter: records the interval, not the waveform, unless the account has a dedicated power quality meter.

A facility that wants sag data has to install an instrument that does half-cycle RMS continuously, at the service entrance and usually at one or two sensitive process feeds. That instrument then produces the event record — timestamp, remaining voltage, duration, phases affected — which can be correlated against what was running and what stopped.

What a voltage sag is not

It is not an outage or an interruption. An interruption is voltage below 10% of nominal. A sag leaves 10–90%, which is why the lights stay on and why the event is so often disputed. This distinction is why a facility can have a perfect reliability record from the utility and a serious sag problem at the same time.

It is not a brownout. A brownout is a deliberate, sustained voltage reduction across a region, lasting minutes to hours. A sag lasts milliseconds and is almost always the by-product of a fault clearing correctly.

It is not usually the facility's fault. The majority of sags originate on the utility's system, often many miles away, and no equipment inside the building caused them or could have prevented them. Large motor starting and transformer energisation do cause internal sags, and those are distinguishable in the record by their characteristic shape and by which phases are affected.

It is not solved by a UPS on the panel. Ride-through solutions are load-specific: a constant-voltage transformer or a dip-proofing inverter on the control circuits often solves the problem for a fraction of the cost of protecting the whole load, because the thing that actually dropped out was a contactor coil, not the motor.

It is not visible in energy data. Repeating this because it is the single most common misunderstanding on this topic. See what is interval data.

For what sags do to industrial equipment once they arrive — the drives that fault, the PLCs that reset, the line stops and what a single event costs a plant — see what a voltage sag costs a production line.