Why Voltage Sags Under Load

Why this matters

Every voltage reading you take at a load is the source voltage minus whatever the path between them stole, and the path only steals while current is flowing. That is why a circuit can read a perfect 241 V with a meter and still starve the equipment on it, and why "I checked the voltage and it was fine" is the sentence that precedes a second trip. Read under load, at more than one point, and the sag stops being a symptom and becomes a locator: it tells you which segment of the run owns the problem, before you open a single lug.

Before you probe anything

Taking these readings means working inside an energized enclosure. That is permitted only under the troubleshooting exception at 29 CFR 1910.333(a)(1), which allows energized work where the test cannot be performed with the circuit de-energized, and it requires a meter, leads and probes rated for the system under 29 CFR 1910.334(c)(2). Do the arc-flash risk assessment first under NFPA 70E-2021, 130.5 and wear what it calls for. Any repair that follows - a lug, a splice, a breaker - is done de-energized, locked and tagged under 29 CFR 1910.333(b)(2), or 29 CFR 1926.417 on the construction side, and proved dead with the live-dead-live sequence in NFPA 70E-2021, 120.5.

If you find charred insulation, do not clear it with compressed air and do not lean into the panel to smell it. Heated and burned wire insulation of the common thermoplastic types releases hydrogen chloride and other irritant decomposition products, which is an inhalation hazard: ventilate the space, stay upwind of the enclosure, and where visible smoke or persistent fumes are present that calls for respiratory protection, not gloves. Never re-torque a lug that is energized or still hot; a lug that has been arcing gets replaced, not tightened.

The one line of physics underneath all of it

The voltage difference between any two points on a conductive path equals the current flowing through that path times the impedance between those two points. No current, no difference. So a no-load reading measures the source and nothing else, and a loaded reading measures the source minus the product of the load current and everything in the way.

Turn that around and it becomes a measuring instrument. If you know the current and you can read the voltage difference across a defined segment, you have measured that segment's impedance without disconnecting anything. That is the whole method, and it works on a service conductor, a lug, a breaker, a splice, or twelve inches of bus.

The call

A commercial customer with a 240 V single-phase compressor that nuisance-trips on start, only in the afternoon, worse on hot days. Two previous visits had replaced the starting components and adjusted the overload, and the fault had come back both times inside a month.

First reading, at the equipment disconnect with nothing running: 240 V. That is the reading that closed the first two visits. Second reading, at the same terminals during a start: 178 V. The load is being offered 178 V for the duration of a start on a system rated 240, a sag of 62 V, which is 26 percent of the 240 V that was there a second earlier.

What was eliminated, and on what evidence

The utility supply. A second meter on the line side of the main during the same start read 240 V no-load falling to 233 V. Seven volts of the total sag happens before the building's wiring begins. That is not nothing, but it is 7 of 62 V of drop measured across the whole path, about 11 percent of it, and it is proportional to a very large amount of conductor. The utility was not the fault.

The equipment. Clamped starting current peaked at 68 A. The nameplate locked-rotor figure for the unit sat above that, and the running current after start settled at nameplate. A load drawing normal current cannot be blamed for the path's inability to deliver it.

Undersized conductors. The runs were checked against their length and size, and the predicted drop across each of them at 68 A was small enough to be a rounding error compared with what was measured. Undersized conductor produces a drop spread evenly along its length, which is the pattern this circuit did not have.

Reading the ladder

The readings were taken during one start, all on the same event, with the current logged alongside them. Everything is measured as the difference across a defined segment.

Source 240 V minus 62 V of drop lands at 178 V at the load, which is what the meter read. The numbers close, which is the first thing to check before drawing any conclusion from a ladder like this - if they do not close, you have a path you have not measured, and that unmeasured path is often the neutral or the equipment grounding conductor.

Now the tell. Four of the five segments drop roughly in proportion to how much conductor they contain. One does not. The subpanel breaker and its load lug carry 63 percent of the entire sag across about a foot of hardware. Divide it out: 39 V at 68 A is about 0.57 ohms in that segment. A foot of conductor at that size is a fraction of a milliohm, so the segment is measuring higher than its length can explain by a factor in the thousands. Use a conductor resistance table for the exact per-foot figure, but you do not need it to see that this is not a length problem.

Opening it de-energized found a discolored, heat-relaxed load lug on the breaker, with the conductor's strands darkened where they entered it. The breaker and the conductor termination were both replaced.

Why the afternoon mattered

The time-of-day pattern was evidence, not noise, and it comes from two effects stacking.

First, the resistance of copper rises with temperature by roughly 0.4 percent per degree C. An afternoon panel in a hot mechanical room is running well above the temperature at which any of it was checked on a spring morning.

Second, and much larger here, a resistive joint is its own heat source. It makes heat in proportion to current squared times its resistance, the heat raises its resistance, and higher resistance makes more heat at the same current. That loop is why the same joint that costs you 39 V during a hot afternoon start can cost noticeably less on a cool morning, and why the fault deepened over three visits rather than staying constant. The behavior of the joint itself is covered in a sibling article on why a loose connection gets hot; the point here is only that a segment's drop is not a fixed property, so a ladder taken cold understates a thermal fault.

That also names the condition that would flip this method's answer. If every segment's drop had been proportional to its length, with no outlier, the finding is not a bad joint at all: it is either genuinely undersized conductors for the load and distance, or a source problem, and the fix is a design change rather than a termination repair. And if the total sag is small at the load but the equipment still misbehaves, the sag is not your fault to chase.

The measurement discipline that made this work

Three habits separate a ladder that names a segment from one that produces an argument.

Measure across the segment, not by subtracting two readings. It is tempting to read 233 V at one point, 228 V at another, and call the difference 5 V. But a meter's error is a percentage of what it reads plus a few counts, so subtracting two roughly 230 V readings to get a 5 V answer can carry uncertainty comparable to the answer itself. Probing directly across the segment puts the meter on a low range measuring a small number, where its accuracy is worth having.

Log the current with every voltage. A drop with no current beside it is uninterpretable. Thirty-nine volts at 68 A and 39 V at 6 A are two completely different faults, and only the pair tells you which.

Take the whole ladder on one event. A start is a transient. Readings gathered across five separate starts, with the equipment warming between them, are five different experiments, and their drops will not add to the total. If you only have one meter, use a recording function or accept that you are measuring segments one at a time and re-verify the total afterward.

How to verify you got this right

Repeat the identical ladder after the repair, during a start, at the same time of day and with the equipment at a comparable temperature. The reason for matching conditions is the temperature effect above: a repair verified on a cool morning against a fault measured on a hot afternoon is not a comparison, and a corrected reading benchmarked against an uncorrected baseline will flatter the fix.

On this system the post-repair ladder showed the same 7 V of utility drop, the same 5 V feeder drop, about 2 V across the new breaker and lug, and the unchanged 8 V and 3 V on the last two segments. Total drop 25 V against 62 V before, so the load now sees about 90 percent of source voltage during its worst moment instead of 74 percent.

If the outlier segment improved but the total is still deep, you have a second joint. That is common, because whatever caused the first termination to loosen - vibration, thermal cycling, an original install habit - was usually applied to every termination in the same panel on the same day.

References

• 29 CFR 1910.333(a)(1) and (b)(2) - de-energizing and safe work practices for electrical work; 29 CFR 1926.417 for the construction counterpart on lockout and tagging of circuits
• 29 CFR 1910.334(c)(2) - test instruments, leads and probes rated for the circuits to which they are connected
• NFPA 70E-2021, 120.5 (establishing an electrically safe work condition) and 130.5 (arc flash risk assessment)
• Conductor resistance and voltage-drop tables in the applicable electrical code and manufacturer documentation
• See related: Why a Loose Connection Gets Hot; What Impedance Adds That Resistance Does Not; Voltage Drop Calculations Reference