The Measurement That Was Right and Meaningless

The complaint that outlived three service visits

A light commercial customer reported a circuit tripping every few weeks. Three visits over four months produced three tickets, all of them competent. Supply voltage measured and in range. Running current measured and well under the rating. Insulation checked. The protective device replaced on the third visit. The trips came back nine days later.

Nothing on those tickets was wrong. Every number was taken with a good instrument, at a sensible point, by someone who knew what they were doing. The problem was that none of the quantities measured could move when the actual fault got worse. This case is reconstructed backwards from the paperwork, because the paperwork is where the answer had been hiding the whole time.

Stop here before the cover comes off

The measurement that finally settled this is taken on live terminations inside an energized panel, and it is the single highest-exposure instruction in this article.

29 CFR 1910.333(a)(1) requires live parts to be de-energized before anyone works on or near them unless de-energizing is infeasible, and its note treats testing that can only be performed with the circuit energized as one of those cases. A drop measurement across a joint under load qualifies, and the justification covers the reading only. Meter, leads and probe tips must carry an IEC 61010-1 measurement-category and voltage rating at or above the panel they are entering, because an under-rated instrument on a fault at this energy level vents as an arc directly at your hands. Wear the arc-rated and insulating protective equipment 29 CFR 1910.335(a) requires for the exposure, keep the free hand out of the enclosure so no path crosses your chest, stand to the side of the panel rather than in front of it, and use insulated probe tips with the minimum exposed metal, since a probe that bridges two lugs is how a millivolt reading becomes a fault.

The instant this becomes a repair - and remaking a termination is a repair - de-energize, apply electrical lockout or tagging under 29 CFR 1910.333(b)(2), or 29 CFR 1926.417 on a construction site, and prove the conductors dead using the live-dead-live sequence in NFPA 70E-2021, 120.5. Do not reach for 29 CFR 1910.147 here; it excludes electrical hazards from work on utilization equipment at 1910.147(a)(1)(ii)(C) and is your standard for the mechanical side, a blocked rotor or a stored spring, not this.

What the record contained

Three tickets, laid side by side:

Say the equipment nameplate gives a rating in the mid-twenties of amps. Every current reading is comfortably below it. Every voltage sits within a few volts of the others and well inside any reasonable utilization band. Read as a set, those nine numbers say the circuit is healthy, and they said it three times.

What the record did not contain

Two absences did more work than any of the present values.

No termination measurement of any kind. Nobody had put a meter across a lug. Every reading in the file was a whole-circuit quantity, taken at a panel or at the equipment.

No load-duration or time-of-day context on the readings. This one the record gave up accidentally. All three visits were logged between roughly 08:40 and 10:20 in the morning, because that is when a first call of the day gets dispatched. Every customer-reported trip in the same log had occurred in the afternoon. Nobody had written that down as a finding; it was simply visible once the tickets were read as a timeline rather than as three separate events.

A gap in a record is evidence. It tells you which hypotheses were never tested, and in this file the untested hypothesis was anything that only appears after hours of continuous load.

The variable nobody had measured

On the fourth visit, with the circuit running under its normal afternoon load, the tech measured the millivolt drop across each of the comparable terminations in the run, using the same instrument on the same range within the same few minutes. Three comparable joints, one current path.

• Joint A: 4 mV
• Joint B: 5 mV
• Joint C: 38 mV

At the 18.4 A flowing at that moment, joint C is dissipating 0.038 V times 18.4 A, which is about 0.70 W, against joint B's 0.005 times 18.4, about 0.09 W. Joint C is putting roughly 7.6 times the heat of its neighbour into a space the size of a lug, and its own resistance works out to 0.038 divided by 18.4, about 2.1 milliohms, against joint B's 0.27 milliohm.

The arithmetic behind three no-fault-found visits

Here is why every earlier measurement was both correct and incapable of finding this.

The entire fault signal is 38 mV. The supply voltage reading it would have to disturb is 238 V. As a fraction, 0.038 divided by 238 is 0.00016, which is 0.016 percent of the reading. A typical field meter's published accuracy on a 238 V reading works out to somewhere around a volt and a half, and 1.5 divided by 0.038 is about 39. The instrument's own uncertainty is roughly forty times the size of the entire thing being looked for. No amount of care at the panel could have surfaced it.

The current reading is no better placed. A joint developing resistance in series with a load does not raise the current, it lowers it very slightly, and the change here is far inside the drift between visits. Note the direction of that: the three current readings actually rose across the four months, 17.8 to 18.1 to 18.4, which is the load working harder as the seasons changed, not the joint improving. Two effects moving in opposite directions inside a difference smaller than the instrument's uncertainty is not a signal.

The time-of-day pattern completes it. Contact resistance at a loose or oxidized joint rises as the joint heats, and the joint heats with sustained current and warm ambient. A morning visit on a cold circuit measures the joint near its best condition. The complaint occurred in the afternoon because that is when the joint was worst, and every visit was scheduled when it was least detectable.

Remaking the joint, and the rule that governs it

With the circuit de-energized, locked out and proved dead as above, the joint came apart. The conductor and the lug both showed discoloration consistent with sustained heating, which is a corroborating observation and not the diagnosis on its own.

Terminations get remade to the manufacturer's published torque value, not to feel. NFPA 70, Article 110 covers electrical connections, and since the 2020 edition it requires that where the manufacturer provides a torque value, the connection be tightened with a calibrated torque tool. That requirement exists precisely because of this failure mode: an under-torqued lug and an over-torqued lug both end up as high-resistance joints, and neither one announces itself at installation. Where the conductor or lug showed damage, the damaged length was cut back rather than reused.

Confirming it against a comparator that matches

The re-check was made under load, on the same instrument and range as the original, at the same point.

• Before: 38 mV at 18.4 A
• After: 6 mV at 18.2 A

Those two currents are not identical, so the raw millivolt figures are not directly comparable. Normalize both to the same basis: 38 divided by 18.4 is 2.07 mV per amp; 6 divided by 18.2 is 0.33 mV per amp. That is a factor of about 6.3 reduction on a matched basis, and the joint now sits in the same range as its neighbours rather than several times above them.

A second confirmation ran the following week: the same measurement repeated in the late afternoon under sustained load, which is the condition all three earlier visits had missed. It held. The customer reported no further trips, and the file now contains a per-joint baseline so the next tech has a comparator instead of a mystery.

An infrared scan would also have found this, and it is faster where it is available - but only with the cover off and the circuit loaded, because an infrared instrument reads the surface in front of it and a closed panel cover shows you the temperature of the cover. It also reads a shiny lug badly unless the emissivity assumption is handled, which a sibling article covers in full.

The habit this changes

The lesson is not "always measure across terminations." It is one level up: before you accept a set of readings as evidence of health, ask whether the quantity you measured is capable of moving when the suspected fault gets worse. If a fault could double in severity without shifting your number by more than your instrument's uncertainty, that number is not a test. It is a formality, and running it three times does not turn it into a test.

Write the answer down either way. The most useful line on that fourth ticket was not the 38 mV, it was the note that the first three visits had all measured whole-circuit quantities and none had measured a joint. That sentence is what stops a fifth visit from repeating the first three.

References

• 29 CFR 1910.333(a)(1) energized-testing gate and its note; 29 CFR 1910.333(b)(2) electrical lockout and tagging; 29 CFR 1926.417 for construction electrical work
• 29 CFR 1910.147(a)(1)(ii)(C) electrical carve-out; 29 CFR 1910.335(a) electrical protective equipment; NFPA 70E-2021, 120.5; IEC 61010-1 measurement categories
• NFPA 70 (National Electrical Code), Article 110, electrical connections and the calibrated-torque-tool requirement introduced in the 2020 edition
• See related: What Emissivity Does to an Infrared Reading; Why a Static Reading Lies About a Loaded Fault; How Averaging Hides a Fault