The Circuit That Read Fine and Failed Under Load

The call, and the thing that happens before any diagnosis

A commercial tenant reports weak heat from a 240 volt single-phase resistive heating load and a smell they describe as hot plastic near the disconnect. A tech was out five weeks earlier, ohmed the circuit, found it within tolerance, and left it. It got worse.

A burning smell at an enclosure is a live fire risk, not a diagnostic clue to go look at. Do not open the cover on an energized panel to see where the smell is coming from, because removing a cover on an arcing fault feeds it air and puts your face at the opening. Clear people out of the immediate area first. Kill power at the furthest upstream disconnect you can reach without opening anything, lock and tag it under 29 CFR 1910.333(b)(2), and if smoke is visible or the enclosure is hot to the touch, keep the room clear and call the fire service rather than opening it yourself. Once the circuit is dead, prove it dead with the live-dead-live sequence in NFPA 70E-2021, 120.5, using a meter and leads rated at least CAT III for the system voltage, then open the cover wearing arc-rated protection appropriate to the task under 29 CFR 1910.335.

On this call the enclosure was warm rather than hot and there was no smoke, so after the lockout and the proving sequence the cover came off. One lug on the load side of the disconnect was discoloured, with the conductor insulation browned for about two inches back from the termination.

Containment before curiosity

The temptation at that point is to snug the lug and re-energize to see if the symptom clears. That answers nothing and destroys the evidence. It also risks the specific failure this joint was already heading toward: a connection carrying that much heat can weld, arc across to the enclosure, or open under load with the cover off and someone standing in front of it.

So the circuit stayed locked out and the sequence was: photograph the joint before touching it, leave the lug exactly as found, and get the numbers that would prove what the joint was doing while it was still doing it. That last part required re-energizing later, deliberately and with the cover on, which is the only defensible reason to take the energized-work troubleshooting exception at 29 CFR 1910.333(a)(1). The exception is conditional on the work genuinely requiring energized conditions, and a voltage drop measurement at working current does.

Charred conductor insulation of unknown composition is an inhalation hazard as well as an electrical one. Do not sand, scrape or wire-brush it to clean the conductor for reuse. Cut the damaged conductor back to clean copper and replace the run, and ventilate the space while you work rather than breathing over the burned section.

What the earlier visit measured, and why it passed honestly

The previous tech did nothing lazy. With the circuit de-energized and locked out, they put an ohmmeter across the circuit at the disconnect and read 10.3 ohms. Their leads shorted together read 0.2 ohms, so the circuit itself was 10.1 ohms. The nameplate on the load is 24.0 amps at 240 volts, which is 240 divided by 24.0, or 10.0 ohms. They saw 10.1 against an expected 10.0, called it within element tolerance, and closed the ticket.

Here is the uncomfortable part: the fault was in that reading. The bad joint contributed about 0.1 ohms cold. At 0.1 ohms it was 1% of the total, sitting inside the same band as lead resistance, element tolerance and the meter's own accuracy. There was no reading discipline that would have separated it. The measurement was not wrong. It was asking a question the fault does not answer.

An ohmmeter interrogates a joint at its own small test current, on the order of a milliamp or less depending on range. The working current here is 20.8 amps. That is a difference on the order of twenty thousand times, and the resistance of a marginal mechanical connection is not a constant across that span.

The measurement that found it

The circuit was re-energized with the cover on, the load put into a full call, and three values taken:

• Line side of the disconnect: 241 volts
• Load terminals at the appliance: 208 volts
• Clamped current on the loaded conductor: 20.8 amps

The drop between those two voltage points is 241 minus 208, or 33 volts. Divide by the current: 33 divided by 20.8 is 1.587 ohms of unintended resistance sitting in series with the load.

Cross-check it from the other direction. The load itself now reads 208 divided by 20.8, which is exactly 10.0 ohms, matching the nameplate. Total circuit resistance is 241 divided by 20.8, or 11.587 ohms. Subtract the 10.0 ohm load and you get the same 1.587 ohms. The two derivations close, which is what tells you the model is right and there is only one significant fault in the run.

Against the cold reading of about 0.1 ohms for that joint, 1.587 ohms is roughly 16 times higher at working current and operating temperature. That ratio is the whole article.

For scale on how far out of bounds this is: the informational note in NEC Article 210 recommends branch-circuit voltage drop be held to about 3%, which on a 240 volt circuit is 7.2 volts. The measured 33 volts is about 4.6 times that recommendation. The note is a recommendation rather than an enforceable requirement, which is exactly why nothing in the installation objected.

Why the overcurrent device was never going to help

This is the part techs find genuinely counterintuitive, and it is the reason the fault ran for weeks. A series resistance does not increase current. It reduces it.

Without the fault the circuit would draw 241 divided by 10.0, or 24.1 amps. With 1.587 ohms in series it draws 20.8 amps. The circuit is protected at 30 amps, so the breaker is seeing 20.8 amps against a 30 amp rating, or 69% of its trip point, and falling. From the breaker's point of view the circuit got lighter. It is doing exactly what it was designed to do and it is structurally incapable of detecting this fault, because the energy is being dissipated inside the conductor path rather than delivered past it.

Put a number on that energy. Power dissipated in the bad joint is current squared times resistance: 20.8 squared is 432.6, times 1.587 ohms is about 690 watts. Roughly 690 watts is being released inside a lug the size of a thumbnail. That is what browned two inches of insulation.

Meanwhile the load is delivering 20.8 squared times 10.0 ohms, or about 4,330 watts, against a rated 5,760 watts. That is 75% of rated output, which matches the tenant's complaint precisely and is also why nobody called sooner. A 25% shortfall in heat reads as "it is not keeping up lately," not as "something is on fire in the panel."

The mechanism, because it predicts where to look next

A mechanical connection carries current through a limited number of actual metal-to-metal contact spots. When clamping force falls, from a lug that was never torqued to spec, from thermal cycling, from creep in the conductor, or from vibration, those spots shrink and an oxide film grows in the gaps. Oxide is a poor conductor, and it thickens with heat.

That creates a feedback loop with a direction: resistance rises, so heat rises, so oxide grows, so resistance rises further. It is slow at first and accelerates. It also explains the two things that make the fault hard to catch. First, a small ohmmeter test current cannot push through the film the way 20 amps can, and the film's behaviour is different at the two currents. Second, the joint's resistance when cold and unloaded is genuinely lower than its resistance hot and loaded, so a measurement taken any way other than hot and loaded understates it.

The direction of that loop is also the useful prediction. The joint does not stabilise. Left alone it goes to an open circuit or an arcing fault, and which one you get depends on geometry rather than on anything you can control.

The repair, and what the repair was not

With the circuit locked out again and proven dead, the damaged conductor was cut back to clean copper, a new listed connector appropriate to the conductor material was installed, and the termination was torqued to the value printed on the equipment using a calibrated torque tool. Where the connection joins dissimilar metals, an antioxidant compound listed for that connector and that conductor combination goes on; a compound not listed for the pair can make the joint worse, so read the connector's own instructions rather than reaching for whatever is in the truck.

What the repair was not: a re-torque of the original lug. A joint that has run at roughly 690 watts has annealed and deformed the conductor and the lug seat, and clamping force on damaged metal does not restore the contact area. Re-torquing a hot joint while it is energized is worse still, and is not a legitimate use of the troubleshooting exception, because tightening a lug is not a diagnostic measurement. It is maintenance, and maintenance goes behind a lock.

The other three terminations on the same circuit were inspected and torqued to spec at the same visit. A joint that failed from thermal cycling has siblings that experienced the same cycles.

Confirming it, and the reading that proves it stuck

Re-energize with the cover on, run the load to full temperature, and take the same three values in the same three places. The circuit now reads 240 volts at the line side, 238 volts at the load terminals, and 23.8 amps clamped. Drop is 2 volts, which is 2 divided by 23.8, or 0.084 ohms across the whole run, and 2 volts on 240 is 0.8%, comfortably inside the 3% figure the NEC note recommends. Delivered power is 23.8 squared times 10.0, or about 5,660 watts, which is 98% of the rated 5,760.

Two follow-ups matter more than that first reading. Take the drop again after the load has been running for at least twenty minutes at full output, because the fault this article is about only appears hot and a joint remade badly will show up on the second reading rather than the first. And record the measured drop and the current in the job record, not just "repaired loose lug." The next tech to open that panel then has a real baseline to compare against, which is the one thing the previous visit could not have had.

References

• 29 CFR 1910.333(a)(1) and (b)(2), the energized-work troubleshooting exception and the electrical lockout requirement for general industry
• 29 CFR 1910.335, personal protective equipment for work on or near exposed energized parts
• NFPA 70E-2021, 120.5, live-dead-live verification
• NEC Article 210, informational note recommending approximately 3% branch-circuit voltage drop
• See related: How to Use Voltage Drop to Find a Bad Connection; What a Resistance Reading Tells You and What It Hides