Why a Loose Connection Gets Hot

Why this matters

A bad connection is the only common electrical fault that gets worse on its own, with no help from the customer, the load, or the weather. Everything else sits where it is until something changes. A degrading termination heats itself, and the heat is what degrades it further, so the curve steepens on its own schedule. That is why the same joint fails again a season after somebody tightened it, why a circuit that measures fine on a Tuesday morning smells hot by Friday afternoon, and why an ohmmeter is close to useless for finding one early.

The moment you smell it

A hot-plastic or burning-insulation smell near electrical equipment is a de-energize-first condition, not a look-first condition. Open the circuit at its overcurrent device before you open any enclosure, lock and tag under 29 CFR 1910.333(b)(2) (29 CFR 1926.417 on construction work), and prove dead with the live-dead-live sequence at NFPA 70E-2021, 120.5. Do not open a device while energized to feel whether it is warm, and do not re-torque a connection that is energized or still hot - the joint you are reaching into is the one that has been arcing.

If there is visible smoke, charring spreading, or flame, that is an evacuate-and-call-the-fire-service event, not a diagnosis: get everyone out, kill the supply from outside the affected area if it is safe to reach the disconnect, and do not open the enclosure.

Overheated wire insulation and device bodies of the common thermoplastic types release hydrogen chloride and other irritant decomposition products when they get hot enough to smell. That is an inhalation hazard and it is answered with ventilation and distance, not with gloves: ventilate the space, stay upwind of the enclosure, and treat visible fume or a persistent sharp smell as calling for respiratory protection before anyone works in it.

The call

A light commercial customer with an electric resistance heating load on a 30 A circuit. Complaint: heat drops out intermittently, restores itself, and there is a faint hot smell in the mechanical closet on cold days. Another shop had replaced the breaker two months earlier, which fixed it for about two weeks.

The complaint pattern is doing real work here before any instrument comes out. Intermittent, self-restoring, and correlated with the coldest days - which is to say, with the days the load runs the longest. Something that only misbehaves under sustained load is thermal, and something that restores itself is not a broken conductor.

What a connection physically is

Two pieces of metal pressed together do not touch across the area you can see. Metal surfaces are rough at a scale you cannot resolve, so contact happens at a scattering of high points, and the true contact area is a small fraction of the apparent one. All the current funnels through those spots. That funnelling is itself a resistance, and it is inversely tied to contact pressure: press harder, more high points flatten into contact, more paths open, resistance falls.

On top of that sits film resistance. Exposed metal grows an oxide or sulfide layer, and those films are poor conductors. A properly made connection breaks through the film by wiping and deforming the metal as it is tightened, which is why the act of tightening is doing something physical and not just holding parts near each other.

So a termination's resistance is a function of pressure and surface condition, both of which change over time. That is the entire reason this failure mode exists.

Conductor material changes the rules, and not by a small margin. Aluminum grows a hard, tenacious oxide the moment fresh metal is exposed, and it creeps under sustained pressure more than copper does. That is why aluminum terminations depend on the connector being listed for the conductor material rather than on technique, and why the joint compound and torque used must be the ones the connector's own instructions specify. Do not reach for a generic antioxidant compound on a connector whose listing does not name it, and do not land aluminum on a terminal that is not marked for it.

The measurement that named it

The circuit was re-energized with the load called and instrumented under the troubleshooting exception at 29 CFR 1910.333(a)(1), which permits energized measurement only where the reading cannot be taken de-energized, with meter, leads and probes rated for the system under 29 CFR 1910.334(c)(2) and the arc-flash risk assessment done first under NFPA 70E-2021, 130.5.

Load current, clamped: 24 A steady. Voltage measured directly across a wire-nut splice in a junction box, probe to probe: 1.9 V.

That pair is the whole diagnosis. Resistance is 1.9 divided by 24, about 0.079 ohms. Power dissipated there is current squared times resistance: 24 squared is 576, times 0.079 gives about 45 W being turned into heat inside a splice you could cover with a thumb.

Put that in proportion. The conductor feeding it, at roughly 0.0012 ohms per foot for its size, dissipates 576 times 0.0012, about 0.69 W per foot at the same 24 A. So the splice is generating as much heat as roughly 65 feet of the conductor that feeds it, concentrated at a single point with almost no surface area to shed it. Read the conductor resistance table in the electrical code for the exact per-foot figure for your size and material; the ratio is the point, and it is not close.

Note what made that measurement work: the drop was read directly across the joint while the real load current flowed. Not subtracted from two readings taken to a common reference, and not taken with the heater off. Both of those alternatives would have returned a number too small to distinguish from meter error.

The loop

Here is why this fault does not sit still.

At 0.079 ohms and 45 W, the joint runs hot. Heat drives three things at the contact interface. It accelerates oxide growth on any exposed metal. It relaxes the metal under load, so the elastic pressure that a torqued screw or a spring connector applies decays over time, which is stress relaxation and it is permanent. And it drives thermal cycling: the joint expands when hot and contracts when cold, and the two metals in a connection rarely expand at the same rate, so every heating cycle works the interface slightly.

Each of those reduces contact pressure or degrades the surface, which reduces true contact area, which raises resistance. Higher resistance at the same current means more heat. More heat means faster relaxation and oxidation.

Run the arithmetic forward on this splice. If the resistance climbs from 0.079 to 0.14 ohms over another season - not a dramatic change, and still a number most techs would call small - the heat at the same unchanged 24 A goes from 45 W to 576 times 0.14, about 81 W. That is a 77 percent increase in heat with no change in the load whatsoever.

Notice which term moved. Heat is proportional to current squared times resistance, and here the current never changed; only the resistance did. So at a fixed load, a connection's heat rises in direct proportion to its resistance, not as a square. The squared relationship belongs to current, and it matters in a different scenario, covered in a sibling article on how heat and current relate in a conductor. Confusing the two leads techs to dismiss a resistance change as small because they are unconsciously expecting the squared term to be what makes a fault dangerous.

Why an ohmmeter would have missed this

At 0.079 ohms, this joint is smaller than the lead resistance of most field meters, which typically sits in the low tenths of an ohm, so without zeroing the leads it does not appear at all. Even zeroed, the ohmmeter tests the interface at a few milliamps and at room temperature, which is not the condition the joint fails in: film resistance behaves differently at a few milliamps than at 24 A, and contact pressure at ambient is not contact pressure at operating temperature. A joint that measures continuous cold and unloaded has told you only that it is continuous cold and unloaded.

Thermal imaging finds these well, with one condition attached: the load must be running at the time. A connection at a fifth of its normal load makes a twenty-fifth of its normal heat, and it will image as unremarkable. Cross-reference the sibling article on reading a thermal image for interpretation; the operative rule here is that an image with no load current logged beside it is not evidence of anything.

Why the breaker swap bought two weeks

The previous shop's repair is worth understanding rather than dismissing. Replacing a breaker means re-making its terminations, which restores contact pressure and wipes fresh metal at that point. If the heat had also been affecting the breaker's own thermal element, a new breaker would stop nuisance tripping for a while regardless of where the actual heat was coming from.

But the fault was at a splice several feet away, untouched. Nothing about the swap changed the splice's resistance, so the loop resumed from where it had got to, and the two weeks was simply how long it took the symptom to become noticeable again. Any repair that makes a self-accelerating fault quiet for a short period should be suspected of having reset a symptom rather than removed a cause.

What the repair had to include

Three things, all de-energized under lock and tag:

Cut back the conductor. The copper at a joint that has been running at 45 W is annealed and often discolored well past the visible damage. Terminating on softened, oxidized conductor puts the new connection at a disadvantage on day one. Cut back to bright, unannealed metal.

Replace the connector, not just re-seat it. A connector that has been heated has taken its own relaxation, and a spring or screw that has lost its elasticity cannot be tightened back into specification.

Torque to the value the device specifies, with a calibrated tool. The termination's whole function is contact pressure, and both errors are real: under-torque leaves insufficient pressure, and over-torque deforms the conductor and the connector so that the pressure decays faster than it otherwise would. Article 110 of the electrical code requires terminations be tightened to the manufacturer's specified torque where one is provided, and that is the value that applies, not a feel.

Confirming it

Re-measure the drop directly across the repaired connection at full load, after the load has run long enough for the enclosure to reach a steady temperature - not two minutes after the repair, which is the most flattering reading the joint will ever give.

A healthy bolted or listed connection carries a drop under load on the order of a few millivolts, small enough that it is a genuine measurement challenge rather than an obvious number. The acceptance figure that governs belongs to the equipment documentation where one is given. Compare against the pre-repair figure taken under the same load and at a comparable temperature, because a post-repair reading taken cold against a pre-repair reading taken hot will look like a triumph regardless of what was done.

Then check the other terminations in the same enclosure at the same time. Whatever produced the first one - a torque habit, a vibration source, a load that runs longer than the install anticipated - was applied to all of them on the same day.

References

• 29 CFR 1910.333(a)(1) and (b)(2) - the troubleshooting exception 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)
• NFPA 70 (National Electrical Code), Article 110 for termination torque and conductor-material listing requirements; conductor resistance tables for per-foot figures
• See related: How Heat and Current Relate in a Conductor; Why Voltage Sags Under Load; What a Hot Spot in a Thermal Image Actually Means