How Thermal Expansion Creates Faults That Vanish Before You Arrive
Why this matters
"No fault found" is the most expensive line a tech writes. It burns a visit, tells the customer you could not find their problem, and leaves the shop nothing to bill against a callback. A large share of those write-ups are the same physical event: a joint, clearance, or seal that was genuinely faulty at temperature and had physically returned to a good condition before the van pulled up. Nothing healed. The metal moved back. Understanding the reset as a mechanical event with a known timescale is what lets you plan around it instead of writing it up as a mystery.
The reset is a measurable mechanical event
Solids grow when heated at a rate set by their material. In round numbers per degree F of rise, per inch of length: steel is about 6.5 millionths, copper about 9.3 millionths, aluminum about 12.8 millionths. Those are small numbers and that is exactly the point - the movement that opens a marginal joint is well under a hundredth of an inch and completely invisible.
The trouble starts where two different materials are joined, because they grow by different amounts over the same temperature swing. Take an aluminum part 4 inches long fastened by a steel fastener spanning the same 4 inches, and put 100 degrees F of rise through it. The aluminum grows about 4 x 12.8 millionths x 100, which is roughly 5.1 thousandths of an inch. The steel grows about 4 x 6.5 millionths x 100, roughly 2.6 thousandths. The difference is about 2.5 thousandths of an inch, and that difference has to go somewhere: into deformation, into higher clamping stress, or into a change in how hard the two surfaces press together.
When the assembly cools, all of it reverses. Every dimension returns. The joint that was loose, tight, gapped, or bound is dimensionally identical to the one that worked fine.
The arithmetic that explains your empty visit
Put your drive time next to the assembly's cooling curve and the "no fault found" write-up stops being mysterious.
Assemblies cool the same exponential way they heat. If the joint in question has a cooling time constant of about 20 minutes, then 20 minutes after shutdown it has closed roughly 63 percent of the gap back to ambient, and 40 minutes after shutdown roughly 86 percent.
Carry the earlier example forward. A 100 degree F rise created about 2.5 thousandths of differential movement. Forty minutes after shutdown, only about 14 degrees F of that rise remains, which leaves roughly 0.35 thousandths of differential - about a seventh of what was there at temperature. If the fault needed most of that 2.5 thousandths to appear, it is completely gone, and it was gone before you finished the drive.
That is the whole story. Forty minutes of travel on a 20 minute time constant is enough to erase almost any expansion-driven fault. Nothing you do with a meter on arrival will find it, because there is nothing there to find.
Which of these erase completely, and which leave marks
Not every expansion fault vanishes without a trace. Sorting them tells you whether to hunt for evidence or plan a return in the failure state.
- Erases completely: contact pressure at a clean, undamaged joint; a running clearance that closed and reopened; a seal that gapped and reseated; a sensor lead under stress that relaxed. These leave nothing at all if the assembly is otherwise healthy.
- Leaves marks over time: any joint that has been cycling this way for months. Repeated micro-movement plus heat plus current or fluid leaves physical evidence even when the current state is fine.
- Leaves marks immediately: anything that reached a temperature high enough to discolor, deform, or char. If you find that, you are not looking at a subtle expansion fault, you are looking at a real overheat.
The evidence that survives the reset
This is the part worth memorizing, because it is what you look for on a call where the fault has already gone.
| Evidence | What it looks like | What it tells you |
|---|---|---|
| Fretting at a joint | Fine dark powder or polished witness patches at a mating face | Repeated relative movement between two surfaces, which is exactly what a cycling expansion mismatch does |
| Oxidation halo at a terminal or fitting | Ring of discoloration around one point in a row of clean ones | Localized heating at a joint whose contact quality is cycling |
| Compression set on a seal | Permanent flat or uneven bite line on one edge only | The seal has been working through more movement than it was sized for |
| Polished or bright wear at a clearance | Shiny band where a moving part contacts under some conditions | The clearance closes at some point in the cycle, then reopens |
| Elongated or ovaled fastener holes | Mounting holes worn to a slot in the direction of expected growth | The assembly has been walking with each thermal cycle |
| Latched fault codes or event logs | An entry with a timestamp, even if the current state is clear | Places the fault in time, which is what lets you match it to run hours |
| A repeated marking or witness position | A witness mark, made deliberately on a prior visit, no longer aligned | Direct proof of movement, and the only entry on this list you can create yourself |
The last row is the one techs underuse. A paint or scribe mark across a joint on visit one costs thirty seconds and answers "did this actually move" on visit two without any instrumentation at all.
What the customer's reset destroys
An expansion fault has a short evidence window and customers close it without knowing. Three habits in particular:
- Resetting immediately and repeatedly. Each reset re-energizes a joint that may be in a poor contact state, which both hides the event and can do real damage.
- Opening the panel or cover "to let it cool." This dramatically speeds the reset and destroys the temperature distribution that was the diagnosis.
- Wiggling, tightening, or reseating anything before you arrive. A joint that has been disturbed will very likely work fine for days, which resets the whole investigation and buys the fault another month.
Those three go on the dispatch card as one sentence: leave it exactly as it is and call us. That instruction is worth more than most of the diagnostic equipment on the truck.
Making it reappear on purpose
You have three ways to get back into the failure state, in order of how much of the real condition they reproduce.
- Run it to the condition. Highest fidelity, because it reproduces both the temperature and the distribution of that temperature across the assembly. Costs run hours.
- Warm restart from partial cool-down. Much faster and reproduces most of the condition, since you start partway up the curve. This is the practical workhorse.
- Local controlled heating of a suspect area, within safe limits and never on anything flammable, pressurized, or containing a fuel. Fastest and lowest fidelity, because you heat one part rather than the whole assembly, so a fault that depends on differential movement between two areas may not appear at all.
Method three is a screening tool, not a proof. A fault that appears under local heating is confirmed thermal; a fault that does not appear under local heating is not ruled out.
A worked example, carried through
A control assembly drops out intermittently on a customer's equipment, always deep into a run, and always works on arrival. Three prior visits, three clean write-ups. All values below are illustrative.
Visit four starts with arithmetic rather than a meter. The customer's log shows the fault appears about 6 hours into a run and clears after about an hour of sit. Drive time from the shop is 40 minutes, and the customer calls after they have tried a reset twice, which adds about 15 minutes. So the earliest realistic arrival is roughly 55 minutes after the event. On a 20 minute time constant that is nearly three time constants, which is about 95 percent of the way back to ambient - the fault is fully erased before anyone could get there. The three prior clean visits were not bad techs. They were an impossible schedule.
Two things change on visit four. First, evidence: a careful look at the assembly finds fine dark powder at one mounting face and a faint oxidation ring around a single terminal in a row of otherwise clean ones. Both survive the reset, and both say repeated movement plus localized heating at that one point. Second, a witness mark is scribed across the joint before leaving.
Visit five is scheduled to arrive at the 5 hour 15 minute mark of a run, before the expected 6 hour fault, and the tech is there when it happens. The witness mark is now visibly offset, confirming real movement across the cycle. At temperature, the joint reads high resistance relative to its neighbours; 45 minutes later, cold, it reads the same as its neighbours.
The fix addressed the mismatch rather than the symptom: a fastener and joint arrangement appropriate for the actual temperature swing, rather than retorquing a joint that will simply walk again on the next cycle. Retorquing alone would have bought roughly one season, which is what the customer had already been buying from a previous contractor.
How to verify you got this right
The proof for an expansion fault is a matched pair of readings on the same point in two thermal states, taken with the same instrument and the same technique. Cold reading, hot reading, same probe, same placement. If the two match, the fault is not thermal expansion no matter how compelling the story. If they diverge, you have the fault on paper and a customer conversation that writes itself.
The failure mode to watch in yourself is accepting a hot reading without a cold one from the same point. A high resistance at temperature means nothing without the baseline, because it might be high in every state and simply be how that joint reads.
What changes the answer
- A very large temperature swing (equipment that goes from a freezing outdoor sit to a full-load run) can produce enough movement to fault a joint that is otherwise built correctly, in which case the answer is an expansion allowance rather than a component replacement.
- A short cooling time constant, common on small light assemblies with good airflow, can put the reset inside 10 minutes and make an on-site catch essentially impossible without instrumentation. Plan to log, not to observe.
- A joint with no dissimilar materials still expands, so a fault can come from a temperature gradient across a single material rather than from a materials mismatch. The mechanism is the same; the fix is different.
- If the evidence shows heat discoloration or deformation, stop treating this as a subtle expansion fault. That is a genuine overheat and it needs a cause found, not a joint improved.
References
- NFPA 70B, recommended practice for electrical equipment maintenance
- Published linear thermal expansion coefficients for common structural and conductor metals
- Trade-standard practice for expansion allowances and dissimilar-metal joints
- See related: Why Thermal Expansion Explains More Faults Than Techs Expect; The Witness Mark or Data Logger Method; How to Measure a Component at Operating Temperature