How to Use a Thermal Image as Evidence Rather Than a Picture

Why this matters

A thermal image is persuasive out of all proportion to what it proves. It is bright, it is obvious, and a customer looking at an orange blob on a blue background does not need to be sold. That is exactly why it gets used badly: the image closes the conversation, so nobody records what would let anyone re-derive it later.

Six months on, somebody asks the only question that matters: is the finding getting worse. By then the file has a temperature in it and nothing else, and the temperature in a thermal image is not a property of the fault. It is a property of the fault together with how hard the equipment was working, how warm the room was, what surface the camera saw, and where the operator stood. This card is organized around the questions your image will be asked later, because each answer can only be captured while you are standing in front of the equipment.

Before the camera comes out

A thermographic survey is taken with the equipment energized and loaded, because an unloaded joint is at room temperature and has nothing to show. That makes it energized work under 29 CFR 1910.333(a)(1), which permits it where de-energizing introduces additional or increased hazards or is infeasible because of operational limitations. A survey that only exists under load qualifies, and that is a determination you make and defend rather than a default. Work to the approach and arc-flash boundaries your employer's electrical safety program sets under NFPA 70E-2021 in the edition it has adopted, which binds you through that program or your contract, and use the electrical protective equipment required by 29 CFR 1910.335(a).

Do not remove a cover, a barrier or a dead front to get a better image. Opening energized gear changes the incident-energy exposure the risk assessment was built on, and doing it to improve a photograph is the weakest justification there is. Where the target is behind a barrier, the answer is an infrared window fitted for the purpose.

Step 4 below asks you to measure load, which usually means a clamp inside the enclosure. That reading carries its own instrument requirements: check the measurement category marked on the clamp and its jaw, inspect the jaw faces, and work inside the same boundaries. The sibling card on what a clamp measurement can and cannot tell you owns that.

The one relationship that makes an image portable

Resistive heating at a defect rises with the square of the current through it. Twice the load through the same bad joint is roughly four times the heating, not twice. That single fact is why an image taken on a mild day at partial load and one taken on a design day at full load are not comparable, and why nearly every "it got better" conclusion in a thermal file is an artifact of load. Two consequences follow, and both drive the steps below.

Compare against a reference in the same frame, not against an absolute number. The rise of a defect above a similar, similarly-loaded component next to it is far more robust than either one's absolute temperature, because ambient, emissivity error and reflected energy affect both alike and largely cancel. The sibling card on what a hot spot in a thermal image actually means owns that comparison; take it as given here.

Normalize that rise to a stated load before you compare visits. Corrected rise equals measured rise multiplied by (reference current divided by measured current) squared.

That correction assumes resistive heating, a defect resistance that has not changed much between visits, and a reference component carrying the same current as the defect. Where the defect is not a resistive joint, a phase is unbalanced, or the reference is a different kind of component, it is an approximation and the record should say so. Pick one reference current for the whole file, typically the circuit's rated or expected full load, and never change it.

Step 1: answer "which component is that?"

A thermal image of a terminal block is a row of rectangles, and in six months nobody, including you, can tell which one. Take a visible-light photograph from the same position immediately after, and note the identifier: panel designation, breaker number, phase, terminal position counted from a stated end.

What you lose by skipping it: the finding cannot be re-inspected. The next tech opens the panel, sees five identical lugs, and either re-surveys everything or guesses.

Step 2: answer "compared to what?"

Get a valid reference into the same frame. On a three-phase feeder, the other two phases at the same point. On a bank of identical contactors or motors, the neighbour. Where nothing comparable exists in frame, capture an unloaded section of the same conductor and say in the record that the reference is not load-matched. Record it as a number, not just a picture: the file may be read later by someone with different software or none.

What you lose by skipping it: the finding reduces to an absolute temperature, carrying ambient, emissivity error and reflection all uncorrected, and cannot be compared to anything.

Step 3: answer "how hot was the room?"

Record ambient air temperature and the reflected apparent temperature you used. Both belong in the record even though the delta-T comparison in step 2 largely cancels them, because they are what lets a later reader judge whether the cancellation was valid. A defect in direct sun, or beside a hot duct, sits in a reflected environment its reference may not share.

What you lose by skipping it: you cannot tell later whether an apparent change between visits was the equipment or the weather.

Step 4: answer "what was it doing?"

Measure the current at the time of the image and write it down with the image. Not "under load", not "running", not the nameplate. The measured amps, on the conductor you imaged, at the moment you imaged it, taken under the gate above. Record what the system was doing too: which stages were on, whether the compressor had just started. Those tell a later reader whether your current was representative or transient.

What you lose by skipping it: everything. Without the load, the image cannot be normalized, and a repeat visit compares two numbers taken under unknown conditions. This is the field that most often turns a good survey into a picture.

Step 5: answer "what did you assume about the surface?"

Record the emissivity value you set and how you chose it. A painted surface, a bare copper lug and an oxidized bus bar behave very differently, and the sibling card on emissivity owns the mechanism. What belongs here is the record, so a later reader who disagrees can recompute rather than re-survey. If you applied a known-emissivity target such as tape or a paint spot, say where: that mark will still be there next visit.

What you lose by skipping it: two surveys of one joint at two emissivity settings produce a trend that is entirely instrument setup.

Step 6: answer "could you actually see it?"

Record the distance, the angle, and whether the image was taken through an infrared window. A target smaller than the camera can resolve at that distance under-reads toward the background, silently, and the sibling card on distance-to-spot ratio owns that geometry. An oblique angle enlarges the measured area and degrades effective emissivity at once. An infrared window carries its own transmission factor that must be entered into the camera or the reading is low.

What you lose by skipping it: an under-read that looks like a mild finding, on the small targets where a hot spot matters most.

Step 7: answer "is it getting worse?"

Compute the delta-T against the in-frame reference, then normalize to your stated reference current. Record all three: measured rise, measured current, corrected rise. The corrected rise is the only one that means anything across visits; the other two let someone check your correction.

What you lose by skipping it: a file of numbers that cannot be lined up, which is what almost every thermal file in the trade actually is.

The worked example: two visits, one lug, opposite conclusions

A three-phase disconnect feeding a rooftop unit. The reference current for the file is the circuit's expected full load, 130 A, stated once and used for both visits. All figures are illustrative and stand in for the ones on your own job.

Visit one, in spring. Phase B lug 118 F, phases A and C 96 F each at the same point, so the measured rise against the in-frame reference is 22 F. Measured current on phase B: 78 A. Corrected rise is 22 multiplied by (130 divided by 78) squared: 1.667 squared is 2.778, and 22 multiplied by 2.778 is about 61 F. Recorded as 22 F measured rise at 78 A, corrected to about 61 F at 130 A. Watch item, re-survey in six months.

Visit two, in autumn. Phase B lug 104 F, phases A and C 91 F, so the measured rise is 13 F. Measured current: 52 A. Corrected rise is 13 multiplied by (130 divided by 52) squared: 2.5 squared is 6.25, and 13 multiplied by 6.25 is about 81 F.

Read the two raw numbers the way a picture file would be read. The rise fell from 22 F to 13 F, a drop of about 41 percent, and the absolute temperature fell too, from 118 F to 104 F. Every visible number improved, and the file would say the joint settled down.

Read the corrected numbers. The rise went from about 61 F to about 81 F, an increase of about 33 percent, on the same joint, against the same reference current, at the same point on the same conductor. The joint is degrading, and it degraded across an interval in which the raw file said it improved.

The autumn survey caught the equipment at 52 A because the day was mild. Load was down to 40 percent of the reference current, and because heating goes with the square of current, that alone cuts the rise to roughly 16 percent of what full load would produce. The defect getting substantially worse was not merely hidden by that; it was inverted.

The failure mode. A shop with the raw file closes the watch item at visit two. The lug fails the following summer, at the load that was never measured. The record shows two surveys, both diligently performed, that jointly proved nothing because neither carried its load.

What would change the answer. Taken at similar currents, the raw and corrected comparisons point the same way and the correction is a formality. It only matters when load differs between visits, which in seasonal equipment it almost always does. And if phase B were carrying a different current from A and C, the in-frame reference is no longer load-matched, the delta-T against it is not clean, and the record has to say so rather than quietly applying the same correction.

What this example still does not settle. A corrected rise is a severity indicator, not a diagnosis: it does not distinguish a loose lug from a corroded face, an undersized conductor or a dissimilar-metal joint. The severity thresholds themselves come from your specification or your client's inspection standard, in the edition adopted.

How to verify the image will still be worth something next year

• Hand the file to someone who was not there and ask them to name the component, the reference, and the load. If they cannot, the image is a picture.
• Confirm the reference current is identical in every entry. One changed reference current invalidates every comparison in the file.
• Re-do one correction by hand. Confirm the direction: a survey taken at lower current corrects to a larger rise, never a smaller one.
• Check the raw numbers survived alongside the corrected ones. A file of corrected values alone cannot be audited or re-corrected.
• Confirm that no cover was removed for the sake of the image, and that any cover that did come off came off under a documented energized-work determination with the boundaries and PPE that determination requires. Removing a dead front changes the incident-energy exposure the risk assessment was built on, and it changes the thermal environment too, so an open-cover image and a closed-cover image are not comparable to each other either.

References

• 29 CFR 1910.333(a)(1) - requirement to de-energize, with the infeasibility and increased-hazard exceptions that cover a survey only available under load
• 29 CFR 1910.335(a) - electrical protective equipment for work on or near energized parts
• NFPA 70E-2021 - approach and arc-flash boundaries, binding through your employer's electrical safety program or your contract in the edition adopted
• Camera manufacturer's documentation for emissivity, reflected apparent temperature, window transmission and spot size
• See related: What a Hot Spot in a Thermal Image Actually Means; What a Distance-to-Spot Ratio Decides About an Infrared Reading