A cool thermal image is not the same as a healthy circuit. Here is the step that separates the two.
During a routine electrical thermal imaging survey, a distribution board read a maximum surface temperature of 36 degrees Celsius. On a route covering dozens of boards in a shift, a reading like that does not slow you down. It was cooler than the room on a warm day, no gradient worth annotating, no discoloured connection, nothing that earns a second look. If the thermal image were the whole story, that board would have been marked clean and I would have moved on.
The reading I could not move past was the one sitting next to it in the record. The circuit was drawing current equal to a fraction of one percent of the device's rated capacity. It was almost unloaded. A component that reads 36 degrees while carrying almost nothing is not the same as a component that reads 36 degrees under full load.
This is where load correction earns its place in a survey, and it is the step I see skipped most often.
What a surface temperature actually tells you
A radiometric camera measures apparent surface temperature at the moment of capture, under whatever load the circuit happens to be carrying at that moment. That last part is the problem. Electrical load is rarely constant. Survey a board at seven in the morning before a plant wakes up and you are measuring it at a fraction of the current it will carry by mid-afternoon.
The heat a resistive fault produces scales roughly with the square of the current through it. Halve the load and you roughly quarter the heat rise. A genuine fault that would be obvious at full load can sit well inside the noise at light load, effectively invisible.
Closing the gap
Load correction closes that gap. By recording the measured current, the device rated or full-load current, and the ambient temperature alongside the thermal reading, you can estimate what the component would run at if it were loaded to 100 percent. The method is well established and appears in most thermography training syllabuses. It takes the measured temperature rise above ambient, scales it by the load factor, and projects a corrected temperature at full load.
On that 36 degree board, the correction told a different story. Projected to full load, the connection estimated out at just over 81 degrees. Measured against a commonly used reference of 75 degrees for conductor operating temperature, that is an exceedance. The image was cool. The corrected figure was not. Without the load reading, the finding does not exist.
Why this changes how you grade a survey
Once you correct for load, you can grade severity in a way that survives the conditions on the day. A raw temperature threshold punishes you for surveying under heavy load and rewards you for surveying under light load, which is backwards. A load-corrected figure, compared against a reference temperature, gives you a severity band that means the same thing whether you scanned at dawn or at peak demand.
The reference temperature itself depends on the device, its listing, and the applicable code and manufacturer documentation for your jurisdiction, and I would send any reader to those sources rather than treat a single number as universal. What holds regardless is the principle. A temperature reading without a load reading is only half a measurement. You cannot separate expected operational warmth from developing resistive heating unless you know how hard the circuit was working when you looked at it.
The discipline that makes it work
Load correction only helps if the inputs are captured at the panel, not reconstructed afterwards. You need the measured current, taken with a clamp meter where the permit to work and safe access allow, the device nameplate or rated current, and the ambient. Miss any one of them on site and you are guessing later, or quietly dropping the correction and reporting the raw image.
The practical failure I see is that load data lives in a separate notebook, a photo of the clamp meter, or nowhere at all, and it never makes it into the same record as the image. By the time the report is being written up, the current reading is gone and the finding defaults to what the camera saw. This is a workflow problem more than a technical one. The correction is simple arithmetic. Getting the three numbers reliably attached to every scan, while you are still standing at the board, is the part that takes discipline.
The takeaway
A thermal image answers one question well. Is this component generating heat right now, under the load it happens to be carrying. That is useful, but it is narrower than the question most asset owners think they are paying for. The question that protects an essential circuit is whether it has margin left when it is working hardest, and that cannot be answered by temperature alone.
If your programme reports surface temperatures without load correction, it is not wrong. It is incomplete. And the assets where that incompleteness bites are exactly the ones running quietly at light load on the day you happened to visit, handing you a clean image and a false sense that nothing needs doing.
