A single thermal survey tells you how an asset looked on one day. A trend tells you where it is going.
A single thermal imaging inspection is a photograph. It tells you how an asset looked on one particular day, under one particular load, in one particular ambient. That is genuinely useful, but it is also the least powerful way to use thermal imaging, because it throws away the one thing that turns an inspection into a prediction. That thing is time.
I was reminded of this on a board I have now imaged across several annual visits. On the most recent scan it looked unremarkable. Judged on that image alone it would have earned a clean line in the report. Judged against its own history, the picture was more interesting, and more useful.
Why the rate of change beats the reading
The absolute temperature of a component on a given day is a noisy number. It moves with load, ambient, time of day and season. Two honest surveys of the same healthy component months apart can produce meaningfully different readings for reasons that have nothing to do with its condition.
What is far less noisy, and far more diagnostic, is how a component behaves relative to itself over repeated inspections under comparable conditions. A connection that is slowly climbing cycle after cycle is telling you something a single snapshot cannot. So is one that spiked, was repaired, and has since returned to baseline. Trending turns a set of disconnected photographs into a direction of travel, and direction of travel is what predictive maintenance actually needs.
The closed loop is the proof
On that same site, a minor anomaly had been recorded on a previous visit. It was actioned, and this year the scan confirmed it had returned to normal. That closed loop, finding, action, verification, is easy to overlook, but it is the clearest evidence a programme is working. Without a trend, you never get it. You find something, you recommend a repair, and then you simply hope. The next one-off survey has no memory of the last, so it cannot confirm the fix held or catch a repair that quietly failed.
How to actually build a trend
Trending only works if you can compare like with like, and that takes discipline set up before the first scan, not after.
Give every inspected item a stable identity. A consistent asset reference and location, carried from one inspection to the next, is what lets you line up this year's image against last year's. Free-text descriptions that change wording every visit break the trail.
Capture under comparable conditions and record the ones you cannot control. Load and ambient will vary, so log them every time. Combined with load correction, this lets you compare corrected figures rather than raw readings that drift with the conditions on the day.
Inspect the whole population, not only the items that already look hot. A component that is trending upward but has not yet crossed a threshold is invisible if you only revisit known problems. The value of trending is catching things before they present, which means scanning things that currently present as fine.
Keep the history somewhere it accumulates. A trend that lives across a dozen separate PDFs in a dozen separate folders is not a trend. It is an archive nobody reads. The history has to sit together, per asset, so the pattern is visible at a glance.
Where this sits in a reliability programme
Thermography described this way stops being an annual compliance photograph and becomes a genuine condition-monitoring stream, sitting alongside vibration, oil analysis and the rest. The competence standards for that already exist. ISO 18436-7 addresses the qualification of thermographic personnel for condition monitoring, and it frames thermography as an ongoing diagnostic activity rather than a one-off inspection. The tooling has caught up to that framing. The practice, in a lot of programmes, has not.
The takeaway
If your thermal surveys are stored as a stack of standalone reports, you own a photo album, not a diagnosis. The individual images may be excellent and still tell you almost nothing about where an asset is heading, because none of them can see the others. Build the history, compare each asset against itself under known conditions, and close the loop on every finding you raise. That is the difference between recording that something was warm once and knowing that something is getting worse.
