What thermography cannot see: five limits facility teams should know

6 min read · Updated 30 July 2026

Every measurement method has a lane, and trouble starts when marketing widens it. A thermal camera measures one thing: infrared radiation leaving a surface. Everything a survey concludes is inferred from that — which is why the five limits below are not defects in any particular camera or crew. They are the physics of the measurement, and they should be visible in how a survey is scoped, scheduled, and reported.

Limit one: no load, no findings.

Heat at a fault is made by current. A loose joint carrying a trickle reads as healthy as its neighbours; the same joint at full load can be the hottest thing in the room. So the scan's value is set before the camera arrives — by what the plant is running at the time. A survey timed for a quiet period produces a confident report about a system that was not really working. What covers it: scheduling under representative load, and a report that records what was running — so the reader can judge what the clean results are actually worth.

Limit two: it reads surfaces, not interiors.

Infrared does not penetrate metal, and most of what matters electrically lives inside an enclosure. Surveyors work around this honestly — opening agreed covers, reading accessible surfaces to infer concealed conditions — but an inference is an inference, and enclosed construction (MV switchgear above all) hides most of its interior from any camera. What covers it: a report that separates direct readings from indirect ones, and states what stayed closed — plus the next limit's remedy, which matters most exactly where enclosures are thickest.

Limit three: it cannot hear insulation failing.

A switchboard fails through its current path or through its insulation — and the camera only watches the first. Partial discharge, the process that erodes insulation toward flashover, produces sound and voltage transients rather than useful surface heat: on a thermal image, a discharging panel usually looks fine right up to the day it is not. What covers it: the PD screen — TEV and ultrasound reading through the closed panel — which is why a serious MV assessment runs both methods, not a camera alone.

Limit four: shiny surfaces lie.

How faithfully a surface reports its own temperature is its emissivity — and bare, shiny metal is the worst offender, radiating poorly and reflecting everything around it. Point a camera at polished busbar and you may read the ceiling lights, or a reflection of the thermographer, dressed up as a plausible temperature. What covers it: technique — reference targets, angle discipline, comparing like surfaces with like — and a trained eye that knows a colourful image is not yet a reading. This limit is the strongest argument that who holds the camera matters more than what the camera cost.

Limit five: a snapshot ages fast.

The scan describes one moment. Faults develop over weeks and months; loads change; repairs happen and quietly regress. A two-year-old clean report is a historical document, not a current condition statement — and a single scan, however good, has nothing to compare itself against. What covers it: a cadence matched to condition and consequence, with each survey reading the same points the same way — which turns isolated images into a trend, and is, declared interest, the reason our assessment is built as a year-on-year programme.

The test these limits give you.

None of this is secret knowledge — any competent provider knows all five. So the limits double as a vendor test: does the proposal talk about load conditions? Does the report separate direct from indirect readings and list what stayed closed? Is there any answer for the insulation side of the board? Does anyone mention emissivity, or is every image taken at face value? And is there a plan for the next scan, or does the engagement end at the invoice? A provider comfortable saying "here is what we will not see" has earned the right to be believed about everything else.