Infrared Thermography for Predictive Maintenance: A Practical Guide

Key takeaways

Infrared thermography reads surface temperature with a thermal camera and turns an abnormal hot (or cold) spot into an early warning of a developing fault.
It is fastest and strongest on electrical connections, motors, cooling systems and thermal processes, where a fault shows itself as heat before it fails.
Emissivity is the number-one thing to get right: shiny bare metal reflects and reads falsely low, so set emissivity to the target or read a high-emissivity surface.
Judge severity by temperature rise above a reference component or ambient, normalised for load, not by the absolute temperature alone.

Infrared thermography is the condition-monitoring technique that uses a thermal imaging camera to read the surface temperature of equipment and find faults that give themselves away as heat. Almost every developing failure changes how much heat a component makes or dissipates: a loosening electrical terminal, an overloaded cable, a failing bearing, a fouled heat exchanger, a stuck steam trap. Because a thermal camera captures thousands of temperature points in one non-contact image, a technician can screen a switchboard, a motor line or a process skid in seconds and spot the one component running hot before it trips or burns out. Done well it is one of the highest-return techniques in a predictive maintenance program.

What the camera is actually seeing

A thermal camera does not measure temperature directly. It measures infrared radiation leaving a surface and converts it to a temperature, assuming a value for how efficiently that surface radiates. Everything above absolute zero emits infrared energy in proportion to its temperature, so a hotter surface glows brighter to the camera even though nothing is visible to the eye. The output is a false-colour image where each pixel carries a temperature, letting you compare one component with its neighbours instantly.

The technique is a surface reading, and that is its main limit: it detects a fault by the heat that reaches the outside of the equipment, not by looking inside. A bearing three layers into a gearbox will show up only once its heat conducts to the housing. That is why thermography pairs so well with techniques like vibration analysis, which reads the mechanical fault directly.

Emissivity: the setting that makes or breaks a reading

Emissivity is how efficiently a surface radiates infrared energy, on a scale of 0 to 1, and it is the single biggest source of error in a thermal reading. A dull, painted or oxidised surface has high emissivity (around 0.95) and reads reliably. Bare, shiny metal such as a polished copper busbar or a stainless housing has low emissivity (often below 0.2), radiates little, and mostly reflects the temperature of its surroundings, so the camera reports a falsely low value on a component that may be dangerously hot.

The practical fixes are simple. Set the emissivity value in the camera to match the target material. Where you can, aim at a high-emissivity feature such as a painted terminal cover or a rubber boot rather than bare metal. For repeat inspection points, a dab of high-emissivity paint or a strip of electrical tape (emissivity about 0.95) gives a stable, honest target every survey. And always be aware of reflections: a low-emissivity surface can mirror a nearby hot object or your own body heat and fool you.

A worked example: classifying an electrical hot spot

Severity in thermography is about temperature rise, not absolute temperature, and it must be normalised for load. Take a three-phase motor starter. Under a measured 70% of rated load, the three line-side lug temperatures read:

L1 = 38 °C L2 = 41 °C L3 = 62 °C (ambient 30 °C)

The three phases carry a similar current, so L1 and L2 act as built-in reference components. The delta-T that matters is L3 against its healthy siblings:

ΔT = 62 − ((38 + 41) / 2) = 62 − 39.5 = 22.5 °C over similar components

A rise of more than about 15 °C over a similar component under similar load is widely treated as a serious fault to correct promptly, so this L3 lug (most likely a loose or corroded connection) is a plan-and-repair item, not a monitor item. The load normalisation matters too: at 70% load the connection is already 22.5 °C hot; heat at a resistive joint rises with the square of current, so at full load the same fault would run far hotter and could reach failure. That is the difference between "there is a warm terminal somewhere" and "re-terminate the L3 lug at the next planned electrical stop before it fails under peak load", the kind of specific lead time that keeps a fault off your electric-motor troubleshooting and controls firefight list.

Where thermography earns its keep

Some of the highest-value survey targets are:

Electrical distribution: loose or corroded connections, unbalanced phases, overloaded conductors and failing breakers, which are a leading cause of both downtime and fire risk.
Motors and drives: overheating windings, bearing housings running hot, and cooling-path blockage.
Mechanical power transmission: overheating bearings, couplings and belts on accessible housings.
Thermal process and utilities: blocked heat exchangers, refractory and insulation loss, steam-trap failures, and tank or vessel levels seen through the shell.

The P-F curve: why lead time is the prize

Thermography pays off because of where it sits on the P-F curve, the interval between the point a failure first becomes detectable (P) and the point of functional failure (F). A loose electrical joint or a degrading bearing generates measurable heat well before it fails, giving days to weeks of warning that lets you order the part, schedule the labour and correct the fault on a planned stop rather than losing the asset mid-shift. It also sets the survey interval: readings must be frequent enough, and taken under representative load, to catch the fault while useful warning remains and while the equipment is energised and working hard enough to reveal a resistive or friction fault.

How to start a survey route

You do not need a program on day one, just a disciplined route. A workable start: rank assets by criticality and consequence of failure, and list the electrical panels, motor lines and thermal skids that hurt most when they stop. Survey them under real load, because a fault under light load hides. Fix the emissivity and record the same inspection points, angle and load each time so the images trend comparably. Set a severity threshold (a delta-T rule as above) that decides monitor, schedule or act. And feed every finding and repair back into your work-order history, so you learn which assets actually fail and build the data for reliability-centered maintenance and a meaningful MTBF and MTTR picture.

As coverage grows, most plants hit the same ceiling: periodic manual surveys miss faults that develop between visits, and the thermal images live in a folder disconnected from the downtime they are meant to prevent. That is where continuous monitoring and machine-health software earn their place, streaming condition data against the same production and stop data operators already generate.

Comparing predictive-maintenance platforms?

See how predictive-maintenance and machine-monitoring tools stack up before you buy.

Read the predictive-maintenance guide

When the goal is not just a thermal warning but linking equipment health to real production loss and closing the loop to a work order, the platform we recommend is Fabrico: it reads OEE and stops directly from the machines and shows the true cause on video, so condition data and downtime data live in one place. Fabrico is EU-built, so your production data stays in EU jurisdiction (ISO 27001 / 20000-1 / 9001, supports audit-readiness). If that fits, you can book a Fabrico demo. The calculators and guides here are free regardless.

FAQ

What can infrared thermography actually detect?

It finds faults that show up as an abnormal temperature: loose or corroded electrical connections, overloaded circuits and unbalanced phases, failing motor bearings, blocked cooling and heat exchangers, steam-trap and insulation problems, and refractory or tank-level issues. It reads surface temperature only, so it detects a fault by the heat it produces rather than by looking inside the part. It is fast, non-contact and can survey a lot of equipment per shift, which makes it a strong first-pass screening technique.

What is emissivity and why does it matter?

Emissivity is how efficiently a surface radiates infrared energy, from 0 to 1. Dull, painted or oxidised surfaces are high (around 0.95) and read reliably; bare shiny metal is low (often below 0.2) and reflects surroundings, so the camera reads a false low temperature. Set the emissivity in the camera to the target, or measure comparatively, and where possible read a high-emissivity spot such as painted housing or apply a known-emissivity tape or paint. Getting emissivity wrong is the most common cause of a misleading thermal reading.

How do I judge how serious a hot spot is?

Use the temperature rise above a reference, not the absolute temperature alone. For electrical work, compare the component against a similar component under similar load (a delta-T between phases), or against ambient air. Common industry practice treats a small rise (about 1 to 10 C over a similar component) as monitor, a larger rise (roughly 4 to 15 C over ambient or a sibling) as a repair to schedule, and a large rise (tens of degrees) as urgent. Always normalise for load, because a fault under light load will look far worse at full load.

Is thermography better than vibration analysis?

They are complementary, not competing. Thermography is fast, non-contact and broad, and it excels on electrical, thermal-process and cooling faults; vibration analysis is the sharper tool for the mechanical health of rotating equipment such as bearings, alignment and imbalance. Strong programs run both, plus oil analysis and inspection, and feed every finding into the same reliability and work-order system so the warnings get acted on in time.