Thermal Imaging for Preventive and Predictive Maintenance

Thermal imaging is the easiest condition-monitoring technique to start and the easiest to do badly. A camera arrives, somebody walks the plant, and within a week there is a folder of orange and purple images with numbers on them. Some of those numbers are close to reality. Some are wrong by a margin that would embarrass anybody who understood how they were produced. The difference has almost nothing to do with the camera and almost everything to do with whether the person holding it understands what the instrument is actually doing. This article is about that understanding, and about the applications, limits and programme discipline that follow from it.

The message up front: a thermal image is evidence to be interpreted, not a measurement to be read off. Most of the maintenance value in thermography comes from comparison, one component against a similar one, or against the same component in a previous survey at comparable load, rather than from absolute temperature figures. Build the programme around comparison, record the conditions, and thermography becomes a genuinely powerful early-warning technique. Build it around the number on the screen and you will chase reflections.

A note on scope before anything else. This article is thermography in depth, as its own discipline. If what you need is the wider survey of how the condition-monitoring techniques compare, which one suits which failure mode, and what each costs to run, that belongs to the sibling guide on condition monitoring techniques: vibration, thermography and oil analysis. Read that for the comparative picture. Read this for the thermal technique itself.

One safety statement, made once: a great deal of the highest-value thermography is done on energised electrical equipment. Inspecting energised equipment, and in particular opening an enclosure on live equipment to get a line of sight, is a hazardous activity that belongs only to competent, authorised people working under a risk assessment with the protective measures, approach limits and permit arrangements set by your own organisation and the law of your jurisdiction. Nothing in this article authorises anybody to open anything. Work from your own electrical safety rules, and if the safe answer is that the panel stays shut until an inspection window is fitted, that is the answer.

1. What a thermal camera actually measures

Every object above absolute zero emits infrared radiation. A thermal imager contains a detector array that responds to the infrared energy falling on it within a particular wavelength band, and produces, for each pixel, a signal proportional to the radiant energy received. That is the measurement. Everything else is arithmetic.

The temperature you see on the screen is the output of a model that converts received radiation into a surface temperature. That model needs inputs, and the camera cannot discover them for you. It uses whatever values are set, including the defaults somebody left in place three years ago. The important inputs are these.

• Emissivity. The efficiency with which a particular surface radiates infrared energy compared with an ideal radiator. Painted, oxidised, dull and organic surfaces radiate efficiently and behave predictably. Bare, polished and plated metals radiate poorly. Since the camera is working backwards from received energy to temperature, a poorly radiating surface that is genuinely hot can appear far cooler than it is. This is not a small correction. It is the difference between finding a fault and walking past it.
• Reflected apparent temperature. Whatever a surface does not emit, it tends to reflect. A low-emissivity surface is, in the infrared, close to a mirror. It will show you the radiation of everything around it: a lamp, a hot motor across the aisle, the sky, the sun, the person holding the camera. The camera cannot tell the difference between energy the target emitted and energy the target reflected.
• Distance and the field of view. A target has to be large enough in the image for the instrument to report its temperature rather than an average of it and its surroundings. A small, genuinely hot component viewed from far enough away is blended with cooler background and under-reported: the pattern may still be visible while the number is meaningless.
• Angle of view. Emissivity is not constant with angle. A shallow, oblique view degrades the measurement and increases the reflected contribution, and awkward access, which is normal in plant rooms and switchrooms, pushes you towards exactly those angles.
• The path between camera and target. Air, humidity, dust, steam, smoke and any intervening glazing or plastic attenuate or block the signal. Outdoors and over distance this matters. Through ordinary glass it defeats the measurement entirely.

The consequence that should change how you work: a shiny metal surface can read wildly wrong, and a reflection can look exactly like a hot spot. Those two facts, taken seriously, separate useful thermography from the output of somebody who bought a camera. There is a simple field test for the second one. Move, and look again from a different angle. A reflection tends to travel across the surface as you move. A real thermal feature stays attached to the component. That single habit removes most false positives before they reach a report.

2. Qualitative versus quantitative thermography

Thermographic work splits into two modes, and being clear which one you are in resolves most arguments about accuracy.

Qualitative, or comparative, thermography looks for an anomaly by difference. One phase against the other two under similar loading. One of six identical pumps against its five siblings. One section of roof against the sections either side. Today's image of a component against the image taken last quarter under comparable conditions. The finding is the difference, and because emissivity and reflection affect the compared items similarly, many of the measurement assumptions largely cancel. You do not need to know the surface temperature to know that one of three identical terminations is behaving differently from the other two.

Quantitative thermography attempts an actual surface temperature. That is a legitimate and sometimes necessary goal, and it demands considerably more: a known or measured emissivity for the specific surface, a measured reflected apparent temperature, controlled distance and angle, recorded atmospheric conditions, a calibrated instrument within its calibration interval, and a competent person who understands the residual uncertainty. It is where process temperatures, compliance measurements and engineering calculations live.

The practitioner's point, and not the one people expect from a technique that produces numbers: most maintenance value comes from the comparative approach. Finding the odd one out is what catches developing faults. Where the surface is uncertain, prefer the comparison and say so in the report. A finding written as "phase B termination on this outgoing way is running materially warmer than phases A and C at similar indicated load, pattern consistent with a high-resistance connection" is useful, defensible and honest. The same finding reduced to a single figure taken off a polished lug is neither.

3. Load dependence, and why a survey without it is worthless

This is the second thing people get wrong, and it is the one that quietly wastes whole surveys. Resistive heating depends on current. A high-resistance connection that will eventually fail may produce almost no detectable thermal signature at light load, and an obvious one at full load. It follows that a thermal survey of an electrical system taken during a quiet shift, on a standby feeder, out of season, or during a partial shutdown can pass a system that has a serious developing fault in it.

Two habits fix this.

• Record the conditions with every image. Indicated load or current, ambient conditions, time of day, what was running and what was not, and anything that had recently been switched or operated. An image without this context cannot be compared with anything, including itself a year later.
• Schedule for representative load. Survey when the system is carrying a load that resembles the load it normally carries, and try to repeat at comparable conditions. In a climate like the Gulf that means being deliberate about season for anything cooling-related, because a chiller hall in February and the same hall in August are different installations as far as a thermal survey is concerned.

Without load context, a survey is not comparable with the last one, and the trend, which is where the real power of the technique lives, never forms. This is also the honest reason some "annual thermal survey" line items on a maintenance contract deliver nothing. The survey happened. It happened at whatever load the day provided, with no conditions recorded, and produced a report that cannot be compared with the previous one.

4. The applications, and what each one actually sees

Thermography is unusually broad, which is part of its appeal and part of the trouble, because breadth encourages people to point it at everything. Here are the applications worth running, with what the camera sees and what it will miss in each.

• Electrical connections and terminations. The classic and highest-return application. A loose, corroded, under-torqued or contaminated connection has elevated resistance, and current through resistance produces heat. The camera sees that heat, often long before anything else notices, and the intervention is usually cheap. It will miss connections it has no line of sight to, and under-report on bare metal.
• Electrical distribution generally. Imbalance between phases, overloaded conductors, degraded cabling, failing breakers and contactors, and overheating inside panels and distribution boards. The comparative approach is strongest here, because three-phase systems come with built-in comparison. It misses faults not yet dissipating heat, and anything behind a closed metal door.
• Motors and drives. Winding and bearing heat, blocked or failed cooling, fouled fins, uneven heating hinting at electrical imbalance, and coupling and alignment symptoms showing as heat at the coupling. It misses early internal electrical faults and reports bearing problems late.
• Mechanical applications. Bearings, couplings, belts, gear casings and misalignment. Honest note, and it matters: thermography usually detects these mechanical faults later than vibration analysis does. By the time a bearing housing is measurably hot, a vibration route or an ultrasonic check would typically have carried the defect for some time. So on rotating equipment thermography complements vibration rather than replacing it. If you have to choose one technique for critical rotating machines, that choice is covered in the condition monitoring techniques guide, and the warning-time argument behind it sits in the P-F curve explainer.
• Steam and condensate systems. Traps blowing through or blocked, passing or misseated valves, buried and lagged line routes, and lost or wet lagging. It misses internal mechanical condition and is confused by recently operated equipment.
• Refrigeration and HVAC. Blockages and restrictions in circuits, coil fouling and uneven coil loading, blocked filters, failed dampers, duct leakage patterns, and refrigerant problems inferred indirectly from the pattern across the circuit rather than observed. Routine thermal checks fit into the task sets in the HVAC preventive maintenance guide. It misses anything needing pressures, flows or gas analysis.
• Building envelope and insulation. Heat loss and gain paths, missing, displaced or wet insulation, thermal bridging, air leakage and glazing performance. A mainstay of energy work, with a real constraint: envelope thermography needs a temperature differential across the envelope to show anything at all, which limits when it can be done and in hot climates usually means working at inconvenient times of day or year.
• Moisture and water ingress. Detected through evaporative cooling and differences in thermal mass, not by seeing water. The camera infers moisture from a thermal behaviour rather than observing it, which matters, because several other things produce the same behaviour and a finding should be confirmed by another method before anybody opens a ceiling.
• Tank and vessel levels. Thermal stratification makes a liquid level visible through the vessel wall where contents and ambient differ and the wall is thin enough. Useful when instrumentation is absent or suspect; it fails on well-insulated vessels and with no differential.
• Refractory and furnace linings. Thinning, cracked or failed refractory, hot spots on shells and ducts, and lost insulation on high-temperature plant. It sees the shell, not the lining, and infers the lining from the shell pattern.

5. What thermography cannot do

Every technique deserves an honest limits section, and thermography needs a long one, because its ease of use invites people to ask it questions it cannot answer.

• It sees surfaces, and only surfaces. Everything the camera reports is the outside of something. Internal condition is always an inference from a surface pattern.
• It cannot see through a closed metal enclosure. This is the single biggest practical constraint in real electrical surveys. Point a camera at a shut switchboard door and you have surveyed a door. This is precisely why infrared inspection windows and ports exist, and why fitting them during a project, or at the next opportunity, is one of the most useful things a maintenance team can specify. Where windows are not fitted, the alternative is gaining a view under your own electrical safety arrangements, which is a scheduling, permit and competence problem well before it is a thermography problem.
• It cannot see through most materials. Glass, most plastics, painted cover plates, lagging and enclosures all block or distort it. A finding through glazing is not a finding.
• It cannot find faults that produce no thermal signature. Plenty of real and serious failure modes do not express themselves as heat at the surface, at least not until very late. If the dominant failure mode of an asset is not thermal, thermography is the wrong technique for that asset no matter how convenient it is.
• It does not diagnose. It localises. The camera tells you that something here is behaving differently. Deciding what is wrong and why is a separate act of engineering judgement that draws on the asset, its history, its duty and the other evidence available.
• It is poor at absolute accuracy on uncertain surfaces. On an unknown, shiny, oblique or distant surface, the number on the screen deserves no confidence at all, even though the instrument displays it with the same authority as any other reading.
• It will happily mislead you. A reflection, a sun-loaded surface, a component that was operated ten minutes ago, friction heat from something adjacent, warm air movement: all of these produce convincing thermal features, and the camera will present them without comment and let you call them faults.

6. Interpretation and reporting: the honest core

A thermal survey produces two possible deliverables. One is a document somebody can act on. The other is a gallery of colourful images with a covering page. The difference is entirely in what the report records.

A useful report states, for every finding, what was seen, under what conditions and load, what the comparison was, what the probable cause is, what action is recommended and by when, and how confident the thermographer is and why. The uncertainty statement is not a hedge. It is the most useful line in the finding, because it tells the engineer reading it whether to schedule work or to schedule another look.

On severity, I will be explicit rather than helpful. This article publishes no severity criteria and no threshold bands. Severity classification depends on the equipment type, the comparison basis used, the applicable standard or testing framework in your industry, your own programme rules and competent judgement. Those frameworks exist and they differ from one another, and a criterion lifted from a web article and applied to equipment it was never written for is worse than no criterion, because it looks authoritative. Take your scheme from your own programme and the applicable published standard, and record in the report which scheme and which comparison basis you used.

The last row is the one that decides whether the programme survives. A finding that does not become work does not exist. Getting from a thermal finding to a raised, prioritised, closed-out job with a cause recorded against the asset is the same discipline described in the failure codes guide, and the system-of-record side of it is covered in the introduction to CMMS.

7. Competence and certification

Thermography is a skill, not a purchase. The instrument is easy, the interpretation is not, and the gap between the two is where bad decisions come from. A camera in untrained hands does not produce no information. It produces confident errors, which is considerably worse, because a wrong finding consumes credibility as well as money and a missed finding leaves the fault in service with a clean bill of health attached to it.

Two distinct things need to be right, and they are governed separately.

• The work. International standards cover thermographic inspection procedure and the interpretation of thermal images: ISO 18434-1:2008, Part 1 on general procedures, and ISO 18434-2:2019, Part 2 on image interpretation and diagnostics. These govern how the inspection is carried out and how the resulting images are read.
• The person. Personnel qualification is not in that series at all. It sits in the ISO 18436 series, which certifies people, not organisations. Within that series, ISO 18436-7:2014 covers thermography and ISO 18436-2:2014 covers vibration, with further parts for field lubricant analysis, lubricant laboratory analysis, acoustic emission at ISO 18436-6:2021 and ultrasound at ISO 18436-8:2013. Parts 1 and 3 address the certification and training bodies themselves. No company can hold an ISO 18436 certification, and any organisation claiming to be "ISO 18436 certified" has misunderstood or is hoping you will.

Carry that distinction into procurement. When you engage a thermography service, the question is who is certified to do the interpretation, at what level and under which scheme, not what camera they own. In-house, budget for training and recertification rather than hardware alone, and accept that a competent thermographer doing fewer, better surveys beats a wider programme run by whoever was available.

8. The standards landscape, accurately

None of the documents below is law anywhere by itself. They bind through contracts, specifications, client requirements and professional practice, and almost all of them are paywalled, so treat what follows as a map rather than a substitute for the published text.

• The ISO condition monitoring and diagnostics family is the natural home for a thermography programme. ISO 13372:2012 fixes the vocabulary. ISO 17359:2018 is the programme-level wrapper that points at the rest. ISO 13379-1:2025 covers diagnostics, which is exactly the step where a thermal image becomes a statement about what is wrong and why, and ISO 13381-1:2025 covers prognostics. ISO 13374, in its published parts on data processing, communication and presentation, is the one integration work collides with when thermal results have to live alongside other condition data.
• Thermography specifically: ISO 18434-1:2008 and ISO 18434-2:2019 for the inspection and interpretation, and the ISO 18436 series, notably ISO 18436-7:2014, for qualifying the person. Beyond those, thermographic inspection practice is governed by the standards and personnel certification schemes that apply in your own field and by your client's specification. Work from yours rather than a generic reference, because electrical testing practice in particular varies considerably between industries and regions.
• Terminology: EN 13306:2017, "Maintenance, maintenance terminology", is a European CEN standard with no ISO twin. Preventive maintenance splits there into predetermined and condition-based, with predictive defined as a form of condition-based. Thermal survey work is condition-based maintenance in that scheme, whichever label the contract uses.
• Dependability vocabulary: IEC 60050-192:2015, Part 192 of the International Electrotechnical Vocabulary, is the formal home of the reliability and maintainability terms behind mean time between failures and mean time to repair. Worth knowing, because arguments about whether a thermography programme improved anything usually turn out to be arguments about what those clocks measure.
• Failure and maintenance data: ISO 14224:2016, third edition, is the taxonomy and data-format standard for collecting and exchanging reliability and maintenance data, and the right model if you want thermal findings analysable years later. Note that OREDA is a proprietary members-only database, not a standard; ISO 14224 is the standard that grew out of that work.

Published texts are available from ISO and IEC . Check the current edition before citing anything in a specification, because parts of these series were revised recently.

9. Where thermography fits in a maintenance programme

Thermography earns its place in four distinct roles, and it is worth separating them because they are budgeted and scheduled differently.

• As a screening and detection technique on a route. Regular thermal rounds over electrical distribution and selected plant, scoped by criticality rather than applied to everything. The thermal route is cheap per asset, which tempts people to include the whole register. Resist that: coverage driven by consequence beats coverage driven by convenience. The ranking that should drive it is in the asset criticality classification guide and, in more depth, the equipment criticality analysis guide.
• As an inspection at commissioning and after electrical work. The genuinely underused application. A thermal check on a panel shortly after it has been worked on, at representative load, catches the connection that was not properly remade while the job is fresh and the contractor still available. The same applies at handover, where a thermal survey belongs in the commissioning pack. Few organisations do this, and the ones that do find things.
• As a complement to other techniques, not a replacement. Vibration and oil analysis detect most rotating-equipment faults earlier, and ultrasound detects several conditions thermography cannot see at all. Thermography belongs alongside those, matched to failure modes that express themselves thermally. Which technique suits which failure mode belongs, again, to the condition monitoring techniques guide.
• As part of a condition-based rather than calendar-based posture. Thermal results should influence what gets done and when, not simply record that a survey occurred. Where that sits in the wider strategy is the subject of preventive versus predictive versus reactive maintenance and, for the prediction layer, the guide to predictive maintenance and failure prediction. I will not re-argue strategy here.

On continuous monitoring: fixed thermal sensing is increasingly practical, and for a small number of very critical, hard-to-access items it makes sense. The trade-offs sit in the guide to IoT sensors for predictive maintenance rather than here. For most organisations, the handheld route done properly delivers far more than a partial fixed deployment done for its own sake.

10. Baselines, trends and the data problem

The power of thermography is not in a single image. It is in the second image, and the twelfth. A trend across comparable surveys tells you whether a warm termination is stable or deteriorating, and that is a completely different decision from the one a single image supports.

Three things make trending possible.

• Fixed inspection points. The same components, from the same positions, through the same windows, at the same settings. Establish these before the first survey, because retrofitting inspection points later invalidates the baseline you already collected.
• Comparable conditions. Repeat at similar load and similar ambient conditions, and record both every time. A survey that cannot be matched to conditions cannot join a trend.
• Results recorded against the asset. Not against the survey. Against the asset, in the maintenance system of record, so that the thermal history of a specific switchboard is retrievable by anyone looking at that switchboard, years later, without knowing which consultant did which survey.

That last point deserves bluntness. In most organisations I have looked at, the thermal images sit in a folder nobody can search, attached to a report nobody re-reads, named after the date of a visit by a contractor since replaced. The technique did its job; the organisation lost the output. The fix is unglamorous and generic: raise the finding as work against the asset, record the outcome and cause against the asset, and attach the image or report to that record rather than to a parallel document library. Any competent maintenance system will do this. What matters is that somebody specified it in the survey scope.

11. How thermography programmes disappoint

The failure patterns are consistent, and every one of them is organisational rather than technical.

• A camera bought and used by nobody in particular. No named owner, no route, no competence requirement, no schedule. The instrument becomes an occasional curiosity.
• Surveys done at light load. Convenient timing produces clean reports and misses the faults that only appear under real duty.
• No baselines, therefore no trends. Every survey is a first survey, and every finding has to be judged in isolation.
• Images with no conditions recorded. Unusable for comparison, and therefore unusable for the technique's main strength.
• Reflections and solar loading called faults. Work orders raised against nothing, followed by a quiet loss of faith in the whole programme when the technician finds a healthy component.
• Enclosures that cannot be opened. The survey covers what was accessible rather than what mattered, and the report does not say so, which is the part that turns a limitation into a misrepresentation.
• Findings that generate a report but no work order. The most common failure of all, and the least excusable, because the detection already happened and was then thrown away.
• The honest one: the technique's low effort makes it easy to do badly and feel productive. A thermal survey produces a satisfying volume of output for a small amount of work. Volume of images is not a measure of anything. Findings that led to corrective work, and faults that did not become failures, are.

The idea to walk away with

Thermal imaging does not measure temperature. It measures radiation, and calculates a temperature from assumptions that you own: emissivity above all, then reflected apparent temperature, distance, angle, atmosphere and load at the moment of the image. Once you accept that, the technique reorganises itself around comparison. Compare like with like, compare similar components under similar load, compare against a recorded baseline, and be candid about surfaces you do not trust. Done that way it is one of the highest-return inspection techniques available to a maintenance team, particularly on electrical connections.

Done the other way it produces a folder of pictures, a few false positives, a couple of missed faults and eventually a quiet decision that the camera was not worth it. The camera was fine. The assumptions were never examined, the load was never recorded, the enclosures were never opened, and the findings never became work.

Final thoughts

If you are starting or fixing a thermal programme, the order I would recommend costs very little. Decide which assets deserve a thermal route, on consequence rather than convenience. Solve the line-of-sight problem for those assets, which usually means specifying inspection windows and planning safe views rather than buying anything. Fix the inspection points and settings before the first survey so that a baseline is possible. Require load and ambient conditions on every image. Require a probable cause, a recommendation and a stated uncertainty on every finding. Take severity criteria from your own programme and the applicable standard, and record which you used. Route every finding into the maintenance system as work against the asset. Then put a competent, certified person behind the camera, and remember that ISO 18436 certifies that person, not your company.

Do that and the technique will pay for itself repeatedly on electrical infrastructure alone, and it will complement vibration and oil analysis on the rotating equipment where those techniques see further ahead. Skip it and you will own a camera, a folder and a slowly growing conviction that thermography does not work. It does. It just does not work by itself.

Related reading: Condition monitoring techniques compared, Predictive maintenance and failure prediction, IoT sensors for predictive maintenance, Preventive vs predictive vs reactive maintenance, Preventive maintenance for HVAC systems, Asset criticality classification, Failure codes: Problem, Cause, Action, What is a CMMS.