A robot thermal image is not a condition-monitoring result

Check what the target occupied in each capture — method added 16 October 2026

A repeated waypoint does not guarantee a repeated thermal measurement. Before interpreting a difference, recover the camera and lens, the exact target region, working distance, viewpoint and focus evidence for both captures. A route can complete successfully while one of those inputs remains unknown. Keep the original files and their capture context available to the reviewer.

FLIR's measurement-distance article, published 12 September 2023 and checked 16 October 2026, relates small-target measurement to resolution and field of view. Digital zoom enlarges the existing pixels rather than improving that measurement capability. Its coverage guidance is manufacturer guidance for the described method, not a universal acceptance rule for every camera and target.

Fluke's undated focus guidance, checked 16 October 2026, explains that focus affects apparent-temperature measurement as well as image sharpness. Record the focus check for each capture. Do not turn an example in that article into a correction factor, an error budget or proof that your own image is adequately focused.

Use the configured camera's instructions to document the target-coverage check and its evidence. Record the selected region of interest separately from a general asset name: two images of the same motor can still select different surface regions. Where a viewpoint or distance tolerance is needed, retain the responsible specialist's approved rule and reference. This guide supplies no tolerance or pixel-count threshold.

For a hypothetical comparison, the baseline distance is recorded as 1000 mm and the later capture as 1500 mm, using the same reference points. The entered distance increased by 500 mm. That arithmetic does not establish how detector coverage changed, whether a temperature is valid or whether the comparison should be rejected. The reviewer still needs the lens, framing, focus, coverage evidence and the remaining operating conditions.

The paired geometry record now preserves those entries for both captures, lists missing evidence and flags changed distance, camera/lens or framing text. It exports the inputs, arithmetic and reviewer note together. Matching entries never produce a comparability pass; different file references are not inspected, and a filled reviewer note is not validated. An unresolved focus or coverage check stays a review issue even if the broader comparison register is complete.

Use the record to assign a specific next action: recover the missing evidence, request a controlled repeat capture or obtain specialist assessment. Do not approach equipment more closely or alter a robot route merely to improve an image without the site's responsible controls. Geometry evidence supplements the existing load, emissivity, reflection and environmental checks; it does not replace them or authorise maintenance or continued operation.

Make the comparison boundary explicit — method added 1 October 2026

Before asking whether a robot's thermal observation has changed, ask whether the two observations answer the same question. A different load, surface treatment, viewpoint or measurement setting can change the comparison boundary. Preserve those differences alongside the images rather than hiding them in an apparently continuous trend.

Fluke's first-party inspection guide distinguishes a reference capture made while equipment is operating as intended, repeated observations of the same component and comparisons of similar components under comparable conditions and loads. It also notes that meaningful anomaly differences vary by component. This is guidance for choosing a comparison method, not an acceptance rule for a particular asset.

FLIR document T810605 identifies emissivity and reflected temperature as important measurement inputs and allows some settings to be local to a measurement tool. Record the settings actually used for each region of interest; a camera-level setting alone may not describe a saved measurement. This documented behaviour is product-specific and does not verify field accuracy.

A practical review packet has two capture references, their dates and operating conditions, the approved comparison method, each material difference and the responsible reviewer's disposition. Our local thermal comparison record keeps these side by side. Its completeness indicator counts missing evidence fields; it does not decide whether an observation is normal, comparable or safe.

Consider a hypothetical inspection in which the baseline was captured at steady production and the next visit occurred during warm-up. The record should show the different operating states and a repeat-capture or specialist-review decision, not normalise the difference by guessing a correction. If the load was not recorded, preserve it as unknown. A reviewer may withhold the comparison even when both image files are readable.

The same discipline applies after a changed camera, lens, target coating or analysis pipeline. Retain the earlier evidence for its original configuration and identify what must be re-established. Agree who owns the next action and where its completion will be recorded. RobotAtom's method organises that handoff; it does not replace qualified thermography, asset diagnosis, site controls or the manufacturer's instructions.

Image collection answers a narrower question

A mobile robot can carry a thermal camera to an inspection point and collect radiometric data. That does not by itself establish whether the measurement conditions were suitable, whether an apparent temperature difference is meaningful or what maintenance action should follow.

The useful requirement is the complete condition-monitoring result: a defined asset and target, controlled collection conditions, recorded measurement assumptions, an agreed comparison and severity rule, qualified review, attributable evidence and a maintenance disposition. Robot mobility supports collection; it does not turn every coloured image or threshold alert into a verified diagnosis.

Separate the four decision layers

A supplier may provide several of these functions in one product workflow, but they remain different claims. A successful robot mission does not verify the temperature result, and a threshold crossing does not establish root cause or required maintenance.

• Robot mission: reach the inspection location and present the sensor at the required pose without bypassing the site's operating or safety controls.
• Thermal capture: acquire the required raw or radiometric data with the camera, optics, settings and environmental inputs identified.
• Thermographic interpretation: compare the observation with an approved reference or rule and assess whether an anomaly exists and how severe it may be.
• Maintenance decision: assign review, escalation, work order, repeat inspection or no-action disposition with the evidence and reasoning retained.

Define the asset, target and operating state

ISO 18434-1:2008 remains current after confirmation in 2023. Its public scope covers infrared thermography for machine condition monitoring, including procedures, severity-assessment criteria, data interpretation, reporting and compensation for measurement influences. That scope supports treating thermography as a controlled programme rather than a camera feature, but it does not validate a particular robot, threshold or maintenance conclusion.

• Identify the asset, component and exact surface or region of interest using the site's controlled asset record.
• State why the measurement matters and which failure, degradation or operating question it is intended to inform.
• Define the required asset state, load, speed, warm-up time, nearby equipment state and other process conditions needed for a useful comparison.
• Record expected surface, coating, insulation, obstruction and background variation instead of assuming every visit presents the same target.
• Name the reference state, comparable asset, baseline or trend used and the conditions under which comparisons are permitted.

Control collection geometry and conditions

• Define the sensor viewpoint, distance, angle, focus, field of view and target framing, including an allowed tolerance where the responsible specialist requires one.
• Identify reflections, hot or cold backgrounds, dust, steam, weather, airflow, sunlight and other conditions that can alter or obscure the observation.
• Record any window, shield or external optic between the camera and target and the evidence used for its transmission treatment.
• Confirm that the route, stopping location, robot body and payload can reproduce the required view without introducing an obstruction or unsafe inspection practice.
• Preserve failed, obstructed or off-condition captures as exceptions rather than silently treating them as normal results.

Keep measurement assumptions with the data

FLIR's current first-party guidance says measured radiation depends on the target's emissivity, reflected radiation and atmospheric absorption. It instructs users to supply parameters including emissivity, reflected apparent temperature and distance, with atmospheric temperature, relative humidity and external-optic properties relevant in supported workflows. These are FLIR technical instructions, not independent proof that a particular field value is accurate.

Do not invent a convenient emissivity, alarm temperature or correction value to complete a record. The responsible thermography and asset specialists should define the method, source and uncertainty appropriate to the surface, camera, conditions and consequence of error.

• Retain the exact camera, lens, robot payload, calibration status, software and analysis versions.
• Store the source measurement settings and environmental inputs with the capture rather than only a rendered image.
• Identify whether the retained file contains radiometric or raw temperature data, a processed export or only display colours.
• Record the asset state, inspection pose, timestamp, route or waypoint version and any operator intervention.

Define interpretation before automating alarms

ISO 18434-2:2019, confirmed in 2024, provides guidance on thermogram interpretation within a machine condition-monitoring and diagnostics programme. Its public scope includes severity criteria, methods, image interpretation, assessment and reporting. It does not prescribe one universal threshold, inspection interval or automated diagnosis for every asset.

• Define the comparison, threshold, persistence and severity rule before reviewing production results, including who approved it and what evidence supports it.
• State which changes require human review and which, if any, may create an automated notification or work order.
• Record false alarms, missed or invalid captures, unresolved anomalies and the reason an observation was accepted or rejected.
• Keep raw data, rendered view, analysis result, reviewer conclusion and maintenance disposition as separate evidence stages.
• Assign the competence, authority and escalation route needed for thermographic interpretation and asset diagnosis.

Treat robot-platform features as manufacturer claims

Boston Dynamics' current Spot 5.1.4 documentation says Spot CAM+IR can capture raw or PGM thermal files and documents the array format, temperature units and a software-version-specific byte-order exception. Its thermal solution page claims scheduled scans, regions of interest, alarm thresholds, trending and work-order integrations. These statements are manufacturer evidence for the named platform and versions; they do not independently verify measurement accuracy, an appropriate threshold, machine condition or maintenance outcome at a buyer's site.

The file-format detail illustrates why configuration control matters. A data pipeline should preserve and correctly interpret the documented format and version before it produces a temperature, visualisation or trend. A technically readable file still needs the measurement and interpretation controls described above.

Validate and maintain the installed workflow

Before operational acceptance, freeze the robot, route, inspection action, payload, camera, optics, software, target definitions, settings, comparison rules, reviewers and maintenance interface. Exercise representative assets and operating states, including reflections, obstructions, invalid measurements, communication loss and the agreed escalation path. Compare robot-collected results with the reference method selected by the responsible specialists and retain every deviation.

Reopen affected evidence after a material change to the robot, route, waypoint, sensor, optic, calibration, software, file parser, asset surface, operating state, environment, threshold, analysis method or maintenance workflow. Preserve the earlier result as evidence for the old configuration rather than carrying its conclusion forward.

Keep every conclusion inside its evidence boundary

ISO is authoritative for the published scopes and status cited here. FLIR and Boston Dynamics are authoritative for their own published instructions, formats and product claims; those statements remain manufacturer evidence until the configured workflow is validated under the buyer's conditions.

RobotAtom records inspection requirements, conditions, versions, source dates, evidence stages, owners, unknowns and required validation. It does not certify thermographic competence, temperature accuracy, machine condition, maintenance decisions, robot safety or regulatory compliance. Qualified thermography, reliability, maintenance, integration and safety owners must decide what applies and whether the evidence is sufficient for the actual application.