Reading Metal Additive Manufacturing Thermal Images: What Does the Colour Mean?

A bright patch in a metal additive manufacturing video can be useful evidence, but it is not automatically a trustworthy temperature reading. Engineers evaluating laser powder bed fusion should first identify what the instrument actually measures, how the image is converted into a temperature, and what conditions limit comparisons between builds. This guide is for people reviewing measurement claims and planning a discussion with a machine or sensor supplier; it does not prescribe settings for producing a qualified part.

Start with the signal, not the colour scale

A thermal camera or pyrometer receives radiation from the surface within a particular spectral range. A coloured image is a display of processed sensor information; the colours depend on the chosen scale. Two screenshots can look equally hot while representing different ranges, and the same raw scene can look different after the display is rescaled. Ask for the underlying measured quantity and its units, the wavelength range, the exposure or integration time, and whether the image shows radiance temperature or a temperature estimated using an emissivity assumption.

NIST's research on emissivity in powder bed fusion explains why an optical detector does not directly measure temperature and why calibration and surface emissivity matter. Its results concern the surfaces and conditions studied there; they are not a universal correction factor for every alloy, machine, or viewing arrangement.

Ask how the surface assumption was obtained

Emissivity describes how a surface emits radiation relative to an ideal reference under specified conditions. Surface texture, oxidation and viewing angle can affect it. A fresh powder layer, a solidified track and a molten region need not behave alike. If a supplier reports an absolute temperature, request the assumed emissivity, how it was determined, the relevant material and process state, and how a change in that assumption affects the result. A single value copied from an unrelated experiment is a hypothesis to examine, not a certificate of accuracy.

Calibration records should identify the reference used, the measurement range and the conditions under which calibration applies. Keep camera settings and image processing versions with each dataset. If a protective window, filter or optical path changes, ask the measurement owner how that change is accounted for before comparing results. A sensor can be repeatable and still be biased; consistency alone does not establish an absolute value.

Check what one pixel represents

In a fast-moving melt pool, image exposure, optical blur and the size of the observed area can mix radiation from regions at different temperatures. A nominal pixel spacing is not by itself proof that the sensor resolves a feature of that size. Inspect the field of view, focus, spatial response and acquisition timing, especially when comparing images from different lenses or machine positions. Request a description of how the instrument aligns the image with the moving laser and how rapidly changing features are handled.

In its study of registering the laser position with the melt pool, NIST describes a specific reference method for its experimental setup. That work illustrates the need for spatial registration; it does not establish that an arbitrary commercial image has the same alignment or uncertainty.

Make comparisons defensible

Before a trial, agree on the exact question. For example: under the same material, camera configuration and analysis method, does a monitored region show a repeatable change when one documented process variable is changed? Preserve the material batch, machine, build location, layer, scan strategy, sensor settings and software processing alongside each observation. Mark any settings that changed between runs. A trend may help identify a condition worth investigating, but a trend alone cannot prove that an internal defect exists or that a part meets a specification.

A practical handover record can contain the sensor owner, calibration date, assumed emissivity, measurement units, field of view, timing source and known exclusions. Attach the raw data where available, not just a screenshot. Note any obscuration or unusual optical behaviour and describe whether the conclusion refers to a relative signal or a calibrated temperature. If evidence is missing, state that limit rather than silently filling in an estimate.

Turn a demonstration into a useful supplier conversation

When reviewing a proposed monitoring system, ask the supplier to show the same sample with its original scale, metadata and processing steps. Request a worked explanation of uncertainty for the actual material and operating conditions. Ask which claims have been validated against independent measurements, which remain development indicators, and how operators can retrieve data later. Agree on who owns the interpretation and what action, if any, a warning should trigger.