Weld Cooling Rate Measurement by Thermography: t8/5 in Practice

Cooling rate measurement by thermography gives you t8/5 without drilling a hole for a thermocouple, and that convenience is exactly where the mistakes come from. An infrared camera does not measure temperature: it measures radiance and converts it under assumptions about the surface, the path and the optics. When those assumptions are wrong the curve still looks plausible, the interval still computes, and the number goes into a report nobody can defend later.

This article is about the measurement engineering. For the metallurgy behind the interval — why the cooling rate governs hardness, toughness and cracking risk — read heat input, t8/5 and microstructure. For why a single PQR coupon is not enough and per-weld measurement is what closes the gap, read why measure T8/5 cooling time on every production weld.

Key takeaways

• t8/5 is the time the material spends between 800 and 500 °C. A camera reports the surface, and surface is not bulk.
• Frame rate has to be set from the fastest joint you weld, not from the camera’s headline figure.
• Emissivity drifts while you measure, because the surface oxidises through the very interval you are timing.
• An emissivity error does not cancel between the two crossings; it biases the interval.
• Arc radiation, fume and spatter are measurement problems, not nuisances, and each has a specific fix.
• A thermographic t8/5 is credible once it has been compared with instrumented coupons and the comparison is on record.

What t8/5 is, and what infrared can see

The interval, and why procedures care about it

t8/5 is the time taken to cool from 800 °C to 500 °C. That window covers the austenite decomposition range for common ferritic steels, so it is a compact proxy for the microstructure the heat-affected zone will end up with. Recognised guidance such as EN 1011-2 provides equations relating heat input, joint geometry and preheat to the resulting cooling time, and the t8/5 cooling time calculator and heat input calculator implement the arithmetic side of that relationship.

An estimate is not a measurement. The value of thermography is that it reports what a particular joint actually did, including the effects a nominal calculation cannot know about: an unexpected fixture acting as a heat sink, a preheat that had decayed, a travel speed that drifted.

Surface temperature is not bulk temperature

A camera sees a surface. The interval that matters metallurgically happens through the section, and the surface leads or lags it depending on thickness, geometry and the heat sink behind the plate. On thin sections the difference is small; on thick sections it is not. So the measurement location has to be fixed, and production numbers have to be compared against a coupon measured the same way — which is what HeatCore AI evaluates per pass, rather than against a textbook value.

Four parameters that decide whether the number is real

Frame rate against the steepest part of the curve

The curve is steepest just after the arc leaves. If the camera samples too slowly, the 800 °C crossing is interpolated across a large temperature step and the interval inherits that error. Set the rate so that several samples fall inside the interval on your fastest joint, then record the rate as part of the measurement, because a later reader cannot infer it.

Spatial resolution and the measurement spot

Each pixel integrates over an area. If that area straddles the fusion boundary, the reported value is a mixture of two thermal histories and the crossings land in the wrong place. Work out the ground footprint of a pixel at your working distance and confirm the region of interest is several pixels inside the zone you mean to measure. A welding camera mounted for process observation is usually positioned for coverage, not for radiometric resolution, so this is worth checking rather than assuming.

Emissivity on a surface that is changing while you measure

Steel oxidises as it heats, and the emissivity of the oxide is different from that of the metal underneath. Within one cooling interval the surface can therefore change its radiative behaviour. Because an emissivity error shifts the apparent temperature, both crossings move, and because they sit on parts of the curve with different slopes, the errors do not cancel. Whether the residual bias lengthens or shortens the recorded interval depends on the shape of the cooling curve at each crossing, not on the sign of the emissivity error alone, so the direction has to be established on the joint being measured rather than assumed.

Spectral band, viewing angle and the arc

The arc radiates strongly and reaches the detector both directly and by reflection off the plate, so the practical answer is to place the region of interest behind the torch and outside the arc’s line of sight. Viewing angle matters separately: apparent emissivity falls away from normal incidence, so a steep oblique view biases every frame in the sequence. Background radiation from hot fixtures and adjacent plate is a third path into the same error.

Setup checklist for a t8/5 measurement

• Region of interest fixed by drawing or fixture mark, not chosen per weld by whoever is operating.
• Frame rate set from the shortest interval expected, and written into the procedure.
• Pixel footprint at working distance calculated and confirmed against the zone width.
• Emissivity determined on the real surface condition, at temperature, against a contact reference.
• Reflected background temperature entered rather than left at a default.
• Viewing angle as close to normal as the fixture allows, and recorded.
• Arc excluded from the region of interest, with ignition and extinction frames discarded.
• Fume extraction running, optics protected, cleaning interval scheduled.
• Temperature range of the camera configured to cover both crossings without saturating.
• Raw sequences retained, not just the computed interval, so the number can be recomputed.

Step-by-step: extracting t8/5 from a thermal sequence

1. Record the whole cooling event, starting before the arc leaves the region of interest and ending well below the lower crossing.
2. Apply the measured emissivity and background, not the defaults, and note that the emissivity you determined belongs to that surface condition only.
3. Extract the time-temperature history for the fixed region of interest, tracking it if the part or torch moves.
4. Reject contaminated frames — arc glare, spatter transits, extraction plume — explicitly, and record how many were rejected.
5. Interpolate the two crossings and state the interpolation method; with a coarse frame rate this choice is a real contributor to the result.
6. Compute the interval and attach the uncertainty, propagated from temperature uncertainty through the local slope of the curve.
7. Store the raw sequence with the result. An interval without the sequence behind it cannot be re-examined when someone disagrees.

Failure modes and fixes

• Frame rate too low on thin sections. The interval is shorter than a few samples and interpolation dominates. Raise the rate or accept that the measurement is only valid on thicker joints.
• Region of interest drifts onto the cap. The number changes with no visible cause. Anchor the region to a tracked feature rather than to fixed image coordinates.
• Emissivity taken from a table. Tables describe idealised surfaces. Determine it on your surface, at temperature, and re-determine it after any change to preparation or shielding gas.
• Saturation at the upper crossing. The camera clips near 800 °C and the interval starts late. Configure the range to cover the whole event with margin.
• Spatter on the optics. Readings fall gradually and nothing alarms. Schedule cleaning, and trend the measured background as a health indicator.
• Only the interval is stored. Nobody can audit a single number. Retain the sequence, the settings and the rejected-frame count together.

Verification against embedded thermocouples

Thermocouples embedded in a qualification coupon remain the reference for this measurement. Instrument the coupon close to the observed region, run the procedure that production will run, and compare the two curves — not only the two intervals. Two curves of different shape that happen to yield a similar interval agree by coincidence, and coincidence does not survive a change of thickness or preheat.

Repeat the comparison across the range of thickness, preheat and heat input the shop actually uses, and keep the disagreement as a number so drift becomes visible. Practitioner background on thermal cycles and on the inspection chain around them is available from TWI Job Knowledge, the International Institute of Welding and the NDT Resource Center.

From a measured t8/5 to a control limit

A validated measurement becomes useful when it turns into a limit the cell can act on. Set the band from your qualification evidence, alarm on excursions rather than reporting them at shift end, and keep the per-weld record alongside the electrical parameters captured by welding parameter monitoring, so an excursion can be traced to what the machine delivered. HeatCam IR-C provides the cylindrical coverage tube and pipe work needs; the same architecture appears in orbital tube welding cooling rate monitoring, where an excursion shows as heat tint and penetration rather than hardness alone.

Where the governing document is a code, the cooling condition is controlled indirectly through preheat, interpass and heat input limits: see the ASME Boiler and Pressure Vessel Code for procedure-qualification variables, AWS D1.1 Structural Welding Code — Steel for structural steelwork, and ISO 13916 for how the associated temperatures must be measured in the first place. Cracking risk is the reason all of this exists, which is covered in hydrogen-induced cracking prevention.

FAQ

What frame rate does a thermal camera need to measure t8/5? Fast enough that several samples fall inside the interval on your fastest joint. Thin sections cooling in a second or two need a much higher rate than thick multi-pass work. Fix the rate from the shortest interval you expect, then record it, because the rate is part of the measurement.

Can t8/5 be measured on the weld cap or must it be taken on the parent metal? Both are used, and they are not the same measurement. The cap cools differently from the fusion boundary and its surface is changing as slag and oxide form. Whichever you choose, fix it in the procedure and keep it identical between qualification and production.

How much does an emissivity error change a measured t8/5? More than most people expect, and it does not cancel. An emissivity error shifts the whole apparent curve, so both crossings move, but they sit on parts of the curve with different slopes. The bias grows as the curve flattens near the lower crossing.

Does arc light corrupt an infrared cooling curve? It can. Radiation from the arc and from the plasma plume reaches the detector directly and by reflection off the plate. The usual answer is to measure behind the torch, outside the arc field of view, and to discard the frames around ignition and extinction.

How is a thermographic t8/5 validated against a thermocouple? Instrument a qualification coupon with thermocouples close to the observed spot, run the same procedure, and judge agreement from how closely the two full curves track each other, not from where they cross 800 and 500 °C. Two curves that cross at nearly the same times but diverge everywhere else have not been validated against each other.

Is a calculated t8/5 acceptable instead of a measured one? For procedure development it is a reasonable starting point, and the recognised equations exist for that. It is an estimate from nominal inputs, so it cannot show what the joint actually did on the day, which is what a measurement adds to the record.

Can t8/5 be measured through welding fume and spatter? Fume attenuates and scatters, and spatter on the optics changes the reading without producing an obvious fault. Extraction, a clear observation angle, protective optics and a scheduled cleaning interval are what make a long production measurement trustworthy.

Therness solution: Welding camera — Thermal and high-speed imaging that watches the arc, the pool and the cooling seam while the weld is made.

Written by Luca Santoro , founder