What does a thermal camera measure in quality inspection?
A thermal camera measures the infrared radiation coming from a surface and estimates surface temperature from it. It is useful in quality inspection wherever a defect shows up as a temperature difference at the surface: heat-sealed packaging, hot-melt glue, energised electrical connections, parts fresh out of a mould. The camera does not see inside the part; it sees only the radiation leaving the surface, which is the surface's own emission and whatever it reflects from its surroundings, combined.
Separating the camera's strength from its limit is the first design step. The product page of the FLIR A50/A70, a typical camera sold for automation, shows it well: its uncooled microbolometer detector, working in the 7.5–14 µm range, has a thermal sensitivity below 35 mK, while its absolute accuracy is ±2 °C or ±2% of reading, for ambient temperatures of 15–35 °C and objects above 0 °C. In other words, the camera can tell apart differences of a few hundredths of a degree but states absolute temperature with an uncertainty of a few degrees. That is why the quality decision should rest, as far as possible, on differences within the same image.
Which defects can a thermal camera detect?
A thermal camera helps when a reliable temperature difference exists between defective and good parts at the moment of measurement. If there is no such difference, or it does not reach the surface, a better camera will not change that.
| Application | Thermal signature | When to measure | Watch out for |
|---|---|---|---|
| Heat-sealed pouches, trays and blisters | An unbroken warm band along the seal; a gap or cold area points to a missing seal | A fixed time after the sealing station | Film reflections, cooling rate, product residue on the seal |
| Hot-melt glue on cartons | Presence, position and continuity of glue lines | Right after application | Glue cools quickly; timing is critical |
| End-of-line electrical test | A hot spot when powered (loose connection, faulty solder joint, heating element) | Under a defined load, after a defined time | The product must be powered safely during the test |
| Moulded plastic and cast parts | An irregular cooling pattern | At ejection from the mould | Small cycle-time differences change the pattern |
| Print, colour, surface scratches | Usually none | Not suitable | A visible-light camera fits better |
The appeal of thermal seal inspection is that it checks every pack rather than a sample. The Sterile Barrier Association's article on heat seal inspection with IR cameras reports that this inspection runs online, allows every single package to be inspected without damaging it, and, depending on the material, can also reveal contamination and other damage to seals. Whether it can replace your existing, validated seal strength tests is a separate validation decision; make it together with your quality team.
How do emissivity and reflections distort the measurement?
Emissivity is how efficiently a surface radiates, on a scale from 0 to 1. FLIR's emissivity guide gives practical values: highly polished metals such as copper or aluminium usually have an emissivity below 0.10; roughened or oxidised metals 0.6 or more; flat-finish paints about 0.90; human skin and water about 0.98. Shiny surfaces behave like mirrors and reflect radiation from hot objects nearby. According to the guide, if the target's emissivity is below 0.5 you are unlikely to get an accurate temperature measurement.
The meaning on the line is clear. On metallised film, an aluminium foil lid or a shiny metal part, the "hot area" the camera sees may be a reflection of the sealing jaw, a heater, a lamp or an operator. Fix the camera angle, move hot sources out of the reflection path of the field of view or shield them, and validate the measurement again whenever the surface type changes. During calibration, sticking a tape of high, known emissivity on one point of the surface and comparing readings against it is a practical method the guide also suggests.
How many pixels do you need on the target?
An accurate temperature reading needs the target to fill more than one pixel. FLIR's guide to measuring distance states that when measuring temperature with a thermal camera you want at least 3 × 3 pixels on the target; do not base an accept or reject decision on a single-pixel reading.
Hypothetical calculation: if your seal band is 5 mm wide, putting at least three pixels across it needs at most about 1.67 mm per pixel. At a horizontal resolution of 640 pixels, that means the camera's field of view can be at most about 1.07 m wide across the line; with a 464-pixel detector the limit drops to about 0.77 m. If several lanes run side by side on the belt, this calculation directly decides how many cameras you need.
Why do measurement timing and line speed matter?
A thermal signature fades. As the part cools after sealing or gluing, the difference between good and defective areas shrinks. So measure after a fixed delay, and define the delay by position rather than time: when the line speeds up or slows down, the same distance corresponds to a different cooling time. Take the trigger from an encoder and record line speed with every measurement. The product's incoming temperature varies too: a tray arriving from cold storage and one at room temperature look different after the same seal. If you can measure incoming temperature, add it to the record.
Motion creates a second problem. A study by Srinivasan Ramanagopal and colleagues, published at Robotics: Science and Systems 2020, shows that uncooled microbolometers suffer from significant motion blur, caused by the thermal inertia of each pixel. On a fast line, a narrow seal band can blur and the temperature difference can spread. Before installation, capture at real line speed, measure how far the edges of the band spread, and if necessary move the measurement to a point where the part slows down or stops.
How do you set the decision rule?
Do not look for a universal temperature threshold. Base the decision on differences within the same image: the seal band's temperature difference from a reference area on the same part, breaks in the profile taken along the band, and the length and position of a break. This approach largely compensates for changes in ambient temperature and in the product's incoming temperature, and it relies on the camera's thermal sensitivity rather than its absolute accuracy.
Derive the limits from measured process variation. Measure known-good parts and defective parts confirmed by the quality team across shifts, products and line speeds, and set the pass, inspect and reject bands from those distributions. Route borderline parts to human review instead of forcing them through a single sharp threshold.
How do you validate a thermal station before go-live?
Calibration shows that the camera reads correctly; validation shows that the station decides correctly. They are separate jobs. During the shadow run, compare the station's decisions with technician checks and keep three counts separately by shift and product: missed defects, needless rejects and unmeasurable frames. To see whether the thermal decision really reflects seal quality, regularly pull samples from both flagged and passed packs and compare them with your existing destructive seal test. If you test only flagged packs, you will never see the defects that escape. Track the unmeasurable-frame rate as a quality indicator too; a sudden rise can point to a change in the camera, the lens or the station's surroundings.
A thermal calibration checklist
| Check | What to do | When |
|---|---|---|
| Reference reading | Place a reference of known temperature next to the measurement area and log its reading; stop automatic rejection if the deviation exceeds the site tolerance | Start of shift and after maintenance |
| Surface card | Record the material, surface finish, viewing angle, distance and emissivity setting for the station | At installation and when the material changes |
| Reflection check | Capture a cold part; any area that looks warm on a cold part is a reflection | At installation and when the station's surroundings change |
| Pixel count and focus | Confirm that at least 3 × 3 pixels fall on the narrowest measurement area and that focus is sharp | At installation and when the lens or mounting changes |
| Measurement delay | Record the distance between the process step and the image, and the line speed | Every measurement |
| Ambient conditions | Record ambient temperature and airflow at the station | Continuously |
| Known samples | Run one confirmed good and one confirmed defective sample per product and log the result | Start of shift |
| Shadow run | Before connecting the reject gate, run a period that only logs decisions, and compare them with technician checks | Before go-live and after major changes |
| Revalidation | Rerun the whole list when the lens, mounting, film, line speed or sealing temperature changes | On change |
Hypothetical example: incomplete seals on tray lids
On a food line, a film lid is heat-sealed onto plastic trays. The thermal camera sits a fixed distance after the sealing station, placed so that the 6 mm seal band covers four pixels across in the image. The decision rule rests on the band's temperature difference from a reference area at the tray's edge and on the length of breaks in the profile along the band. During the trial week, the quality team prepares trays with product residue deliberately left on the seal band and compares their thermal images with confirmed good trays.
In the second week, a product with a metallised lid film enters the line, and the camera starts seeing a reflection of the sealing jaw on good trays as well. This product needs its own station card, a different viewing angle and newly set decision bands; until then it is inspected manually.
Common mistakes
- Treating a bright pixel as proof of a defect.
- Deciding on an absolute temperature threshold instead of differences within the same image.
- Entering the emissivity setting once and not updating it when the material changes.
- Not calculating how many pixels fall on the target.
- Assuming measurement timing is independent of line speed.
- Not testing motion blur at real line speed before installation.
Whether the camera is thermal or visible-light, the general steps of acceptance testing, line integration and monitoring after go-live are the same; we describe those steps on our visual quality inspection service page.