Visual inspection in production is often done by a person: they look, measure and decide. That works while volumes are small and the criteria are simple. When there are many parts, the defects are small or every part has to be measured, camera-based quality inspection is often more reliable and cheaper. A camera does not get tired, checks every part the same way and sends the result straight to a spreadsheet or a PLC.
2D or 3D inspection
The first question is whether the defect is visible in a photo or has to be measured in space.
- 2D inspection. Colour, contrast, contours in a plane, text, codes, whether a part is in place. Industrial cameras with suitable lighting do the job.
- 3D inspection. Height, depth, flatness, deformation, dimensions in space, dents and bumps that do not show up in a photo. This calls for 3D scanners and 3D cameras.
3D technologies differ. ToF cameras are fast and see the whole scene, but their accuracy is at the millimetre or centimetre level. Structured light scanners such as Photoneo PhoXi achieve sub-millimetre accuracy but scan more slowly and are sensitive to motion. According to the manufacturer, Photoneo MotionCam-3D can also scan moving objects. For more, see 3D machine vision and ToF 3D cameras.
What can be checked
Machine vision is not only for detecting defects. A camera can check:
- dimensions: length, diameter, height, distances between features;
- shape: flatness, bending, deformation;
- surface defects: scratches, cracks, dents, missing material;
- completeness: whether all components are present and in the right place;
- marking: text, QR codes and barcodes, instrument display readings;
- sorting by a measured parameter.
Shiny surfaces without matting spray
Shiny metal surfaces reflect light, so 3D scanners often see "holes" or false points on them. The usual fix is to coat the part with matting spray, but that is impractical in production. We scan shiny surfaces without spray. In general terms, this is achieved by combining several exposures (HDR), choosing the scanner angle and position (the scanner can be moved by a robot), filtering out reflection artefacts and merging scans from several poses. A trial will show whether this works for your part.
Examples from our work
- Sorting logs by thickness (sawmill, 2023). A camera measures the log diameter, and logs are sorted accordingly.
- Reading meter displays on a metrology bench: 4 cameras read the LCDs of water meters once per second, and the readings are compared with reference values and written to Excel. More in reading instrument displays with cameras.
- Wagon inspection (pilot project, 2026). A robot with a 3D scanner and cameras checks wheel profiles, axle boxes and springs. Each finding is linked to a scan and a photo, and the inspector makes the decision. AI segmentation is used to find the axle boxes.
When a camera will not replace a person
Not every defect can be seen reliably. If a defect is invisible even to a person, if the criteria are hard to define in words or numbers, or if the part's position varies a lot every time, the cost and complexity of the system go up. That is why the purpose of a trial is not only to show what works but also to state clearly what the camera will not check. Sometimes the best solution is for the camera to check what can be measured, while a person decides only the unclear cases.
How a project runs
- Task definition. What counts as a defect, what the limits are, what the throughput and cycle time are.
- Samples. Good and bad parts: the more real defect samples, the better.
- Trial with real parts. Before any larger investment, we test cameras or 3D scanners on your parts and show what the system sees and what it misses.
- Equipment selection. 2D or 3D, camera, lens, lighting, mounting, and whether a robot is needed.
- Software. Measurement algorithms, neural networks where needed, the operator panel.
- Commissioning. Results are compared with manual inspection, and the criteria are agreed with the quality department.
Where the results go
- To a spreadsheet or database: measurements and the decision for every part, for traceability.
- To the PLC: a "good / bad" signal, sorting or part rejection. We do the PLC and HMI work ourselves; see our automation services.
- To the operator panel: the image, the result and the reason a part was rejected.
We start quality inspection projects with a trial on your real parts. Find out more in our machine vision services and projects. Send us a few good and bad parts or photos of them: get in touch, and we will tell you what a camera can check.