What Does AI Visual Inspection Actually Catch?

AI visual inspection is a quality control method that uses cameras and trained software to compare each part against expected good and defective examples, catching problems a human inspector would otherwise have to spot by eye. What it catches depends entirely on whether the defect is visible in an image at all.

Surface defects

Cracks, pits, scratches, dents and porosity that break the surface are the most common category. These defects create a visible change in texture, shadow or reflectivity that a trained system can learn to recognize. TARKO's metal parts crack and surface inspection application is built around exactly this category.

Shape and dimensional deviation

A part that has warped, shrunk, or come out of a mould with a slightly wrong profile can be compared against a reference part and flagged when it falls outside tolerance. This applies whether the deviation comes from a moulding process or a stamping process. TARKO's plastic molding inspection application checks for exactly this.

Missing or incomplete features

A missing hole, an unformed rib, or a cavity that did not fully fill are all visible as an absence where something should be present. Short shots on moulded parts are a specific, well-known example of this category.

Assembly and completeness checks

Whether the right part is in the right place, whether a fastener is present, or whether a label is correctly applied are all checks that come down to comparing what is in the frame against what should be there.

Fill and moulding-specific defects

Short shots, where a cavity does not fully fill, and flash, where excess material escapes at a parting line, are a defect category specific to injection moulding. Both are visible as a deviation at the edge or boundary of the part, which is what makes them a good fit for camera-based detection. See TARKO's short shot and flash detection application for how this is scoped.

How catch rate is actually measured

A defect catch rate is reported against a specific, agreed set of defect types and severities, not as a single number that applies to every possible flaw. A system tuned to catch fine surface cracks will report a different number than one tuned to catch only gross shape errors. When a vendor quotes an accuracy figure, the useful follow-up question is which defect types and what severity threshold that figure was measured against.

Multiple defect types on one part

A single camera station can often check for more than one defect category at once, for example checking a moulded part's shape and its fill quality in the same pass. This depends on all the relevant regions of the part being visible from the same camera position, which is confirmed during setup rather than assumed.

What AI Visual Inspection Does Not Catch

A defect with no visual signature at the surface, such as an internal void deep inside a solid part, will not show up in a standard camera image no matter how well the model is trained. Chemical composition, electrical continuity and internal stress are examples of properties that are not visible optically. These require a different inspection method entirely, such as X-ray, ultrasonic testing or electrical test, in place of visual inspection.

Defects that are visible but occluded, for example a crack on the underside of a part that never faces the camera, also will not be caught unless the part is presented to the camera in a way that exposes that surface. This is a fixturing and camera placement problem more than a software problem, and it is usually solved during setup rather than after.

Where the line actually falls

A useful rule of thumb: if a trained human inspector could catch the defect by looking at the part under good lighting, AI visual inspection is a realistic candidate. If a human inspector would need to cut the part open, weigh it, or run an electrical test to find the defect, visual inspection is the wrong tool, and pairing it with a different test method makes more sense than expecting a camera to do that job.

If you are not sure which category your defect falls into, TARKO's free feasibility audit reviews the specific defect types you are trying to catch and tells you plainly whether visual inspection is a fit.