Industrial cameras, optics, lighting and vision algorithms capture and judge the appearance, dimensions, hole positions, characters and defects of a part and output a result. The point is that the judgement drives sorting, release or isolation and can be recorded.

Six common classes: presence and assembly completeness, dimensions and geometry, holes and hole position, appearance and surface defects, characters and codes, and color difference. Which items and to what level follows the material, surface, tolerance and defect characteristics.

Plane dimensional items can be assessed: length, width, outside diameter, center distance, spacing, angle. Dimensional inspection is sensitive to locating, camera, lens, lighting, algorithm and environment, so the optics are derived backwards from the tolerance and repeatability is confirmed on sample parts.

It can be assessed, but hole diameter is a fine measurement item, a different level of difficulty from hole presence. It usually needs a suitable lens (telecentric where required), dedicated lighting, stable locating and a measurement algorithm, sometimes with a laser or displacement sensor.

Yes. Hole position is a relative position judgement and is usually easier than hole diameter, provided the locating datum is clear and the part arrives in the same position every time. Hole presence and hole count are easier still. The three levels are assessed separately.

It can be assessed for a specific surface and defect type, but the result depends on scratch width, length, depth and direction, reflection, material, lighting angle, camera resolution and background. Tests with OK samples, NG samples and real defects are required; no unverified detection limit is quoted here.

Part of the visible burrs can be inspected. Burr position, size, direction, color, occlusion, reflection and shape all affect the result. Very small burrs, hidden burrs and complex three-dimensional burrs may need multi-angle or 3D vision, laser, other sensors or a mechanical check.

Yes, but the defect types and the degree that counts as a reject have to be agreed first. Appearance inspection depends more on lighting, material, color, reflection and contrast than dimensional inspection, so the same phrase can mean completely different schemes.

Yes, and it is one of the more established classes: distinguishing two states needs less accuracy than a measurement. Where a single sensor solves it reliably, a sensor is the better answer.

It can be assessed. Common items are the presence of parts, screws, seals, pins and connectors, part orientation, and whether something is reversed, missing or wrong. It is judged position by position — this feature should be here or should not — and suits tests on real assemblies.

Yes, but two things have to be separated: OCR reads what the characters say, while character inspection judges whether they are present and meet the requirement. Whether a readable but badly printed character is a reject depends on the customer's criteria and is agreed at proposal stage.

It can read the encoded content. Again, reading and judging are different: a code inspection judges whether the code or QR code meets the reading or quality requirement.

2D handles plane images: plane dimensions, outline, characters, surface defects, presence, position and some hole positions. 3D can assess height, depth, three-dimensional form, spatial position and three-dimensional surface defects where the scheme supports it. Neither is always better.

From the requirement backwards: product requirement and tolerance first, then the method, optics and algorithm, then verification on real samples. Resolution, lens, field of view, working distance, locating accuracy, distortion, vibration, temperature and speed all affect the final result.

That question cannot be answered without the part and the tolerance. The same system performs differently on different features and different tolerance bands, so no fixed figure is quoted here.

The complete cycle is inspection time plus loading and unloading time plus data processing plus sorting. More inspection items, more cameras and more complex algorithms make it longer. The final cycle follows the project test.

Yes. In-line inspection places the station after the machine and feeds it with the handling unit or a conveyor: real time, less manual carrying, interlocked with the line and suited to continuous production. Where the line cannot be changed yet, an independent off-line station is a reasonable first step.

Yes. The inspection cell and the machine exchange signals through the PLC for picking, presenting, judging and returning. The station normally sits on the unload side so the part does not leave the existing material flow.

Yes. The robot picks, carries, flips, places and sorts; vision inspects, judges and outputs the result. Robot plus vision inspection plus OK / NG sorting is a common combination, but note that the robot does not judge quality: the judgement comes from the optics and the algorithm.

Possible destinations are an NG bin, an NG conveyor, robot sorting, a cylinder pusher, a manual re-check station, a rework station, re-inspection or an isolation area. Which one is used is settled with the customer's production flow at proposal stage — an inspection system cannot output only "NG".

Data interfaces and MES integration can be designed to the project requirement. Recordable fields include part ID, time, equipment, inspection items, result, dimensional data, NG reason and inspection images. Where there is no real MES interface in the project, the page does not describe a completed data loop.

Part name and photos, drawings, size and weight, material, color and surface state; inspection items, dimensions, tolerances and the smallest defect to be found; OK samples, NG samples, inspection position, quantity, hourly output and target cycle; machine count, loading method, locating method, and whether a robot, automatic sorting or MES is required.

Because the physical precondition of inspection is a difference in the image between a good and a bad part. How large that difference is, and whether it can be captured stably, only real samples can confirm. With drawings and no samples the scheme stays an assumption.

Yes. The main issue is reflection: bright metal can produce specular reflection that lowers defect contrast and can create defects in the image that are not there. This belongs to the lighting design and usually needs a dedicated scheme and sample tests.

Yes, with a lighting angle, polarization, diffusion or other scheme chosen for the case, and sometimes a change in the relative position of camera and light. The scheme still follows the sample test; the page does not claim that any reflective part can be done.

Yes. Black surfaces absorb light and lower the image contrast, so a matching light and lighting method is needed to bring the feature out. It is usually harder than a light-colored part and has to be confirmed on samples.

No detection limit is promised in advance. Whether a scratch can be seen depends on its contrast against the background, the resolution of the optics, the lighting and the surface state; only tests on real defects answer how small it can be.

Automatic inspection is consistent, repeatable, recordable and can cover every part; manual inspection handles complex, ambiguous cases that need experience, and changes parts fastest. In practice automatic inspection carries the repetitive, standard judgements while people handle exception review and complex decisions.

Typical cases: the criteria cannot be stated clearly; good and bad parts cannot be distinguished in the image; no samples can be provided for testing; internal structure has to be inspected; a dimension must be touched to be determined; or the batch is very small with extremely frequent changeover. Those cases are better served by people, sensors or dedicated measuring equipment.