check
check
check
check
check
check
Inspecting presence, shape, height and assembly completeness is a specification task before it is a technology choice. Define the inspection features and their tolerances, fix the part presentation and cycle time, then choose the simplest sensing that answers the defined question: a photoelectric sensor for pure presence, a 2D vision system for features visible in an image, and 3D sensing — ToF or structured light — when height, depth or completeness depends on geometry. Then validate with real good and bad samples before production reliance: KJT Sensors' vision guidance asks for samples, defect photographs and inspection criteria, and false-reject behavior is tuned against known parts, not assumed. Measuring ranges, accuracy and interfaces are confirmed for the specific model and project.
Feature → tolerance → technology, in that order. "Is it there" (presence), "is it the right shape" (2D feature), "is it the right height" (depth), "is it complete" (presence + geometry together) each demand different sensing.
A simple presence question deserves a simple sensor: a photoelectric switch answers "present or not" at a fraction of a vision system's cost and complexity.
Completeness is usually a geometry question: missing components change height or profile before they change appearance — 3D sensing answers it directly.
Tolerances pick the precision class: the tolerance you must hold determines whether a switching sensor, 2D vision, or micrometer-class structured light is required.
Validation uses real samples: good and bad parts, defect photographs, defined acceptance criteria — the manufacturer's own project intake. False rejects on running lines have their own diagnosis ({{URL_T10}}).
Presentation and cycle time are constraints, not afterthoughts: part position variation and takt time bound what any technology can do.
Write the inspection as features with tolerances — the specification every later decision hangs on:
| Feature question | What it actually asks | Tolerance to define |
|---|---|---|
| Presence | Is a part / component there at all? | Position window; minimum target size |
| Shape | Is the visible form correct (contour, hole, print)? | Dimensional tolerance in the image plane; acceptable cosmetic variation |
| Height | Is a surface / component at the right level? | Height tolerance; reference plane |
| Assembly completeness | Is everything present and seated — nothing missing that matters? | Which components are inspection-relevant; missing-part signature; seated-position tolerance |
KJT Sensors' vision selection guidance frames the same discipline: first confirm what is being tested — appearance defects, dimension, flatness, height, installation state or classification — then the material, colour, reflectivity, surface curvature, defect size and line speed (manufacturer-stated).
A presence question — "did the part arrive, is the component installed" — is answered by a photoelectric sensor at the inspection point, in through-beam, retro-reflective or diffuse arrangement. It is fast, cheap and sufficient — when presence is genuinely the whole question. The sensing-mode trade-offs (beam arrangement, background, contrast) are covered in the photoelectric mode comparison ({{URL_F03}}).
When the question is about visible form — contour, hole position, print, surface marks — a 2D vision system inspects features in the image. Its dependency is engineered light and repeatable presentation. KJT Sensors' vision defect-detection systems cover appearance flaws, in-plane dimensions and classification across documented applications (steel wire, ceramics, welding, lithium-battery shells, mirrors, bottles, metal, wood — manufacturer-stated).
When the question involves depth — a surface at the wrong level, a component not seated, a volume short — the answer is 3D: ToF cameras for spatial tasks and structured-light cameras for close-range precision. KJT Sensors' anchors (manufacturer-stated): the ToF camera (iToF, RK3588, 6 TOPS, 56 fps, 100 klux, 103° × 81° FOV, real-time point clouds) and the structured-light area-scan 3D camera with micrometer-level inspection over a 60 × 50 mm field of view. Completeness inspections frequently land here: a missing clip or unseated gasket changes geometry while looking nearly identical in a top-down image.
The architecture comparison between 2D and 3D — how the technologies differ in lighting dependence, presentation tolerance and output — is its own guide ({{URL_C08}}); this article fixes what to inspect first.
Three constraints bound every choice:
Presentation: how repeatably does the part arrive — fixtured, on a conveyor with known wander, random orientation? Presentation variation consumes tolerance and pushes toward 3D (geometry is presentation-tolerant) or better fixturing.
Cycle time: parts per minute and where the inspection fits in the takt. Vision processing and 3D acquisition take time; the feasibility check is per application (56 fps is a camera frame rate, not a per-part inspection guarantee).
Environment: lighting (critical for 2D), ambient light, cleanliness of the optical path, vibration, and the space available for sensors and lighting.
KJT Sensors' vision project intake is explicit: provide samples, defect photographs and inspection criteria (manufacturer-stated). The validation logic:
Known-good parts establish the baseline image/geometry and its natural variation (part-to-part, colour, finish).
Known-bad parts — each defect type you must catch — confirm each failure mode is actually detectable by the chosen sensing at the defined threshold.
Acceptance criteria written as measurable rules, not adjectives: "hole present within ±0.2 mm of nominal position," not "looks right."
Threshold tuning against the sample set, including the worst-case acceptable part and the smallest rejectable defect.
An inspection validated this way earns its reliability; one commissioned without samples borrows it. When a validated line still rejects good parts in service, the false-reject diagnosis applies ({{URL_T10}}).
