Missing Part Inspection as Poka-Yoke for Assembly

Yes. A missing part is essentially "a part that should be there is not in place", which is the same type of judgement as missing-part inspection: define the position where the part should be on the image and judge whether an object matching the features exists in that area. Distinguishing "missing part" from "wrong part" needs finer acceptance criteria — a missing part is judged by presence/absence, while a wrong part also requires comparing model, orientation, or color.

First, determine this product How many parts there should be in total and where each part is located, then verify each item in the image one by one. In engineering practice, a "part list + position map" is usually provided first, the list is converted into an inspection area list, and finally a presence criterion is set for each area. The clearer the list, the more stable the inspection.

Yes, but reflection is a problem that must be solved. The usual measures are changing the lighting method (low angle, dark field, polarization), adjusting the mounting angle, or switching to an algorithm that is insensitive to reflection (for example, doing regional statistics first and then judging, rather than relying on edges). Which one to use must be tested on the actual workpiece.

It depends on the task type. For binary tasks such as presence/absence judgement, a few dozen OK samples are usually enough to start; tasks such as appearance anomaly detection that must learn the full picture of the normal state have noticeably higher sample requirements. The key with samples is not piling up quantity, but Covers variations in real incoming material (batch, surface condition, lighting fluctuation).

The judgement threshold needs to be set in combination with the on-site tolerance: the cost of an escape (judging NG as OK) and of a false call (judging OK as NG) is usually asymmetric, so the threshold should be biased toward whichever is less acceptable, rather than simply taking the middle value. The output method is selected according to the production line's existing control system: use I/O when only a release / rejection signal is needed; use TCP / Modbus / S7 / Profinet when inspection data must be sent back to a host system.

A common approach is zone-by-zone verification: according to the product's assembly structure, divide the position of every required part in the image into independent inspection zones, and judge "present / absent" independently in each zone. In this way, even if the illumination or angle in one zone is not ideal, it will not affect the judgement of the other zones. For products whose position is fixed, use fixture positioning or template matching to align the coordinate system first, then judge by fixed zones; for products whose incoming posture fluctuates, first use a positioning algorithm to find the datum features, then unfold the inspection zones according to relative positions. The difference between the two approaches is whether every piece needs to be repositioned, not the level of algorithm complexity.

View the more specific object characteristics, acceptance criteria and optical considerations by object; covers 8 common object types including screw presence/absence inspection, nut presence/absence inspection, gasket presence/absence inspection and O-ring presence/absence inspection.

What visual inspection replaces is repetitive visual judgement, not manual labor for everything. The typical division of labor is: vision performs part-by-part full inspection and judgement, while people handle re-judgement of borderline samples, changeovers and exceptions, and maintenance of the optics and tooling. How many people are actually involved depends on the degree of automation and the re-judgement strategy.

Send us photos of the workpiece, the list of parts that should be fitted and a description of their positions, and our engineers will determine which items are suited to visual verification and which require tooling changes or additional illumination.