Detecting the Wrong Part or Wrong Orientation

Not necessarily. If fixture positioning is used, the product attitude is fixed and the inspection areas can be hard-coded; if the incoming material attitude fluctuates, the datum features must be located first before the inspection areas are defined. The former is simpler to implement, the latter places lower demands on tooling, and the trade-off depends on the existing production line conditions.

It depends on whether the difference can be imaged stably and measurably. The size of the difference itself is not the only criterion; imaging resolution, illumination method and lens choice all affect what can actually be resolved. The only reliable way to judge is to image both types of part under the same optical conditions and compare them.

Set up one inspection recipe (inspection area + acceptance criteria) for each model, and call the corresponding recipe when switching. Recipe switching can be triggered by a PLC signal, triggered by a code reading result, or selected manually by the operator, depending on the information flow on site.

False calls come mainly from two sources: unstable difference features and improperly set judgement thresholds. The former is solved by improving imaging and feature selection, the latter by adjusting the threshold in line with site tolerance. We recommend a small-batch trial run first and setting the threshold after collecting misjudged samples.

For wrong-part poka-yoke, we recommend coordinating with the upstream production line: if the risk of mixed material comes from the loading step, the best solution is still poka-yoke at loading; visual inspection then serves as the final safeguard, at lower cost and with more reliable release. In mixed-model production, the model signal can be provided by the PLC or read by the equipment itself from the workpiece (such as code reading, OCR). The former is simple to implement; the latter is more independent but relies more on the consistency of incoming material identification.

The technical core of wrong-part inspection is choosing the right distinguishing feature. For the same part, if two models differ only by 0.2 mm in outer diameter, distinguishing them by external dimensions is very difficult; but if the two models differ in color, or are the same part assembled in a different orientation, color or orientation features are far more stable. The engineering sequence is: first list all the models/states that need to be distinguished, then find the most obvious visual difference between them, and finally confirm whether that difference can be imaged consistently under the lens. If all three conditions are met, the inspection is feasible; if the distinguishing feature itself is too weak, the tooling or material must be changed first, rather than forcing an algorithm to do it.

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

Please send several pieces of each model of part that must be distinguished, and state which combinations must "never be confused"; the engineer will assess whether stable distinguishing features exist.