False Signals and Mistake-Proofing Verification Frequency

Why it matters

Poka-yoke is a mistake-proofing device that helps prevent an error from occurring or from passing unnoticed. Physically restricting the orientation of a part is not the same function as detecting a wrong placement and giving a warning. The distinction between types was built in M9; here we deal with how far the device's decision can be trusted and how often its operation will be verified. A device being in place does not prove that it works correctly under every condition. Part changes, contamination, setting drift or a connection problem can affect detection. Verification is genuinely testing the function with test parts of known conformity or with test conditions safely prepared for the task. The absence of a signal in production does not replace this test.

To evaluate the accuracy of the decision, the true state of the part must be known from an independent basis. If a truly defective part passes, it is a false acceptance; if a truly conforming part is stopped, it is a false stop. The denominators of these two results are different. False acceptance is divided by the truly defective parts, not by all trials; false stop is divided by the truly conforming parts. The total error rate can be calculated separately but does not by itself explain the product's risk of escaping. In a trial containing many conforming parts, a low total error rate can hide that a significant share of the defective parts escaped. For this reason the table must show not only how many signals were given but which true state each signal corresponded to.

The verification interval affects the production quantity that may remain uncertain after the last conforming function check. The quantity found by multiplying production rate by time does not say that all parts in that interval are defective. It is the upper quantity of the product scope that may need to be protected after the last reliable check. The calculation rests on the assumptions of constant rate, complete traceability and no additional production once the problem is noticed. If there is intermediate stock, a different product flow or a delay in stopping, the boundary can widen further. Shortening the interval does not by itself reduce the probability that the function makes a wrong decision; it gives the opportunity to catch the problem earlier and leave less product uncertain. The two effects do not replace each other. This boundary rests on the assumption that the last verification reliably tested the relevant failure mode; an intermittent error, or one not visible in verification, can affect older production as well.

The frequency decision is made together with the false acceptance finding, the effect of the error and the strength of traceability. For a high-impact defect, loosening the detection threshold merely to reduce needless stops is not appropriate. While the cause of stopping conforming parts is investigated, the probability of passing a defective part is also retested. The number of trials, the product variants covered and the conditions must be visible in the report. Seeing no escape in a small sample does not prove there will be none in all future production. The results here are the observed rates of training data; they are not a universal adequacy threshold or a ready verification calendar for all processes. The decision is made keeping local risk and the limits of the evidence explicit.

The identity, defect and usage condition of the reference are preserved. A worn reference is renewed; if the basis measurement changes, the classification is reviewed. The consistency of the instruction between workers is tested; an unclear result does not count as success.

Formula