Mistake-proofing: designing the error out instead of asking for more care

Prevention, detection and attention are not equivalent options

There is an order of preference worth stating aloud in every review. Best is a design or fixture that makes the error physically impossible. Next is a device that detects the error and holds the process before the part moves on. After that comes detection at a later station, which contains the escape but has already paid for the defect. Last, and barely a countermeasure at all, is a line in the work instruction asking for care. Teams drift down this list because the lower options are quicker to write and consume nobody's engineering time, and the drift is visible in any corrective action file.

A device that slows the job will be defeated

Operators bypass mistake-proofing for rational reasons: the sensor false-triggers, the fixture is awkward to load, the check adds time to a cycle they are held to, or the device blocks a legitimate variant nobody told engineering about. Tape over a sensor is a symptom of a badly specified device rather than a badly behaved operator, and treating it as misconduct removes the information while leaving the device defeated in a subtler way. Design reviews should include whoever will run the station, and should ask directly what they would do on a bad day when the schedule is behind.

Checking that the checker still works

Detection devices fail silently. A proximity switch drifts, a light curtain gets knocked out of alignment, a weight check has its tolerance widened during a difficult run and never restored. Unless somebody verifies the device by deliberately presenting a bad condition and confirming rejection, the plant is relying on a control whose status is unknown, and the discovery normally arrives with a customer complaint attached. Build verification into the start-of-shift routine, record who performed it, and treat the deliberately faulty sample as a controlled item, because the commonest reason verification lapses is that the master defect went missing.

Errors are made by processes; defects are what escape

Separating the two changes the conversation. An error is a slip anyone could make under the conditions provided; a defect is an error that reached the next customer because nothing caught it. Blame collapses the distinction and buries the error data, since nobody volunteers a mistake that ends in a disciplinary meeting. Plants where reporting is safe collect a far richer list of near misses, and near misses are exactly the raw material this work needs, because they identify the confusable parts, the ambiguous orientations and the steps that are easy to omit under pressure.

The countermeasure costs least when it is drawn rather than retrofitted

Retrofitting is normal and necessary, but the same protection designed into the part or the tool is usually simpler, more reliable and invisible to the operator: an asymmetric locating feature, a connector that cannot mate the wrong way, a moulded stop that prevents reverse insertion. That requires defect history to reach the people doing design work, which in many companies it never does. A standing review where recurring assembly errors are fed back into drawings and tooling specifications is unglamorous and achieves more than any increase in inspection.

Frequently asked questions

Is retraining a valid response to a recurring assembly error?

Only alongside something physical. Training addresses knowledge, and most recurring assembly errors are slips made by people who already know the correct method and were interrupted, rushed or presented with two parts that look alike. Where the corrective action file for a station shows repeated retraining entries against the same failure, that is evidence the countermeasure is inadequate rather than evidence the operators need another session with the trainer.

How do we stop operators disabling a detection device?

Find out why they did it, in a conversation carrying no penalty. Almost always the device misfires, adds time, or blocks a variant it was never told about. Fix the underlying nuisance, and where the device must stay strict, make defeating it obvious rather than trying to make it impossible: a tamper-evident cover, a status shown on the area display, a start-of-shift verification that would reveal the bypass. Devices survive because they are trusted, not because they are locked.

Do we need sensors, or can mistake-proofing be purely mechanical?

Mechanical is generally preferable where it is achievable. A locating pin that makes the wrong orientation physically impossible needs no power, no calibration and no verification routine, and it cannot be bypassed with tape. Sensing is for conditions that cannot be made physically impossible, such as a missing fastener or an out-of-range tightening torque. The presence of electronics is not a measure of quality here; the hierarchy is about whether the error can occur at all.