Every plant that supplies an automotive OEM has poka-yoke in its control plan. Far fewer have poka-yoke that would survive the shift when the regular operator is absent, the line is behind schedule and the part looks nearly right. That gap — between poka-yoke as documentation and poka-yoke as physics — is where customer complaints come from.
This guide is about the engineering side: what to error-proof, how to choose the mechanism, how to interlock it so it cannot be skipped, and what it costs to build in.
Start from the defect, not the technology
The starting point is not "where can we add a sensor" but "which defects reach the customer, and how are they made?" Take your warranty returns, customer complaints, internal rejects and the high-RPN lines of the process FMEA, and for each failure mode write down the physical error that produces it. Not "wrong bracket fitted" but "two brackets are the same size and differ only by a hole position 6 mm apart, and they sit in adjacent bins."
Once the error is described physically, the countermeasure usually becomes obvious — and it is frequently not a sensor. Separating the bins, or changing one bracket so the wrong one physically will not seat, costs nothing per cycle and cannot fail.
The four levels, from strongest to weakest
Rank every proposed countermeasure. If you can move it up a level, do that instead of buying instrumentation.
| 1. Elimination | The error becomes impossible. Asymmetric locating features, a fixture that only accepts the part one way round, a connector that will not mate wrongly, a common part replacing two similar ones. Nothing to maintain, nothing to calibrate, nothing to bypass. |
|---|---|
| 2. Prevention | The operation cannot proceed until the condition is right. The fixture will not clamp with the part reversed; the nutrunner will not enable until part presence is confirmed; the next station will not accept an untorqued assembly. |
| 3. Detection at source | The error is detected in the same cycle, at the station that caused it, and the station stops. The defect exists for seconds and never moves downstream. |
| 4. Detection downstream | End-of-line test or inspection catches it. Necessary as a net, but by then you have already built value into a defective unit and possibly a queue of them. |
A useful review question for any existing poka-yoke: if this device silently failed this morning, when would you find out? At level 1 the question is meaningless. At level 4 the honest answer is often "at the customer".
Choosing the detection method
For errors that cannot be designed out, match the mechanism to what you are actually verifying.
| Part presence and position | Inductive, capacitive or photoelectric sensors; mechanical probes. Cheap, robust, fast. Watch out for sensors that see the fixture rather than the part, and for the case where a missing part reads the same as a wrongly seated one. |
|---|---|
| Correct variant | Contact probes on a differentiating feature, a profile sensor, vision, or an RFID tag on the carrier. For high-mix lines, driving the station from the build order via barcode or RFID is usually more reliable than sensing the part. |
| Fastener tightening | Torque-and-angle monitoring with an accept window, plus counting. Torque alone misses cross-threading and stripped threads; the angle trace is what catches them. |
| Press and insertion operations | Force-displacement signature monitoring with an accept envelope. Detects missing components, misaligned parts, wrong depth and wrong component in one measurement. |
| Leak or flow integrity | Pressure decay or mass-flow testing with a calibrated master leak. Needs temperature and dwell control to be repeatable. |
| Surface, print and cosmetic | Vision with controlled lighting. The lighting design, not the camera, decides whether it works. |
| Sequence and completeness | Station logic that refuses to release the part until every required step has returned an OK — usually the highest-value and lowest-cost poka-yoke on a manual line. |
Interlocking: the part that makes it real
A device that lights a lamp when something is wrong is an indicator, not a poka-yoke. What separates them is the interlock — the physical or logical consequence of a NOK result. Effective patterns:
- Clamp interlock. The fixture will not clamp, or will not release, unless the part is correct and complete.
- Tool enable. The nutrunner, dispenser or press is enabled by the controller only when the preconditions are met, so the wrong part cannot even be worked on.
- Release interlock. The conveyor stop or lift-and-locate will not release the carrier until the station reports OK. The defect physically cannot travel.
- Reject-only exit. A NOK part can leave only through a locked reject chute or bin whose open command is logged, so it cannot rejoin the flow by being put back on the line.
- Downstream re-check. The next station confirms the previous operation was completed, so a bypass at one station does not go unnoticed.
Traceability turns error-proofing into evidence
Once a station is measuring, storing the result costs very little and changes what you can do when a customer complains. The practical minimum per unit: serial or batch identifier, station, timestamp, the measured values with the accept window applied, the operator or shift, and the component batch where it matters.
That record lets you bound a containment to the units actually affected instead of the whole month's despatch, answer a complaint with measured data, and see drift in the process before it produces rejects. For safety-critical and regulated parts it is usually a customer requirement rather than an option.
Keep the data model simple. One record per unit per station, written at the moment of the result, in a form your plant can query without a specialist. Sophisticated systems that nobody can query during a containment are worse than a plain database.
Where error-proofing actually earns its money
Build the business case on what the defect costs, not on the device price. Compare, per defect escaping: scrap or rework cost, line stoppage, sorting and containment labour, premium freight, the customer's charge for a complaint, and the commercial consequence of a repeat. Against that, a ₹1.5 lakh force-monitoring station is usually settled by a single avoided containment.
Annual benefit = (defects prevented per year × cost per escaped defect) + (sorting and rework hours avoided × loaded hourly cost). Payback in years = device cost ÷ annual benefit.
Illustrative only — substitute your own figures. If a station prevents 30 escapes a year at an assumed ₹12,000 each, the benefit is about ₹3.6 lakh a year, and a ₹1.5 lakh device pays back in roughly five months. The figures are assumptions; the method is the point.
Indicative cost bands
| Mechanical elimination | Fixture or part design change. Often the cheapest and always the strongest — frequently under ₹50,000 and sometimes free at the design stage. |
|---|---|
| Sensor-based presence and position check | Roughly ₹25,000–1.5 lakh per station depending on the number of points and the controller. |
| Torque-angle monitored fastening | Roughly ₹1.5–6 lakh per spindle including controller and data output. |
| Force-displacement press monitoring | Roughly ₹2–8 lakh per station including load cell, encoder and evaluation. |
| Vision inspection | Roughly ₹2–12 lakh per station; the lighting and fixturing decide reliability more than camera resolution. |
| Traceability layer | Roughly ₹2–10 lakh per line for marking or reading, data collection and reporting. |
Common failures to design against
- Detection without consequence. A lamp or buzzer with no interlock. Under production pressure it will be ignored.
- Sensing the fixture, not the part. The classic false-pass: the sensor confirms the clamp closed, not that the component is there.
- NOK parts that rejoin the line. No locked reject route, so the defect goes back into the flow while the operator investigates.
- Never-challenged devices. No master-part check, so a failed-on sensor passes everything until a complaint arrives.
- Bypass switches with no logging. If a bypass is genuinely needed for maintenance, make it key-controlled, time-limited and recorded.
- Poka-yoke added at end of line only. You detect the defect after building the full value into it, and you cannot tell which station caused it.
Retrofitting to an existing line
Do it in defect-cost order rather than all at once. Rank failure modes by escaped cost, take the top three, and at each one work down the four levels — can it be eliminated, prevented, detected at source, and only then detected downstream? Fit the countermeasure, run known-bad master parts through it, confirm it rejects them, and add the daily challenge check to the shift start-up sheet. Then take the next three. A line error-proofed this way is measurably better within weeks; a line-wide project usually is not.
We design and build error-proofed stations, fixtures, monitored fastening and press stations, test benches and the traceability layer that ties them together — as part of a new line or retrofitted to what you already run. Send us your top failure modes and we will tell you which ones can be designed out before anyone buys a sensor.