Every plant has a story about the operator who assembled the part backwards for a week. The useful question is not who did it, but which device allowed it, and why that device was chosen over a cheaper one that would have made the error impossible.
The hierarchy
Error-proofing devices rank by how hard they are to defeat. At the top, the part physically cannot be fitted the wrong way. At the bottom, a screen asks the operator to confirm something they have already done. Most plants operate at the bottom and wonder why defects persist.
Level 1: geometry
Asymmetric locating pins, a relief cut that only accepts one orientation, a fixture that will not close if a part is missing. Geometry has no power supply, no PLC and no bypass. It is also the cheapest to maintain and the hardest to retrofit, which is why it belongs in the design review rather than the improvement programme.
Level 2: presence sensing
A proximity sensor confirms that the right component is in the fixture before the cycle starts. The weakness is the sensor's field: an inductive sensor detects metal, not identity. Two different steel brackets look identical to it.
For identity, use a mechanical key, a colour sensor, a vision check or a coded carrier. For presence only, a simple sensor is fine and costs 30 to 80 US dollars per point.
Level 3: sequence interlock
The PLC will not release the carrier until every station has signalled completion. This prevents skipped operations but not wrong ones. It also introduces the most common failure mode in error-proofing: the maintenance bypass key that is left in the run position.
Level 4: parameter capture
Torque, press force, dispense weight and test results are recorded against the serial number. This is the only level that produces evidence, which is why automotive customers demand it. It catches wrong values but only after the operation, so it needs a rework path.
Sensor selection by failure mode
| Check | Device | Typical failure |
|---|---|---|
| Part present | Inductive or photoelectric | Detects the wrong metal part |
| Part orientation | Mechanical key, vision | Wear on the key, lighting drift |
| Fastener seated | Torque transducer | Wrong torque window, socket wear |
| Clip engaged | Force-displacement monitor | Sensor drift, no calibration |
| Label correct | Scanner with database check | Duplicate serial numbers |
Design rules that keep it working
- One device per error, not one device per station. If a station has three possible errors, it needs three checks.
- Fail to the safe state. A disconnected sensor should stop the line, not allow it.
- Log every bypass, with a name and a reason. A bypass register that nobody reads is a decoration.
- Test the device with a deliberately wrong part during commissioning and again every quarter.
The economics
A defect that reaches the customer costs 50 to 500 times the cost of catching it at the station, depending on the industry. That ratio is the argument for spending on level 1 and level 2 devices rather than adding another inspection station at the end of the line.
The counter-argument is equally real: over-instrumenting a stable process adds maintenance burden and false stops. Apply error-proofing where the process has actually produced defects, or where the failure mode is catastrophic, and leave the rest alone.
