Error proofing, or poka-yoke, is designing a process so a mistake either cannot happen or is caught the moment it does, before the part moves on. Automated error proofing uses sensors and machine vision to check every part at the station where the error could occur, and to stop, reject or lock out the process when something is wrong.
This article is part of our machine vision guide. It is written for continuous improvement, quality and manufacturing engineers deciding when a fixture or checklist is enough and when an automated check pays off.
In short
- Prevention beats detection: a fixture that makes the error impossible is the strongest fix.
- Use automated checks where prevention isn’t practical: many variants, subtle defects, or high volume.
- Check at the station where the error happens, not at the end of the line.
- A check that operators can bypass, or that rejects good parts, will be bypassed.
The levels of error proofing
Error-proofing methods are usually ranked from strongest to weakest. Work down the list only when the stronger options don’t fit.
| Level | What it does | Example |
|---|---|---|
| 1. Prevention by design | The error is physically impossible | An asymmetric locating pin so a part only fits one way |
| 2. Prevention by control | The process won’t start or continue if conditions are wrong | A press that won’t cycle until a sensor confirms the part is seated |
| 3. Immediate detection | The error is caught at the station and the part is stopped | A camera confirms all clips are present before the part leaves the cell |
| 4. Downstream detection | The error is caught later, before shipping | End-of-line inspection or test |
| 5. Human checks | People look for the error | Visual inspection, checklists, sign-offs |
Levels 1 and 2 are mostly mechanical and controls design. Machine vision and sensors live at levels 2 and 3, and they replace level 5 when human checks keep missing errors.
When an automated check beats a fixture or a checklist
- Many variants on one line. A fixture can’t tell a 12 mm screw from a 14 mm screw, or the left-hand bracket from the right-hand one. A camera can check the right part for the current recipe.
- Subtle or cosmetic errors. Missing O-rings, reversed labels, short shots, wrong color. These are hard to prevent mechanically and easy for people to miss after a few hours.
- High volume or fast cycle. At hundreds of parts an hour, even a 99% attentive inspector lets defects through.
- You need a record. An automated check can store an image and a result for every part, which a checklist can’t.
- The error is upstream of a costly step. Catching a missing component before welding, potting or final assembly saves the value added afterward.
Common automated error-proofing checks
| Error type | Typical check | Technology |
|---|---|---|
| Missing component (screw, clip, seal, label) | Presence and count | 2D vision, presence sensors |
| Wrong part or variant | Part identification against the recipe | 2D vision, barcode or data matrix read |
| Wrong orientation | Feature position or angle | 2D vision |
| Incomplete seating or height | Height or step measurement | Laser profiling, 3D vision, displacement sensors |
| Wrong count in a kit or package | Count against target | Vision counting |
| Torque or press-fit out of range | Force or torque signature | Process monitoring with DAQ |
Most of these are simple vision tasks. The engineering effort goes into lighting, part presentation and handling the reject, not into the camera itself. See how vision inspection systems work for the components.
Designing an error-proofing check that people keep using
- Start from the defect data. Pick the error that escapes most often or costs most when it does, not the easiest to automate.
- Put the check where the error happens. Every station between the error and the check adds cost to the part you reject.
- Define what happens on a fail. Stop the cycle, divert the part, or lock the station until a lead resets it. A light that flashes and is ignored is not error proofing.
- Test it with real good and bad parts. Include borderline parts. A check that rejects good parts gets turned off.
- Verify it every shift. Run a known-bad “challenge part” at start-up to prove the check still works.
- Add it to the control plan. Record the check, its reaction plan and its verification frequency.
Measuring the result
Track the defect the check targets at the next downstream point and at the customer. A working error-proofing check should drive that defect close to zero, while false rejects stay low enough that nobody is tempted to bypass it. If you need to justify the investment, the cost of poor quality for that defect is usually the clearest number.
Have an error that keeps escaping?
Sciotex designs automated error-proofing and inspection checks for assembly and production lines, and we test your sample parts free of charge to show what a check can catch. Designed, programmed, built and supported in the US since 1997.
Frequently asked questions
Error proofing, or poka-yoke, is designing a process so a mistake cannot happen or is detected immediately at the station where it occurs, before the part moves on.
Poka-yoke prevents or immediately catches an error where it happens. Inspection usually finds defects later. Automated checks at the station are a form of poka-yoke; end-of-line inspection is detection.
When prevention by fixture isn’t practical: many part variants, subtle or cosmetic errors, high volume, or when you need an image and result recorded for every part.
Run a known-bad challenge part at start-up or each shift, confirm the device detects it and triggers the reaction plan, and record the result.
False rejects, slow resets and unclear fail messages. A check must be reliable on good parts and easy to recover from, or it will be turned off.