Poka-Yoke

Thinking Tools6.1 Designing and Testing the SolutionAlso known as Mistake-Proofing

Developed by Shigeo Shingo (Zero Quality Control: Source Inspection and the Poka-yoke System, English edition, 1986).

Designs physical or procedural constraints so an error cannot be made, or is caught the moment it happens, before it becomes a defect.

Three devices against the wrong seat bracket: fit any of them and see what still gets through

Two brackets that differ by 6 mm sit side by side at a seating station, and fitters pick the wrong one about 80 times in 10,000 chairs. The plant's first fix, a warning label, is on. Switch each device on or off: the line says how many wrong brackets still reach customers, how often the line stops, and what the device relies on. Only one device removes the error itself. Illustrative.

Warning label on the bins

Line-stop sensor

Locating-pin fixture

  • Switch the devices on or off.No device: all 80 wrong brackets per 10,000 chairs are fitted, and all 80 reach customers, who find out when the seat tilts weeks later.A warning: escapes fall from 80 to 24 per 10,000 and stay there, because on a busy shift about three fitters in ten stop reading the label. It still relies on someone noticing.A detector: 4 per 10,000 reach customers, but the line stops 76 times, four minutes each, about five hours per 10,000 chairs. The error is still made every time.Warning and detector: about 1 per 10,000 reaches customers, and the line stops 23 times (92 minutes). Better, and the wrong bracket can still be picked and fitted.Prevention: the short bracket cannot sit in the seat plate, so it cannot be fitted. No errors, no escapes, no stops, and nothing for anyone to notice, for a $1,200 fixture.Zero escapes, all from the fixture: no wrong bracket can be fitted, so the label no longer prevents anything. Keep it only as a picking aid.Zero escapes, and the sensor never fires, since the wrong bracket cannot be fitted. Move the sensor to a station that has no fixture, as the plant did.Zero escapes, but only the fixture is doing the work. The label and the sensor are cost with nothing left to catch.
  • No device: all 80 wrong brackets per 10,000 chairs are fitted, and all 80 reach customers, who find out when the seat tilts weeks later. (label: Off, sensor: Off, fixture: Off)
  • A warning: escapes fall from 80 to 24 per 10,000 and stay there, because on a busy shift about three fitters in ten stop reading the label. It still relies on someone noticing. (label: On, sensor: Off, fixture: Off)
  • A detector: 4 per 10,000 reach customers, but the line stops 76 times, four minutes each, about five hours per 10,000 chairs. The error is still made every time. (label: Off, sensor: On, fixture: Off)
  • Warning and detector: about 1 per 10,000 reaches customers, and the line stops 23 times (92 minutes). Better, and the wrong bracket can still be picked and fitted. (label: On, sensor: On, fixture: Off)
  • Prevention: the short bracket cannot sit in the seat plate, so it cannot be fitted. No errors, no escapes, no stops, and nothing for anyone to notice, for a $1,200 fixture. (label: Off, sensor: Off, fixture: On)
  • Zero escapes, all from the fixture: no wrong bracket can be fitted, so the label no longer prevents anything. Keep it only as a picking aid. (label: On, sensor: Off, fixture: On)
  • Zero escapes, and the sensor never fires, since the wrong bracket cannot be fitted. Move the sensor to a station that has no fixture, as the plant did. (label: Off, sensor: On, fixture: On)
  • Zero escapes, but only the fixture is doing the work. The label and the sensor are cost with nothing left to catch. (label: On, sensor: On, fixture: On)

An example

An office chair plant kept shipping chairs with the wrong seat bracket. Two brackets, 6 millimeters apart in length, sat in neighboring bins at the seating station, and about 80 chairs in 10,000 got the wrong one. The wrong bracket held for a few weeks and then let the seat tilt, so the plant learned about each one from a customer.

The first fix was a bright label on each bin. Escapes fell by about two-thirds and stayed there; on a busy shift, people stopped reading labels. The second was a sensor that measured the bracket and stopped the line when it was wrong. Escapes fell to a handful, but the line now stopped about 75 times in 10,000 chairs, and each stop cost four minutes. The third was a steel fixture for the seat plate with a locating pin placed so the short bracket could not sit in it. The wrong bracket could no longer be fitted, so there was nothing to catch. The fixture cost $1,200, and the plant moved the sensor to another station.

Why it works

Most quality systems find defects after they are made: inspect the product, catch the bad ones, rework or scrap them. That works, at a cost, and it lets some defects through, because inspection is done by people and machines that miss things. Mistake-proofing moves the effort to the moment the error would be made, and changes the work so the error cannot happen or announces itself at once.

The difference between the three levels is where they act. A warning still depends on someone noticing. A detector lets the error happen and stops the process before it becomes a defect. A preventing device, such as a fixture or a guide pin, makes the wrong action physically impossible, so no attention is required and nothing has to be caught. The best devices are cheap and simple, because they are built into the work itself.

The thinking behind it

Shigeo Shingo Shigeo Shingo, a Japanese industrial engineer who worked with Toyota, is known for single-minute exchange of dies and for poka-yoke: simple devices that make a mistake impossible or detect it at once. In Zero Quality Control (English edition, 1986) he argued for inspecting at the source, catching errors before they become defects. The distinction underneath is between an error and a defect: an error caught where it happens never becomes a defect, and so never needs finding later.

How to run it

  1. Pick one recurring error. Choose a mistake with a real cost and a known place in the process, such as the wrong bracket at one station.
  2. Go and watch it happen. Stand at the station and see how the wrong action becomes possible: similar parts, poor light, a rush.
  3. Find the moment of error. Name the exact step where the mistake is made, before anything downstream sees it.
  4. Try to make it impossible. Look first for a fixture, shape, pin, or sequence that physically blocks the wrong action.
  5. Otherwise detect it on the spot. If prevention is not possible, add a device that stops the process the moment the error occurs.
  6. Keep warnings as a last resort. Use labels, lights, or alarms only where nothing can block or stop the error.
  7. Measure escapes before and after. Count defects reaching the customer and stops at the station for a month on each side of the change.

Where it goes wrong

The first failure is stopping at the warning. A label or alarm feels like action and costs nothing, and it reduces errors without ending them. If the fix depends on someone reading something, keep looking.

The second is a detector that becomes the process. A sensor that stops the line seventy times a shift catches the error but leaves its cause in place, and the line pays for every stop. Track stops as well as escapes, and treat a busy detector as a sign that prevention is still needed.

The third is a device people defeat. A fixture that slows the job gets bypassed with a shim. Design with the operators, and check after a month that it is still in use.

When to use something else

  • Forcing Function. When the error is in an office process or software, and the fix is making a later step wait on an earlier one.
  • Jidoka. When the aim is a line that stops itself on any abnormality, so people can fix causes as they appear.
  • Failure Mode and Effects Analysis. When the job is ranking many possible failures by severity, likelihood, and detection before choosing where to act.

Where it sits in SOLVE eX

Go to the source

  • Zero Quality Control: Source Inspection and the Poka-yoke System, Productivity Press, English edition, 1986

Back to the catalog