When I review an action plan and the fix starts with "train the operator to," I already know the ending. The defect drops for two weeks and comes back. Nobody did anything wrong. The error was still possible.
Poka yoke means mistake proofing. The idea, from Shigeo Shingo inside the Toyota Production System, is simple: human error is inevitable, and defects do not have to be. Below are examples you can adapt this week, ranked from least to most reliable. At the end is the test that decides whether any of them actually works.
Level one: the ones that warn
These are the easiest to install and the weakest, because they depend on somebody reacting.
- A light or alarm when a component is missing from the station kit.
- Color coding by model on bins, tooling and the work instruction, so a mix up jumps out.
- A scale that flags a box that weighs less than it should when the pack is closed.
They are a fine first step. The trouble is that by hour seven of a shift, one more alarm is just noise. Which is why there is a second level.
Level two: the ones that stop the line
Here the process stops itself when it detects the wrong condition. It no longer depends on anybody reacting.
- A presence sensor that will not let the press cycle unless the part is seated.
- A counting driver that will not release the part until every screw is in.
- A scanner that locks the station if the label does not match the order.
They are better, and they still have a weakness almost nobody plans for: the sensor fails too. Hold on to that thought, because the final test lives there.
Level three: the ones that make the error impossible
This is the level to reach when the defect is critical. Here the error is not detected: it does not fit.
- An asymmetric locating pin on the fixture, so the part only goes in one way.
- A hard stop that keeps the look alike part from the other model out of that tooling.
- A kit tray with one cavity per component: if something is left over, you see it before you close.
Notice that almost all of them are mechanical. A pin, a guide, a stop or a geometry change in the tooling costs less and fails less than a sensor. The best ones I have seen cost less than an hour of downtime was worth.
Before the test, one more place people forget because it does not look like a plant.
The office counts too
Poka yoke applies anywhere there is a repetitive process. A required field on a form, a locked cell in the costing spreadsheet, or a dropdown that only accepts valid part numbers are poka yoke under another name. The data entry error that reaches the floor was usually born at a desk.
The test that decides whether it works
Here is what almost nobody does. You validate a poka yoke by trying to push the defect through on purpose. Take the bad part, the wrong orientation or the other model’s label, and prove it cannot pass. If you did not try, you did not validate. You have a device you believe works.
Then you decide what happens when the device itself fails. A disconnected sensor nobody notices is worse than none, because it creates false confidence. So every poka yoke goes into the start of shift check, with its known bad master part stored right next to it.
Start here
Take the defect that repeats most on your line. Find the station where it is born, which is almost never where it is caught. Ask what makes the error possible, not who made it. And look for the physical fix first.
My problem solving workshop ends with proposed devices for real cases from your operation, each with its level and its validation test. If you already have the defect in mind, tell me about it.
What are the types of poka yoke?
Three levels: warning (signals and depends on someone reacting), control (stops the process) and prevention (makes the error physically impossible). The more critical the defect, the higher the level you need.
How do you validate a poka yoke?
By trying to push the defect through on purpose with a known bad master part and proving it cannot pass. Repeat it at the start of every shift to confirm the device still works.
Does a poka yoke have to be electronic?
No. The most reliable ones are usually mechanical: a locating pin, a hard stop or a kit tray with one cavity per component. They cost less and fail less than a sensor.