Assembly fixtures and poka-yoke: error-proofing at the station

Prevention first, detection second

Poka-yoke means mistake proofing. The idea is to design the station so an error cannot be made, or so it is caught immediately at the point where it happens. The distinction matters because prevention and detection cost different amounts and give different guarantees.

Prevention is physical. An asymmetric locating pin means a bracket only fits one way. A nest that is deeper on one side means a part cannot sit upside down. A shutter that only opens when the previous operation is complete means the operator cannot skip a step. These devices do not depend on attention, which is why they survive a busy shift.

Detection is electronic or visual. A proximity sensor confirms that a part is seated. A vision system checks that a label is present and correct. A torque controller confirms that a fastener reached its target. Detection is useful where prevention is impractical, but it can be defeated by a determined operator, a muted alarm or a sensor that has drifted.

The practical rule is to prevent where the geometry allows it and detect where it does not. A station that relies on three alarms and no mechanical guide is fragile. A station with one well-placed mistake-proof nest and one confirmation sensor is usually enough.

Locators decide what the fixture builds

A fixture does not hold a part; it locates it. The difference is that location comes from contact with defined surfaces, and those surfaces should match the datum scheme on the drawing. If the drawing locates on two holes and a face, the fixture should too. If it clamps on a cosmetic surface instead, the assembly may measure well in the fixture and badly in the customer's hand.

Over-constraint is a common fault. A part that is located on four pads and then clamped will rock, and the clamp will pull it to whichever pad is highest. Three pads, or a pad and two pins, define a plane without fighting each other. The same logic applies to a shaft: two centres locate it, and a third support should be adjustable or floating.

Foolproofing belongs in the locator design. Offset the pins, add a key, make one nest deeper, or shape the pocket so only the correct variant enters. If two variants are similar, the fixture should reject the wrong one rather than rely on a colour code that fades under factory light.

Access matters as much as accuracy. The operator has to load, seat and clamp without trapping fingers, and the tool has to reach every fastener without the fixture in the way. A perfect locator that adds four seconds to every cycle will be modified on the floor, and the modification is rarely reviewed.

Sensors, interlocks and the bypass problem

A sensor is only part of a control. The station needs a sequence that will not advance until the sensor is satisfied, and a reason for the operator to keep the sensor working. If the sensor is easy to tape over, it will be taped over on the shift when it misreads.

Choose the sensing method for the failure. A proximity sensor confirms presence but not orientation. A through-beam sensor confirms that a part has passed a point. A vision check confirms a label, a colour or a feature, but it needs lighting and a stable part position to be repeatable. A load cell or a force sensor confirms that a press-fit reached force, which a position sensor cannot.

Interlocks should fail safe. If the sensor cable breaks, the station should stop rather than run. That sounds obvious, but a normally open sensor wired into a PLC input can read as satisfied when it is disconnected. Specify the polarity, the diagnostics and the behaviour on loss of signal.

Fastening and torque control

Screwed joints are a frequent source of escapes because the failure is invisible after assembly. A cross-thread, a missing washer or a screw driven at the wrong torque can all look acceptable on the bench.

A controlled torque tool with a count and a sequence is the baseline. The tool should record the torque and angle for each fastener against the serial number, and the station should not release the part until every fastener in the sequence is complete. A simple click wrench relies on feel and gives no record.

Poka-yoke at the joint can be physical as well. A captive screw cannot be forgotten. A washer that is pre-assembled to the screw cannot be omitted. A boss that only accepts the correct screw length stops a longer screw from bottoming out. A cross-thread is harder to prevent, but a floating driver and a lead-in chamfer reduce the chance.

Torque values should come from the joint design, not from a generic table. The drawing needs the value, the tolerance and the tightening strategy. If the joint uses a thread-forming screw into plastic, the driving torque and the strip torque both matter, and the window narrows as the boss ages.

The errors a station should expect

The table lists the common assembly errors and the device that catches each one. It is a starting point for a station review, not a substitute for watching the line.

Keeping the fixture honest

A fixture wears. Locating pins lose diameter, pads frett, clamps lose force and swarf collects in the places that matter. None of this shows up in a dimensional report until parts start to drift, and by then a batch may be out.

Put the critical locators on a check schedule. Measure pin diameters, check pad heights and confirm clamp force at a stated interval. Clean the nest at every shift change if the process produces chips or dust. Record the checks, because a fixture that is adjusted without a record cannot be traced to the parts it built.

Treat fixture changes as engineering changes. A reamed hole, a shimmed pad or a new clamp can alter the assembly and invalidate the first article. The change should be reviewed, the fixture drawing updated and the affected parts re-qualified. A station that drifts one shim at a time is a slow route to a quality escape.

Train the operator to stop the line when the device is not working. That permission is part of the design. If stopping the line is punished, the operator will find a way to run, and the error-proofing will exist only on the audit checklist.

Validate the station before volume

Run a challenge test. Take a known bad part, or deliberately load a part the wrong way, and confirm that the station rejects it. Do this for every device, at the start of production and at a defined interval after. A device that has never been challenged is an assumption.

Check the fixture against the drawing and the control plan. The locators should match the datums, the clamp positions should not distort the part, and the inspection points should be reachable. Run a small batch and measure the assembly, then compare the results with the tolerance stack-up.