13.5 Preventive Action, Mistake-Proofing, and Read-Across
Key Takeaways
- Preventive action eliminates the cause of a potential nonconformity, so it is triggered by data analysis -- trends, near-misses, high-priority failure modes, and lessons from elsewhere -- rather than by an event that already occurred.
- Near-misses, meaning defects caught by something other than the control that was supposed to catch them, are the highest-value preventive input because they identify a failing control before any escape occurs.
- Poka-yoke devices are classified by mechanism as contact (shape or physical property), fixed-value (a required count of actions or parts), and motion-step (required sequence), and by response as control devices that stop the process or warning devices that rely on a person.
- The preventive control hierarchy runs from eliminate, to prevent with a control-class poka-yoke, to detect at the source, to detect downstream, to warn, and finally to train and remind -- so retraining as the sole action is the weakest available response.
- Read-across (horizontal deployment) applies the action to every other part, process, shift, line, site, and supplier where the same weakness could exist, and the action record must document the areas evaluated, including those found not applicable and why.
13.5 Preventive Action, Mistake-Proofing, and Read-Across
Preventive Action Acts on Something That Has Not Happened Yet
Section 13.4 established the distinction between a correction (fix this part), a corrective action (eliminate the cause of a nonconformity that occurred), and a preventive action (eliminate the cause of a potential nonconformity so it never occurs). This section covers the preventive branch in operational detail, because it is the branch technicians find hardest: there is no defective part to point at, no customer complaint to investigate, and no obvious trigger.
The practical consequence is that preventive action must be triggered by data analysis rather than by an event. Somebody has to go looking.
| Corrective action | Preventive action | |
|---|---|---|
| Trigger | A nonconformity that occurred | Analysis identifying a potential nonconformity |
| Evidence at the start | A defective part, a complaint, an audit finding | A trend, a near-miss, a high-risk failure mode, a similar failure elsewhere |
| Question asked | "Why did this happen, and how do we stop it recurring?" | "What is likely to happen, and how do we stop it occurring at all?" |
| Effectiveness measure | Recurrence rate of a known defect | Absence of a defect that never appeared, demonstrated through leading indicators |
[!IMPORTANT] Effectiveness of preventive action is measured on leading indicators. You cannot prove a defect did not happen because of your action. What you can demonstrate is that the precursor went away: the capability trend flattened, the near-miss rate fell to zero, the tool-wear curve stayed inside its band, the PFMEA occurrence rating dropped with data to support it.
Selecting Data Analysis Techniques to Identify Potential Failures
The Body of Knowledge expects a technician to select analysis techniques, not merely to name them. Match the technique to the kind of signal you are hunting.
| Signal you are hunting | Technique | What a hit looks like |
|---|---|---|
| Slow deterioration before a specification violation | Control chart run rules and Cpk trending over months | Six consecutive declining Cpk points; a trend of seven increasing subgroup averages |
| Growing variability with a stable mean | Range or standard-deviation chart trending | Widening R chart while X-bar holds; capability erodes from spread |
| Concentration of small problems | Pareto of near-misses, of rework reasons, of downtime causes | One category carrying 40 percent of near-misses |
| A relationship between a process input and an output | Scatter diagram and correlation | Burr height rising with tool cycles; leak rate rising with ambient humidity |
| Wear-out and life limits | Tool-life and maintenance-interval analysis; reliability trending | Failures clustering at a repeatable cycle count |
| Untested failure modes in the design or process | FMEA review, prioritized by Action Priority (Section 13.2) | High-severity modes whose only control is operator vigilance |
| Risks proven at another plant, line, or product | Read-across review (below) | A corrective action at site B that applies to an identical process at site A |
| Systemic weaknesses not yet in product data | Internal audit observations and OFIs | Repeat observations against the same requirement in different areas |
Near-Misses Are the Highest-Value Preventive Input
A near-miss is a nonconformity that was caught by something other than the control that was supposed to catch it — a defect found at final inspection that the in-process check should have caught, a wrong component spotted by an alert operator rather than by the fixture, a mislabeled lot noticed during a routine count. Near-misses are free information about a control that is not working, and they arrive before the escape. Organizations that only log confirmed defects discard the most predictive data they have.
Assigning Responsibility
A preventive action with no named owner is a suggestion. The action record must name the person responsible for improving the process, the specific action, the due date, and the method and date of effectiveness verification. Responsibility normally sits with the process owner — manufacturing or process engineering — with quality supplying the analysis and verifying the outcome, because a preventive action almost always changes the process rather than the inspection.
Error-Proofing and Mistake-Proofing (Poka-Yoke)
The strongest preventive actions do not depend on anyone remembering anything. Poka-yoke, developed by Shigeo Shingo, designs the process so the error either cannot be made or cannot pass undetected.
Prevention Versus Detection
| Class | What it does | Example |
|---|---|---|
| Prevention poka-yoke | Makes the error physically impossible | An asymmetric locating pin so the part cannot be loaded backwards; a connector that only mates one way |
| Detection poka-yoke | Allows the error but catches it immediately, at the source | A light curtain that stops the cycle if a hand enters; a torque monitor that flags a fastener below setpoint |
Prevention is always preferred, because detection still depends on the detector working.
The Three Poka-Yoke Mechanisms
| Mechanism | Principle | Shop-floor example |
|---|---|---|
| Contact method | Uses shape, dimension, or physical property to detect a deviation | A fixture pin that will not accept a part machined with the hole in the wrong location; a limit switch that confirms the part is seated |
| Fixed-value (constant number) method | Checks that a fixed number of actions or components occurred | A kit tray with exactly eight fastener wells; a nutrunner that will not release until eight torque events are counted |
| Motion-step (sequence) method | Checks that steps were performed in the required sequence | A machine that will not cycle until the first-piece buy-off is scanned; a light-directed pick sequence that will not advance out of order |
Control Versus Warning Devices
- A control device stops the process or prevents the next step. It does not rely on human response.
- A warning device signals with a light, a buzzer, or an alarm and relies on a person to react.
Control devices are stronger. A warning device on a high-speed line simply produces a bin of defective parts accompanied by a noise.
Where Error-Proofing Fits in the Hierarchy
The preventive design hierarchy, strongest first:
- Eliminate the possibility in the product or process design (no fastener, no error).
- Prevent the error with a control-class poka-yoke (it cannot be done wrong).
- Detect the error at the source, immediately, with automatic stop.
- Detect it downstream by inspection.
- Warn and rely on a person.
- Train and remind — the weakest control, and the one most often proposed.
"Retrain the operator" is level six. When a corrective or preventive action plan offers retraining as its only action, it has proposed the weakest available control for a problem that reached the point of needing a formal action.
Procedural Changes and Lessons Learned
Making the Change Stick
A preventive action that is not embedded in a document dies with the person who thought of it. Embedding normally means several documents change together:
- The work instruction or SOP (what the operator does), at a new revision level, with training recorded against the new revision.
- The control plan (what is checked, how often, and the reaction plan).
- The PFMEA (occurrence and detection ratings re-scored with evidence, not optimism).
- The inspection plan or checklist, if the verification method changed.
- The drawing or specification, via engineering change, if the product definition changed.
Document control then does the rest: superseded revisions are withdrawn from the point of use, and only the current revision is available at the workstation (Section 1.2).
Lessons Learned
A lessons-learned process captures what was discovered so that the next program does not rediscover it. To be useful it must be:
- Specific: "Countersink depth on 0.062 inch aluminum skins cannot be held with a fixed-stop microstop in a hand drill; use a power-feed unit" is a lesson. "Improve drilling quality" is not.
- Retrievable at the moment of decision: lessons belong in the design checklist, the FMEA library, the standard control-plan template, and the process-selection guide — not in a report nobody opens.
- Fed back into the FMEA library, so that the next PFMEA for a similar process starts with the known failure mode already listed and already rated.
Read-Across (Horizontal Deployment)
Read-across — also called horizontal deployment or yokoten — is the discipline of applying a lesson from one place to every other place it could apply. It is the single highest-leverage step in the preventive-action process and the one most often skipped, because the team that solved the problem considers the job finished.
Defining the Read-Across Population
After any corrective or preventive action, the team asks a structured set of questions:
| Question | Read-across candidates |
|---|---|
| Which other parts use this same feature, material, or tolerance? | Sister part numbers, the rest of the family |
| Which other processes use this same machine type, tooling, or fixture concept? | Other cells with the same machine model |
| Which other operators or shifts perform this operation? | Second and third shift, weekend crew, temporary staff |
| Which other lines, cells, or plants run a similar process? | Other sites, contract manufacturers |
| Which other suppliers provide a similar component by a similar method? | The rest of the approved vendor list |
| Which other documents contain the same defective instruction or ambiguity? | Every procedure copied from the same template |
The read-across review is documented as part of the action record, including the areas evaluated and found not applicable and why. An empty read-across section is a legitimate audit finding, because it shows the question was never asked.
[!CAUTION] The most expensive words in quality are "that was a different line." A shop that fixes a fixture-location error on line 1 and leaves the identical fixture on line 2 untouched has converted a solved problem into a scheduled one.
Worked Example: From Near-Miss to Read-Across
Situation: Over four weeks, final inspection catches three brackets with a missing 4 mm dowel pin. All three were caught before shipment, so no customer ever saw one. No formal nonconformance was raised in two of the three cases because the parts were simply sent back for the pin to be pressed in.
Step 1 — Recognize the signal. Three catches at final inspection means the in-process control failed three times. These are near-misses, and they are preventive-action input, not routine rework.
Step 2 — Analyze. A Pareto of final-inspection catches shows missing-pin events are the top category. A check of the assembly station shows the pin is picked from an open bulk bin and there is no verification that it was installed.
Step 3 — Choose the control level. Retraining assemblers is level six. The team instead installs a fixed-value poka-yoke: pins are issued in a kitted tray of exactly the shift quantity, and a contact-method sensor in the downstream fixture confirms pin presence and will not release the part if it is absent. The sensor is a control device, not a warning light.
Step 4 — Embed the change. The work instruction is revised and re-trained, the control plan adds the sensor as the control method with a reaction plan, and the PFMEA detection rating is re-scored with the sensor capability recorded as evidence.
Step 5 — Read across. Four other bracket part numbers use the same pin and the same fixture concept, and a sister plant runs the same assembly. All are evaluated; three receive the same sensor, one is confirmed not applicable because the pin is installed by an automatic press with existing verification, and the rationale is recorded.
Step 6 — Verify effectiveness. The leading indicator is the near-miss count at final inspection, monitored for three months alongside the sensor's own reject log. Zero missing-pin catches across the read-across population, with a nonzero sensor reject count proving the sensor is actually catching attempts, is the evidence that the action worked.
Common Exam Traps for CQT Candidates
[!CAUTION] Trap 1: Calling rework a preventive action. Pressing in the missing pin is a correction. Preventing the omission is the preventive action.
Trap 2: Waiting for a nonconformity to trigger preventive action. Preventive action is triggered by analysis: trends, near-misses, FMEA priorities, and lessons from elsewhere.
Trap 3: Offering retraining as the action. Training is the weakest control in the hierarchy; prefer elimination, then prevention-class error-proofing, then detection at the source.
Trap 4: Choosing a warning device where a control device is feasible. A warning relies on human response and produces defective parts with an alarm attached.
Trap 5: Confusing the three poka-yoke mechanisms. Contact uses shape or physical property, fixed-value counts a required number, and motion-step enforces sequence.
Trap 6: Skipping read-across. The action record must show which other parts, processes, shifts, lines, sites, and suppliers were evaluated, including those found not applicable and why.
Trap 7: Trying to prove a negative. Preventive-action effectiveness is demonstrated with leading indicators such as near-miss counts, capability trends, and error-proofing reject logs, not by the absence of a defect that never appeared.
An assembly cell installs a fixture sensor that will not release the part unless a dowel pin is detected in the bore. Under the standard poka-yoke classification, which mechanism and which response class does this device represent?
After a corrective action eliminates a fixture-location error on stamping line 1, the team closes the action. Six weeks later the identical defect appears on line 2, which runs a different part number on the same fixture concept. Which element of the process was omitted, and what should the action record have contained?
Over four weeks, final inspection catches three assemblies with a missing dowel pin. No unit reached the customer, and in two of the three cases the part was simply returned to the cell for the pin to be pressed in, with no record raised. How should a quality technician characterize these events?
You've completed this section
Continue exploring other exams