In PFMEA, Should "Inspection" Always Result in a Lower Score for Detection (D)? —— A Five-Step Method for Correctly Assessing and Correcting Detection Scores
When the PFMEA for a steering bracket at an automotive parts factory was submitted to the client, the severity and occurrence columns were acceptable, but the detection (D) column almost entirely fell within the 2-4 range. The reason given was uniformly "final inspection," "self-inspection," or "poka-yoke devices." The client auditor simply asked, "What is the detection control for the missed welding item? At which process step does the detection occur?" The on-site response was final visual inspection, while the FMEA form listed a poka-yoke device.
The consequences were clear: in the AP (Action Priority) ranking, the true risk of missed welding, which should have been prioritized, was ranked in the bottom third. Improvement resources were entirely focused on a few items that had "high scores but were inherently controllable." The team was not lazy; the issue was their understanding of detection, which was limited to whether or not an inspection was in place.
1. Detection Measures Not Just "Whether There Is an Inspection," but "Whether It Can Be Intercepted"
Detection answers the question: when a failure cause has occurred and a failure mode has been generated, how early and reliably can the current control measures identify and prevent it from flowing out? This is determined by three factors: the timing of detection, the maturity of the detection method, and the evidence supporting the effectiveness of the method. Focusing solely on whether an inspection exists is equivalent to looking at only half of the second factor.
Another often overlooked aspect is the division of labor: preventive controls affect occurrence (O), while detection controls affect detection (D). Filling in preventive descriptions such as "operator follows SOP," "incoming inspection has been passed," or "equipment is regularly maintained" in the detection column will artificially lower the D score and inflate the O score—both columns will be distorted, making the risk ranking meaningless.
2. A Comparison Table: Aligning D Scores with Detection Methods
The table below categorizes scores based on "detection timing + method characteristics + evidence requirements." The specific scoring tiers should follow the criteria adopted by the company (AIAG-VDA manual or client requirements), but the evolution direction is consistent: the closer to the occurrence point, the more automated, and the harder the evidence, the lower the score.
| D Score | Detection Timing | Method Characteristics | Required Evidence |
|---|---|---|---|
| 1-2 | At the occurrence point, failure cannot flow out | Poka-yoke devices, automatic line stop or automatic rejection | Poka-yoke function verification records (at the start of each shift, each changeover) |
| 3-4 | Within the process, detection and judgment occur simultaneously | Online measurement or automatic judgment by equipment, automatic isolation | MSA of automated measurement systems, judgment and isolation rules |
| 5-6 | In this process or the next process | Manual measurement (gauges + drawing requirements), 100% or sampling inspection | GR&R qualified, inspection work instructions, judgment criteria |
| 7-8 | Final inspection, before warehousing | Manual visual inspection, qualitative judgment by go/no-go gauges | Attribute consistency (Kappa), visual inspection environment and training records |
| 9-10 | At the customer site, during use, or undetectable | No detection control, or method does not match failure mode | None—high scores inherently require no evidence |
The key to reading this table is not to memorize the score tiers, but to remember one phrase: the same method can have different scores depending on the timing. Final visual inspection is scored 7-8, even though it is "an inspection"; online automatic measurement within the process can score 3-4, even though it is "just an inspection."
3. Five-Step Method: Accurately Assigning D Scores
Step 1: Lock down the occurrence point, then ask about the detection point. Detection is scored based on the failure cause/failure mode, and before scoring, it must be clearly stated "which process step generates this failure mode." If missed welding occurs at the welding station, even if it is detected at the final inspection, the detection control should describe "what intercepts it from flowing out of the welding station to the final inspection."
Step 2: Only write detection controls in the detection column. Anything that "prevents occurrence" (SOP, training, maintenance, poka-yoke design itself) should be categorized as preventive control; anything that "detects and handles after occurrence" (measurement, visual inspection, automatic alarms, isolation) should be categorized as detection control. Clear distinction between the two columns is essential for accurate scoring.
Step 3: Score based on timing and method. First, determine at which process step the detection occurs, then assess whether the method is automatic or manual, and whether there are judgment and isolation actions. Finally, assign the score. "Final visual inspection" should be scored 7-8, and "final inspection is important, so score it 4" is not acceptable.
Step 4: Verify evidence, and adjust scores if evidence is insufficient. For D ≤ 4, evidence of poka-yoke verification or MSA of automated measurement systems must be provided; for D 5-6, a qualified GR&R is required; for manual visual inspections scored 7-8, attribute consistency and training records must be available. If evidence cannot be provided, the score should be increased by one tier—evidence is the receipt for the score, and scores without receipts do not count.
Step 5: Use AP to verify the direction of measures. Projects with high S, high O, and high D will naturally be ranked at the top. Priority measures should focus on "reducing detection scores": upgrading final visual inspections to online automatic judgments or poka-yoke devices can drop the D score from 7-8 to below 3, which is much more effective than repeatedly emphasizing "strengthening inspections." Remember: severity cannot be reduced by control methods; only design changes can achieve this. D score reductions must be implemented in tandem with the measures, not just by changing the numbers in the table.
4. Four Common Misratings
- Assigning low scores to any "inspection." Scores 2-4 correspond to automatic judgment and isolation within the process. Final manual visual inspections do not meet this criterion, and artificially low scores can bury high-risk items.
- Filling preventive controls in the detection column. This approach artificially lowers D scores while inflating O scores, distorting the AP ranking in two places.
- Writing the detection process as the occurrence process. Treating "self-inspection at the welding station" as the detection for missed welding at the welding station masks the fact that it is only intercepted at the final inspection.
- Changing scores without changing the process. During reviews, adjusting the D score to 3 to pass the audit while the production line still relies on visual inspections will bring the same issues to light in the next client audit or complaint.
Detection is not just about labeling inspection methods, but about assessing whether the "gate" can effectively prevent failures from flowing out. Scores that align with timing, methods, and evidence are the only ones that can be trusted in PFMEA AP rankings.
Detection is not about "whether there is an inspection," but about "whether the failure can be intercepted before it flows out." Scores that align with timing, methods, and evidence are the only ones that can be trusted.
Knowledge code: 8.3.1
Version: v20261014
Author: QTank QTank is dedicated to providing systematic knowledge, methodologies, and practical tools for quality management professionals, helping companies continuously improve their quality capabilities.