PFMEA Practice Clarification (Part 3) | Does Inspection Interception Count When Scoring Occurrence? —— Analysis of the Relationship Between Occurrence (O) Scoring and Problem Detection Methods

One: A Real-World Issue That Divides the Scoring Team

During PFMEA review meetings, one of the most common scenes of debate is the occurrence scoring: the defect rate of a certain process is actually not low, but someone in the team insists, "We conduct a full inspection at the final stage, and all defects are intercepted, so they don't reach the customer. The occurrence score shouldn't be that high." Another group counters, "Inspection can intercept defects from flowing out, but it can't prevent them from occurring. The occurrence score should be what it is." Each side sticks to their argument, and in the end, either a number is decided on a whim, or a middle value is taken, leaving no one satisfied.

Almost every company that conducts PFMEA encounters similar dilemmas: the "current process control" column in the table clearly lists several problem detection methods—first article inspection, patrol inspection, SPC control chart, final full inspection, poka-yoke—and when scoring the occurrence (O), should the score be based on the defect rate after these methods have intercepted the defects, or the defect rate without any interception?

Getting this question wrong can distort the entire PFMEA risk ranking: high-risk items that need to be prioritized for improvement are suppressed with low scores, and resources are misallocated to less critical areas. This article will clarify the standard basis, judgment criteria, and practical steps.

Two: Standard Answer: Occurrence Only Recognizes Preventive Controls, Not Detection Interception

To start with the conclusion: The occurrence (O) score reflects the likelihood of the failure cause itself occurring, and only considers the effectiveness of preventive measures. Problem detection methods are considered detection controls and are scored separately in the detection (D) column.

The PFMEA occurrence evaluation table in the current AIAG-VDA FMEA manual (published in 2019, a unified standard for American and German systems) clearly states the scoring criteria in the first sentence: "Potential Failure Causes rated according to the criteria below. Consider Prevention Controls when determining the best Occurrence estimate." — When determining the occurrence estimate, the focus is on preventive controls (Prevention Controls). The entire occurrence evaluation table, from 10 points to 1 point, is anchored on preventive controls:

• 10 points: No preventive controls, the failure cause is almost certain to occur (approximately 1 in every 10 pieces);
• 7 points: Preventive controls are "effective," approximately 1 in every 100 pieces;
• 1 point: The failure cause is eliminated through preventive controls—reliant on design methods such as part geometry, fixtures, and tooling, making the failure cause physically impossible to occur.

Notice that the table does not mention "inspection," "interception," or "sampling inspection" at all. This is because inspection, testing, SPC alarms, visual inspections, and other such methods are categorized under another type in the new manual—detection controls (Detection Controls). They correspond to a separate detection (D) evaluation table, scored based on "maturity of detection methods, detection opportunities, and detection capability."

The fourth edition of the AIAG FMEA manual (2008, old version) follows the same principle: occurrence (O) is the "likelihood of a specific cause or mechanism occurring," assessed based on current preventive controls; detection controls are not factored into the occurrence score and are evaluated in the detection (D) column. The old version's occurrence table has clear frequency ranges: 10 points for approximately 1 in every 10 pieces (failure is almost inevitable), 7 points for approximately 1 in every 100 pieces, 5 points for approximately 1 in every 2000 pieces, 3 points for approximately 1 in every 100,000 pieces, and 1 point for approximately 1 in every 1.5 million pieces (preventive controls almost certainly prevent the failure from occurring). Note that these anchor points describe the frequency of occurrence of the failure, not the number of defects intercepted. The VDA FMEA method in the German system also separates "preventive measures" and "detection measures" for evaluation. In other words, from American to German systems, from old to new versions, the standard is consistent: interception effectiveness does not offset the occurrence frequency.

Three: Distinguishing: What Methods Can Reduce Occurrence, and What Methods Can Only Reduce Detection

In practice, people often get confused because the term "problem detection methods" is too broad. To determine whether a method can affect the occurrence score, simply ask one question: Does it prevent the failure cause from occurring, or does it detect the failure after it has occurred?

Mnemonic: "Preventing it from happening" is prevention, "detecting it after it happens" is detection.

Here are two specific examples:

Example One: Poka-yoke Fixture. At a certain assembly station, a positioning pin is used to prevent parts from being installed incorrectly, making it physically impossible to install them the wrong way—this is a preventive control. If this poka-yoke design is reliable enough, the occurrence score can be adjusted downward, even to 1 point (the cause is eliminated).

Example Two: Final Full Inspection. Defects from the same process will flow to the final inspection and be 100% detected and reworked—this does not change the probability of "installing parts incorrectly." Workers will still install them incorrectly, but the time of detection is advanced. Therefore, the occurrence score remains unchanged, and the credit for the full inspection is reflected in the detection score (D score, which is low, indicating strong detection capability).

A common pitfall here is that many people consider "poka-yoke devices" as detection methods. In fact, the implementation of poka-yoke needs to be examined in detail—poka-yoke that prevents the cause from occurring (such as physical poka-yoke, interlocks that stop the machine upon anomaly) is a preventive control and can reduce the occurrence score; poka-yoke that only alarms or detects defects (such as automatically removing nonconforming products but the cause continues to occur) is a detection control and can only reduce the detection score. The definition of O=1 in the manual is very clear: "Failure is eliminated through prevention control"—elimination through prevention, not interception.

Another common gray area to analyze: Is the SPC control chart a preventive or detection method? SPC itself is detection—it cannot prevent defects from occurring, but it can detect process anomalies in a timely manner through statistical rules, shortening the time from "occurrence to detection." Therefore, its direct effect is reflected in the detection score (D). However, note that if SPC is accompanied by a reaction plan, and the process is immediately stopped and adjusted upon an anomaly alarm, bringing the process back to a controlled state and thus avoiding more defects, this "alarm + timely correction" combination has some preventive attributes in practice. Under sufficient team argumentation, its preventive part (preventing the anomaly from continuing, preventing batch defects) can be moderately reflected in the occurrence assessment. The judgment standard remains the same: does it directly reduce the probability of the failure cause occurring? Only observation and alarm methods are categorized as detection; methods that can prevent the cause from continuing are considered preventive.

Four: Why Can't Interception Effects Be Factored into Occurrence

Someone might ask: if full inspection 100% intercepts defects, and the customer receives no defects, why does the occurrence score still need to be high? There are three reasons at different levels.

First, logically, it would result in double counting. The occurrence (O) score reflects the "likelihood of the failure cause occurring," while the detection (D) score reflects "our ability to detect it." These are independent dimensions. If the interception effect is counted in O, it means the same method is credited twice in O and D: O is reduced, D is improved, and the RPN (in the old version) or action priority (AP) (in the new version) is underestimated, systematically masking the risk. The new manual's use of AP instead of RPN is precisely to address this—AP maps the combination of S/O/D in the order of "severity, occurrence, detection" to high/medium/low. Even if the detection capability is very strong (D is very low), a high occurrence plus high severity will still trigger high-priority improvement actions. The standard designers do not allow detection to "whitewash" the occurrence frequency from the beginning.

Second, the data does not match. The occurrence score should be consistent with historical defect data: if a certain cause produced 50 actual defects over the past three months, with an average of 5 defects per 1000 pieces (1/200), the occurrence score should be around 5 points. Even if all these defects were intercepted by inspection and none flowed out, the occurrence frequency remains 1/200, and the occurrence score will not become 1 point just because "all were detected." If the score does not match the actual data, it will be exposed during the audit.

Third, it can mislead management decisions. The significance of the occurrence score is to inform you of "how prone this process is to errors," thus deciding whether to implement root-cause solutions in terms of process, equipment, and poka-yoke. If the occurrence score is artificially lowered, this item will disappear from the risk ranking seen by management, and root-cause solutions will never be initiated. Defects will continue to occur and rely on inspection to be caught—quality costs will remain high, and there is always a risk of batch defects due to a single inspector's oversight. Inspection is the last line of defense, not a "get-out-of-jail-free card" for the production process.

Five: How to Score Occurrence: Five Practical Steps

Once the principles are clear, scoring becomes straightforward. Follow these five steps, each with a clear basis:

Step One: Lock Down the Failure Cause (FC). In the new PFMEA, the occurrence (O) score is associated with the failure cause row, not the failure mode row. First, determine the specific cause you are evaluating (e.g., "operator fails to install a washer"), and evaluate each cause individually.

Step Two: Assume No Inspection, Estimate the Occurrence Frequency. This is the most critical mindset shift—pretend there is no inspection or alarm in the process, and estimate how often this cause occurs based on historical data (PPM, defect rate, after-sales data) or similar process experience. Use the occurrence table's anchor points (10 points ≈ 1/10, 7 points ≈ 1/100, 5 points ≈ 1/2000, 3 points ≈ 1/100,000, etc.) to determine the baseline score.

Step Three: Inventory Preventive Controls, Adjust Scores Based on Effectiveness. List the preventive measures in the current process control: poka-yoke fixtures, interlocks that stop the machine, automatic control of parameters, standard work instructions, process design improvements, etc. Adjust the score according to the preventive control descriptions in the occurrence table—stronger preventive controls that can prove the cause can be eliminated result in lower scores; no preventive controls mean maintaining the baseline score or even higher.

Step Four: Do Not Adjust Occurrence for Detection Methods. Methods such as first article inspection, patrol inspection, SPC, final full inspection, and testing are all listed in the "current process control - detection" column and scored according to the detection (D) evaluation criteria. D and O do not interfere with each other.

Step Five: Cross-Verify, Let the Data Speak. After scoring, cross-check the historical defect data: does the frequency range corresponding to the occurrence score match the actual occurrence frequency? If there is a significant discrepancy, return to Step Two for a re-evaluation until the team agrees on the "actual frequency of the cause occurring."

Let's walk through an example of a tightening process. A company's critical process is bolt tightening, and historical data shows a torque deficiency rate of approximately 1 in every 500 pieces. The current process includes: torque automatic control on the tightening gun (preventive), first article torque inspection (detection), SPC torque monitoring (detection), and final torque sampling inspection (detection).

• Step One: Lock down the failure cause—“torque drift of the tightening gun leading to torque deficiency”;
• Step Two: Assume no inspection, and based on historical data of approximately 1 in every 500 pieces, use the occurrence table's anchor point (5 points ≈ 1/500) to set the baseline score to 5 points;
• Step Three: Inventory preventive controls—torque automatic control on the tightening gun is a preventive measure, and regular calibration is effective, so adjust the score slightly to 4 points (if an intelligent tightening gun with angle monitoring is used, which can further prove the cause is eliminated, the score can be further reduced);
• Step Four: First article inspection, SPC, and final inspection are all listed in the detection column, scored in D (the combination of SPC and full inspection has strong detection capability, so D can be scored very low), but they do not affect O=4;
• Step Five: Cross-verify with data—4 points correspond to approximately 1/2000, which deviates from the actual 1/500. After team re-evaluation, it is concluded that torque automatic control can significantly reduce the drift probability, and the score is maintained at 4 points with the basis noted in the remarks.

Throughout the process, the three inspection methods were not involved in the occurrence scoring—their value is already reflected in the D score.

Six: Common Misconceptions and Self-Checklist

List the recurring issues in practice for self-check during reviews:

One more tip for the review facilitator: large discrepancies in occurrence scoring often stem from inconsistent information within the team about the "actual frequency of the cause occurring." Don't rush to decide the score; first, lay out the data—do you have defect statistics, similar line experience values, and are preventive controls really effective? Once the data is aligned, the scores will naturally align.

Seven: Conclusion

The occurrence score answers the question, "Will the cause occur?" The detection score answers, "If it occurs, can we detect it?" These two questions cannot be offset against each other. Counting the interception effect of problem detection methods in the occurrence score is the most common misuse of PFMEA scoring logic and a hidden source of risk underestimation. Remember this: Inspection can intercept defects from flowing out, but it cannot prevent defects from occurring. Occurrence deals with production, detection deals with discovery, and each is evaluated separately to form the true risk.

Occurrence only considers prevention, detection only evaluates detection, and interception does not offset occurrence.

Knowledge code: 8.3.1

Version: v20260808

Author: Quality Think Tank Quality Think Tank is dedicated to providing systematic professional knowledge, methodologies, and practical tools for quality management practitioners, helping companies continuously improve their quality capabilities.