PFMEA Practice Clarification (Part 2) | Why Are Failure Modes Listed in the Welding Process When They Are Detected in the Final Inspection? —— The "Occurrence Process" and "Detection Process" of Failure Modes

1. Review Site: Failure Modes Misattributed to Processes

While reviewing a PFMEA, the page for the final inspection process was filled with failure modes: "poor welding," "missing material," "scratches," "dimensional deviations" — almost all the defects from the entire workshop were piled under the final inspection. The review expert asked, "Are these defects generated in the final inspection?" The engineer was taken aback: "No, they are detected in the final inspection." The expert followed up, "Then in which process were they generated, and why are they not listed in the PFMEA for that process?" The meeting room fell silent.

This is another common mistake in PFMEA preparation: attributing failure modes to the process where they are detected, rather than where they occur. Processes like final inspection, assembly, and testing, which are good at detecting failures due to comprehensive inspection records, often become the "shelters" for failure modes. Conversely, the PFMEAs for the processes where failures actually occur — such as welding, injection molding, and painting — remain blank. The principle for attributing failure modes is clear: they should follow the process where they are generated, not the process where they are detected.

2. Standard Logic: Failure Modes Are the Failures of Process Outputs

PFMEA is a tool for analyzing each process step. The AIAG-VDA manual refers to each process step as a "focus element," and the definition of a failure mode (FM) is the failure of the output from this focus element (process) — that is, which characteristic of the product or semi-finished product processed by this step does not meet the requirements.

This definition determines the attribution of failure modes: FM belongs to the process that produced it. The output of the welding process is the "weld seam," and a poor weld seam is a failure mode of the welding process. The output of the injection molding process is the "molded part," and a missing material in the molded part is a failure mode of the injection molding process. The output of the painting process is the "coating," and a scratched coating is a failure mode of the painting process. What is the output of the final inspection process? It is the "inspection conclusion" — the output of the final inspection cannot be a poor weld, missing material, or scratches; it only detects these defects.

Therefore, when determining which process a failure mode should be attributed to, the question is not "where was it seen," but "which process produced this state?" Seeing "poor welding," trace it back to the welding process; seeing "missing material," trace it back to the injection molding process. The process that detects it does not produce it; it is merely a "witness."

3. Separate Occurrence and Detection, Each Has Its Own Account

Separating "occurrence" and "detection" corresponds to two completely different columns in the PFMEA:

A failure can travel through multiple processes from its occurrence to detection: welding process generates poor welding → assembly process detects fracture under stress → final inspection confirms the defect through non-destructive testing. The division of labor in the PFMEA for these three processes is as follows:

• Welding Process: List the failure mode "poor weld seam," failure causes (current drift, welder technique, incoming weld wire quality), frequency O, and preventive measures.
• Assembly/Final Inspection Process: List "non-destructive testing" and "fracture check during assembly" as detection controls in the detection column of the welding process row, and evaluate D — the earlier and more reliably it is detected, the lower the D score.

Note that although the detection controls are written in the row of the welding process, they describe "in which stages the failure can be detected" — the detection stages can be the current process, downstream processes, or even at the customer end. This is the meaning of one of the three elements of detection opportunity in the AIAG-VDA manual: the shorter the chain from occurrence to detection, the better the detection opportunity, and the lower the D score. Detection opportunity is about "where it is detected," which belongs to D; failure mode is about "where it occurs," which belongs to FM. Two separate accounts, clearly defined.

4. The Cost of Misattributing Processes

Writing failure modes in the detection process may seem like just a misplaced row, but in reality, it misaligns the entire analysis chain:

First, a vacuum in risk analysis for the occurrence process. If the PFMEA for the welding process does not list "poor welding," it means the actual risks of welding have never been analyzed — the causes are not identified, the frequency is not evaluated, and preventive measures are not proposed. Risks do not disappear just because they are registered in a different place; they simply become invisible in the risk ledger.

Second, inability to conduct cause analysis. Failure causes (4M) act on the occurrence process: current drift occurs in welding equipment, and missing material occurs in injection molding machines. Writing the failure in the final inspection process leaves the cause column empty — the final inspection process has neither welding current nor injection molding parameters, so the cause can only be listed as "upstream material defect," pushing the cause to the previous process and breaking the analysis chain.

Third, all measures are misaligned. Preventive measures for poor welding (error-proofing of welding parameters, regular calibration of equipment, incoming weld wire inspection) should be listed under the welding process. If they are listed under the final inspection process, they become "strengthen final inspection" — using detection as a substitute for prevention, defects will still occur, just with an additional interception step. This is the structural reason why many companies update their PFMEAs annually but see no reduction in defect rates.

Fourth, control plans and special characteristics become disordered. Control plans inherit control measures from the PFMEA. If the PFMEA misattributes processes, the inspection station settings, error-proofing device layouts, and special characteristic control points in the control plans will all be misaligned. During customer audits, a comparison with the process flow chart will immediately expose the issue.

Fifth, frequency and detection scores are contaminated. If a failure is written in the detection process, frequency O can only be evaluated in the detection process — the score reflects "detection frequency" rather than "occurrence frequency." Detection score D is also unclear about whom to evaluate. All three scores (S/O/D) become distorted, and risk prioritization loses its meaning.

5. Why It Is Always Wrong: Three Common Misconceptions

Misconception One: Copying the detection process directly from the nonconformance report. Inspection records and nonconformance statistics are usually registered by inspection station (detection point). Teams preparing PFMEAs often take shortcuts by copying the failures from the nonconformance report directly into the detection process table. Solution: Nonconformance data only provides "what the failure is," not "where it occurs" — the occurrence location must be traced, not copied.

Misconception Two: Mistaking "detection point" for "occurrence point" during reverse analysis. Many companies use reverse FMEA (tracing risks from occurred nonconformances) to update their PFMEAs. They stop tracing at the inspection record, mistakenly assuming "detected in final inspection = generated in final inspection." Solution: Reverse tracing should continue up the process flow chart until the process that generated the failure is identified.

Misconception Three: Organizing analysis by inspection station rather than by process. Teams are accustomed to dividing analysis objects by "inspection station" because inspection stations have readily available data. However, the analysis unit for PFMEA is the process (each step in the process flow chart), not the inspection station. Solution: Use the process flow chart (PFD) as the framework, analyze each process step, and treat inspection stations as nodes on the flow chart.

6. Correct Approach: Process Flow Chart Framework + Reverse Tracing

Step One: Use the process flow chart (PFD) as the framework. Each row in the PFMEA corresponds to a process step on the flow chart. The flow chart must be complete before the PFMEA can be prepared — an incomplete flow chart will result in misaligned PFMEAs.

Step Two: Ask "Will the output be nonconforming?" for each process. For each process, ask what problems might occur with its output (product/semi-finished product state) and list the failure modes under the "process that produced it."

Step Three: Trace the occurrence process before listing the failure. When encountering any failure, first ask "which process produced this state," and do not list it if it cannot be traced — the occurrence process must be confirmed before listing it in the PFMEA.

Step Four: List detection information in the detection column. Write "in which stages and how early it can be detected" in the detection control column of the corresponding failure mode, and evaluate D. The same failure mode can list multiple detection stages (current process inspection, downstream inspection, final inspection), and the best detection opportunity should be used to evaluate D.

On-site Verification Question During Review: Randomly point to a failure mode and ask three questions — "Which process produced it?" "Where was it detected?" "Are these two processes the same?" If the answers to the last two questions are different, and the failure is listed in the detection process, it is misattributed.

Let's walk through the four steps with an example of an injection molded part. A certain injection molded part was found to have a missing material defect during the assembly process, and the nonconformance report registered it under the "assembly process."

• Step One (Flow Chart Framework): List each process step according to the process flow chart — injection molding, de-gating, assembly, final inspection.
• Step Two (Process Output Inquiry): The output of the injection molding process is the "injection molded part," which may have missing material, shrinkage, or flash. The output of the assembly process is the "assembly," which may have missing parts or incorrect assembly — list the failure modes for each process separately.
• Step Three (Trace the Occurrence Process): The missing material defect is generated in the injection molding process (material temperature, holding pressure, material shortage in the barrel), not in the assembly process — move "missing material (shrinkage)" from the assembly process to the injection molding process.
• Step Four (Detection Information in the Detection Column): The detection of missing material in the assembly process is listed as a detection control in the detection column of the injection molding process row: "visual inspection at the assembly station can detect missing material," and evaluate D accordingly. At the same time, the injection molding process should add failure causes (low material temperature, insufficient holding pressure, abnormal material level in the barrel) and frequency O, along with preventive measures.

After the revision, the injection molding process, which was previously blank, becomes a complete analysis with "missing material failure mode + four causes + preventive measures + detection methods." The PFMEA for the assembly process retains only the failure modes it outputs. The same nonconformance data, organized in two different ways, results in vastly different risk analysis quality.

7. Common Misconceptions and Self-Check List

8. Conclusion

Failure modes are the "outputs" of processes, not the "witness statements" of processes. Poor welding belongs to the welding process, missing material belongs to the injection molding process, and scratches belong to the painting process — even if they are all detected in the final inspection. The place where they are detected is the domain of detection controls, evaluated for detection capability; the place where they are generated is the domain of failure modes, evaluated for frequency, analyzed for causes, and improved for processes. Remember this: failure modes follow the production process, and detection controls follow the detection stages; each to its own, no risk is overlooked.

Failure modes are attributed to the occurrence process, and detection controls to the detection stages; two separate accounts, each in its place.

Knowledge Number: 8.3.1

Knowledge code: 8.3.1

Version: v20260808

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