8D Closed, but New Project Falls into the Same Pit? —— Five Steps to Feed Failure Experience Back into FMEA

1. 47 8Ds Closed on Time, but the New Project Still Fails in the Same Way

A certain automotive electronics company supplies vehicle power modules to a whole vehicle customer. In March 2024, the customer reported that the module failed to start at low temperatures, with a batch defect rate reaching 1.2%. The 8D team used the 5Why method to trace the root cause, which was identified as: a specific model of MOSFET had a higher on-state voltage drop at low temperatures, combined with a small design margin in the drive circuit. Any fluctuation in the incoming material batch would trigger the failure. The solution was two-pronged—switching to a more stable supplier batch and increasing the design margin of the drive voltage. The 8D was closed on time, the customer's verification passed, and everyone breathed a sigh of relief.

In June 2025, the company's new platform module for another customer entered the PV trial production phase, and the low-temperature start test failed again in the same way. The engineering team reviewed the previous 8D report and found that while the new project used a different MOSFET model and recalculated the drive parameters, the logic for determining the design margin was identical to the previous version. The lessons from the previous 8D had never been incorporated into the new project's design input.

The quality department subsequently conducted a special audit, and the data was not encouraging: out of the 47 8Ds closed in 2024, only 6 were updated in the corresponding product's PFMEA, with an update rate of less than 13%. Among these 47 issues, 9 were "new failure modes," and 3 of them reappeared within 18 months after changing the model or production line. In other words, the fire was put out, but the fire prevention plan was never revised, leading to another fire in a different location.

The economic cost was also significant: the first fire took six weeks to extinguish, with travel, screening, air freight for replacement parts, and customer claims totaling several hundred thousand yuan. The second time, although it did not result in a batch complaint, the PV trial production was delayed by a month. The same lesson, due to not being fed back into the design input, cost the company twice.

This is not an isolated case. Many companies have a fast 8D process—covering all steps from D1 to D8—but when they open the PFMEA, it is still the old version from before mass production. The actual failure modes that occurred are not listed in the "potential failure modes" section. The 8D addresses the issues of the last batch, while the FMEA aims to prevent the next product. The lack of a "feedback loop" for experience is the most hidden and expensive gap in the recurrence prevention system.

2. Why 8D Conclusions Must Be Fed Back into FMEA

First, let's look at the positioning of both tools. FMEA is a "pre-emptive" risk identification tool: before the design and process are finalized, it identifies potential problem areas and adds preventive and detection measures in advance. 8D, on the other hand, is a "reactive" problem-solving tool: when a failure has already occurred, it identifies the root cause and implements corrective actions. One is for fire prevention, and the other is for firefighting, logically they are naturally upstream and downstream.

The issue is that most companies treat them as two separate processes: the FMEA is completed and reviewed during the APQP phase, then archived and forgotten; the 8D is initiated when a customer complaint occurs and similarly archived after closure. This leads to an absurd situation—FMEA's "potential" failure modes are guessed based on experience, while the real failures verified by 8D are never used to update this list. Real failures are the most authoritative test for FMEA: why were they not listed? If listed, why were they given a low occurrence rating? Why did the detection methods fail? Each question is a free lesson to improve the quality of the FMEA.

The AIAG-VDA FMEA manual also includes this feedback loop in its methodology: use "lessons learned" to continuously update the base FMEA and family FMEA, so that new projects can start their APQP on the shoulders of history, rather than guessing risks from scratch. The reasons for the feedback loop breaking down are usually threefold: first, treating FMEA as a one-time deliverable that no one cares about after the project ends; second, the 8D closure criteria do not include "synchronizing FMEA updates," leaving no one responsible; third, engineers want to do it but lack the skills—8D conclusions are written in plain language, such as "MOSFET has a high on-state voltage drop at low temperatures, insufficient drive margin," while FMEA requires a structured expression of "failure mode—failure cause—current control—action," and no one leads the translation effort.

Understanding this, it becomes clear that feedback is not just "one more piece of paperwork," but a way to ensure that the money spent on firefighting is turned into bricks for the next fire prevention effort.

3. Five Steps to Feed Failure Experience Back into FMEA

Step One: Add a "Feedback Review" to the 8D Closure Gate. Include "whether FMEA needs to be updated" in the 8D closure checklist as a mandatory signature item after D7 and before formal closure, to be confirmed by the quality engineer and product engineer. If no update is needed, provide a reason; if an update is required, assign a responsible person and set a completion date, and do not close the 8D until the update is complete. This step, though simple, is the foundation of the entire mechanism—without it, the following four steps are just empty words. Specifically, this item should be a mandatory field on the closure form: if no update is needed, provide a reason; if updated, attach the modification record, and sign for traceability. The review should be conducted after the corrective action has been verified, at which point the root cause and measures have been confirmed, making the content reliable.

Step Two: Use the "Five Questions" to Screen, Not Every 8D Needs to Update FMEA. Feedback is not about stuffing all 47 8Ds into the FMEA—too many entries will turn the FMEA into an unreadable log. During the closure review, go through the following five questions for each 8D, and only those that hit any of the questions are worth including:

Step Three: Translate 8D Language into FMEA Language. This is the most challenging part—get the translation wrong, and everything that follows will be wrong. The corresponding relationships are as follows:

When translating, adhere to two boundaries: the failure mode should be written in the "occurrence process," not the "discovery process"—in this case, the root cause is in incoming material and design, so the PFMEA should be updated in the corresponding processes. The failure cause should not be vague terms like "operator error" or "nonconforming incoming material," but specific mechanisms that can be addressed with preventive measures.

Step Four: Update the PFMEA and Synchronize Linked Documents. Add or revise the failure mode in the corresponding process row, and re-evaluate S/O/D based on real data: S is rated based on the impact on the customer and end user, O is adjusted based on actual occurrence frequency, and D is reassessed to see if the current detection methods are truly effective, and if not, the score should be lowered. New measures should not just be listed in the "recommended actions" column for future review, but should be directly implemented in the control plan, work instructions, and inspection specifications, forming a synchronized update package of "PFMEA—control plan—on-site documents." If the mechanism is design-related—such as component selection, derating, or topology—it should also be fed back to the DFMEA and the family FMEA and base FMEA should be updated accordingly, so that new projects can directly inherit the lessons. Documents should be controlled and released in the same batch to avoid a new gap where the FMEA is updated but the on-site operations continue using the old methods.

Step Five: Use Metrics to Ensure Habit Formation. Once the mechanism is established, it needs to be quantifiable and auditable. Two metrics are suggested: first, the "FMEA update rate," which is the proportion of closed 8Ds that have completed FMEA review and updates, with a target of over 80%. This should be tracked monthly, and any rate below the target should be investigated to identify the bottleneck. Second, the "new project historical failure coverage rate," which checks during the APQP phase gate review whether the PFMEA covers similar historical failures from the past two years. If not, the project should not pass the gate. A quarterly review of the consistency between the FMEA and 8D logs should also be conducted, focusing on whether the failure modes are fully supplemented, the scores match the real data, and the measures are implemented in the control plan, to prevent the mechanism from reverting.

4. The Five Easiest Pits to Fall Into

Pit One: Stuffing All 8Ds into the FMEA. Over-correcting by including all one-off, single-point, and fully resolved issues in the FMEA will make it an unreadable dictionary, drowning out truly important failure modes. Always go through the "five questions" before including any 8D.

Pit Two: Only Updating the PFMEA for the Affected Product, Leaving the Family and Base FMEAs Unchanged. If only the single product's PFMEA is updated, new projects will still refer to the old version of the base FMEA during APQP, and the lessons will not be passed on. General mechanisms that apply across models and processes must be synchronized to the higher-level FMEAs; otherwise, the feedback loop is only half complete.

Pit Three: Not Re-evaluating O/D During Updates, Keeping the Scores as Original. If the failure has already occurred and flowed out in batches, but the occurrence (O) score is still 2-3 and the detection (D) score is "inspection has intercepted," it is equivalent to denying reality with the scores. O should be adjusted based on actual occurrence data, and D should be re-evaluated based on whether it actually intercepted the failure—this time, if it did not intercept, it indicates that the original detection score was inflated.

Pit Four: Only Updating the FMEA, Not Synchronizing the Control Plan and On-Site Documents. The FMEA is the "risk archive," while the control plan is the "execution directive." If the measures are only included in the FMEA and not the control plan, they will never be implemented on-site; if included in the control plan but not the work instructions, the implementation will be flawed. Updates must be controlled and released in the same batch.

Pit Five: Delegating the Update Task to Clerks or New Engineers. The core of FMEA updates is engineering judgment—whether the mechanism is general, whether the scores are reasonable, and whether the measures are appropriate. This requires the personal involvement of product engineers and process engineers. Delegating the task to someone without the technical background will result in documents that are correctly formatted but contain inaccurate content, which is worse than not updating at all.

5. One Sentence Summary

The difference between firefighting and fire prevention is just one step—feeding failure experience back into the FMEA. Without this step, recurrence prevention will always be one step short.

Feeding failure experience into the FMEA is the true closure of 8D.

Knowledge code: 5.2.1

Version: v20260904

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