FMEA

Severity × Occurrence × Detection, prioritise by RPN.

What it measures

An FMEA walks through a design or a process element by element and asks, for each one: how can it fail, what happens when it does, what causes it, and would we catch it. The two types ask those questions about different things, and conflating them is the most common way FMEAs go wrong.

DFMEA: failure modes of the design

• Unit of analysis: a design element and its function. For a cell: the vent, the tab, the separator selection, the electrolyte formulation, the can crimp geometry.
• Failure mode: the way the element fails to perform its function, assuming it is built exactly as designed. The vent opens above its specified pressure. The tab cross-section cannot carry the rated current at temperature. Manufacturing escapes are deliberately out of scope here.
• Causes: design weaknesses. Wrong material choice, insufficient margin, tolerance stack-up, missing requirement.
• Controls: design verification. Simulation, tolerance analysis, design reviews, DVP&R testing (abuse tests, cycle life, vibration). Detection asks whether the verification program would expose the weakness before production, not whether an inspector would catch a bad part.
• When and who: during design, owned by the design team, ideally before tooling is cut, because the fixes are design changes.

PFMEA: failure modes of the process

• Unit of analysis: a process step. Mixing, coating, calendering, notching, stacking, welding, electrolyte fill, formation.
• Failure mode: the way the step can produce a nonconforming output even though the design is sound. Cold weld, coating weight out of window, misaligned stack, contamination introduced at fill.
• Causes: process variation sources. Parameter drift, worn tooling, operator error, material lot variation, environment (humidity, dust).
• Controls: prevention first (interlocks, poka-yoke, SPC on the driving parameter), then detection (inline checks, end-of-line tests). Detection asks where in the flow the bad part would be caught, and a catch many steps downstream scores worse than a catch at the station.
• When and who: before launch and continuously after, owned by process and quality engineering, updated every time the process changes or an escape proves a score wrong.

The scoring, common to both

Each failure mode gets three judgments on 1 to 10 scales. Severity rates the worst credible effect (in DFMEA, the effect on the end user and on safety; in PFMEA, the effect on the downstream process, the end product, and the customer). Occurrence rates how often the cause arises given current prevention. Detection rates how likely current controls catch it before it escapes, where 10 means essentially undetectable. The classic priority is RPN = S × O × D, from 1 to 1000. The current AIAG-VDA handbook replaces raw RPN with an Action Priority table (high, medium, low looked up from the S-O-D combination) because equal products hide unequal risks: a 9-2-2 mode at RPN 36 deserves more urgency than a 3-4-3 at the same product. Whichever ranking you use, the cycle is identical: score, rank, act, re-score, and the number has to move.

How to read the output

Read the worksheet left to right: element or step, its function, the failure mode, the effects (which set S), the causes (which set O), the current prevention and detection controls (which set D), then the priority and the action list. Then read it as a portfolio. High-S rows are design conversations regardless of how rare they are; severity only moves by changing the design or the effect path. High-O rows want prevention: error-proofing, parameter control, more robust process windows. High-D rows, the failures you would not see, are often the cheapest big wins, because adding a detection point drops risk without touching process physics. The portfolio view also exposes the classic imbalance: a sheet whose every action is "add inspection" is treating detection as the only lever, and a sheet whose every action is "retrain operator" is not acting at all.

Finally, check the linkage. A DFMEA failure mode whose cause is "process cannot hold the tolerance" should hand that characteristic to the PFMEA as a special characteristic, and the PFMEA's detection plan becomes the control plan. The two documents describe one product; when they disagree about what matters, one of them is stale.

A real use case

DFMEA. During cell design, the vent scores S 9 (thermal event if it fails to open), O 4 (a known sensitivity of the crimp geometry to tolerance stack-up), D 5 (the abuse-test matrix exercises nominal vents but not worst-case-tolerance vents). RPN 180, and the severity makes it high Action Priority regardless. The action is design-side: a tolerance analysis on the crimp stack plus worst-case vent samples added to the DVP&R abuse tests. Detection falls to 2: the verification program now exposes the weakness before tooling.

PFMEA. On the assembly line for the same cell, the tab-weld step scores "cold weld, intermittent contact" at S 8 (field failure risk), O 5 (known parameter sensitivity), D 6 (first catch is the end-of-line resistance test, hundreds of cells downstream). RPN 240, top of the process sheet. Severity is fixed by physics and occurrence is already being worked through weld parameter control, so the action targets detection: an inline weld-resistance check at the station. D falls from 6 to 2, re-scored RPN 80, and when an electrode lot later drifts weldability, the station catches it in minutes instead of after a day of production. Same product, two FMEAs, two different kinds of fix, and both numbers moved because both sheets were re-scored after the actions landed.

Common mistakes

• Mixing the two types: putting "operator welds at wrong parameter" in a DFMEA or "tab cross-section undersized" in a PFMEA. Each document loses its meaning when it absorbs the other's failure modes.
• Scoring without acting. An FMEA whose top priorities have no actions and owners is paperwork, not risk management.
• Treating an RPN threshold as gospel. Equal RPNs hide unequal risks; severity-9 modes deserve attention at any product. This is exactly why AIAG-VDA moved to Action Priority.
• Writing it once. An FMEA frozen at launch describes a line that no longer exists; every process change, escape, and 8D should update the relevant rows.
• Negotiating scores down to dodge the action threshold. The scale stops meaning anything and the riskiest modes hide in plain sight.
• Ignoring the DFMEA-to-PFMEA handoff: special characteristics identified in design that never arrive in the process sheet or the control plan.
• Forgetting detection placement. "We catch it at end of line" and "we catch it at the station" can both be true, and they are very different D scores.