Failure mode and effects analysis (FMEA) walks through a design or process step by step and asks, for each part: how could this fail, what would that do, and what would cause it? It grew out of US military reliability work (MIL-P-1629, 1949, later MIL-STD-1629A), spread through aerospace and the car industry, and is now standardised internationally as IEC 60812:2018. Each failure mode is rated on three scales and the ratings are multiplied into a Risk Priority Number (RPN). Where the other methods in this series ask “how likely?” and “how bad?”, FMEA adds a third question that security teams should recognise at once: would we detect it before it hurts?
1 · Severity, Occurrence, Detection
| Rating | Question | 1 means | 10 means |
|---|---|---|---|
| Severity (S) | How bad is the effect of this failure? | No noticeable effect | Hazardous — safety or regulatory failure without warning |
| Occurrence (O) | How often will the cause happen? | Extremely unlikely | Almost certain / very frequent |
| Detection (D) | How likely are current controls to catch it before the effect is felt? | Almost certain to be detected | Cannot be detected |
▸ Detection runs backwards: a high number means poor detection. That is what lets it multiply into the RPN — a failure nobody would notice raises the priority, which is exactly how an unmonitored security control should be treated.
Each organisation (or industry handbook) defines what every step from 1 to 10 means for its products or services. The anchors above are the common shape of those scales, not a quotation from a standard.
2 · The Risk Priority Number
Failure modes are sorted by RPN, actions are planned for the top of the list, and each action is re-rated afterwards to show the new RPN. The before-and-after table is what makes FMEA a working document rather than a one-off assessment.
IEC 60812:2018 describes how FMEA and FMECA (FMEA with criticality analysis) are planned, performed, documented and maintained. Its third edition added alternative ways of calculating an RPN and a criticality-matrix method — that is, a plain severity-by-occurrence grid of the kind used in the 5×5 risk matrix — because ranking by the bare product has known problems.
3 · Ties and Traps
The multiplication creates the same issues as a 5×5 score, at a larger scale:
- Very different failures tie. S10 × O6 × D2 = 120, and so does S2 × O6 × D10. One is a catastrophic failure you would see coming; the other is a nuisance nobody notices. 24 different combinations produce 120.
- Only 120 values exist. The 1,000 possible combinations collapse onto 120 distinct RPNs, bunched at the low end; there is no 11, 13 or 999.
- Severity can be outvoted. A failure that could kill someone (S10) but is rare and easy to detect (O1, D1) scores 10 — below a cosmetic defect rated S4, O5, D5 (100).
- Fixed thresholds mislead. “Act on anything above 100” ignores the S10 above and over-reacts to a pile of mid-scored nuisances.
- Ordinal, again. 1–10 ratings are ordered labels; an RPN of 200 is not twice as risky as 100.
4 · Beyond RPN: Severity First
The car industry’s joint AIAG & VDA FMEA Handbook (first edition, 2019), now the reference for automotive suppliers, replaced RPN ranking with Action Priority (AP): each S-O-D combination is assigned High, Medium or Low priority from a lookup table, weighting severity first, then occurrence, then detection. The point is the same as IEC 60812’s criticality matrix: never let a good detection score bury a severe failure. (The AP table itself is part of the handbook and is not reproduced here.)
You can get most of the benefit without the handbook:
- Review every S 9–10 failure mode whatever its RPN — a severity-first rule.
- Then sort by S × O (severity and occurrence only), which IEC 60812 offers as an alternative to the full product.
- Use D to choose the action, not the priority: a high D says “add monitoring”; a high O says “remove the cause”.
5 · Worked Example: Backup and Restore
FMEA shines on processes and controls. Take the control that decides the series’ ransomware risk — restoring from backup — and analyse how it fails:
| Failure mode | S | O | D | RPN | Action → new rating |
|---|---|---|---|---|---|
| Backups encrypted along with production | 9 | 4 | 7 | 252 | Immutable, offline copy → O 2, RPN 126 |
| Backup job fails silently for weeks | 8 | 5 | 6 | 240 | Alert on failed or missing jobs → D 2, RPN 80 |
| Backup admin credentials in the same domain | 9 | 3 | 8 | 216 | Separate identity and MFA → O 1, RPN 72 |
| Restore takes longer than the recovery-time objective | 7 | 6 | 3 | 126 | Quarterly timed restore tests → O 3, RPN 63 |
Note what the severity-first rule adds: three of the four failure modes are S 8–9, so all three are reviewed regardless of their rank. The “silent failure” row also shows FMEA’s particular value in security — its fix is detection, a lever the other scoring methods never ask about.
Try it: risk priority number
Bottom line
- Three questions, not two — severity, occurrence and detection; detection is the one security risk registers forget.
- RPN = S × O × D, 1 to 1,000 — but only 120 values exist and 24 combinations tie at 120.
- Severity first — IEC 60812’s criticality matrix and AIAG-VDA’s Action Priority both exist to stop a good D score hiding a severe failure.
- Use D to pick the fix — high D means add monitoring; high O means remove the cause.
- Best on processes and controls — backups, change management, access reviews — where “how does this fail?” is the right question.
Sources & method. IEC 60812:2018, Failure modes and effects analysis (FMEA and FMECA), edition 3.0 — scope and changes as published by IEC (cited, not quoted). AIAG & VDA FMEA Handbook, 1st edition (2019) — Action Priority described, its table not reproduced. Historical lineage: MIL-P-1629 (1949) and MIL-STD-1629A (1980). RPN counts are arithmetic over all 1,000 S-O-D combinations. Scale anchors and the backup example are illustrative. Original analysis; no third-party graphics or text reproduced.
← Back to Cyber security