algo-mfg-fmea
用 FMEA 找出失效模式、排優先順序並降低風險
中文翻譯
---
name: "algo-mfg-fmea"
description: "進行 FMEA，有系統地找出、排定優先順序並降低潛在的失效模式。當使用者需要評估產品或製程風險、排定矯正措施的優先順序，或建立風險登記表時使用，即使他們說的是『failure mode analysis』『risk assessment』『what could go wrong』或『RPN calculation』。"
metadata:
category: "WP-48 製造演算法"
tags: ["manufacturing", "fmea", "risk-analysis", "quality"]
---
# FMEA（Failure Mode and Effects Analysis，失效模式與效應分析）
## 概述
FMEA 有系統地找出潛在的失效模式、它們的效應、原因和現有的管制措施。每個失效以嚴重度（S）、發生度（O）、偵測度（D）各用 1-10 分評分。RPN = S × O × D 用來排定先處理哪些風險。AIAG-VDA FMEA（2019）用 Action Priority（AP）矩陣取代 RPN。
## 何時使用
**觸發條件：**
- 設計新產品 / 製程，並主動辨識風險
- 有系統地評估現有的失效模式，以排定優先順序
- 符合汽車（IATF 16949）或醫療器材（ISO 13485）的品質要求
**不該使用的情況：**
- 針對已知問題做根因分析時（用魚骨圖 / 5-why）
- 對不良資料做統計分析時（用 SPC 或柏拉圖）
## 演算法
```
IRON LAW: Severity Can NEVER Be Reduced by Design Changes
Severity is determined by the EFFECT on the customer. A brake failure
is always severity 10, regardless of how unlikely or detectable it is.
FMEA reduces risk by: lowering Occurrence (better design/process) or
improving Detection (better testing/inspection). NEVER inflate
Detection scores to lower RPN artificially.
```
### 階段 1：輸入驗證
定義範圍：設計 FMEA（DFMEA）或製程 FMEA（PFMEA）。組成跨部門團隊。準備：製程流程圖或系統方塊圖。
**關卡：** 範圍已定義、團隊已組成、有參考用的圖。
### 階段 2：核心演算法
1. 針對每個功能 / 製程步驟，列出所有潛在的失效模式
2. 每個失效模式都要找出：對客戶的效應、根本原因、現有的預防管制、現有的偵測管制
3. 評分：嚴重度（1-10）、發生度（1-10）、偵測度（1-10）
4. **經典 RPN：** RPN = S × O × D。優先處理高 RPN。
5. **AIAG-VDA AP：** 用 S-O-D 組合矩陣指派 Action Priority：High、Medium、Low。
6. 為 High 優先的項目定義建議措施，並指定負責人和目標日期
### 階段 3：驗證
審查：所有功能 / 步驟都涵蓋了嗎？嚴重度分數符合實際對客戶的影響嗎？偵測度分數實際嗎（不要過於樂觀）？
**關卡：** 涵蓋完整、評分實際、高優先項目都有指派措施。
### 階段 4：輸出
回傳 FMEA 登記表，附上排好優先順序的措施。
## 輸出格式
```json
{
"fmea_items": [{"failure_mode": "seal leak", "effect": "water damage", "cause": "material degradation", "severity": 8, "occurrence": 4, "detection": 6, "rpn": 192, "ap": "high", "action": "add pressure test at final inspection"}],
"summary": {"total_modes": 45, "high_priority": 8, "medium": 15, "low": 22},
"metadata": {"type": "PFMEA", "scope": "assembly line 3"}
}
```
## 範例
### 輸入輸出範例
**輸入：** 咖啡機沖煮模組，功能：「把水加熱到 93°C」
**預期：** 失效模式：過熱（S=7, O=3, D=4, RPN=84）、加熱不足（S=5, O=4, D=3, RPN=60）、完全不加熱（S=8, O=2, D=2, RPN=32）。
### 邊界情況
| 輸入 | 預期 | 原因 |
|-------|----------|-----|
| S=10，不管 O 和 D | 一律高優先 | 攸關安全的失效不管 RPN 多少都要處理 |
| RPN=100 (S=10,O=1,D=10) vs (S=1,O=10,D=10) | RPN 相同，風險差很多 | 這就是 AIAG-VDA AP 要取代純 RPN 的原因 |
| 沒有現有管制 | D=10（無法偵測） | 誠實的評估才會推動改善 |
## 注意事項
- **RPN 會誤導**：S=10,O=1,D=10 算出的 RPN=100（災難性但罕見、無法偵測），和 S=1,O=10,D=10（無關緊要但頻繁）差很多。AIAG-VDA AP 矩陣就是處理這個缺陷。
- **評分的一致性**：沒有校準的話，不同的團隊成員打分會不一樣。要用附範例的評分標準，並由團隊一起校準。
- **偵測 ≠ 預防**：偵測度分數低（偵測得好）並不會防止失效，只是把它攔下來。降低發生度要比改善偵測度更優先。
- **活的文件**：設計 / 製程有變更、出現新的失效資料，或實施矯正措施時，FMEA 都必須更新。靜態的 FMEA 價值會遞減。
- **範圍蔓延**：想涵蓋所有東西的 FMEA 會變得無法管理。先專注在關鍵功能或風險最高的區域。
## 參考資料
- AIAG-VDA AP 矩陣與評分表，見 `references/aiag-vda-ap.md`
- S/O/D 評分標準，見 `references/scoring-rubrics.md`英文原文
---
name: "algo-mfg-fmea"
description: "Conduct FMEA to systematically identify, prioritize, and mitigate potential failure modes. Use this skill when the user needs to assess product or process risks, prioritize corrective actions, or build a risk register — even if they say 'failure mode analysis', 'risk assessment', 'what could go wrong', or 'RPN calculation'."
metadata:
category: "WP-48 製造演算法"
tags: ["manufacturing", "fmea", "risk-analysis", "quality"]
---
# FMEA (Failure Mode and Effects Analysis)
## Overview
FMEA systematically identifies potential failure modes, their effects, causes, and current controls. Each failure is scored on Severity (S), Occurrence (O), and Detection (D) on 1-10 scales. RPN = S × O × D prioritizes which risks to address first. AIAG-VDA FMEA (2019) replaces RPN with Action Priority (AP) matrix.
## When to Use
**Trigger conditions:**
- Designing new products/processes and identifying risks proactively
- Systematically evaluating existing failure modes for prioritization
- Meeting automotive (IATF 16949) or medical device (ISO 13485) quality requirements
**When NOT to use:**
- For root cause analysis of a known problem (use fishbone/5-why)
- For statistical analysis of defect data (use SPC or Pareto)
## Algorithm
```
IRON LAW: Severity Can NEVER Be Reduced by Design Changes
Severity is determined by the EFFECT on the customer. A brake failure
is always severity 10, regardless of how unlikely or detectable it is.
FMEA reduces risk by: lowering Occurrence (better design/process) or
improving Detection (better testing/inspection). NEVER inflate
Detection scores to lower RPN artificially.
```
### Phase 1: Input Validation
Define scope: Design FMEA (DFMEA) or Process FMEA (PFMEA). Assemble cross-functional team. Prepare: process flow diagram or system block diagram.
**Gate:** Scope defined, team assembled, reference diagrams available.
### Phase 2: Core Algorithm
1. List all potential failure modes for each function/process step
2. For each failure mode, identify: effect on customer, root cause(s), current prevention controls, current detection controls
3. Score: Severity (1-10), Occurrence (1-10), Detection (1-10)
4. **Classic RPN:** RPN = S × O × D. Prioritize high RPNs.
5. **AIAG-VDA AP:** Use the S-O-D combination matrix to assign Action Priority: High, Medium, Low.
6. Define recommended actions for High-priority items with responsibility and target dates
### Phase 3: Verification
Review: are all functions/steps covered? Do severity scores match actual customer impact? Are detection scores realistic (not overly optimistic)?
**Gate:** Complete coverage, realistic scoring, actions assigned for high-priority items.
### Phase 4: Output
Return FMEA register with prioritized actions.
## Output Format
```json
{
"fmea_items": [{"failure_mode": "seal leak", "effect": "water damage", "cause": "material degradation", "severity": 8, "occurrence": 4, "detection": 6, "rpn": 192, "ap": "high", "action": "add pressure test at final inspection"}],
"summary": {"total_modes": 45, "high_priority": 8, "medium": 15, "low": 22},
"metadata": {"type": "PFMEA", "scope": "assembly line 3"}
}
```
## Examples
### Sample I/O
**Input:** Coffee machine brewing module, function: "heat water to 93°C"
**Expected:** Failure modes: overheating (S=7, O=3, D=4, RPN=84), under-heating (S=5, O=4, D=3, RPN=60), no heating (S=8, O=2, D=2, RPN=32).
### Edge Cases
| Input | Expected | Why |
|-------|----------|-----|
| S=10, any O and D | Always high priority | Safety-critical failures require action regardless of RPN |
| RPN=100 (S=10,O=1,D=10) vs (S=1,O=10,D=10) | Same RPN, very different risk | This is why AIAG-VDA AP replaces pure RPN |
| No current controls | D=10 (no detection) | Honest assessment drives improvement |
## Gotchas
- **RPN is misleading**: RPN=100 from S=10,O=1,D=10 (catastrophic but rare, undetectable) is very different from S=1,O=10,D=10 (trivial but frequent). AIAG-VDA AP matrix addresses this flaw.
- **Scoring consistency**: Without calibration, different team members score differently. Use scoring rubrics with examples and calibrate as a team.
- **Detection ≠ prevention**: A low Detection score (good detection) doesn't prevent the failure — it only catches it. Prioritize Occurrence reduction over Detection improvement.
- **Living document**: FMEA must be updated when design/process changes, new failure data appears, or corrective actions are implemented. A static FMEA provides diminishing value.
- **Scope creep**: An FMEA that tries to cover everything becomes unmanageable. Focus on the critical functions or highest-risk areas first.
## References
- For AIAG-VDA AP matrix and scoring tables, see `references/aiag-vda-ap.md`
- For S/O/D scoring rubrics, see `references/scoring-rubrics.md`