Risk-based quality planning

Problem class

The 2019 AIAG-VDA FMEA Handbook eliminated the Risk Priority Number (RPN) and replaced it with Action Priority (AP) — a change that fundamentally altered how the automotive and adjacent industries prioritize quality risks. The old RPN had well-documented flaws: an airbag failure (S=10, O=2, D=2, RPN=40) would rank lower than a cosmetic scratch (S=3, O=7, D=7, RPN=147). FMEA as a checkbox exercise — done once during development, never updated — remains the dominant failure mode.

Mechanism

Systematic identification, evaluation, and mitigation of potential failure modes through DFMEA (Design), PFMEA (Process), and FMEA-MSR (Monitoring & System Response). Also encompasses HACCP hazard analysis (food), ICH Q9 quality risk management (pharma), and ISO 14971 risk management (medical devices).

Why RPN was eliminated. The new AP system uses a decision table yielding High/Medium/Low priority where severity is the dominant factor — severity 10 (safety risk) yields AP=High even if Occurrence=2 and Detection=2. Approximately 1,000 S×O×D combinations map to H/M/L categories. High AP means mandatory corrective action (MUST); Medium means SHOULD; Low means COULD.

ICH Q9(R1) revision (2023). The pharma risk management standard was substantially revised with an 850% increase in mentions of "formal/formality" (from 4 to 34). Key additions include a dedicated section addressing subjectivity and bias in risk assessments, three defined decision-making approaches (highly structured, less structured, rule-based), a new subsection on supply chain/product availability risks, and clarification on applying risk management with emerging digital technologies. Adopted by FDA (May 2023), EMA (effective July 2023), and Swissmedic.

AI transformation — the most technically advanced AI application in this tech tree. An XGBoost model integrated with FMEA achieved R² = 0.985 and MAPE = 2.84% on historical failure mode data, with SHAP analysis confirming severity as the most influential parameter. Deep learning models on aviation operational data achieved ~95% fault prediction accuracy with dynamic risk evaluation replacing static expert estimates. A Random Committee ML-FMEA improved correct classification from 77.47% to 90.09% and AUC-ROC from 80.9% to 91.8%. Cambridge University Press published an LLM-integrated FMEA framework for automated failure mode identification, risk analysis, and CAPA recommendation. A 2026 Springer paper demonstrated RAG combined with open-source LLMs (DeepSeek R1, Gemma 2, Mistral Saba) for automated FMEA table generation from industry standards. One manufacturer trained ChatGPT on existing FMEA data and reported "significantly reduced time and cost of FMEA process while increasing accuracy and thoroughness."

Required inputs

• Document Control
• Process knowledge (process flow diagrams)
• Historical quality data (complaints, field failures, warranty data)

Produced outputs

• DFMEA, PFMEA, and FMEA-MSR documents with Action Priority ratings
• Control Plans linked to PFMEA results
• Inspection planning driven by risk analysis
• Design verification/validation planning inputs
• CAPA prioritization criteria
• Process validation scope and risk-based testing plans

Industries where this is standard

• Automotive (AIAG-VDA FMEA mandatory; 2019 handbook is current standard)
• Medical devices (ISO 14971 risk management — harmonized standard for EU MDR)
• Pharmaceuticals (ICH Q9(R1) — revised 2023)
• Food safety (HACCP hazard analysis — conceptually identical to PFMEA)
• Aerospace (AS9100D Clause 6.1)

Counterexamples

• FMEA as a checkbox exercise — done once during development, never updated.
• Copy-paste FMEAs from similar products without process-specific analysis.
• Not connecting FMEA results to Control Plans and inspection planning.
• Overestimating detection effectiveness — "manual checks considered 'strong' when they are unreliable."
• Not documenting risk acceptance rationale for Medium/Low AP items.

Representative implementations

• AIAG-VDA FMEA Handbook (2019) — joint automotive industry standard replacing separate AIAG and VDA approaches; used by all major OEMs and tier 1/2 suppliers globally.
• XGBoost + SHAP FMEA integration — R² = 0.985 on historical failure mode prediction.
• Random Committee ML-FMEA — improved correct classification to 90.09%, AUC-ROC to 91.8%.
• RAG + LLM FMEA (DeepSeek R1, Gemma 2, Mistral Saba) — automated FMEA table generation from industry standards, published in Springer 2026.
• ICH Q9(R1) adoption — FDA (May 2023), EMA (July 2023), Swissmedic.

Common tooling categories

FMEA software (standalone or QMS-integrated), risk management platforms, ML/causal AI tools for failure prediction, LLM-assisted FMEA generation tools, control plan management systems.

Regulatory anchors

ISO 9001:2015 Clause 6.1, IATF 16949 Clause 8.3.5.1, AS9100D Clause 6.1, ISO 14971 (medical devices), ICH Q9(R1) (pharma, revised 2023), AIAG-VDA FMEA Handbook (2019), FDA 21 CFR 820.30(g).