QualityMethod

AI drafts and checks PFMEA / DFMEA

Added

Role

quality / process / design engineer

The problem

FMEAs are slow to build and easy to get wrong. Teams copy old spreadsheets, operation numbers drift from the process flow and control plan, severity ratings go missing, and design and process FMEAs are not linked, so risk analysis is incomplete and launches slip.

How AI helps

An LLM drafts functions, failure modes, effects and causes from a process flow diagram or design specs, suggests Severity / Occurrence / Detection ratings for the team to challenge, checks the PFMEA against the control plan (operation numbers, prevention and detection controls), and flags gaps against the AIAG-VDA 7-step structure (e.g. missing severity, causes placed at the wrong level, actions with no owner).

Step by step

1. Gather inputs: process flow diagram (operation numbers and steps), drawing/key characteristics, current control plan, past 8Ds/NCRs and any existing master FMEA.
2. Ask the AI to structure the analysis (AIAG-VDA steps 2–3): process items, process steps, work elements (4M) and their functions/requirements.
3. Have it draft failure modes, effects (at plant, customer and end user) and causes for each step (step 4).
4. Ask for suggested S/O/D ratings with a one-line justification each, labelled as suggestions (step 5).
5. Have it cross-check the draft against your control plan: matching operation numbers, a prevention and detection control for every cause, nothing in the control plan missing from the FMEA.
6. Ask it to flag gaps: missing severity, high-severity items with weak detection, actions with no owner or date, DFMEA characteristics not carried into the PFMEA.
7. Run the team review: challenge every line, set the final ratings and Action Priority (AP), assign actions (step 6).
8. Record the final FMEA in your system and keep it alive with 8D and change inputs (step 7).

Tools

• General LLM (ChatGPT, Claude, Copilot, Gemini) for drafting
• Spreadsheet or FMEA software for the final document
• Specialist FMEA AI tools exist, e.g. Omnex Systems FMEA AI Agent (vendor case)

Human review: cautions

• The FMEA stays a cross-functional team exercise: AI output is a first draft to challenge, never a finished FMEA.
• Never accept AI S/O/D ratings as-is. Severity comes from the customer/effect, Occurrence from your own process data, Detection from your actual controls.
• Check every failure mode against the real line (gemba) and your lessons learned; an LLM does not know your machines.
• Customer-specific requirements (CSRs) and safety/regulatory effects must be reviewed by a qualified engineer.
• Do not paste confidential customer drawings into public AI tools; use an approved, private tool.

You are a senior quality engineer expert in the AIAG-VDA FMEA handbook (2019). Draft a PFMEA for review by our cross-functional team.

Part / product: [e.g. machined aluminum housing, part no.]
Customer and critical/special characteristics: [list, with drawing tolerances]
Process flow (operation number, step, machine/equipment): 
[paste, e.g. 10 Receive bar stock / 20 CNC turning / 30 Deburr / 40 Wash / 50 Final inspection / 60 Pack]
Current control plan: [paste or 'none']
Known past issues (8Ds, NCRs, scrap reasons): [paste]

Do this:
1. Structure analysis: for each operation, list the work elements (Man, Machine, Material, Method, Environment, Measurement).
2. Function analysis: the function/requirement of each step and work element, tied to the product characteristics.
3. Failure analysis: a table with columns Op no. | Process step | Failure mode | Failure effect (plant / customer / end user) | Failure cause (work element) | Current prevention control | Current detection control | Suggested S | Suggested O | Suggested D | Action Priority (H/M/L) | Recommended action.
4. Give a one-line justification for each suggested S, O and D rating, and mark them clearly as SUGGESTIONS for the team to confirm.
5. Check against the control plan: list operation-number mismatches, causes with no prevention or detection control, and control-plan items missing from the PFMEA.
6. Gap list: missing severities, high severity with weak detection (D 7+), special characteristics without a dedicated control, and open questions for the team.

Rules: use only the information I give you plus general engineering knowledge, and say which is which. Never invent measured data, Cpk values or occurrence history; if something is missing, list it as a question. Output is a draft for team review: do not change anything in any system.

Sources

• https://www.omnexsystems.com/case-study/fmea-ai-agent-for-a-global-manufacturer