PFMEA Analysis for Machining Process
This guide shows engineers and buyers how to run a PFMEA analysis for machining process work: which operations to walk, how to score severity, occurrence and detection, and what controls to put on the shop floor. Read it before you release a drawing to a machine shop.
In this article
- 1
- 2
- 3
- 4
- 5
- 6
- 7
What matters most
What PFMEA analysis for machining process work actually covers
A PFMEA is a structured walk through a process, one operation at a time. For each operation you ask three questions: what could go wrong, what happens to the part if it does, and how would we catch it. Then you score how bad the effect is, how often the cause shows up, and how well the current detection catches it. The three numbers combine into a risk priority number, and the highest numbers get action first.
On a machining process the walk is short but the failure modes are specific. A roughing pass can leave chatter marks that a finish pass cannot remove. A deep pocket can pull chips back into the cut and break a 6 mm end mill. A reamed hole can come out oversize because the reamer ran dry. PFMEA analysis for machining process work is about naming those before the first workpiece is clamped.
It is not a quality department form. The people who write it are the process engineer, the programmer and the machinist who will run the job. If those three are not in the room, the severity scores are guesses.
The output is two documents. The PFMEA table lists the risks and the actions taken. The control plan lists what the operator checks, how often, and with what gauge. One without the other does not survive an audit or a first article.
- 1ScopeOne part number, one routing, one revision of the drawing.
- 2InputsDrawing, tolerance stack, routing, fixture concept, material spec.
- 3OutputsRanked risk table plus a control plan with gauges and frequency.
- 4OwnerProcess engineer, with the programmer and lead machinist signing.
Severity, occurrence and detection on real cuts
Severity is about the effect on the customer or the next operation. A scratch on a cosmetic face is a 2 or 3. A press-fit bore that loses 0.03 mm of interference and loosens in service is an 8 or 9. For safety-related features in automotive or medical work, severity stays high no matter how unlikely the cause is. You do not average that away.
Occurrence is about the cause, and it should come from data you already have. If your shop has run 4,000 aluminium brackets on the same fixture with two scrapped for chatter, that is a low occurrence score. A first-run titanium part on a new fixture with no history gets a cautious score until you have 30 pieces of data.
Detection is the score teams argue about most. A vernier caliper on a 0.02 mm true position callout is weak detection, because the gauge cannot resolve the requirement. An in-process probe that checks the feature after the finish pass is strong detection. A CMM check at final inspection sits in the middle: it finds the problem, but only after the whole lot is cut.
Multiply the three. A high severity times a high occurrence times weak detection produces the number that stops the job. When severity is 9 or 10, work the occurrence and detection scores down even if the product stays moderate. Severity is the one number you cannot fix with a better gauge.
- 1Anchor severity to the drawingMatch it to the tolerance or the function, not to how hard the cut is.
- 2Use your own historyOccurrence scores from your scrap data beat generic tables.
- 3Ask what the gauge can seeIf the gauge cannot resolve the tolerance, detection is weak.
- 4Never average severity downA rare but severe failure still needs a control.
The failure modes that repeat in CNC work
Dimensional drift shows up as a bore that walks out of tolerance over a run. The usual causes are thermal growth in the spindle, tool wear past the offset window, and fixture clamp pressure that changes after the first ten parts. Detection is a probe check or a timed operator check with a bore gauge.
Surface defects are chatter, cutter marks, burn marks and scratches. Chatter usually traces to tool overhang, weak workholding or a speed that sits on a natural frequency. On a deep pocket with a 6 mm cutter, keep the length-to-diameter ratio under 4 where the geometry allows and reduce radial engagement instead of pushing feed.
Tool failure is the most expensive mode because it damages the part and the fixture. Small taps and long end mills break first. Countersink and tap depth monitoring, load monitoring on the spindle, and a tool life counter tied to the material all help. A broken 3 mm tap in a finished part usually scraps it.
Geometric errors come from fixture location. A part located on a rough cast surface moves when the clamps tighten. The fix is a machined datum in the first operation, then everything downstream locates on that datum. Burrs and edge condition matter too: a burr on a sealing face becomes a leak the customer finds, not you.
- 1DimensionalBore drift, step mismatch, flatness loss after unclamping.
- 2SurfaceChatter, cutter marks, burn, scratches on cosmetic faces.
- 3ToolTap breakage, end mill snap, insert chipping on interrupted cuts.
- 4GeometricLocating errors, datum shift, burrs left on functional edges.
How the PFMEA links to the quality system
The PFMEA is one node in a chain. The design FMEA sets the critical characteristics. The process PFMEA turns those into operation-level risks. The control plan turns the risks into checks the operator performs. Work instructions then describe the check in the words the operator uses. If any link is missing, the chain breaks at the machine.
For automotive work under IATF 16949:2016, the PFMEA and control plan are reviewed together during the audit, and the auditor will trace a critical characteristic from the drawing to a control plan line to a record. For medical work under ISO 13485:2016, the same logic applies with more emphasis on process validation and traceability. Shops holding both, plus ISO 9001:2015 and ISO 27001:2022, usually keep these documents in one revision-controlled system.
There is a practical test. Pick a critical feature on the part. Ask the machinist which gauge they use and how often. Ask to see the last three records. If the answer matches the control plan and the records exist, the system is working. If the machinist shrugs, the PFMEA is paperwork.
Keep the language plain. A control plan line that says verify conformance means nothing at 2 a.m. on a night shift. Verification method, sample size, frequency, and the reaction plan when it fails: write all four.
- 1Trace one feature end to endDrawing to PFMEA to control plan to record. If it breaks, fix the chain.
- 2Write the reaction planSay what happens to parts already cut when a check fails.
- 3Keep revisions tiedA drawing change without a PFMEA update is a gap an auditor will find.
Six steps to build the PFMEA
Work in this order. Each step feeds the next.
- 1Fix the scope and the routingWrite down the part number, drawing revision, material and full operation list from raw stock to final inspection. Split operations where the setup changes. A 4,000 mm rail and a 40 mm bracket do not share a risk profile, so do not share a sheet.
- 2Walk each operation and list failure modesFor every operation, ask what the feature is and how it could miss. Cover dimension, surface, tool and geometric modes. Expect 3 to 6 modes per operation on a complex part and 1 to 2 on a simple one.
- 3Score severity first, from the drawingRead the tolerance and the function. A bearing bore at Ø40 H7 with a 0.025 mm band scores higher than a clearance hole at ±0.2 mm. Record the score with a one-line reason so the next reviewer understands it.
- 4Score occurrence from your dataPull the last 12 months of scrap and rework for similar parts. If the shop has no history, score conservatively and flag the item for a first-article review with a full dimensional report.
- 5Score detection and name the gaugeWrite the actual gauge, not the word inspection. A 0–25 mm micrometer at 0.001 mm resolution on a 0.010 mm band is weak. A CMM with a 0.002 mm stated accuracy on the same feature is stronger. In-process probing after the finish pass is strongest.
- 6Rank, act, and write the control planSort by the product of the three scores. For anything in the top group, add a control: change the cutter, add a spring pass, add a probe check, or tighten the tool life limit. Then move each control into the control plan with a frequency and an owner.
- 7Re-score after the first articleRun the first article, measure the critical features, and update the occurrence and detection scores with what you learned. A first run on a new fixture almost always changes at least one line.
- 8Set a review triggerReopen the PFMEA when material or supplier changes, when a fixture is rebuilt, when a tool geometry changes, or when scrap on that feature crosses your internal limit. A fixed annual review alone is too slow.
Which control fits which risk
Match the control to the failure mode instead of adding inspection everywhere.
| Risk | Weak control | Stronger control | When it pays off |
|---|---|---|---|
| Bore drift over a run | Operator check every 20 parts | Probe check after finish pass | Runs above 50 parts, tight band under 0.03 mm |
| Chatter in a deep pocket | Listen and adjust speed | Shorter cutter, reduced radial engagement | Pocket depth over 3× cutter diameter |
| Tap breakage in blind holes | Visual check after tapping | Torque monitoring plus depth control | Taps under M4, stainless or titanium |
| Datum shift on castings | Clamp and measure after | Machined datum in operation one | Rough or as-cast locating surfaces |
| Burrs on sealing faces | Deburr by hand | Controlled chamfer cycle in the program | Leak-tight faces and medical parts |
| Cosmetic scratches | Final visual inspection | Protective film plus dedicated trays | Visible Class A surfaces |
Start the PFMEA before the first chip
Pick the three features that decide whether the part works, score them honestly, and put a gauge and a frequency behind each one. That is the whole method. Everything else is paperwork that supports it.
Questions engineers ask
How long should a PFMEA take for a machined part?
For a part with 8 to 12 operations and a few critical features, expect a half day to a full day for the first draft with the process engineer, programmer and lead machinist in the room.
Revisions after the first article take under an hour. The first draft is the slow part because you are pulling scrap history and fixture concepts together.
Do I need a PFMEA for a one-off prototype?
Not a full one. For a single prototype, a short risk list covering the three or four features that decide whether the part works is enough, plus a first-article dimensional report.
If the prototype is the front end of a production program, start the full PFMEA at the same time so the production controls inherit what you learned.
What risk priority number should trigger action?
Most teams set a threshold from their own history rather than a fixed number. A common rule is to act on any item whose score sits in the top 20 percent of the sheet, plus any item with severity 9 or 10 regardless of score.
The threshold matters less than the discipline. If nothing ever crosses it, the scoring is too generous.
How does detection scoring change with in-process probing?
In-process probing on a 5-axis machine checks the feature before the part leaves the fixture. That moves the detection score down sharply because you catch the error while the setup is still live.
The catch is program time. Probing adds cycle time, so reserve it for critical features rather than every dimension.
Can a supplier run the PFMEA without the customer's drawing stack?
Partially. A shop can score machining-level risks such as tool breakage and chatter from the routing alone. It cannot score severity properly without the function and the tolerance stack.
Send the drawing, the critical characteristic list and any known service loads. That is what turns a generic sheet into a useful one.
What documents should I ask for with a first article?
Ask for the dimensional report, the material certificate, the control plan line for each critical feature, and the inspection record showing the checks were done.
GreatLight provides inspection reports on request, with 100% inspection before shipment covering raw material check, in-process monitoring and final inspection.
Send the drawing and get a risk review with the quote
Upload your part files and we will return a quotation with free DFM analysis within 12 hours, including the machining risks we see on your critical features.
12-hour quote100% inspectionNDA on request
More machining notes from our shop floor
We publish setup notes, tooling trials and inspection data from the factory floor.
