Skip to content
FMEA and Control Plan | Part 4 of 6

Rating Severity, Occurrence and Detection

How to rate the three FMEA criteria from 1 to 10 with your own anchors, with data and without wishful thinking.

Every cause in an FMEA gets three ratings from 1 to 10. Ten is always bad. The numbers look precise and are judgments, so the way a team arrives at them matters.

Three questions

  • Severity: how bad is the effect? It is rated for the worst effect in the chain, whether that falls on your plant, the customer plant or the end user.
  • Occurrence: how often does the cause lead to the failure mode, given the prevention that exists today?
  • Detection: how sure is the check that would find the cause or the failure before the part leaves?

A guide in your own words

If your customer prescribes rating tables, use them. If not, write a guide the whole team understands. The one below is a simplified working aid with five anchors per criterion. A full guide describes all ten levels.

Rating Severity Occurrence Detection
10 A person can be hurt without warning On almost every part No check exists
8 Product loses its main function About once per shift A look at some parts
6 A secondary function is lost About once per week Measurement on samples by plan
3 Some customers notice a blemish About once per year Automatic check, parts locked out
1 No noticeable effect Cannot happen Cause detected before a part is made

Frequencies are easier to judge than rates. The hose assembly cell H2 makes 400 assemblies per day, so once per week is about 1 in 2,000 and once per year is 1 in 92,000.

Anchors

A guide works when the team adds anchors from its own products: one known case per level. Cell H2 agreed these in its first rating session:

  • Severity 10: a fitting comes off under pressure.
  • Severity 8: particles make a valve stick and the machine stops.
  • Occurrence 6: a worn saw blade, about once per week in the sister cell.
  • Detection 8: the operator looks at some parts.
  • Detection 3: a test bench with a sensor that holds failed parts.

With anchors, a new row is rated by comparison. Is it rarer than the worn blade?

Severity belongs to the effect

Severity does not change when you add a check. A hose that comes off is a 10 whether one in a hundred or one in a million does it. Only a change that removes the effect lowers the number.

Occurrence needs data

Occurrence is rated for the cause with its current prevention. Where a comparable process exists, its data sets the number. The sister cell had four leaks at the crimp in a year. Whoever proposes a 2 for the causes behind them has to explain what is different now.

Detection is about certainty

The most common mistake is to rate detection by whether a check exists. The question is how sure it is.

A look by a person is a weak check and deserves a 7 or 8. A gauge used on every part by plan reaches 4 or 5. Ratings of 2 or 3 need an automatic check that keeps the part from moving on.

“The operator would see it” is not a detection control. It has no method, no frequency and no reaction. Rate it 9.

How to run the rating

Rate all severities of one step first, then all occurrences, then all detections. Let everyone show a number at the same time and discuss the outliers. When two people say 3 and 7, they usually mean two different causes. Split the row.

Three rows from cell H2

  • Wrong hose type, similar reels side by side: severity 10, occurrence 3, detection 7.
  • Particles in the bore, cleaning shot skipped: 8, 5 and 10.
  • Wrong quantity in the box, counted by hand: 3, 6 and 5.

Which of these needs action first? The old answer was to multiply the three numbers. The next part shows why that answer was dropped and what replaced it.

Newsletter

One Lean idea.
Every two weeks.