Identifying a failure mode and changing the design before it happens.
All notes

Operational excellence

FMEA step by step: find what could go wrong before it does

Work through a process FMEA for order fulfilment. Rate severity, occurrence and detection, then see what a risk score can hide and where the action-priority lookup fits.

Teams use failure mode and effects analysis (FMEA) to examine how a process could fail, what the effects would be and which controls address the causes or detect the failures. They record that reasoning before choosing actions and revise it as they learn about the process.

The food-distribution example below shows why the team must examine severity alongside any total score. Its highest-severity failure has the lowest risk priority number (RPN).

The vocabulary, once

Use separate fields for the failure, its effect, its cause and the current control:

  • Failure mode How the step fails. "Wrong item picked."
  • Effect What the customer or next step experiences. "Customer receives the wrong product."
  • Cause Why the failure mode happens. "Similar items stored side by side."
  • Control What currently prevents the cause or detects the failure. "Visual check at packing."

A simple test: effects are about the customer, modes are about the step, causes are about the conditions.

The seven steps

The 2019 AIAG and VDA handbook, now the reference for automotive and widely used elsewhere, organises FMEA into seven steps:

Step What you do
1. Planning and preparation Scope, team, timing, what's in and out
2. Structure analysis Break the process into steps (and, for each, the people, machines, materials and environment involved)
3. Function analysis State what each step is supposed to achieve
4. Failure analysis Link failure modes, their effects and their causes
5. Risk analysis Record current controls; rate severity, occurrence and detection
6. Optimisation Decide actions, owners and dates; re-rate after they're done
7. Results documentation Record and communicate what was found and changed

Agree the scope and intended functions before rating failures. Otherwise, the team may assign scores to different understandings of what the process must do.

The example: order fulfilment at a food distributor

In this fictional example, a distributor picks, labels, packs and ships mixed food orders to shops. Some products contain allergens. The team records these lines in its process FMEA:

Step Failure mode Effect S Cause O Current control D RPN
Pick Wrong item picked Shop receives the wrong product 5 Similar items stored side by side 6 Visual check at packing 5 150
Pack Damaged in transit Refund and re-shipment 6 Poor palletising 4 Customer complaint 8 192
Dispatch Late dispatch Missed delivery slot 4 Wave planning overload 6 End-of-day report 4 96
Label Batch code missing Traceability lost in a recall 8 Printer ribbon runs out 2 Operator glance 5 80
Pack Allergen product in the wrong carton Allergic reaction for a consumer 10 Similar outer packaging 2 Scan at packing 2 40

Rating: keep the scales concrete

Every organisation tunes its own scales, but they should describe observable things, not feelings. A simplified version:

Score Severity (of the effect) Occurrence (of the cause) Detection (by current controls)
9 to 10 Safety or legal consequence Happens very often, or no prevention in place Almost certainly won't be caught before the customer
7 to 8 Customer seriously affected, product unusable Happens often Caught only by chance or by the customer
4 to 6 Customer noticeably affected Happens occasionally Caught by a manual check
2 to 3 Minor annoyance Rare, with good prevention Caught by an automatic check
1 No noticeable effect Practically impossible Failure is prevented or always caught

Use the applicable rating tables and record the evidence supporting each score. Rate severity for the effect, occurrence for the cause, and detection for the relevant current control. Resolve a disagreement if it could change the action priority.

Under this teaching scale, the customer complaint receives a detection score of 8 because the failure has already reached the customer. The operator's glance receives 5 as a manual check. These are illustrative choices within the stated bands, not official AIAG and VDA ratings.

The trap: ranking by RPN

The traditional way to prioritise is the risk priority number: RPN = S × O × D. Sort by RPN, work from the top. In the example, that puts the allergen line last.

RPN = S × O × D S = severity Damaged in transit 192 S 6 Wrong item picked 150 S 5 Late dispatch 96 S 4 Label missing batch code 80 S 8 Allergen pack in wrong carton 40 S 10
The allergen failure has the highest assigned severity, 10, but the lowest RPN, 40. Review that severity directly before using the score to choose actions.

This isn't a quirk of the example. RPN has well-known flaws:

  • It treats the three factors as equal. A severity of 10 counts no more than an occurrence of 10. But a rare catastrophe and a frequent nuisance are not the same risk.
  • The scores are rankings, not measurements. A detection of 6 isn't "twice as bad" as a 3. Their product does not measure the probability or cost of the effect.
  • Very different risks get the same RPN. 10 × 2 × 2 and 2 × 5 × 4 both give 40.
  • It invites gaming. Teams learn that lowering any factor lowers the RPN, and choose the easiest one, often detection.

Reliability researchers have pointed this out for over twenty years. One well-cited review called RPN prioritisation seriously flawed from a technical perspective.

How the action priority lookup works

The AIAG and VDA handbook uses action priority (AP) tables to assign high, medium or low priority to combinations of severity, occurrence and detection. Two failures can have the same RPN and different combinations of ratings. Use the handbook table to look up each combination and check what action it calls for. The example shows what the RPN ranking can hide without assigning official AP categories.

For each high-severity effect, review the evidence behind prevention and detection. A low occurrence or detection score needs support from the actual controls and operating conditions.

The fictional team checks the allergen line's low occurrence and detection ratings. The scan works only if every carton passes through it, and the night shift sometimes skips it under pressure. The team therefore reviews those scores and prioritises the allergen control. It also reviews wrong picks and transit damage against their effects and current controls. Record the safety decision and its evidence separately from the RPN ranking.

Optimisation: the actions

Good actions follow the same order as the logic: remove the cause, then prevent it, then detect it.

Failure mode Action Owner and date Proposed S / O / D, subject to verification
Allergen product in the wrong carton Store allergen products in a separate zone; the scanner blocks packing if an allergen item goes into a non-allergen order Warehouse manager, end of Q1 10 / 1 / 1
Wrong item picked Scan-to-pick on every line; separate look-alike items Operations lead, end of Q2 5 / 3 / 2
Damaged in transit Palletising standard with a photo check; stretch-wrap settings fixed Shipping supervisor, next month 6 / 2 / 4

For the same effect, retain its severity while checking whether the action changes occurrence or detection. Test the scanner's block with a wrong item, a missed scan and the recovery procedure before accepting the proposed ratings.

Mistakes that turn FMEA into paperwork

  1. Doing it alone. One engineer filling in a template misses most of the failure modes. The people who do the work know where it breaks.
  2. Doing it once. An FMEA should change whenever the process does, and whenever a failure happens that it didn't predict.
  3. Scores without evidence. Use the agreed scales and investigate differences that could change the required action.
  4. Actions without owners and dates. "Improve training" isn't an action. "New pick-path training for night shift, owned by Ana, by 15 March" is.
  5. Stopping at detection. Check whether the team can prevent the cause. Compare the control's coverage, operating burden and response to a failure before selecting it.

Where FMEA fits

After a failure, update the FMEA with the cause evidence and any control that failed to work as expected. Carry accepted changes into the control plan from the DMAIC project, with an owner and a check that the control still works in use.

The same thinking, for projectsProjects have failure modes too. Here's how to build a risk register that actually drives decisions.

FMEA Explained: How to Stop a Failure Before It Happens (Worked Example) A furniture factory prepares to make a new wooden chair. A different process and different ratings from the order fulfilment example above.

Sources

The distributor, its process and ratings are illustrative.

NextHoshin Kanri vs OKRs: turning strategy into what teams do on Monday

Discuss an operations problem

I work with operations and transformation teams on operational excellence, digital transformation programs, supply chains and industrial AI. Send me the problem you’re working on and the result you need.

Abolfazl ShirkavandImage preview unavailable