FMEA: anticipating failures before they happen

In one line

FMEA (Failure Mode and Effects Analysis), called AMEF in Spanish, is a technique that anticipates how a product or process can fail by scoring every possible failure from 1 to 10 on three axes — Severity, Occurrence and Detection — and multiplying them to decide what to tackle first.

What it is

FMEA was born, like 8D, in the military and manufacturing world of the mid-twentieth century, and ended up consolidating as a standard tool in the automotive industry: the AIAG (Automotive Industry Action Group) FMEA manual is today the reference that automakers and suppliers use to rate and prioritize failure risks, alongside 8D itself — this site has an article dedicated to 8D that explains that shared origin.

There are five types of FMEA (system, design, process, service and software), but they all share the same underlying logic: for each function or part of a product or process, the team asks what could fail, and scores that potential failure on three factors:

• Severity (S): how serious the effect of that failure would be if it did happen — from a defect the customer barely notices (1) to a safety risk or a legal violation (10).
• Occurrence (O): how often the cause that generates that failure shows up — from a remote cause that is almost never seen (1) to a cause that fails almost inevitably (10).
• Detection (D): the factor that gets confused the most. It does not measure how easy the failure is to detect — it measures the probability that the failure will not be detected before it reaches the customer or the next process. That is why 10 means "very hard to detect, it will slip through to the customer without anyone noticing" and 1 means "practically impossible for it to get through without some control catching it first". The worse the current inspection or control system, the higher the Detection score.

Multiplying the three factors gives the RPN (Risk Priority Number) — called NPR in Spanish-language literature:

RPN = Severity × Occurrence × Detection

The result runs from 1 (the best possible case: 1×1×1) to 1,000 (the worst case: 10×10×10). The higher the RPN, the sooner you need to act on that failure mode. FMEA is considered a "living document": it is not filled in once and filed away, but reviewed periodically (the source recommends at least once a year) as the design, the process or the available information changes.

What it is for

It is used to decide, with an objective criterion, which failure mode deserves time and resources first when there are many potential risks and you cannot tackle them all at once. Unlike a reactive tool such as 8D — which starts after the problem has already reached the customer — FMEA is preventive: it is applied earlier, at the design stage of a product or a process, to anticipate failures that have not happened yet. That is why you should start it as early as possible in development (ideally during design conceptualization) and not treat it as a last-minute formality.

How to apply it

1. Form a multidisciplinary team — design, quality, production and maintenance, depending on which type of FMEA you are doing.
2. Define the scope: which product, system or process (and down to what level of detail) is going to be analyzed.
3. For each function or part within that scope, list the potential failure modes — all the ways in which that part could fail to perform its function.
4. For each failure mode, describe the effect (the impact it would have on the customer or the next process) and score the Severity.
5. Identify the most likely cause of that failure mode and score the Occurrence.
6. Write down the current controls that exist to detect the failure before it moves forward (inspections, tests, poka-yokes) and score the Detection based on how good those controls are.
7. Calculate the RPN for each row and sort the table from highest to lowest.
8. Define actions on the highest RPNs first — lower the severity by redesigning, lower the occurrence by eliminating the cause, or improve detection by adding a control — and score again after applying them, to confirm that the RPN went down.

Real example

The source documents the case of a refrigerator factory in Peru that had been using 8D to tackle failures already reported by customers in four imported electrical components (timer, compressor, fan motor and thermostat). As part of the permanent solution, the team decided to replace two of those components (timer and thermostat) with a new electronic module — and before taking it into production, it applied a design FMEA to anticipate what could go wrong with the change, instead of finding out later in the market as had happened with the original components.

One of the rows in that FMEA table analyzes the damper, the part that regulates the flow of cold air between the freezer and the fresh-food compartment:

• Function: keep the temperature of the fresh-food compartment between 0 and 6 °C.
• Potential failure mode: the damper does not adjust the airflow properly and the fresh-food compartment ends up at temperatures that are too high or too low.
• Effect: the product does not control the temperature as it should — the customer perceives it as the refrigerator not cooling well or freezing too much.
• Severity = 8 (direct and noticeable impact for the customer).
• Cause: the factory calibration of the damper does not take into account all the real operating conditions of the product.
• Occurrence = 5 (a cause of medium frequency, neither rare nor almost guaranteed).
• Current controls: the only control is taking samples of the component in the laboratory — 100% of the units are not checked.
• Detection = 8 (precisely because it depends on sampling and not on a full inspection, there is a high probability that a unit with this problem will get through undetected).
• RPN = 8 × 5 × 8 = 320.

That 320 was among the highest values in the whole FMEA table of the study, and the team assigned it a concrete action with an owner and a date: re-evaluate the working parameters of the refrigeration system before releasing the change. That is exactly the function of the RPN — turning a long list of possible risks into a short, ordered list of what to solve first.

Template to use

One row for each potential failure mode. These are the columns and what goes in each one:

Column	What to write there
Function	What that part, component or process step is supposed to do.
Potential failure mode	The specific way in which it could fail to perform that function.
Effect	What consequence that failure would have for the customer, the user or the next process.
Severity (1-10)	How serious that effect is if it were to happen.
Cause	Why that failure mode would happen — the specific technical origin.
Occurrence (1-10)	How often that cause is expected to show up.
Current controls	What inspection, test or control exists today to detect the failure before it moves forward.
Detection (1-10)	How likely it is that this failure will NOT be detected by those controls before it reaches the customer or the next process (10 = hard to detect).
RPN	Severity × Occurrence × Detection. Sort the table from highest to lowest RPN to prioritize actions.

So you do not have to build it by hand, there is an Excel FMEA spreadsheet with the formulas already in place: it uses the same columns as this table, calculates the RPN and the priority, and the new RPN after the action. The values it comes with are examples and need to be deleted.

Calculate your RPN

Enter the three scores for a failure mode of your own and see the RPN and the priority in real time:

Benefits

• It is preventive: it is applied before the failure happens, not after it has already reached the customer (unlike an 8D or an RCA).
• It gives an objective, numeric criterion for prioritizing when there are many potential risks and limited resources to tackle them all.
• It leaves a living document with the team's knowledge about the risks of a product or process, reusable in future similar designs.
• It works for product design as well as for process, service, system or software — the same S × O × D logic adapts to any of them.

Limitations to keep in mind

• The three scores are subjective: they depend on the team's judgment and experience, so two teams analyzing the same failure mode can arrive at different numbers.
• Without historical failure data, the Occurrence score also ends up being an estimate rather than hard data.
• It is a desk analysis: it reduces the risk of overlooking something, but it does not replace real tests and validations on the product or process.
• If the team fills it in once and never reviews it again, it loses its value as a "living document" and becomes outdated as soon as the design or the process changes.

In summary

FMEA scores every possible failure mode on three axes — how serious (Severity), how frequent (Occurrence) and how hard to detect before it reaches the customer (Detection, where 10 is the worst, not the best) — and multiplies them into an RPN that objectively orders where to start acting. It is the preventive counterpart of reactive tools such as 8D: instead of investigating a failure that already happened, it anticipates which ones are most likely and most serious before they occur.