FMEA: the full method, from scoring to a worked example
FMEA step by step: the 3 types, Severity, Occurrence and Detection scoring, the RPN calculation and a worked example, with its honest limits.
FMEA is a preventive analysis method that consists of systematically reviewing everything that could go wrong on a product, a process or a piece of equipment, before it happens. The name says exactly what it does: it analyses the ways a function can fail (the failure modes), what those failures cause (the effects), and how much each one should worry you (the criticality). It is known in French industry as AMDEC.
The essentials
FMEA is not a form to fill in, it is a structured discussion between people who know the installation. Its value rests on one simple indicator, the RPN = Severity x Occurrence x Detection, which lets you rank dozens of failure modes. But this indicator has a known weakness: by multiplying three scores together, it can let a rare but serious risk slip through unnoticed. A useful FMEA is reviewed regularly and corrects this limitation by hand.
What FMEA does, and what it does not
FMEA answers one question: for each function of a system, in what ways can it stop being fulfilled, what are the consequences, and how much should you worry about it?
The result is not a prediction. It is a ranking: a list of failure modes ordered from the most worrying to the least, together with the actions to launch first. FMEA does not tell you when a failure will occur. It tells you where to concentrate the prevention effort for a given budget.
It is also teamwork. An FMEA completed by a single person at their desk mostly reproduces that person's blind spots. The method is only worth anything through the confrontation of viewpoints from maintenance, production, quality and sometimes the supplier.
The three types of FMEA
There are three uses, depending on what is being analysed. The scoring method stays the same, but the failure modes and the scales change.
- Design FMEA (product): it applies to the design of a product or a component. It looks for the failures that would affect its function for the end customer. This is the FMEA of design offices.
- Process FMEA: it analyses a manufacturing process. It looks for what, within a step (dosing, welding, filling, heat treatment), can produce a nonconformity. This is the one that prepares quality audits and feeds the handling of recurring nonconformities.
- Machine FMEA (equipment): it covers a piece of production equipment. It looks for the failure modes that would cause a stoppage, a drift or a hazard. This is the FMEA of maintenance and reliability.
The same site may run all three, on different scopes. In maintenance, it is the machine FMEA that structures the approach.
The method step by step
Break the system down into functions
You do not start from the equipment but from what it has to do. A conveyor moves, positions, regulates a speed. Each function will be analysed separately. This breakdown avoids missing failures that are not visible on the mechanical drawing.
List the failure modes
For each function, you enumerate the ways it can stop being delivered: complete stoppage, degradation, unintended operation, drift out of tolerance. A failure mode describes a state, not a cause.
Describe the effects
What happens downstream if this mode occurs? Line stoppage, scrap, a risk to an operator, a nonconformity delivered to the customer. The effect drives the severity score.
Search for the causes
For each mode, you trace back to the plausible root causes: wear, lack of lubrication, a setting error, corrosion. A single mode often has several causes, to be handled as separate lines.
List the current detection means
How, today, is the problem noticed before it produces its effect? An inspection round, a sensor, an alarm, a quality check. The absence of any detection means is a critical piece of information, often the one with the heaviest consequences.
Score Severity, Occurrence and Detection
Each line gets three scores on a scale of 1 to 10. S measures the severity of the effect (1 = negligible, 10 = hazard or major stoppage). O measures how often the cause appears (1 = very rare, 10 = almost permanent). D measures the difficulty of detection (1 = certain to be caught before the effect, 10 = undetectable before the failure).
Calculate the RPN
The Risk Priority Number is the product of the three scores: RPN = S x O x D. It ranges from 1 to 1000. It has no physical meaning, it is simply a sorting tool.
Rank
You sort the lines by descending RPN. The highest values point to the modes to tackle first. You usually set a threshold above which action is mandatory.
Define the actions
For each priority line, you decide on an action that works on O (make the cause more reliable) or on D (improve detection). Acting on S almost always means a redesign: this is rarer and more expensive.
Re-score after action
Once the action is done, you re-score to check that the RPN has actually dropped. An FMEA without re-scoring never proves that the actions were worth anything.
A worked example
Bearing seizure on a bagging conveyor
Take the machine FMEA of a food-processing line. Function analysed: drive the belt of the bagging conveyor.
Failure mode: seizure of the drive motor bearing. Effect: complete stoppage of the bagging line for several hours. Cause: lack of lubrication. Current detection: none. The failure is noticed at the moment of the stoppage.
Initial scoring: Severity 7 (long line stoppage, no safety stake), Occurrence 6 (two seizures in three years), Detection 8 (no way to anticipate). The RPN is therefore 7 x 6 x 8 = 336, a high value that puts this mode at the top of the list.
Action decided: install temperature and vibration monitoring on the bearing housing, and make the lubrication round more reliable.
Re-scoring: severity stays at 7, because the consequence of a seizure has not changed. Occurrence drops to 4 thanks to the more reliable lubrication, and detection to 3 thanks to the early measurement. The recalculated RPN is 7 x 4 x 3 = 84. The mode leaves the priority zone, and the reasoning stays traced for the next review.
Where FMEA sits in the reliability approach
FMEA comes in upstream. It is the one that produces the fine-grained knowledge of failure modes and their criticality, before other tools make use of it.
On the output side, its results feed the equipment criticality matrix, which aggregates this information at fleet level to rank equipment against each other. They then feed risk-based maintenance, which decides on intervals and campaigns from that ranking. Moving from analysis to operational action is detailed in prioritising maintenance on real risk.
Seen this way, FMEA is not a standalone deliverable: it is the analytical building block that makes the trade-offs taken afterwards defensible.
The limits you need to know
- The RPN aggregates and can hide a serious but rare risk.A mode scored S 10, O 1, D 1 gives an RPN of 10, very low, even though it is a potentially fatal hazard. Always correct the ranking by hand: any high severity deserves a review, whatever its RPN.
- The scales are subjective.Two teams will score the same mode differently. Without written definitions for each level, the RPN mostly measures the mood of the group on the day of the meeting.
- An RPN does not compare from one FMEA to another.The scales differ from one study to the next. An RPN of 200 here is not the same as an RPN of 200 elsewhere.
- It is teamwork, not a form.An FMEA completed alone reproduces the blind spots of a single person. The value comes from the confrontation of viewpoints.
- An FMEA that is never reviewed is worthless.It describes the state of knowledge at a given date. Without re-scoring after action or periodic revision, it becomes an archive document.
FAQ
What is the difference between design FMEA and process FMEA?
Design FMEA analyses the design of an object and its failures with respect to the end customer. Process FMEA analyses a manufacturing process and looks for what, within a step, can generate a nonconformity. The first belongs to the design office, the second to quality and production.
What is the RPN really worth?
It is a good sorting tool, not an absolute measure of risk. It lets you compare modes within a single study and bring out priorities. It should never be used on its own: a high severity forces action even when the RPN stays low.
How often should an FMEA be revised?
At every change to the process or the equipment, after any significant incident, and failing that at a periodic review, often annual. An FMEA frozen for several years describes an installation that no longer exists.
Who should take part in an FMEA?
At a minimum maintenance, production and quality, supplemented as needed by safety, methods or the supplier. A facilitator guarantees the rigour of the scoring. An FMEA carried by a single function is systematically challenged at the first trade-off.
Should FMEA come before or after the criticality matrix?
FMEA comes upstream: it produces the fine-grained analysis of failure modes that then feeds the criticality matrix at fleet level. In practice, the two approaches feed each other over successive revisions.
Sources
Sources and references
IEC 60812:2018, Failure modes and effects analysis (FMEA and FMECA): webstore.iec.ch/publication/26359
Written by Adama CamaraAI Consultant · Industry · view profile
Published on August 8, 2026
Maintenance strategy
Industrial equipment criticality: building a matrix that earns its keep
How to rate severity and probability on a real plant, and why the probability axis is what makes most criticality matrices fail within two years.
Maintenance strategy
Preventive maintenance plan: building it from real risk
How to build a preventive maintenance plan from equipment criticality and real condition, instead of inherited intervals that no one questions any more.
Maintenance strategy
Preventive, predictive and condition-based maintenance: differences and examples
What these three maintenance policies actually cover, a concrete example of each, the data they demand and how to choose the right one.
Quality and compliance
Why the same nonconformities keep coming back, and what AI changes
Why the same nonconformities keep coming back despite 8D and CAPA, and how AI surfaces the recurrences that case-by-case handling never sees.