FMEA — Failure Mode and Effects Analysis

FMEA stands for "failure mode and effects analysis"; the French term is AMDEC, which adds criticality. It is a preventive approach: it is applied before production, to deal with risks on paper rather than on the shop floor.

Each line follows the same chain: failure mode, effect for the customer, cause, control in place. The reasoning is worthwhile only if it goes down to the perceived effect — "the dimension is out of tolerance" is not an effect, "the part does not fit at the customer" is one.

Three scores from 1 to 10: severity of the effect, occurrence of the cause, detection — the last one being inverted, since a high score means poor detection. Their product gives the risk priority number (RPN), between 1 and 1000.

The RPN has a known and rarely stated flaw: two lines with the same product can describe unrelated situations. A severity of 10 together with an occurrence of 2 and a detection of 5 gives 100, exactly like a severity of 4, an occurrence of 5 and a detection of 5 — yet the first puts safety at stake and the second does not. Hence the replacement of the RPN, in the 2019 AIAG-VDA joint handbook, by a ranked action priority that looks at severity first.

The practical consequence comes down to one sentence: a high severity is dealt with whatever the value of the product of the three scores.

An FMEA serves to spend money in the right place before knowing where the problem will land. Its value is measured by what it prevents, hence by what will never appear in an indicator — whence its status as first casualty of tight schedules.

The moment to launch it is determined by what can still be changed. A design FMEA run on a frozen design can only propose inspections; a process FMEA run after the line is installed, only routing tweaks. Run early enough, the same analysis leads to a design modification or an error-proofing device built into the equipment — the only actions that remove the failure mode instead of monitoring it.

Four usual triggers: a new product, a new process or piece of equipment, a transfer to another site, and a major customer complaint. The last is the most instructive: it almost always reveals a line missing from the table, or an occurrence scored well below reality. One useful clarification, finally: FMEA does not rank observed causes — that is the role of the Pareto and the Ishikawa.

Process FMEA on an assembly station. Four lines among several dozen, with their severity (S), occurrence (O) and detection (D) scores, and the resulting RPN.

Gasket badly positioned, effect: leak found at the customer. S = 9, O = 2, D = 5, i.e. an RPN of 90.

Wrong screw reference, effect: insufficient mechanical strength. S = 8, O = 3, D = 7, i.e. an RPN of 168.

Scratch on visible face, effect: cosmetic return. S = 4, O = 7, D = 4, i.e. an RPN of 112.

Tightening torque out of range, effect: loosening in service. S = 7, O = 4, D = 6, i.e. an RPN of 168.

Sorted by decreasing RPN, the table proposes the screw and the tightening torque at 168, then the scratch at 112, and finally the gasket at 90: the most serious line comes last, behind a cosmetic defect. That is not an anomaly, it is the normal behaviour of a product of three scores, where a low occurrence is enough to push a high severity down.

Reread starting from severity, the order reverses — gasket, screw, tightening torque, scratch. That is the action-priority logic of the 2019 AIAG-VDA joint handbook, and here it gives an entirely different work plan.

The table also points to the lever. The scratch has an occurrence of 7: it is the frequency that must be treated. The screw has a detection of 7, hence an inspection that does not see much. The gasket carries a severity of 9 that will not move — a severity is reduced only by a design change, never by an additional inspection. The possible action there bears on detection: an error-proofing device brings D from 5 to 2, and the RPN from 90 to 36. The severity, for its part, stays at 9 for the whole life of the product.