The Psychology of Quality and More
Failure Mode and Effects Analysis (FMEA): How to understand it
How to understand it
Many problems are caused by systems which fail in unexpected ways, which can result in significant costs. An example of this could be where a new roofing compound is decomposed by acid rain, with the result that the manufacturers have to pay substantial warranty costs, as well as gaining a reputation for poor products. A detailed analysis of the possible ways in which a system might fail, and the possible effects of these failures may thus save significant future costs.
Failure Mode and Effects Analysis (commonly called FMEA) takes the dual step of first finding out how an item can fail, and then finding what effect this failure might have, as in Fig. 1. This second step thus helps to identify the importance of a failure mode, allowing identification of the key failure risks which must be addressed.
Fig. 1. Failure mode and effect
Criticality is a measure of importance that can be applied both to failure modes and to effects, allowing prioritization of remedial actions. A simple way of measuring failure mode criticality is to use the likelihood that it will occur in a given period. The criticality of a failure effect is the likelihood of that effect occurring due to any failure mode (see the illustration). Criticality may be further refined by also taking into account any other items which are considered to be important, such as severity of failure or chance of discovery by a customer.
Taking this extra step is performing Failure Mode, Effects and Criticality Analysis, or FMECA, although this often still referred to simply as FMEA.
Fig. 2. Failure criticality
A limitation of FMEA is that although it goes into a lot of detail about the failure of individual components, it does not take combinations of failures into account. For example, the failure of a lift shaft drive may be inconvenient, but if the braking system has also failed, it could be catastrophic.
As with other numerical methods, figures are best derived from either actual measurement or controlled experiments. If these are not available, then estimates should be treated with appropriate caution.
And the big