ABSTRACT
Failure Mode, Effects & Criticality Analysis (FMECA) is a widely-used tool for system safety and reliability evaluations; one popular approach is outlined by MIL-STD-1629A. MIL-STD-1629A Criticality Analysis requires estimates of the Failure Mode Ratio (α) and the Failure Effect Probabilities (βi) for each failure mode. To maximize impact on the design, FMECAs are initiated early in the Product Development Process (PDP), before reliability data is available for the new product. Typically, initial criticality estimates are derived via a combination of failure mode/effect data from similar legacy products and from engineering judgment. Later in the PDP, actual data on the new product becomes available. Such a situation is ideal for employing Bayesian methods. The data from which the Criticality parameters are estimated is Multinomial. The Dirichlet distribution is chosen to represent prior uncertainty in the Criticality parameters, thus taking advantage of the Bayesian conjugacy property for the Dirichlet-Multinomial pairing. This paper describes the application of Bayesian techniques to FMECA Criticality analysis, the structure of a Bayes-enabled FMECA, briefly outlines some expert elicitation approaches for developing prior distributions for αi and βi, and demonstrates the use of evidence in the form of field data to update those prior estimates.
