Failure analysis with the aid of bayesian networks

In this paper, a model for failure analysis using the theory of Bayesian Belief Networks (BBN), will be presented. For the initial knowledge base, information about cause-effect-relations is used to build up the network. These relations have been acquired by the application of a slightly modified failure mode and effects analysis (FMEA) in earlier stages of product design. The known inference strategies, which operate on these networks, make use of the conditional probability tables which are attached to the edges of the network. Since these numbers are not available, an alternative inference mechanism was developed. This approach is based on the principle of stochastic simulation, but uses certainty factors instead of conditional probabilities. The method yields good estimates for the exact probability values, which are used to identify the real origins, responsible for an occurred malfunction. Due to the learning capabilities of BBN, the model is able to progressively adapt itself to changes in the problem domain