ABSTRACT
ABSTRACT: Bayesian Networks (BNs) provide an excellent framework for modeling system performance, particularly in near-real time applications when it is necessary to update models in light of observations. However, BNs can be very demanding of computer memory and inference can become intractable if care is not taken to optimize their topology. In this paper, efficient BN formulations for modeling system performance are presented. First, formulations are developed for series and parallel systems. Then, results are extended to general systems for which the minimal link and/or cut sets are known. Finally, an optimization algorithm is developed to automate the generation of efficient BN formulations for modeling system performance.
