An efficient framework for the conversion of fault trees to diagnostic Bayesian network models

Complex aerospace systems´ cannot afford downtime to diagnose problems $the interruption of mission critical functions and prohibitive cost of lost business are unacceptable. Such systems are characterized by having many components and require a team of experts to diagnose problems after they occur or to assemble a knowledge database suitable for rapid model based diagnostics. In this paper we present an efficient and largely automated method for developing diagnostic Bayesian network models. The models are created by exploiting existing domain knowledge in the form of reliability fault trees and diagnostic observation lists. The algorithms for conversion of the trees and databases into Bayesian network models have been embedded in a C++ software tool and tested on examples of fault trees ranging from 10 to 800 nodes, which were developed for satellite systems