A computational tool for Bayesian networks enhanced with reliability methods

A computational framework for the reduction and computation of Bayesian Networks enhanced with structural reliability methods is presented. During the last decades, the inner flexibility of the Bayesian Network method, its intuitive graphical structure and the strong mathematical background have attracted increasing interest in a large variety of applications involving joint probability of complex events and dependencies. Furthermore, the fast growing availability of computational power on the one side and the implementation of robust inference algorithms on the other, have additionally promoted the success of this method. Inference in Bayesian Networks is limited to only discrete variables (with the only exception of Gaussian distributions) in case of exact algorithms, whereas approximate approach allows to handle continuous distributions but can either result computationally inefficient or have unknown rates of convergence. This work provides a valid alternative to the traditional approach without renouncing to the reliability and robustness of exact inference computation. The methodology adopted is based on the combination of Bayesian Networks with structural reliability methods and allows to integrate random and interval variables within the Bayesian Network framework in the so called Enhanced Bayesian Networks. In the following, the computational algorithms developed are described and a simple structural application is proposed in order to fully show the capability of the tool developed.