A novel fault propagation path identification inference algorithm using parent nodes filter

Abstract

Bayesian network (BN), which is an effective probabilistic graphical model with strong characteristics of practicality and applicability, is often used for real-world faults diagnosis. However, most of the existing BN-based studies merely analyze how to diagnose root cause and do not have the capability of identifying fault propagation path. To address this problem, a novel Gaussian Bayesian network-based faults propagation path identification inference algorithm using parent nodes filter is proposed. For a specific fault node, the set of its parent nodes is firstly reduced according to parameter weights and value fluctuations. Then estimates of the maximum conditional probability for possible subsets of reduced parent set are computed based on decomposition of conditional probability and dichotomy. After estimation, the most probable cause can be determined by the closest estimate to actual value. Through such layer-by-layer inference, fault propagation paths in the network can be finally tracked. Experimental results have demonstrated that the novel approach is capable of identifying fault propagation paths effectively, which provides higher adaptability and faster speed.