Multi-state analysis functional models using Bayesian networks

Abstract

Multilevel Flow Modeling (MFM) model maps functionality of components in a system through logical interconnections and is effective in predicting success rates of tasks undertaken. However, the output of this model is binary, which is taken at its extrema, i.e., success and failure, while in reality, the operational status of plant components often spans between these end. In this paper, a multi-state model is proposed by adding probabilistic information to the modelling framework. Using a heat exchanger pilot plant as a case study, the MFM model is transformed into its fault tree [1] equivalent to incorporate failure probability information. To facilitate computations, the FT model is transformed into Bayesian Network model, and applied for fault detection and diagnosis problems. The results obtained illustrate the effectiveness and feasibility of the proposed method.