Propagation of Uncertainties Bayesian Belief Networks: A Case Study in Evaluationof Valve Reliability

摘要

In this paper we introduce the concept of Bayesian belief networks (BBN) anddiscuss its application to investigate the reliability of a subset of valves as they are currently used. A BBN is a graphical model that encodes causal and probabilistic relationships among variables of interest. It is a visually effective method for modeling complex systems, which may contain both qualitative and quantitative variables. It models the relationship among the components and processes that make up the system and has a intuitive representation of information flow. When used in conjunction with statistical techniques, a BBN model of a complex engineering system has several advantages for data and decision analysis over existing methods such as fault trees and event trees. First, a BBN encodes dependencies among all variables and, therefore, can accommodate scenarios where data are missing. Second, a BBN can encode causal relationships to gain understanding about a problem domain, and to predict the consequences of intervention. Third, since a BBN combines both causal and probabilistic relationships, it is an ideal representation for combining prior knowledge and experimental data. Finally, using Bayesian statistical methods, the BBN approach to modeling a system provides an efficient and principled approach for avoiding overfitting of data. We have used Bayesian belief networks to construct a model of a valve from prior knowledge and existing data. Further, we have extended the method to evaluate the reliability of the valves within the framework of their application. Within this context, we consider several plausible scenarios that can occur during the operation of the valve and evaluate their effect on its parts and its overall reliability.