SYSTEM-RISK SENSITIVITY ANALYSIS IN BAYESIAN NETWORKS

Importancemeasuresarewell-knownandimportanttools which are widely used in risk-informed decision making. Theiroutstandingtraditionaldenitionshavemade them useful in many applications related to risk and reliability aspects of dierent systems. These perfect traditional denitions help researchers to nd the most important components in a system, and consequently, to detect and obviate weaknesses in system structure and operations. Generally, these measures are based on fault tree technique. Although fault tree is a powerful tool to study risk, reliability, and structural characteristics of systems, Bayesian networks have indicated explicit advantages over it in modeling and analysis abilities. Classical fault tree is not suitable in analysis of large systems that include aspects such as: common cause failure, redundant failure, uncertainty, and some kind of complex dependencies such as sequentially dependent failures, while these aspects are not negligible in large modern systems anymore. So, the perfect definitions of importance measures are restricted to limitations of fault tree. Bayesian networks, on the other hand, have become a widely used method in dierent kinds of statistical problems, including fault diagnosis, reliability and safety assessment, and updating safety systems failure probabilities. In addition, Bayesian networks due to their modeling and analytical abilities, are capable of accommodating the mentioned aspects easily and straight forwardly. In this paper, we extend the traditional denitions of importance measures to Bayesian networks resulting in more capable importance measures in terms of modeling and analysis. Theimportancemeasures that are extended to Bayesian networks in this research are the most important and widely used ones that some of them are used in famous methods named probabilistic safety assessment. The extended importance measures are: Risk achievement worth, Risk reductionworth,Fussell-Veselyimportancemeasure,Birnbaum importance measure, and Dierential importance measure. The results of implementing the new achievements on a real-world case study prove the desired effectiveness.