Structural and System Reliability

Chapter 15 describes the use of the Bayesian network (BN) methodology for reliability assessment and updating of structural and infrastructure systems. A brief review of the BN as a graphical representation of random variables and an efficient framework for encoding their joint distribution and its updating upon observations is presented. D-separation rules describing the flow of information within the network upon observation of random variables are described and methods are presented for discretizing continuous random variables, thus allowing the use of efficient algorithms applicable to BNs with discrete nodes. Efficient BN models for components, systems, random fields, and seismic hazard are developed. For time- or space-variant problems, the dynamic Bayesian network is introduced. This model is used in conjunction with structural reliability methods (FORM, SORM, simulation) to develop enhanced BNs to solve reliability problems for structures under time-varying loads. Detailed examples are presented, including post-earthquake risk assessment of a spatially distributed infrastructure system and reliability assessment of a deteriorating structure under stochastic loads. The chapter concludes with a discussion of the potential of the BN as a tool for near-real-time risk assessment and decision support for constructed facilities, and the need for further research and development to realize this potential.