ABSTRACT
Infrastructure systems serve a pivotal role in communities, which highlights the importance of making proper decisions on their operation and maintenance. Their performance depends on multiple factors with inherent uncertainties, e.g. earthquakes, floods, and deterioration, calling for a probabilistic approach. While it is not straightforward to formulate the high-dimensional probability distribution that covers all these factors, Bayesian network (BN), by graphically representing their probabilistic relationship, can facilitate the modeling. However, the conventional strategy of BN quantification is often infeasible for large-scale systems. This is because it stores the probability values of all basic mutually exclusive and collectively exhaustive (MECE) events, whose number exponentially increases with the number of components. This paper shows how this issue can be addressed by employing the recently proposed matrix-based BN (MBN), whose matrix-based quantification eliminates the restrictions of (1) storing instances in the unit of basic MECE events and (2) incorporating all existing events. Once the MBN is quantified, one can analyze various types of systems, e.g. transportation networks, oil distribution networks and power systems, and perform various tasks of inference, e.g. system reliability, component importance measure and optimization.
