Cascading failures of interdependent infrastructure systems have become increasingly critical as revealed by recent natural disasters and human disruptions. By determining how interdependencies affect the fragility of components within infrastructure systems, I propose that one can identify the most critical components, determine where to invest resources, and decrease the time it will take to regain operational status. This dissertation presents a modeling approach and the accompanying sets of algorithms that enable computationally efficient probabilistic modeling of large infrastructure systems while considering interdependencies between networks. The proposed method creates a computationally tractable, representative Bayesian network of the system, with which exact inferences over the network are possible. Once the Bayesian network is constructed, inference analyses can be performed over a range of component state and hazard event scenarios to identify vulnerabilities across networks. The model is applied to analyze component criticality within infrastructure systems. These component importance measures can be used to prioritize components for repair, replacement, and reinforcement. The proposed methodology is applied to assess critical water services in the city of Atlanta, Georgia, including dependencies of the water distribution system on the power distribution system.
Dr. Iris Tien
Dr. John Crittenden, Dr. Leonardo Duenas-Osorio (Rice University), Dr. John Taylor, Dr. Brani Vidakovic (ISYE)