Development of Seismic Reliability Analysis Methods for Large-scale Infrastructure Networks

Abstract

With the advancement of technology and the densification of modern society, infrastructure facilities are closely interconnected, thereby forming a vast infrastructure network. Seismic damage to individual structures can result in socio-economic costs to the entire infrastructure network. To quantify the risk of networks and ensure efficient operation and maintenance, there is a need for network seismic reliability analysis. To this end, the seismic failure probability of structures should first be assessed, and then the network reliability is evaluated under different combinations of structural conditions. It is challenging to apply such network seismic reliability analysis for large-scale networks due to common source effects of earthquakes throughout the network, interdependent seismic demands, and It is challenging to apply such network seismic reliability analysis for large-scale networks. Monte Carlo Simulation (MCS) has been used to overcome these limitations, but still has several limitations, including inefficiency for low probability events and difficulty in probabilistic inference. This dissertation proposes three main methodologies for seismic network reliability analysis. The first approach introduces Bayesian networks (BNs) and junction trees (JTs) to evaluate network reliability and quantify the complexity. Based on the JT constructed from the dual representation of a given network, the reliability of directed acyclic networks can be evaluated by one-way message passing. Even for a cyclic network, the reliability can be accurately assessed using a set of equivalent directed acyclic subnetworks through cycle decomposition. Meanwhile, although it is common to quantify the complexity of network reliability analysis only by the number of components, the network topology also affects the actual computational complexity. Numerical examples demonstrate that the proposed method can not only evaluate network reliability and component importance measures in real time, but also quantify the complexity using the maximum clique size in JT. Second, a centrality-based selective recursive decomposition algorithm (CS-RDA) is proposed to identify critical components that play a key role in terms of connectivity based on the network centrality, thereby (1) simplifying the network for multi-scale approaches and (2) significantly increasing the convergence of recursive decomposition algorithm (RDA). Compared to other RDAs, CS-RDA can achieve the target bound width using significantly fewer subgraphs. The efficiency and accuracy of CS-RDA are demonstrated by numerical examples including large-scale highway bridge networks. The application examples also investigate the trade-off between efficiency and accuracy with respect to the degree of network simplification. Finally, a variance-reduction sampling method is proposed to enhance the scalability and efficiency of direct MCS. The binary limit-state function for network connectivity is reformulated into more informative continuous limit-state functions that quantify how close each sample is to the network failure event. The proposed functions facilitate the construction of intermediate relaxed failure events, thereby enabling network reliability analysis using subset simulation. Furthermore, a single implementation of subset simulation can generate the network reliability curve by configuring each intermediate failure domain as a network failure event under a given earthquake intensity. Numerical examples demonstrate that the proposed method can accurately and efficiently evaluate network reliability curves in terms of k-terminal reliability and maximum flow, as well as two-terminal reliability.