Simulation supported Bayesian network approach for performance assessment of complex infrastructure systems

With increasing traffic demand, aging infrastructure and higher user expectations, bridge network managers seek tools by which they can maintain the normal service of bridge networks. This paper presents a novel modeling framework for assessing the dependability and performance of bridge transportation networks to support maintenance decisions that balance both owner and user perspectives. Bridge conditions, characterized by their residual capacity, directly impact network travel times. This relationship necessitates extending beyond traditional binary state representations to consider multiple performance and service levels. We adopt expected travel time as a key indicator of network availability, requiring multistate modeling approaches. However, the combination of multistate modeling and complex network interdependencies makes performance assessment particularly challenging.
To address these challenges, we propose a framework that integrates Bayesian Networks (BNs) with Markov Chain-based traffic modeling tools, enabling scalable analysis of large flow-based networks. We introduce an adaptive n-layer decomposition approach that leverages traffic dynamics to recursively partition complex networks into manageable subnetworks while preserving operational dependencies. These subnetworks can then be simulated independently to estimate expected travel times and maintenance costs under various maintenance strategies, significantly reducing computational complexity. The resulting simulation data drives the quantification of the BN model.
We demonstrate our methodology on a case study of a highway network with 20 bridges located in Los Angeles County, California. We illustrate how our framework supports maintenance decision-making by evaluating the propagation of bridge-level decisions and their impact on system-wide travel times and maintenance costs across different scenarios. Validation through exhaustive simulation of the case study network demonstrates strong alignment between predicted and simulated results. Additionally, a sensitivity analysis of key model parameters affecting BN quantification confirms the framework’s robustness.