Robustness of a hybrid simulation-based/Bayesian approach for the risk assessment of a real-world road network

This paper proposes a hybrid approach for the Bayesian modelling of the performance of a real-world road network subjected to seismic hazard, based on preliminary Monte-Carlo simulations. While several computational bottlenecks are limiting the application of Bayesian Networks to simplified infrastructure systems, Monte-Carlo simulations are used here for the identification of the most common and typical damage configurations, in order to design a simplified Bayesian Network formulation, i.e. the thrifty-naïve formulation. The practical appeal of this approach is demonstrated on a complex road network in the French Pyrenees, where the corresponding Bayesian Network has the ability to predict various system performance indicators and to update predictions based on field observations.