Bayesian networks for the multi-risk assessment of road infrastructure

• Main Author: Gehl, P.
• Other Authors: D'Ayala, D.
• Published: University College London (University of London) 2017
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.756088

The purpose of this study is to develop a methodological framework for the multi-risk assessment of road infrastructure systems. Since the network performance is directly linked to the functional states of its physical elements, most efforts are devoted to the derivation of fragility functions for bridges exposed to potential earthquake, flood and ground failure events. Thus, a harmonization effort is required in order to reconcile fragility models and damage scales from different hazard types. The proposed framework starts with the inventory of the various hazard-specific damaging mechanisms or failure modes that may affect each bridge component (e.g. piers, deck, bearings). Component fragility curves are then derived for each of these component failure modes, while corresponding functional consequences are proposed in a component-level damage-functionality matrix, thanks to an expert-based survey. Functionality-consistent failure modes at the bridge level are then assembled for specific configurations of component damage states. Finally, the development of a Bayesian Network approach enables the robust and efficient derivation of system fragility functions that (i) directly provide probabilities of reaching functionality losses and (ii) account for multiple types of hazard loadings and multi-risk interactions. At the network scale, a fully probabilistic approach is adopted in order to integrate multi-risk interactions at both hazard and fragility levels. A temporal dimension is integrated to account for joint independent hazard events, while the hazard-harmonized fragility models are able to capture cascading failures. The quantification of extreme events cannot be achieved by conventional sampling methods, and therefore the inference ability of Bayesian Networks is investigated as an alternative. Elaborate Bayesian Network formulations based on the identification of link sets are benchmarked, thus demonstrating the current computational difficulties to treat large and complex systems.