Bayesian networks based hierarchical vulnerability evaluation of long-span structures

Abstract

Vulnerability analysis of long-span structures explores the weakness regions, which should be concerned during structural design and operation stages. However, the vulnerability analysis procedure under regular loads is conventionally a tough task in virtue of structural complexity and uncertainties. Therefore, a Bayesian networks (BNs) framework has been developed for hierarchical vulnerability evaluation of long-span structures under regular loads. External loads, structural systems and components are defined as network nodes, and mechanical and risk causalities are simultaneously considered during network establishment. The causality strength between two nodes is quantitatively expressed by a conditional probability table. The state probability inference of nodal variables is accomplished after inputting the observed state of a damaged component as the evidence into the established BN. A new component importance coefficient is defined and calculated on the inferred nodal state probabilities. A component vulnerability index is further defined to predict the most likely failure sequence of components. In addition, a system vulnerability measure is proposed for evaluating the safety risk of the system. The proposed method has been verified against an experimental space truss model and an actual cable-stayed bridge. The component importance of the truss members and the cables was well evaluated with the predictions of their most likely failure sequences. The estimated system vulnerability could indicate the safety risk of the long-span structures due to the damaged components