Structural System Reliability Analysis Based on Improved Explicit Connectivity BNs

Abstract

The study proposes an improved Explicit Connectivity Bayesian Networks (ECBNs) for system reliability assessment. The framework combines Analytic Hierarchy Process (AHP) with the traditional ECBNs. AHP is adopted to consider probabilistic dependencies between system components. Judgment matrix is constructed and weight vector, which meets the requirement of a random consistency check, is extracted to describe a conditional probability information of BNs intermediate nodes. The framework is especially suitable for the system with rare damage filed data. In addition, considering the multiple failure modes for the system component, the multi-dimensional Performance Limit State (PLS) function is proposed to estimate the marginal probability for the BNs root nodes. PLSs are properly modeled as interdependent random variables instead of deterministic quantities. Failing to properly account for the dependencies between PLSs, the non-conservative failure probability results will be obtained. Finally, the system reliability can be calculated through the BNs Junction Tree forward inference algorithm, and the most vulnerable components in the system can be identified through backward diagnose inference. The improved ECBNs theory is first applied to the system reliability evaluation for a reinforced concrete bridge to verify its validity.