A LIMIT-STATE-SURFACE-BASED DISCRETIZATION OF BAYESIAN NETWORK VARIABLES FOR STRUCTURAL SYSTEM RELIABILITY ASSESSMENT

This paper proposes a discretization framework for Bayesian network–based structural reliability analysis involving continuous random variables. The method identifies important points for reliability estimation, including design points on limit state surfaces, and discretizes continuous random variables such that the estimation accuracy is maintained in the vicinity of these points. The proposed approach enables accurate and stable estimation of system failure probability under a limited number of discrete states. Its effectiveness is demonstrated through a benchmark problem and a frame structure example with multiple collapse modes, including sensitivity analysis and posterior inference conditioned on collapse.