An efficient discretization scheme for a dynamic Bayesian network in structural reliability analysis

Abstract

Using a dynamic Bayesian network (DBN) to estimate the failure risk of a component or system that deteriorates with time has several advantages. A DBN discretizes the probability distribution of variables and thereby increases the efficiency of computing resources and reduces computation time. However, it is important to devise an optimal discretization scheme because the size of the model grows exponentially as the number of discretized intervals increases. In this paper, we propose an optimal discretization scheme for a DBN used to model the time-varying deterioration of a turbine blade component. The results of estimating the reliability indices with the DBN were verified by comparing them with the results of a Monte Carlo simulation. In addition, compared with a log-transformed discretization method, our DBN discretization method shows a significantly increased computation speed.

Structural reliability analysis of aircraft wing rib fatigue cracking using surrogate dynamic Bayesian network

“…We used a heuristic scheme formulated by combining a conventional log-transformed scheme and a beta distribution function. 35 Given that the log-transformed scheme was densely located in the small crack region, it was good for representing the initial crack length distribution. The beta distribution function was used to depict the near-critical crack length region that determined the POF.…”