Reliability prediction used dynamic Bayesian networks for redundant system with time‐lag correlation

Abstract

AbstractIn the reliability prediction process of a redundant system, active units and standby units are assumed independent generally. Due to the interaction among units, failure of the active unit gives the standby unit a potential interior shock, then the standby unit fails after a period of operation. Therefore, the correlation between the active unit and the standby unit is triggered by the failure of the active unit and also has time‐lag. In this paper, a correlation estimated method with two parameters is proposed based on cross‐correlation analysis, where the two parameters are time‐lag interval and time‐lag correlation coefficient, the discriminant conditions of independent, simultaneous correlation and time‐lag correlation are given. Subsequently, the Conditional Probability Table (CPT) of Dynamic Bayesian Network (DBN) is updated by adding time‐lag intervals and time‐lag correlation coefficients. Then, DBN is optimized into Time‐lag Dynamic Bayesian Network (TDBN). Meanwhile, the modeling process and calculated method of TDBN are given. To reduce the complexity of TDBN, the maximum time‐lag correlation coefficient and the time‐lag interval corresponding are regarded as the main parameters to estimate time‐lag correlation. Finally, through the simulation analysis and the case of Chinese train control system (CTCS‐3), it could be seen that the reliability prediction of TDBN is more accurate than DBN, TDBN improves the reliability prediction accuracy of train and driver interface module to 0.8135 either.