System operational reliability evaluation based on dynamic Bayesian network and XGBoost

This paper proposes a methodology to evaluate system operational reliability. The dynamic Bayesian network (DBN) and XGBoost are integrated within an evaluation framework. The component dependencies are established by DBN considering maintainability. XGBoost is used to map the multidimensional monitoring data from sensors into component states. The monitoring nodes are added to the DBN to introduce the influence of state diagnosis results on system operational reliability. The conditional probability tables (CPTs) of the monitoring nodes are obtained based on the confusion matrix. In order to demonstrate the methodology, the state diagnosis experiment for the generator is conducted. Another case is presented to evaluate the operational reliability of the marine electrical propulsion system through simulation method. The proposed model archives the reliability evaluation integrating monitoring with statistical failure data. Meanwhile, the DBN-based framework shows applicability to diagnosis models based on machine learning.