Bayesian Approach for Reliability Evaluation and Remaining Useful Life Prediction

Abstract

For redundant system modeling, the probabilistic graphical model is considered the predominant approach for predicting system reliability. The reliability of the system, determined through an empirical study like event tree analysis (ETA), fault tree analysis (FTA), and Bow-tie models uses the assumption that units are independent at different points in time. These models do not account for the time-lag connection between units or cannot repeatedly deduce system reliability when the states of the units change. The primary approach for predicting system reliability from a structural learning standpoint is the Bayesian network (BN) methodology. The BN can assess uncertain events and establish the correlation between units using probability functions. The directed acyclic graph BN employs the conditional probability table (CPT) to establish the causative relationship. The occurrence of time-lag correlation in a redundant system is a result of a causal relationship that is initiated by the failure of the active unit. Therefore, it may be argued that Bayesian networks (BN) are better suited for predicting system reliability in redundant systems. Additionally, BNs do not consider the dynamic effects that may arise from the units in different stages within the system. The dynamic Bayesian network (DBN) is an extension of the BN that incorporates the temporal aspect by considering the failure of a unit as a first-order Markov process. This chapter details the BN and DBN along with the approaches of BN in the field of reliability analysis and remaining useful life (RUL) prediction. © 2024 by The Institute of Electrical and Electronics Engineers, Inc. All right reserved.