System Reliability Modeling And Assessment By Bayesian Networks And Observation Data

With the increasing demand of the industrialization of intelligent equipment and the pace of industrialization,customer requirements,system structure,and working environment of various systems such as military systems and industrial products are becoming increasingly complex.As the basis of system reliability assessment,optimization,and maintenance decision making,constructing reliability model of a system has become a critical issue in the field of system reliability engineering.The traditional system models,such as fault trees and reliability block diagrams,are unable to model the uncertainty associated with the system structure function and failure dependence among components.The uncertainty of a system structure function will lead the relation of states between the system and its constituents to be probabilistic,and the parameters in the system structure function are unknown.On the other hand,ignoring the dependency among components will cause potential risk to system reliability assessment.Bayesian networks(BNs)can not only represent the uncertainty and dependency among variables,but also bidirectionally calculate the probability of its node's state.It is,therefore,of great significance to choose the BN as system reliability modeling tool.On one hand,the BN-based method can greatly deal with the uncertainty associated with failure dependence of components.On the other hand,the BN-based method can dynamically update the reliability of a specific system by aggregating observation data.This dissertation devotes to reliability modeling and assessment of engineered systems by BNs.The main contributions and innovations are summarized as follows:(1)Development of a BN-based reliability modeling and assessment framework for both deterministic and probabilistic relation between system's state and components' states.Firstly,the graphical structure of a static BN is built by connecting the nodes among different hierarchies.A reliability model based on dynamic Bayesian network(DBN)is,then,developed by copying static BN as multiple time slices and linking the degradation behaviors of components over time.Furthermore,the state distribution and reliability of the system at different time instants are evaluated.Two numerical examples,along with an engineering case,are given to demonstrate the effectiveness of the proposed framework in terms of modeling deterministic and probabilistic relations among components.(2)Development of a reliability structure function parameter estimation method with incomplete observation data.Once the DBN is constructed as the system reliability model,the unknown relation among components' states can conducted by estimating the unknown parameters in the conditional probability tables(CPTs)between system and subsystem nodes.To integrate the available state observations and achieve a consistent estimation at different time instants,a customized Expectation-Maximization algorithm with parameter modularization is proposed.Two illustrative examples are given to demonstrate the effectiveness of the proposed method in terms of accurately learning the relations of the states of a system.(3)Development of a reliability modeling and assessment method for k-out-of-n systems with functional dependence.By considering the functional dependence between functional components in a k-out-of-n structure and other support components in an engineered system,a new system structure,namely modular k-out-of-n system with functional dependence,is introduced,and its general definitions are given.The graphical structure of the DBN is constructed by taking account of the functional dependence among components,and the CPT parameters are generated automatically by a customized algorithm.Based on the corresponding DBN model,the reliability function of the entire systems can be assessed.A case study of a fixed pulse instrumentation radar equipment in transmitter system is used to demonstrate the effectiveness of this method.