Research On Multi-state System Reliability Analysis Based On Evidential Networks

With the increasing advance of industry 4.0,the development directions of modern engineering systems and large-scale manufacturing equipment are turning into large-scale and complicated structure,precise and intelligent function,and high reliability of performance.Under the influence of extreme service environment and severe operating conditions,the complex systems own more diverse failure mechanisms and present multiple failure modes or degradation states,which cannot be analyzed precisely by traditional binary system reliability theory,leading to ascending pressure in system reliability analysis and bringing more huge risks to practical system operations.Meanwhile,due to the inaccuracy of data,lack of information,and limited subjective cognition level of engineers,the impact of epistemic uncertainty has to be taken into consideration when analyzing the reliability of such multi-state systems,which further aggravates the complexity of reliability analysis of these systems.In recent years,evidential networks(ENs)has been used for reliability analysis of systems considering epistemic uncertainty and has been widely applied in many practical engineering fields,as it is generally believed that it can deal with epistemic uncertainty well.However,when the system presents multi-state characteristic,system reliability analysis based on evidential networks will always produce combination explosion of states during the epistemic uncertainty propagation,which means that the cost of system reliability analysis will increase exponentially with the increase of the number of components and that of functioning states,namely the "dimensional disaster".Therefore,aiming at handling with the challenge of "dimensional disaster" faced by evidential networks in the reliability analysis of multi-state systems with epistemic uncertainty,some related research has been done in this thesis as follows:(1)An improved evidential network method based on interval probabilistic transformation is proposed,which can significantly alleviate the combinat ion explosion in reliability analysis of multi-state systems based on evidential networks.Firstly,according to two extreme belief mass distribution rules,two BBA probabilistic transformation methods based on confidence interval are utilized at the root nodes for redistribution of the interval form of prior evidence to the monomorphic ones.And then,two boundaries of belief masses in monomorphic states are obtained.After that,the confidence intervals of the system reliability can be calculated through the uncertainty propagation of the evidential network.The results of two multi-state numerical examples have shown that the proposed method not only can improve computational efficiency but also can guarantee calculational accuracy,which provides a reasonable and effective method in solving the reliability pr oblems of multi-state system with epistemic uncertainty.(2)A method for reliability analysis and performance evaluation of multi-state system based on improved evidential networks is presented.This method utilizes a proposed extreme BBA probabilistic transformation method based on utility function to preprocess the prior evidence information of root nodes,so as to alleviate the "dimensional disaster" faced by the evidential networks in system reliability analysis.Specifically,the functioning states of each component are sequenced according to utility function at first.Then,the basic belief assignments(BBAs)of each component are reassigned in terms of modified commonality function,through wh ich the BBAs defined in the power set space are represented by two extreme BBA distributions defined in the frame of discernment.Finally,the reliability intervals of the system states are calculated through evidential network,whilst the confidence interval of system performance level is computed by expected utility function.Two multi-state system cases are investigated to demonstrate the effectiveness and efficiency of the proposed method in reliability analysis and performance evaluation of multi-state system with epistemic uncertainty,as by using the proposed method,not only the“dimension disaster” in calculation can be alleviated effectively,but also a confidence interval of system performance without information loss can be obtained.