Reliability Assessment And Remaining Useful Life Prediction For Engineering Systems By Fusing Multi-Source Imprecise Information

In recent years,with the proliferating progress of high-end equipment,low reliability and frequent failures have become the salient challenges to be overcome.It is,therefore,urgent to conduct the system reliability assessment to reveal the dynamic system degradation behaviors,find out the potential reasons for system failures,and provide a suite of guidance for reliability growth,fault warning,and maintenance decisions.Nevertheless,traditional reliability assessment methods using a large amount of failure data are inaccessible for high-end equipment as they are highly reliable products and the cost of system-level reliability tests is unaffordable.In fact,during the whole lifecycle of a system,some pieces of reliability information related to the whole system and its components can be gathered from multiple sources while they might be imprecise,heterogeneous,and conflicting with each other.Properly fusing the multi-source imprecise information(MSII)provides an alternative for the reliability assessment of high-end equipment.In this dissertation,we aim at conducting the MSII fusion from three different lifecycle stages of a system,i.e.,the reliability assessment stage,RUL prediction stage,and the reliability design stage.The primary research contributions and innovative outcomes are itemized as follows:(1)Development of a conflict measure and an informativeness metric for MSII.The proposed method utilizes the interval theory to quantify the epistemic uncertainty of MSII and leverages the constrained optimization to assess the system reliability bound by using each individual piece of MSII.The reliability bounds by all pieces of MSII are converted into mass functions by using the interval-to-mass transformation method.Therefore,a two-dimensional conflict measure,which combines the conflict factor and Jousselme distance in evidence theory,is proposed to measure the conflict between mass functions.A Bhattacharyya distance-based method is developed to further quantify the informativeness of each piece of MSII to the system reliability estimate.As demonstrated by the illustrative examples,the two-dimensional conflict measure outperforms the traditional conflict measures in measuring conflicts among MSII.Moreover,the results of both the proposed conflict measure and informativeness measure are critical for further calibration and fusion of MSII.(2)Development of two reliability assessment methods by fusing conflicting and non-conflicting MSII,respectively.The constrained optimization is designed to assess the system reliability bound by fusing the non-conflicting MSII,whereas the two-stage optimization is designed to fuse conflicting MSII for system reliability assessment.In the first-stage optimization,upper and lower bounds of the degradation model parameters are determined by minimizing the conflict between the predicted and elicited MSII.The second-stage optimization is,then,conducted to identify the upper and lower bounds of the system reliability given that the degradation model parameters are constrained in the bounds obtained from the first-stage optimization.As demonstrated by the illustrative examples,both the constrained optimization and the two-stage optimization can be utilized for non-conflicting MSII and the system reliability results are the same.Nevertheless,when the MSII are conflicting with each other,only the two-stage optimization should be utilized to assess the system reliability bounds.(3)Development of an RUL prediction method by fusing imprecise observations.An interval particle filtering(PF)is proposed to predict RUL but fusing the imprecise observations quantified as interval data.The IPF is built on three pillars: 1)An interval contractor(IC)that mitigates the error explosion problem when the epistemic uncertainty in the interval-valued observation data is propagated;2)An interval likelihood function for assessing the weights of particles;3)An interval kernel smoothing(IKS)algorithm for estimating the unknown parameters in the IPF.To examine the performance of interval-valued RUL results,four performance metrics were extended under the interval framework.As demonstrated by the illustrative examples,the RUL prediction result cannot converge to its ground truth without introducing the IC method.(4)Development of an RUL prediction method by fusing condition monitoring data and expert knowledge.A mixture of Gaussians-evidential hidden Markov model(Mo GEHMM)is proposed to fuse expert knowledge and condition monitoring information for RUL prediction under the evidence theory framework.The evidential expectationmaximization algorithm is implemented in the offline phase to train the Mo G-EHMM parameters based on historical data.In the online phase,the trained model is used to recursively update the health state,reliability,and the probability mass function of RUL.A numerical metric is defined based on the Bhattacharyya distance to measure the RUL prediction accuracy of the developed method.The results from illustrative examples demonstrate that despite imprecisions in expert knowledge,the performance of RUL prediction can be substantially improved by fusing expert knowledge with condition monitoring information.(5)Development of a multi-objective redundancy allocation for multi-state systems by fusing imprecise information.The proposed method utilizes the evidence theory to quantify the epistemic uncertainty of component state assignment.An evidential network(EN)model is developed to address the epistemic uncertainty propagation from components to system reliability evaluation.The resulting multi-objective redundancy allocation problem is resolved via a modified NSGA-II,in which a set of evidential Pareto dominance criteria is put forth to compare any pair of interval-valued solutions.As observed from the case studies,the traditional Pareto front is extended as an intervalvalue Pareto front when considering the epistemic uncertainty of component state assignment,and the interval-valued Pareto front is consistent with the true one.