Particle Filtering-based Degradation Modeling And Remaining Useful Life Prediction For Multi-Component Systems

With the advances of technology,some sophisticated systems with a number of components require high reliability,long life,long-term online operation.To ensure the long-term and highly reliable operation of such systems,the degradation of the system should be characterized so as to accurately predict the remaining useful life(RUL)of the system.However,on the one hand,due to the limitation of measurement accuracy of monitoring technology,error arises between the measurement results and the real degradation state of a system.On the other hand,due to the limited accuracy of clustering,classification,and regression algorithms,subjective judgements from experts,etc.,epistemic uncertainty is inevitable in terms of identifying degraded state.Additionally,due to factors such as load sharing mechanism among components,components exposing to the same environment,and close physical location of components,the degradation processes among components are s-dependent.Therefore,how to accurately characterize the multi-component degradation behaviors and the correlation between components in the presence of measurement error is of great importance.In this thesis,by taking account of the measurement errors,the degradation modeling and remaining useful life prediction methods for multi-component systems under a constant environmental factor and external shock environments are studied.By considering the epistemic uncertainty associated with degradation data,the remaining useful life prediction of the system based on the belief particle filtering algorithm is developed.The major works of this thesis are summarized as follows:(1)Development of a degradation model and a remaining useful life prediction method for a multi-component system by considering a constant environmental factor.A multi-component degradation model is developed by constructing a state space model where components in a system share the same external environment and working conditions.By fusing the degradation data of all the components,the unknown parameters in the model can be estimated online by the particle filtering algorithm.On this basis,the remaining useful life of each component in a system can be predicted.The case study shows that the RUL prediction results of the multi-component degradation model under constant environmental influence are significantly different from the case where the dependency among components are not considered.(2)Development of a multi-component degradation model and a remaining useful life prediction method under a shock environment.By considering the external shock and the internal failure of components,the remaining useful life prediction model under a shock environment is developed.First,the failure modes of multi-component system failure mode in the shock environment are introduced,including the hard failure mode caused by external shock and the soft failure mode due to cumulative damage.Second,the impact of the shock environment is considered to model the dependency of the degradation behaviors of multiple components in the shock environment.Third,to address the high-dimensional parameter estimation issue,the particle filtering algorithm is combined with Markov Chain Monte Carlo(MCMC)to estimate the unknown model parameters.On this basis,the remaining useful life prediction for multi-component systems suffering shock and degradation is derived.The proposed method is validated by a numerical example.(3)Development of a degradation model and a remaining useful life prediction method for systems with epistemic uncertainty.In view of the epistemic uncertainty of the degradation data of systems,this thesis uses the belief function theory to describe and quantify the epistemic uncertainty,and introduces the belief particle filtering algorithm,which combines the belief function theory and the particle filtering algorithm,to predict the RUL for systems.The belief particle filtering algorithm aims to develop the sampling rules under the belief function theory,and predict the possible sets of states in the future for systems.Then,the lower and upper bounds of the remaining life distribution are calculated by applying the belief and plausibility functions,respectively.The results from the case study show that the belief particle filter algorithm can effectively predict the state of systems under epistemic uncertainty.Moreover,the more of the degradation data with epistemic uncertainty are collected,the narrower bounds of the remaining life distribution can be achieved.It indicates that the epistemic uncertainty associated with the RUL prediction is reduced.