The System Degradation Modeling And Remaining Useful Life Prediction Considering Multi-Indicator Random Correlation

The development of industrial big data has introduced prediction and health management as new research topics in the field of intelligent manufacturing.With the increasing complexity of the system structure,using only a single degradation indicator often does not fully reflect the potential degradation state and fault type of the system.Meanwhile,there may be a correlation between multiple degradation indicators that jointly reflect the law of system degradation and failure.Therefore,multiple degradation indicators are important to predict the remaining useful life of complex systems.The performance degradation of multiple indicators surpasses the respective threshold,causing multiple system fault types and failure modes,such as redundancy,fusion,and competition among different fault combinations,which introduces a new challenge when predicting the remaining useful life of the system.In this study,a multi-indicator stochastic correlation system was considered as the research object.Based on the real-time monitoring of state information,the research is performed from four aspects: the multi-indicator fusion method,degradation modeling,failure mode division,and remaining useful life prediction.Numerical simulations and case data were used to verify the effectiveness of the model.The main conclusions and results are as follows:(1)A distributed fusion method is proposed to predict the remaining useful life for a class of nonlinear multi-indicator stochastic systems with measurement errors.First,a statespace model of the nonlinear Wiener process was established,and the distributed Kalman filter algorithm was used to fuse and filter multi-measurement data.Next,the parameters and degradation states of the state–space model were estimated and updated online in real time using the expectation maximum and smoothing filter algorithms.Moreover,the distribution of the system’s remaining useful life is obtained according to the estimated state–space model considering the random failure threshold factor.Finally,numerical and case experiments demonstrate the correctness and effectiveness of the distributed fusion prediction method.(2)An adaptive remaining useful life prediction method is proposed based on multiple constructed indicators,considering multiple correlation relationships between multiple features and degradation indicators.First,the feature values in the multisensor signals were extracted and selected,and the relationships between multiple features and indicators were analyzed.Multiple indicators were then constructed to characterize the system degradation,and different degradation patterns of multiple indicators were considered to establish a degradation statespace model involving several indicators.Finally,using the definition of the first hit time,the corresponding remaining useful life distributions were derived for the three failure modes.Two datasets from the PHM2012 Challenge and the XJTU-SY were used as examples for experimental verification and analysis.(3)A remaining useful life prediction framework is proposed for a multi-indicator system that considers the effects of random shocks and stochastic dependence.First,considering the influence of random shocks and measurement errors,a degradation state-space model was established using the stochastic correlation of multiple degradation indicators.Next,the hidden state and unknown parameters of the state-space model are jointly estimated using the expectation-maximization algorithm in conjunction with a strong tracking filter algorithm based on online monitoring data.Then,considering the failure modes of multiple degradation indicators of the system,the distribution of the remaining useful life of the system was adaptively calculated and updated according to the estimated state-space model.Finally,the accuracy of the remaining useful life prediction model is verified for a system affected by random shocks through numerical experiments.The adaptability and effectiveness of the proposed method were verified using the FD001,FD003,and FD004 datasets in the C-MPASSS dataset and high-temperature furnace data as examples.(4)A unified multi-failure mode division framework and the remaining useful life prediction model under different failure modes for multi-indicator systems.First,the combination relationship between different fault types caused by multiple indicators outweighing their thresholds was analyzed,and different redundancy,fusion,and competition failure modes were defined.Next,a formal fault type definition and the remaining time before occurrence under different failure modes are provided.A distribution calculation model for the remaining time of different fault types was derived.The corresponding system’s remaining useful life prediction model under multi-fault competition was established,and then degradation modeling,parameter estimation,and remaining useful life distribution calculations were performed using the state-space model.Finally,the validity of the remaining useful life prediction model according to multiple failure modes was verified through numerical experiments.The applicability and feasibility of the proposed method were proved using the XJTU-SY bearing and C-MAPSS datasets as two examples.(5)A remaining useful life prediction method for multi-component hierarchical systems that considers multi-indicator random correlation and external environmental influence is proposed.First,multiple degradation indicators of multiple components are constructed,and then the degradation state of the components is characterized by the fusion of multiple indicators of the components.Meanwhile,the random correlation and structural correlation between the components and the external environmental influence of measurable and unmeasurable factors are considered,and a degradation state space model of the multicomponent system is constructed.Next,the parameters in the model were estimated and updated step-by-step.According to the structural relationship between the multiple components,the remaining useful life of the multicomponent system under different structures was derived and calculated.Finally,the effectiveness and adaptability of the proposed method are verified using a numerical example and gearbox data from the University of Huddersfield in the UK.