Research On Remaining Useful Life Prediction And Maintenance Decision For Complex Electromechanical Systems In Service

Promoted by the “Made in China 2025” plan,the traditional equipment manufacturing industry starts to deeply comprise the industrial and information techniques,and turns to be the highly sophisticated equipment manufacturing industry with intelligence.The complex electromechanical systems,as the critical parts of the manufacturing process,are also developed towards the goal of function richness,high reliability,accuracy and efficiency.However,due to the multiple degradation trends and complex failure mechanisms of the electromechanical systems,traditional prognostic and health management methods cannot fully meet the requirements for the rapid development growth of the manufacturing industry.To guarantee the reliability and the safety of systems,at the same time,to optimize the efficiency and the cost,the research of developing an intelligent health management method which is suitable for the service process of complex electromechanical systems is required urgently.With the utilization of the advanced sensoring detection technique,industries try to achieve the system degradation monitoring,analysis,and prediction,so as to make and optimize the system maintenance policy,and to further control the degradation tendencies,identify and predict failures,resist risks,and minimize costs.This is a critical way for the industries to solve the life cycle health management issue of the complex electromechanical systems,which is also a huge challenge in the field of the reliability engineering.The dissertation tries to solve two critical issues of the life cycle health management for the complex electromechanical systems,which are the remaining useful life(RUL)prediction,and the maintenance decision.With the fully consideration of the fusion of both statistical failure data and condition monitoring information,the degradation tendencies and failure features of multiple periods(early failure period,random failure period,and wear-out failure period),the influence of condition monitoring frequency,maintenance policy,maintenance effect and dependency among components on the system availability and maintenance cost,this dissertation mainly focuses on theoretical research and application validation on the RUL prediction and maintenance decision optimization of the complex electromechanical systems with multi-stage and multi-level.This study tries to provide a concrete scientific proof for the management decision of manufacturing industries.The main contributions and contents are as follows:(1)Establishment of the data-driven multi-segment RUL prediction models for the service processTo describe the failure modes under different periods of the system service process and implement the degradation condition evaluation and RUL prediction,this dissertation developed two data-driven multi-segment RUL prediction models under the minimal maintenance hypothesis:1)When only statistical failure data is available,a NHPP-based failure intensity model with multi-segment and an adjustment factor is developed to flexibly fit different segments of the bathtub-shaped failure intensity curve.It is shown from a case study that this model is advanced on describing the failure intensity tendency in early failure period,random failure period,and wear-out failure period,and the RUL prediction effect is also verified.2)When both statistical failure data and the condition monitoring information are available,a PIM-based multi-segment model is developed.Compared with the first NHPP model,the new model can better describe the synthetic effect of system degradation and condition features on the failure intensity.With both numerical simulations and real applications,the goodness of fit and the RUL prediction accuracy are verified.(2)Development of a parameter estimation method with the combination of the statistical process control(SPC)and the fuzzy clustering techniques for the multi-segment modelTo address the parameter estimation issue of the developed multi-segment models due to the increasing of parameter numbers,this dissertation developed a modified maximum likelihood estimation(MLE)method by combining SPC and the fuzzy clustering techniques.First,the modified control chart was built by the Bootstrap method to accomplish the segment partition for the failure process.To reduce the disturbance of the random signal,the ordered fuzzy clustering method was implemented to cluster observations into different segments.Then,the MLE can be applied to estimate parameters for each segment.With both numerical simulations and real applications,the prediction accuracy and the computing efficiency of the developed method were analyzed.(3)Establishment of the maintenance framework in service and the maintenance decision model with the consideration of different failure periods.According to different degradation tendencies and failure modes under different periods in service,a maintenance framework including early failure period,random failure period,and wear-out failure period was developed.Focusing on the fundamental factors of the maintenance policy,this dissertation mainly studied on the maintenance decision modeling and optimization for random failure period and wear-out failure period.1)For random failure period,a preventive maintenance policy with regular inspection by the SPC was developed.First,a PIM-based RUL prediction model was developed to evaluate the probability of both sudden failures and degradation failures.Taking the inspection interval and the failure intensity threshold as optimization variables,a maintenance decision model was developed to minimize the expected cost per unit time,so as to achieve the maintenance optimization for random failure period.2)For wear-out failure period,a rolling horizon based dynamic predictive maintenance policy was developed under the continuous condition monitoring.Considering the influence of the imperfect maintenance on the system RUL,the degradation threshold and the predictive maintenance time were taken as optimization variables.Then,the short-term averaged cost was minimized,so as to achieve the optimal maintenance schedules.(4)Establishment of a dynamic group maintenance model under multi-objective optimization with the consideration of component dependency.Based on the developed maintenance framework,the structure dependency and the economic dependency among system components are further considered.Taking maintenance grouping structure and maintenance time as optimization variables,a multi-objective based dynamic group maintenance model was developed to maximize the availability and minimize the maintenance cost simultaneously.Based on the traditional rolling horizon optimization technique,this dissertation first developed a maintenance downtime model dependent on maintenance type,maintenance effect and maintenance threshold,then combined with the proposed maintenance models of different failure periods,the system availability and the maintenance cost can be balanced.In the case of a motorized spindle,the advantages of the developed group maintenance model was proved.