Research On Data-driven Fault Prognostic Method In Closed-Loop Lifecycle Management System

In recent years,industrial equipment has become more and more large-scale and complicated,and the data that reflects the mechanism and operation state of the operation process is also generated.How to use these massive data to meet the increasing demands of system reliability has become an urgent problem,and the data driven fault prognostic method has become an important solution,which has been paid more attention in recent years.However,the data driven method is often limited by historical data.Once the actual operating environment changes,its prognostic performance will also fluctuates.With the continuous development of Internet of Things,information and communication technologies,relevant technologies and standards can be used to eliminate the information barriers between all stages of the equipment life cycle,to form the closed-loop management of information flow and knowledge flow in the whole life cycle of product equipment,and the closed-loop life cycle management(CL2M)opens a new way to solve this problem.Based on the research of the data-driven fault prognostic method and the key problems in CL2M system,in this thesis,the fault prognostic maintenance architecture in the CL2M system is designed,the dynamic modeling algorithm is proposed,and the concrete subsystem layer solution is realized.The main content of the thesis is as follows:1.The characteristics of data in the use stage of product equipment in CL2M system is studied,and the failure prognostic maintenance architecture is designed.Firstly,the architecture and research contents of CL2M system are summarized.Then the characteristics of the data in the CL2M system and data-driven fault prognostic method in practical application are analyzed,on this basis,the combination of the two is analyzed,the necessity of application of data-driven fault prognostic method in CL2M system is emphasized.Finally,the failure prediction maintenance architecture of CL2M system is designed.2.The data-driven fault prognostic process is studied.The classical Support Vector Regression(SVR)algorithm and the novel gradient enhancement tree algorithm XGBoost are introduced,that lay the foundation for verifying the universality of the CL2M fault prognostic maintenance architecture.In order to improve the performance of prediction model,Artificial Fish Swarm-Particle Swarm Optimization(AFS-PSO)hybrid Optimization algorithm is introduced.3.On the basis of the failure prognostic architecture in the CL2M system,the limitation that adaboost algorithm can only be applied their own training data is broke,CL2M system dynamic modeling method is proposed,and the prediction modelupdate link is completed,the dynamic CL2M modeling is completed,and then a subsystem layer failure prediction scheme is designed.On the basis of the data-driven method modeling process,the different scenarios of fault diagnosis and fault prognostic are considered respectively,and the dynamic modeling of CL2M fault diagnosis and the dynamic modeling of CL2M fault prognostic are realized.4.The failure prognostic of the fault prognostic architecture subsystem in the CL2M system is tested.IEEE PHM 2015 Prognostic Challenge data set and NASA Ames lithium battery data set are introduced,respectively based on SVR and XGBoost algorithm,and AFS-PSO optimization algorithm,and the dynamic modeling of CL2M fault prognostic is specifically realized.Finally,compared with the other four modeling schemes,the results of the two experiments show the advantages of the dynamic modeling method proposed in this thesis.