Font Size: a A A

Prognostics Method Study Of Complex Engineering System

Posted on:2021-02-11Degree:DoctorType:Dissertation
Country:ChinaCandidate:R H JiaoFull Text:PDF
GTID:1368330605454585Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
With the rapid development of technology and modern industry,the equipment in the fields of machinery,energy,petrochemical,transportation and national defense is getting larger,more integrated and more automated.Along with these,the requirements for the safety and reliability of equipment operation are increasing,which makes the corresponding maintenance strategy change from traditional breakdown maintenance and preventive time-based maintenance to condition-based maintenance.However,most of the systems and equipment still rely on preventive time-based maintenance strategy,which not only consumes resources but also has low efficiency.Therefore,it is necessary to develop the technology of prognostics and health management(PHM)to realize condition-based maintenance for systems,which has attracted increasing attention recently.Remanining useful life(RUL)can be estimated in terms of history measurement data,which is useful for improving maintenance schedules to avoid catastrophe and save cost.The RUL prediction is the core process of PHM.Based on the demand of prognostics and health management technology for complex engineering system,this study aims to comprehensively improve the safety and reliability of complex engineering systems.It focuses on the theoretical research,application verification of the health state assessment and RUL prediction methods,and tries to provide an important decision support for the realization of condition-based maintenance for complex engineering systems.The main contents of this work include following aspects:(1)To solve the problems that the degradation state of complex engineering systems and the severity of fault are difficult to be observed,an integrated and visualized scheme for health indicator construction,health state identification,and safety level assessment is proposed.In order to construct a health indicator that can effectively reflect the health degradation trend of complex engineering systems,an unsupervised health indicator construction method based on deep belief network(DBN)is proposed.Then the continuous hidden Markov model is employed to achieve accurate recognition of health state.Following that,a fuzzy comprehensive safety level evaluation method based on health indicator and health state recognition results is proposed to assess the safety level of complex system.At last,a visualization platform is designed to reflect the current health state and safety level of the system conveniently and intuitively.(2)Aiming at the characteristics of complex engineering system with complex structure,numerous parameters,and strong nonlinearity,a RUL prediction strategy combining data-based method and model-based method is proposed.Firstly,based on the proposed deep recurrent neural network which can extract fine-grained feature and coarse-grained feature,the degradation trend of health is accurately identified.Then the state space model of degradation features can be consrtucted based on the health indicators.At last,the particle filter algorithm is introduced to predict the RUL of the system.Besides,since the particle degradation phenomenon in the particle filter algorithm will lead to an inaccurate prediction result,a method of using conditional variational autoencoders to improve the particle filter algorithm is proposed,which is able to improve the prediction accuracy.(3)To solve the problems of variable coupling,multi-source and multi-symptoms of faults in complex engineering systems,a framework of fault identification and RUL prediction under multi-failure mode is proposed.In order to identify the degradation trend of multiple faults,a Gap-DBN model which can accurately extract the degradation characteristics of the system is proposed.Following that,the degradation features of multiple fault types can be separated,and each fault type can be described by a support vector data description model.Finally,the current identified fault type is modeled and the particle filter algorithm is used to estimate the RUL value and the confidence interval,so as to realize fault identification and RUL prediction under multiple fault modes.
Keywords/Search Tags:complex engineering systems, remaining useful life prediction, safety assessment, deep learning, multiple fault modes
PDF Full Text Request
Related items