Research On Key Technologies For Data-Driven Methods Of Remaining Useful Life Prediction Of Equipment

As the rapid increasing investments on the manufacturing of complex equipment in engineering areas such as the aerospace,the high-speed railway,and the nuclear power,the reliability and safety technologies involving the whole critical components and systems have been transforming from the previous sample condition monitoring to Prognostics and Health Management.Thus,to realize the proactive failure prognosis the reliability modeling and the remaining useful life(RUL)prediction have been playing the key roles to enable the maintenance activities transformation from the preventative to the predictive one.Meanwhile,due to the complexity of the system framework,the complicate degradation mechanism,and the significant coupling mechanisms between components,it is of great challenge to implement the effective and accurate health assessment for complex equipment,which is still one of the significant engineering topics.Motived by the current deficiencies in the reliability modeling and RUL prediction,in this dissertation the failure mechanisms and the fault patterns are thoroughly analyzed,based on which the research evolves according to the main framework combining the continuous-time Markov chain and the proportional hazards model(CTMC-PH).The method enabling the RUL prediction of the system featuring the multiple degradation states and the complex degradation mechanisms is proposed,as well as the corresponding calculation of health characteristics and the method of model parameter estimation.To demonstrate its practical superiorities,the theoretical method is applied to one of the critical power electronic components,i.e.,the power MOSFETs(Metal-Oxide-Semiconductor Field-Effect Transistors).Additionally,it is also proposed an RUL prediction method for the continuous degradation system subject to the hard failure and the dependent competing risks.The effectiveness and accuracy are comprehensively verified using the numerical case studies,which extends the CTMC-PH framework for practical applications.It demonstrates that the proposed methods provide a new perspective and a series of effective methods for the health assessment of engineering equipment.The main contributions are summarized as follows.1.A new method of the RUL prediction for complex equipment featuring the multiple degradation states and the complex degradation mechanisms is proposed.The system degradation and failure behaviors are integrated into the CTMC-PH framework and described using the whole life cycle transition probability matrix,based on which the health characteristics including the reliability function,the mean residual life,and its corresponding confidence interval are developed with the explicit closed-form.The effectiveness and accuracy are comprehensively verified using the numerical case studies,which significantly extends the CTMC-PH framework.One of the superiorities is that the explicit health characteristics are quite computationally efficient,which can be feasibly applied to the practical systems with the requirement of the critical power consumption.2.With the degradation and failure data of power MOSFETs from an accelerated testing experiment,the practical application of the developed theoretical health assessment method is demonstrated on the RUL prediction of power electronics.The degradation path of the power MOSFETs is modeled with the CTMC,which is then incorporated into the PH model as the covariate process to describe the hazard rate of the time to component failure.Regarding the discretization technique of the continuous degradation path,a k-means method is developed so that the degradation path can be represented using several degradation states.Also,the model parameter estimation method is developed.The results of the case study show that the developed CTMC-PH framework is applicable for the health assessment of power electronics,which also aims to provide an illustration for other practical applications.3.A new health assessment method for continuous degradation system subject to hard failure is proposed.The non-monotonic and the monotonic degradation behavior of the system are driven by the Wiener process and the Gamma process,respectively,which are then incorporated into the PH model as the covariate process to describe the hazard rate of the time to system failure.Regarding the health characteristics calculation,the discretization techniques involving the continuous degradation path and the continuous monitoring time are proposed,based on which the failure and the degradation behavior of the system can be integrated into the whole life cycle transition probability matrix,and the health characteristics are provided in the computationally efficient explicit form.Finally,the effectiveness of the proposed method is verified with a numerical study.4.A health assessment method for continuous degradation system subject to the dependent competing risks is proposed.The dependent relationship is facilitated using the degradation path,and the soft failure occurs when the degradation path exceeds the predefined critical level.The degradation path is incorporated into the PH model to describe the hazard rate of the hard failure.The failure and the degradation behavior of the system can be integrated into one model framework using the whole life cycle transition probability matrix and the health characteristics can be feasibly calculated.Finally,the effectiveness and accuracy of the proposed method are verified using the numerical study.