Research On Fault Prognosis Methods Of Health Management For Flight Vehicle Key Components

Through the monitoring of the working status of key components,diagnosis and prognosis for potential or existent faults,health management(HM)technologies can provide guidance and advice for the maintenance of system.It is important to ensure safe operation of system,avoid accidents,and maximize the performance of system.Fault prognosis is an important theme in the architecture of vehicle HM,it plays a very crucial role for reducing the use and maintenance costs,improving security and reliability.There are some problems such as that the disturbance from uncertain factors,lack of prior knowledge,low prediction accuracy,and uncertainty management,etc.Accordingly,this dissertation takes the key component of helicopter transmission system and lithium-ion battery in spacecraft as objects and researches on fault prognosis methods oriented to vehicle HM.The detailed contents and innovative work can be summarized as follows.1.The phenomenon of phase space warping caused by slow-time variable process of mechanical damage are deeply analyzed,and a damage tracking method based on phase space warping and smooth orthogonal decomposition is presented.Fast-time system dynamics is coupled to a slow-time damage evolution in a hierarchical dynamical system.By quantifying the phase space warping,the tracking function and tracking matrix for damage evolution is built.Degradation and operation variety in the tracking matrix are separated by smooth orthogonal decomposition.The research shows that the method based on phase space warping can track the evolution trend of the rotating machinery fault without the influence of operation condition variety,meanwhile,it has large amount of computation and difficult to quantitative the tracked trend.2.Relevance vector machine(RVM)based degradation state recognition method under the complex conditions is deeply studied.(1)An adaptive threshold model for health monitoring of bearings under varying operation conditions is proposed.RVM is used for regression of the relationships between the adaptive parameters of the threshold model and the statistical characteristics of vibration features.The health status of bearings can be indicated via detecting whether vibration features exceed the adaptive threshold.(2)A Thevenin equivalent circuit model is presented.By identifying the relationship of state of charge(SOC)compare to battery terminal voltage,charging/discharging current and cycles,the SOC estimation model under various operation conditions can be constructed with RVM.The research shows that the RVM regression method has good sparsity,and can solve the problem of fitting during the progress of recognition model building and the overfitting phenomenon.Validity of the methods is validated with bearing and lithium-ion batter data.3.On the basis of RVM regression method,a prediction method combining degradation model and RVM is proposed.Take the historical observation data for least square fitting to determine the appropriate degradation model.And RVM are used for extracting relevance vectors from feature series,the sparse data sets consisted of these relevance vectors are used to fit the degradation data and extrapolate by the degradation model.According to the time of feature series reach a specified failure threshold,remaining useful life of the equipment can be estimated.The research shows that the prediction frame combined the degradation model and RVM requires a higher consistency of the trend of feature and fault degradation.It has higher prediction accuracy for medium or short-term forecasting,and with a small amount of calculation for on-line prediction.4.In order to overcome the shortcomings of the basi particle filter(PF)in selection of the priori prediction model and the determination of the initial distribution of model parameters,a kind of general prediction model as well as a prediction framework of PF based on tracking degradation rate for predicting degradation trend is presented.In the proposed method,the statistical rule of historical data is utilized to guide the tracking of degradation rate and simplify the prediction process.Prediction method is validated by remaining useful life prediction for bearing and lithium-ion battery.The research shows that,compared with the basic PF prediction method,the proposed method has the simplified implementation process and the better generality,and with complete utilization of the useful information in historical data,it can obtain higher prediction accuracy and valuable uncertainty information.In summary,this dissertation is concerned with the key problems of fault prognosis in the vehicle health monitoring.Several efficient methods based on phase space warping,relevance vector machine and particle filter is developed for solving the problems of damage tracking,degradation state recognition under complicated operation,improving the accuracy of prediction of remaining useful life,and providing valuable uncertainty information for prediction.Finally,the effectiveness of the proposed method is verified by the experiment.