Research On Remaining Useful Life Prediction For The Air Turbine Starterof Civil Aircraft

Starter is an important accessory of aero engine.In the process of operation,it transmits torque rejection and drives the engine to accelerate from standstill to idling.It can be seen that the dynamic characteristics and remaining useful life of the starter have a significant impact on the reliability of the engine and the aircraft.Therefore,the research on remaining useful life prediction for the air turbine starter of civil aircraft has become a very challenging and theoretical research topics.The thesis mainly completes the following four aspects of work:Firstly,through the principle analysis of the Relevance Vector Machine(RVM),there are three problems that need to be solved urgently,namely the poor accuracy of long-period prediction,the lack of dynamic updating of the prediction physical model,and the slow model training speed.Then based on Bayesian theory,the calculation method of the physical model of the correlation vector machine is deduced.Secondly,remaining useful life prediction for the air turbine starter based on relevance vector machine and empirical mode decomposition(RVM-EMD)is proposed.The EMD technology is used to decompose the non-stationary sample data,and the decomposed Intrinsic Mode Functions(IMFs)are evaluated and selected,then the stationary modeling data is obtained.In addition,the Subtraction Clustering Method(SCM)is introduced to perform clustering analysis on the filtered IMFs and extract the cluster centers,thereby ensuring that the fusion framework always maintains a good matching performance and effectively improves the prediction accuracy of the entire model.Then,an optimized incremental learning relevance vector machine model is proposed.The optimized algorithm can not only refresh the RVM framework model timely by using the incremental learning technology,but also perform an accurate assessment of the quality of the most original sample data by means of Sample entropy technology,so as to calculate more accurate data with a longer period.Finally,the proposed two algorithms,RVM-EMD and Optimized-RVM,which have positive significance for improving the prediction model,are verified in the simulation experiment,and the different algorithms are compared and analyzed in the same experimental conditions.The results prove that the two algorithms mentioned above are more effective.