Life Prediction Of Aero Engine Based On The Construction Of Health Assessment Index And Nonlinear Regression

As one of the most important parts of an aircraft,once an aero engine fails,it will directly affect flight safety,cause economic losses and even endanger the lives of personnel.However,aero engines are often operated for a long time under severe conditions such as high temperature,high pressure and high load.Therefore,research on aero engine failure prediction and health management is of great significance for ensuring the safe and reliable operation of engines and reducing maintenance costs.Based on condition monitoring parameters,this paper adopts a data-driven method to study the health index(HI)construction and remaining useful life(RUL)prediction problems in the research of aircraft engine performance degradation.The main contents are as follows:(1)Aiming at the problems of complicated aero-engine health monitoring data and different characterization capabilities of engine performance degradation,the gray correlation analysis method is utilized to construct HI of aero-engine.Through parameter screening and weighted fusion,the monitoring parameters that have a higher correlation with the engine degradation trend can be strengthened,and vice versa.The experiment verified the good characterization ability of the HI obtained by this method for the engine degradation trend.(2)Aiming at the problems such as high noise content and unobvious degradation trend of aero engine health monitoring data,the kernel canonical correlation analysis algorithm based on stacked denoising auto encoder(SDAE-KCCA)is applied to realize the construction of aero engine HI.The SDAE network is used to capture the deep features related to the engine degradation state,so that the KCCA algorithm reduces the dimensionality of the deep features to obtain the HI that best characterizes the engine degradation trend.Through experimental comparison,the effectiveness of the HI constructed by this method is verified.(3)Aiming at the problems of complex structure of aero-engine system,strong uncertainty,and few historical monitoring data,a full-order time power gray model(FOTP-GM(1,1))is used to predict engine RUL.The model overcomes the lack of historical performance degradation data through gray theory,and can adaptively update the performance degradation trajectory according to the input data to improve the prediction effect.Taking into account the differences of different engine failure thresholds,the sliding window Euclidean distance comparison method is applied to determine the HI failure threshold.In the RUL prediction experiment,this method showed a good prediction effect.(4)Aiming at the problems of aero-engine system with strong nonlinearity and lack of effective information that can characterize its degradation trend,the radial basis kernel function and polynomial kernel function are mixed to construct a kernel ridge regression(KRR)algorithm.After optimizing the algorithm through the gray wolf optimization algorithm,a RUL prediction model is established.This model has a strong nonlinear fitting ability and generalization performance,which can obtain a more accurate fitting effect.In addition,a life adjustment function based on the KL distance life is used to comprehensively determine the failure threshold of the test engine,thereby reducing the RUL prediction error.Through experimental comparison,it is verified that the method has high prediction accuracy.(5)Aiming at the different degradation modes of different aero-engines,the high cost of acquiring monitoring parameters,the few historical state parameter samples,and the difficulty in learning degradation trends,a combination of data migration and neural network methods are utilized to achieve aero-engine RUL prediction.First,compared with the data migration algorithm based on joint adaptation distribution,the data migration method based on dynamic time warping is selected to improve the similarity of different engines HI and complete the expansion of the training sample set;secondly,a bidirectional gated recurrent unit with self-attention(Bi GRU-SA)model is established to fully learn the engine Degradation trend information;finally,experiments are carried out on data with multiple working conditions and different time series lengths,verifying that this method has higher prediction accuracy and better prediction effect than long short-term memory(LSTM)and time convolutional network(TCN)models.