Research On Multi-sensor Data-driven Approach To Aero-engines Health State Assessment And Remaining Useful Life Prediction

The health assessment and remaining useful life(RUL)prediction are often based on the multi-sensor data.The existing research always neglected the correlation and imbalance of multi-sensor data.The uncertainty representation for RUL prediction model is also lacking,which will lead to the problem that degradation information is not sufficiently mined and the accuracy of the prediction results model does not provide a valid reference for management decisions.This thesis studies the autocorrelation and imbalance of multi-sensor degradation data of aero-engines,extracts effective degradation features,fuses the features to obtain health index(HI)curves that can characterize the health state of aero-engines,and construct RUL prediction models to provide data support for the formulation of maintenance strategies.The main research contents include the following three aspects:(1)An aero-engine health condition assessment method based on common dynamic principal component analysis and multi-sensor linear weighted fusion model is proposed.In order to address the autocorrelation and imbalance of the multi-sensor data,on the basis of principal component analysis,common dynamic principal component analysis(CDPCA)is developed for feature extraction.A linear weighted fusion model considering four metrics: time correlation,monotonicity,consistency,and robustness is then utilized to fuse the degradation features and construct a one-dimensional health index(HI)curve to assess the health state of aero-engines.The results of the case study show that the HI curves obtained by the method proposed in this paper outperform single sensor data and HI curves constructed by other methods in all metrics.(2)An aero-engine RUL hybrid point prediction method that takes into account the variability of the initial degradation state is proposed.For the case of aero-engines with varying degrees of initial degradation,the K-means clustering method is used to classify and identify the initial degradation state of aero-engines based on the constructed HI curves.The RUL prediction model is trained separately for engines in different initial degradation states to improve the overall prediction accuracy.A hybrid RUL prediction model combining long shortterm memory(LSTM)network and support vector regression(SVR)is proposed developed to predict the RUL.The results of the case study show that the RUL prediction method based on initial degradation state recognition proposed in this paper not only outperforms the prediction results without initial degradation state recognition but also outperforms other single machine learning and deep learning prediction models.(3)An interval prediction method for aero-engines based on kernel density estimation is proposed.The prediction errors of the point prediction model at different degradation states are fitted to the distribution using kernel density estimates,the inverse cumulative distribution function is calculated and the prediction interval belonging to each degradation state is calculated.In the prediction period,the engine is first identified as to which degradation state it is in at this time and the prediction interval of the degradation state is added to the point predicted value,which provides the RUL prediction model the ability to output intervals.Finally,a case study is utilized to verify the performance of the prediction intervals constructed by the proposed methods in this paper on interval coverage probability(PICP),interval width(PINAW),and coverage width-based criterion(CWC)with existing method.In conclusion,the health state assessment and RUL prediction method proposed in this paper accurately assesses the health state of the aero-engine and provides the prediction interval of RUL,which enriches the theory of aero-engine PHM.The research provides powerful data support for the formulation of aero-engine maintenance strategy,reduces the failure rate and operation and maintenance cost of aero-engine,and has important theoretical significance and practical engineering application significance.