Research On Remaining Life Prediction Of Turbofan Engine Based On Feature Fusion And Wiener Proces

As the core component of aircraft power system,the stability of turbofan engine’s related performance is an important cornerstone to ensure voyage safety.However,the long-term exposure of the engine to high temperature,high pressure and high vibration will inevitably lead to the deterioration of its performance,which will seriously threaten the safety of life and property.In order to avoid safety risks and save maintenance costs at the same time,it is imperative to establish an efficient prediction mechanism for the Remaining Useful Life(RUL)of turbofan engines.Thanks to the rapid development of sensor technology and data statistical analysis in recent years,the research method of predicting the RUL of turbofan engine by using stochastic degradation model is surging.In this paper,based on the data-driven method,combined with signal processing,feature extraction and various fusion methods,and using different forms of Wiener random degradation process,two prediction models,ES-Kpca AE-Wiener and ES-Kpca AEoff Non Wiener-on Non Wiener,are designed,and the RUL of turbofan engine is obtained.The main research contents of this paper include:1.Screening and processing of monitoring signals of turbofan engine.Aiming at the poor quality of monitoring data of multi-source sensors in turbofan engine,this paper analyzes and screens the monitoring data of multi-source sensors from three angles: expert experience,evaluation index and data visualization.Then,in order to solve the problem that the standard exponential smoothing method has poor data smoothing effect in the initial stage,this paper adopts the improved form of exponential smoothing method(ES)to denoise the filtered data,and explores the influence of the two exponential smoothing methods before and after the improvement on the processing effect under different smoothing coefficients.Experiments show that the improved exponential smoothing method can retain more performance degradation information of the lower turbofan engine in the initial stage.2.Study on degradation modeling and residual life prediction of turbofan engine based on feature fusion and Wiener process with linear drift coefficient.In order to solve the problem of poor prediction accuracy,a single feature extraction method is mostly used to obtain the Health Index(HI)in the research of engine RUL prediction by using performance degradation modeling analysis method.In this paper,Kernel Principal Component Analysis(KPCA)and Auto-Encoder(AE)are used to extract features from normalized data to obtain feature vector groups,and the weighted average method is used to effectively fuse the feature vectors to construct a Composite Health Index(CHI).Then,combined with Wiener process with linear drift coefficient,the degradation model of the CHI is analyzed,and the RUL of turbofan engine is obtained.The root mean square error,scoring function and accuracy of the engine at different times are 31.9572,2947.446 and 51.42%,respectively.The experiment shows that the CHI obtained by the weighted average method can better characterize the performance degradation of equipment and avoid falling into the dilemma of failing to build HI.3.Study on degradation modeling and RUL prediction of turbofan engine based on feature fusion and Wiener process with nonlinear drift coefficient.Aiming at the limitation of constructing CHI by weighted average method and the uncertainty of engine in actual operation,this paper first uses off-line model off Non Wiener process with nonlinear drift coefficient to evaluate the useful life of engine,and puts forward a method to determine the fusion coefficient of each feature vector by minimizing the mean square error between the predicted value and the real value of engine useful life,and constructs a new CHI.Furthermore,combined with the on-line model of Wiener process with nonlinear drift coefficient(on Non Wiener),the RUL prediction of turbofan engine is realized.The root mean square error,scoring function and accuracy of the engine at different times are 11.8492,550.4628 and 72.64% respectively.The experiment shows that the CHI evaluation value obtained by this method is higher,and combined with Wiener process with nonlinear drift coefficient,the RUL prediction accuracy is better.