Research On Aero-engine Reliability Analysis Based On Deep Learning

As an important method of assessing the reliability of aero-engines,the reliability of aero-engines has a significant impact on civil aviation safety.In order to carry on the aero-engine reliability analysis more accurately,based on the operational reliability analysis and the complex reliability analysis theory,this dissertation deal with the non-linear and time-varying problems between the state parameters of aircraft engines and the operational reliability by deep learning.And this method makes corresponding improvements for the reliability analysis of aero-engine.The details are as follows:1)The framework of aero-engine reliability analysis is studied.Combining with its characteristics,this paper studies the reliability analysis of complex system and reliability analysis respectively,and how to collect and analyze the monitoring information of aero-engine on this basis.This paper also studies the condition monitoring method of aero-engine and deep learning and puts forward to adopt Denoising Autoencoder(DAE)and its training algorithm-greedy layer-wise pre training.2)Based on the improvement of DAE,the gas path fault analysis for aero-engine has been researched.The problem is that state parameter is non-linear and it is susceptible to noise pollution.In order to solve the problem,a new gas path fault analysis method based on DAE is proposed.Based on the DAE,this method uses the improved firefly algorithm(FA)to improve the Radial Basis Function(RBF)network to analyze faults.An example is given to compare the proposed method with the original DAE,FRBF,Support Vector Machine(SVM)and RBF.The results show that the proposed method has the highest analytical accuracy and robustness is better than the other methods.3)Aero-engine performance degradation assessment based on DAE has been studied.Then,aiming at the form and regularity of aero-engine performance degradation,aero-engine performance degradation assessment based DAE on method is proposed.Aimed at the collected parameters of aero-engine monitoring,the DAE is adopted,the algorithm of greedy layer-by-layer training is used,and the performance degradation assessment is carried out.An example is given to compare the proposed method with Back Propagation(BP)neural network and SVM.The results show that the proposed method is improved in accuracy,robustness and can prevent the problem of overfitting with small sample.4)Studied the prediction of remaining useful life of aero-engine based on deep learning integrated network.Aiming at the problem that the remaining useful life of aero-engine needs a more nonlinear model and the cumulative effect is not considered,integrated network based on DAE and Deep Belief Network(DBN)has been proposed.FA is selected to optimize the weight of the output of the result.By comparing with the results obtained from DAE,DBN,BP network and SVM,the proposed method is more accurate.