Resesrch On Gas Path Fault Diagnosis Technology For Civil Aero-engine Based On Deep Feature Mining

The civil aero-engine is multi-fault aeronautical machinery.Fault diagnosis for the engine can not only avoid catastrophic failure,but also ensure flight safety.However,problems such as: large individual differences in the engine,few fault samples,and high data noise make existing fault diagnosis methods unable to meet the needs of airlines.Therefore,it is of great significance to develop a method for fault diagnosis for civil aero-engine.In this paper,the following research work is carried out on the fault diagnosis technology of civil aero-engine based on deep feature mining.First of all,aiming at the random noise and gross error existing in the performance parameters,the research on the preprocessing method of gas path parameters has been developed.A method to remove gross errors of grouped parameters by using GRB criterion is proposed.The denoising method based on Empirical Mode Decomposition(EMD)and Singular Value Decomposition(SVD)is used to denoise the data whose rough error has been removed.Firstly,the trend component of the data is extracted from the data by using EMD algorithm,and a trend component extraction method based on the correlation is proposed.Than the SVD algorithm is used to denoise the detrened data.Finally,the simulation signal and the actual exhaust temperature margin(EGTM)sequence are used to verify the above method.The results have proved the effectiveness of the above method.Secondly,the gas path fault diagnosis method of civil aero-engine based on SDAE has been studied in this paper.According to the analysis of Customer Notification Report(CNR),significant changes can be found in performance parameters when the engine fault occurred.In order to extract the characteristics of the change of performance parameters,an state feature extraction model is established by using Stack Denoising Auto-Encoder(SDAE)in this paper,and a method for determining the number of nodes in hidden layer based on the feature extraction capability of single Denoising Auto-Encoder(DAE)has been proposed.Then the established feature extraction model is used to extract the features of the fault samples,which will be used as the input of the support vector machine(SVM)for fault diagnosis.The method has been applied to multiple fault classification of engine,and the experimental results show that the method has a high diagnostic accuracy.Then,the gas path fault diagnosis method of civil aero-engine based on CNN is studied.By analyzing the actual fault data of engine,the relationship between the parameters also has a significant impact on fault diagnosis.In order to take into account the correlation and time sequence of parameters,the Convolutional Neural Network(CNN)is used to establish the state feature extraction model in this paper.Then the established model is used to mine the deep feature of fault samples,which will be used as the input of SVM for fault diagnosis.The method has been applied to multi-fault classification of engine,and the results show that the method has an excellent diagnostic ability.What's more,the influence of structure parameters of CNN on diagnostic accuracy is analyzed.Finally,according to the research content of this paper,the core business components of the gas path fault diagnosis of civil aero-engine is designed and developed on the basis platform of customizable aero-engine health management and maintenance decision support system,which provides technical support for the performance parameters preprocessing and gas path fault diagnosis of the civil aero-engine.