Aero Engine Fault Diagnosis Technique Based On Data Mining

In recent years,the average growth rate of domestic civil aviation fleet is about 10%,showing a gradually stable development trend,but the number of mechanical accidents in civil aviation accidents also shows an upward trend,and the accident caused by engine failure accounts for more than 50% of the total failure.From the data shows that most of the problems in the process of aircraft in service related to aero engine.The research on the fault diagnosis of aero engine can better guarantee the safety and reliability of aircraft in service.This paper will focus on the problem that the accuracy in the current aero engine fault diagnosis is difficult to meet the demand,and carry out the research on the aero engine fault diagnosis based on data mining technology.Aiming at the problem of noise in aero engine performance parameters,the wavelet threshold denoising method is studied.In the process of fault location,considering the problem that traditional convolution operation cannot effectively carry out global feature interaction,the convolutional neural network fault diagnosis method based on attention mechanism is studied.For problems with small samples or no samples,cross-model fault diagnosis methods based on transfer learning are studied.The specific methods are as follows:In consideration of the problem that the noise in the aero engine performance parameters will drown the useful signals that can represent the faults to a certain extent,the wavelet threshold denoising method is introduced to denoise the random noise in the aero engine gas-path performance parameters.In view of the problem that the wavelet threshold denoising effect is not good,the problems of additional oscillation and constant deviation existing in the soft and hard threshold functions are analyzed.Based on the idea of compromise,a new threshold function is proposed to reduce the constant deviation and additional oscillation while retaining the overall trend and detail characteristics of the signal.Finally,the effectiveness of the proposed improved compromise threshold function wavelet threshold denoising method is verified by using the shared data set and the actual performance parameters of the aero engine.At present,convolutional neural network has been widely in many fields.However,the current fault diagnosis method based on convolutional neural network usually only deals with local regional features,and it is difficult to effectively capture their long-range interaction,resulting in low diagnostic accuracy.Therefore,attention mechanism is introduced to enhance feature interaction and improve the accuracy of multi-feature correlation fault diagnosis.In order to solve the problem that typical attention mechanisms lack sufficient interaction on input features in specific dimensions,multiple attention mechanism is proposed by integrating non-local modules with channel attention mechanisms,which are embedded into the convolutional neural network to form MACNN.Finally,the validity of the proposed fault diagnosis method is verified by using the engine service data of an airline.In order to solve the problem of poor accuracy of traditional machine learning in the fault diagnosis task of cross-model aero engine,a transfer learning method was introduced,and a large number of fault samples of an aeroengine were taken as the training set to complete the corresponding pre-training.In the training process,the batch nuclear-norm maximization method is used to constrain the correlation degree of the data in the source domain and the target domain,so that the model performs data mining in different domain simultaneously.The model is fine-tuned to realize the migration of fault diagnosis model across models.Finally,the validity of the proposed fault diagnosis method based on transfer learning is verified by using different types of aero-engine fault data sets.