Study On The Classification Method Of Aeroengine Gas Path State Based On QAR Washing Data And Neural Network

The aero engine is the core component of the aircraft and its safety status directly affects the entire cruise process of the aircraft.The implementation of health management of aero-engine not only ensures engine safety to a greater extent,but also shifts the traditional engine maintenance mode from regular maintenance to condition-based maintenance.In addition,the cost of airlines can be reduced significantly.This paper focuses on the performance of engine gas path,using traditional neural network and convolutional neural network to develop methods for classifying the state of gas performance in the engine.The main research contents of this thesis include the followings:In view of the lack of real operating data of the engine in the current research,this paper will cooperate with airlines to obtain QAR data and perform pre-processing,such as outlier deletion and smoothing processing,on gross errors and measurement errors that are contained in QAR data.At the same time,this paper will also extract stable points that can reflect the best flight condition of the whole cruise from the QAR data.Combine with stable points and washing records in the maintenance report provided by the airline,a data set which can be utilized to study the classification diagnosis of aero engine performance will be constructed.Based on the aero-engine status classification diagnosis data set,we explore the classification accuracy of the method that is combined with the feature selection and the MLP neural network.The feature extraction algorithm can extract the relevant parameters that are most beneficial to the classification from the classification diagnosis data set and avoid the information interference of the unrelated parameters in the data set.The extracted relevant parameters can be used as the input of the MLP neural network.In addition,the data set would be divided into the training set and the test set.After the division,the optimal classification model is trained by the training set,and then the accuracy rate can achieve 86.5%in the test set.Aiming at the successful application of current deep learning methods in many fields,it is also attempted to introduce it into the engine status classification research in this thesis,mainly using convolutional neural network method.The convolutional neural network has a more powerful feature extraction function than the traditional neural network.Therefore,the process of feature selection and parameter selection is no longer needed.At the same time,the dual-tree complex wavelet method is also introduced to combine with the convolutional neural network.The dual-tree complex wavelet method can further filter out possible errors in the data set and implement multiple decompositions on the data set to expand the data set.The combination of convolutional neural networks and dual-tree complex wavelets will ultimately be verified on the aero-engine status classification dataset and achieve 92%accuracy.