Finite Element Analysis And Fault Diagnosis Of The Aeroengine Bearing

Aeroengine,a complex non-stationary system,is the power source of the aircraft flight.As the aeroengine often works in high temperature,high pressure,high load and other complex environment,its performance and structure are required to be very high.Aeroengine bearings,the key components of the engine rotor system,are related to the safety of the engine and even the entire aircraft.Therefore,it is very necessary to do some researches on modal analysis,feature extraction and selection,fault diagnosis and reliability assessment of aeroengine bearings.These studies can provide the theoretical basis for the structure design and failure mechanism of bearings.At the same time,they can accurately and timely detect the potential and existential faults to guarantee the safe operation of the equipment.Before the signal data processing,modal analysis is carried out on bearingsU solid modelings in different fault conditions and different types to provide a certain amount of prior knowledge for the subsequent fault diagnosis of vibration signals.Advantages and disadvantages of feature extraction based on wavelet packet and Hilbert spectrum and singular value decomposition(SVD)are compared.And then they are used for the fault diagnosis and the reliability evaluation respectively.Reducing the dimension of singular value by genetic algorithm(GA)can not only ensure the integrity of fault information,but also provide sensitive input for training and testing of neural networks.Combing Hilbert spectrum and SVD with GA,a fault diagnosis model of wavelet neural network(WNN)is build.And GA is aslo applied to optimize the weights and thresholds of WNN.The final feature vector is input into BP neural network,RBF neural network,WNN and the optimized WNN for training and testing respectively.The reliability index is redefined from the view of signal's similarity.The reliability of bearings in three states is evaluated based on subspace similarity.The state matrix of the information is obtained by wavelet packet method.The Subspace of state matrix is constructed by Kernel Principal Component Analysis(KPCA).The kernel of sub space inner product matrix is obtained based on SVD.The first nuclear protagonist normalized is treated as the operational reliability index.The process that the failure makes the reliability decrease effectively is reflected by reliability curve.In order to make researches of this thesis become systematic and practical,the GUI graphical user interface of the fault feature extraction system based on Hilbert spectrum and SVD and the fault diagnosis system based on WNN and RBF Neural Network of bearings,is set up.