Feature Extraction And Recognition Of Dual-rotor Misalignment State Based On Optimized DBN By Improved FOA

The internal and external dual-rotor are the key parts of modern advanced aeroengine.The misalignment is the main factor that affects the normal operation of the dual-rotor system of internal and external,and the misalignment fault is inevitable.Accurate identification of the misalignment state is an important way to ensure the safe and stable operation of the dual rotor system of aeroengine.At present,the main content of the research work on the misalignment fault diagnosis of dual-rotor system is to identify whether there is any misalignment fault and the degree of misalignment.Then,the research work seldom identifies the type(position)and the degree of the misalignment of the dual-rotor system,that is,the misalignment state,because the current traditional "shallow method" is difficult to accurately identify the misalignment state.In order to effectively and accurately identify the misalignment fault in the process of fault diagnosis of dual-rotor system,the deep belief network(DBN),which can effectively identify high-dimensional nonlinear and non-stationary signals,is studied from the following four aspects in this paper,combining with the data essential characteristics that can be deeply mined.1.Aiming at the vibration signal of dual-rotor misalignment state,the feature extraction and classification recognition ability of DBN are studied.In this paper,we study the feature extraction performance of RBM under different iterations,different number of hidden layer nodes,different learning rate and momentum to reflect the feature extraction ability of depth confidence network.The experimental results show that the number of iterations,the number of nodes in the hidden layer,the learning rate and the momentum term of RBM affect the feature extraction of the vibration signal of the depth confidence network,and the reasonable selection range of each parameter is obtained.At the same time,the classification and recognition ability of DBN is reflected by the classification results of DBN in different network depths.The experimental results show that DBN has the ability of vibration signal classification and recognition under different network depths,and the network depth with the best classification and recognition ability is obtained.2.Aiming at the problem of DBN structure parameter optimization,we use the fruit fly optimization algorithm(FOA)to optimize the structure parameter of DBN and realize more accurate fault identification of DBN network model.An improved adaptive FOA algorithm is proposed to solve the problems of fixed search step,easy to fall into local optimal solution and limited application of the algorithm.The experimental results show that the algorithm can obtain the optimal parameters quickly and accurately,which is suitable for the optimal selection of DBN network structure parameters.3.Aiming at the problem of non signal DBN input vector construction,a fault state identification method based on VMD improved FOA optimization DBN is proposed by usingthe variable mode decomposition(VMD)with better fault feature extraction ability as the method of DBN input vector construction.The experimental results of single sensor show that the method is more suitable for the identification of misalignment than other methods.4.In view of the limitation of single sensor collecting vibration data in the practical application of aeroengine vibration signal collection,this paper uses D-S evidence theory to identify the fault of multiple sensors based on VMD improved FOA optimized DBN single sensor fault state recognition method,and aims at the problem of evidence conflict in D-S evidence theory,An improved D-S evidence theory based on mutual information measure is proposed.Finally,a multi-sensor fault state identification method based on VMD improved FOA optimized DBN and improved D-S evidence theory is proposed.The experimental results show that the proposed multi-sensor fault state recognition method based on VMD improved FOA optimized DBN and improved D-S evidence theory improves the recognition rate of misalignment fault state and enhances the fault-tolerant ability of the diagnosis method.