A Study On Fault Diagnosis Method For Gearbox Based On LSTM

As one of the most common part of mechanical equipment,the gearbox is widely applied.It is also a component with a high failure rate,whose reliability will affect the performance of transmission system and the entire device.By analyzing the vibration signal of the gearbox,the working status and fault could be monitored and diagnosed,so that the maintenance and replacement of the gearbox can be carried out in time,which ensure the reliable and safe operation of the equipment.In this paper,the feature extraction and pattern recognition methods based on the LSTM cyclic neural network model are discussed,which take the vibration signal of the gearbox as the research object.Firstly,a new signal processing method SWD decomposition is introduced and applied to extract features of gearbox vibration signal.Then the SWD decomposition is used to obtain the time-frequency spectrum of vibration signal,which is send into the LSTM model for fault pattern recognition.To get rid of the defect of traditional fault diagnosis mode that are excessively dependent on artificial feature extraction,an end-to-end fault diagnosis and pattern recognition method based on DB-LSTM model is proposed.Finally,CNN-LSTM model for the fault diagnosis issues with strong background noise is proposed,and it combine with CNN's ability to extract spatial features and LSTM's ability to extract the time series related information to automatically extract the depth features and improve the antinoise ability of the model.The validity and superiority of the proposed methods are verified by experimentally analyzing the convergence speed,generalization ability,accuracy and stability of the model.The main research contents of the paper are as follows:1)A gearbox fault diagnosis method based on SWD-LSTM model is proposed.First,a new signal processing method named as SWD decomposition is introduced,and simulated signal is applied to analyze the frequency discrimination ability of the SWD method.Then,a fault diagnosis method for gearbox compound fault based on SWDAVDIF is proposed with the theory of morphological filtering,and the method is verified by simulation signal and experimental signal.Finally,a fault diagnosis model for gearbox fault pattern recongnition based on SWD-LSTM is proposed.2)Combining the artificial intelligence method named as Deep Learning(DL),a gearbox fault diagnosis model based on DB-LSTM is proposed,which can process the original signal and extract the intrinsic expression of the signal.First,the effect of networks with different input size and network structure on the recognition task are analyzed,and the input size and the network structure of the model are designed.Second,the DB-LSTM's noise immunity is analyzed,and the superiority of the proposed method is verified by analyzing the accuracy,stability,convergence speed and generalization ability of the model.3)The gearbox fault diagnosis method based on CNN-LSTM is proposed to solve the problem that the recognition effect of fault diagnosis model is degraded by strong background noise.The model's feature extraction layers are composed of a CNN layer with spatial feature extraction capability and a LSTM layer with timing feature extraction capability,so that the model has the ability to extract deep fault features from low SNR vibration signals.The CNN-LSTM is applied to the gearbox fault diagnosis,and the superiority of the method is verified from the aspects of accuracy,stability,generalization ability and convergence speed.Finally,the depth features extraction ability of the proposed model is demonstrated by visualizing and analyzing the depth features.