Gear Fault Diagnosis Method Based On Adaptive Signal Decomposition And Deep Learning

With the development of science and technology and changes in social needs,modern industrial production has increasingly higher requirements for mechanical equipment.As a very important transmission device in mechanical equipment,gears are often operated in a relatively harsh environment.If they are not well maintained or properly operated,the gears are prone to wear,deformation,and broken teeth.Once the gear fails,the mechanical equipment will not be able to operate normally,which will lead to a reduction in production efficiency,and may even cause a series of serious accidents.Therefore,for today’s engineering applications and academic research,the development of an efficient and accurate gear fault diagnosis technology is of great significance.In order to design a more scientific method for gear fault diagnosis,this thesis deeply studies the mechanism of gear vibration,as well as the time-domain and frequency-domain characteristics of different fault gear vibration signals.These theories provide sufficient theoretical basis for subsequent data processing methods and classification model design.In this thesis,the vibration signal of gearbox is studied,and two fault diagnosis schemes based on adaptive signal decomposition and deep learning technology are proposed.The first scheme uses the time domain data of the gear vibration signal,and the second scheme uses the frequency domain data of the gear vibration signal.The main contents of the two schemes are as follows:(1)In order to solve the problem that the components of the vibration signal are complex and the effective information is interfered by a large number of disorderly signals,this thesis proposes a method to reduce the noise of the gear vibration signal by using Empirical Mode Decomposition(EMD)according to the distribution characteristics of various components of the gear vibration signal in frequency domain.In combination with the advantages of Convolutional Neural Networks(CNN)and Long Short-Term Memory(LSTM),this thesis designs a CNN-LSTM to classify noise reduction signals.The final experimental results show that the accuracy of this method for gear fault diagnosis is 98.93%.(2)Since the frequency domain characteristics of the vibration signals of gears in different states are mainly reflected in the rotation frequency,rodent frequency and harmonic frequency,etc.,a fault diagnosis method based on the combination of Variational Modal Decomposition(VMD)and Wide+Narrow Visual Field Neural Networks(WNVNN)was proposed.First,use the variational modal decomposition to extract the features of the original signal,and then use the wide and narrow field of view neural network proposed in this thesis to classify it.Subsequent experimental results show that,compared with other classification schemes based on deep learning proposed recently,the scheme proposed in this thesis not only has higher classification accuracy,and the average classification accuracy is 99.62%,and furthermore the classification performance is more stable.