Research On Motor Bearing Fault Diagnosis Based On Time-frequency Graph And Deep Learning

The motor bearing is a key component in the generator equipment,and the working condition of the bearing will directly affect the generator equipment,and even the normal operation of the entire unit.Although the bearing vibration signal can directly reflect the operating data of the unit,because it is easily affected by various factors and the vibration signal is easy to cross-couple,directly collecting the eigenvalues of the bearing vibration signal will lead to lower bearing performance due to inaccurate eigenvalue acquisition.Fault diagnosis accuracy.In view of the above problems in bearing fault diagnosis,this paper combines timefrequency transformation with deep learning,and proposes a fault diagnosis method for motor bearings based on time-frequency diagram and deep learning.The research content of this paper can be divided as follows:First,in view of the difficulty in extracting the eigenvalues of bearing vibration signals and the inability of convolutional neural networks to properly perform diagnosis,a method was proposed to convert one-dimensional bearing vibration signals into two-dimensional timefrequency images and input them into convolutional neural network training.By analyzing the vibration characteristics of motor bearings and the causes of various bearing faults,the processing capabilities of short-time Fourier transform,wavelet transform and S transform on bearing signals are studied and compared,and the time-frequency transform method is used to improve the characteristics of motor bearing vibration signals.Compared with other timefrequency processing methods,the time-frequency map formed by the wavelet transform of the complex Morlet wavelet basis function has better eigenvalue expression ability for bearing vibration signals.Then,in view of the low accuracy of bearing fault diagnosis of Le Net-5 network under a single working condition,an improved method of Le Net-5 network is proposed.In this method,a new feature extraction layer is added to the Le Net-5 network to form a parallel feature extraction layer,and convolution kernels of different sizes are set in the two feature extraction layers to improve the network model’s ability to extract tiny features..And replace the activation function for the network model,add the Dropout layer and the adaptive parameter algorithm,to avoid the instability of the model due to too many features after adding the feature extraction layer.Through the experimental comparison,it is proved that the time-frequency graph is more suitable as the input signal to participate in the bearing fault diagnosis,and based on this,the fault diagnosis ability of the improved convolutional neural network is verified.Finally,in order to solve the problem of low accuracy of the network model due to the large difference in the distribution of eigenvalues of the training data when the improved network model is used for fault diagnosis of bearings across working conditions,a transfer learning algorithm and an improved Le Net-5 network are proposed.A combined network approach for bearing fault diagnosis across operating conditions.By using the maximum mean difference algorithm,the method calculates the distribution moments of the source domain data set and the target domain data set under different working conditions,reduces the distribution distance of the eigenvalues of the source domain and the target domain,achieves the effect of domain adaptation,and realizes the cross-working of the network model.Condition fault diagnosis.The experimental results demonstrate the effectiveness of the improved network model combined with transfer learning for bearing fault diagnosis across working conditions.