Research On Fault Diagnosis Method Of Rolling Bearing Based On Deep Learning

As a key component of rotating machinery,the running state of rolling bearing is directly related to the reliable operation of the whole equipment.With the continuous large-scale and complexity of rotating machinery,the vibration signal of rolling bearing also shows nonlinear and non-stationary characteristics,which increases the difficulty of traditional methods to extract fault features.Deep learning network can realize the autonomy of data feature extraction and the intelligence of classification,which has very important theoretical and practical value in the field of fault diagnosis.Aiming at the limitation of traditional fault feature extraction method,this paper constructs the fault diagnosis model of rolling bearing by using the framework of deep learning,mainly studies the structure improvement and parameter optimization method of deep convolution neural network.Based on the improved deep convolution neural network model,the vibration signal is input into the model to train the feature extraction ability of the model,then input the extracted feature into the classifier to complete the fault diagnosis process.The research content of this paper is as follows:(1)Because the two-dimensional structure(convolution kernel,pooling kernel,etc.)of the convolution neural network is not consistent with the one-dimensional characteristics of vibration signal,this paper adjusts the two-dimensional structure of conventional convolution neural network to one-dimensional,proposes an improved one-dimensional convolution neural network(1DCNN)to adapt to the input of one-dimensional vibration signal,and then uses 1DCNN to extract the rolling bearing vibration signal fault characteristics.In this process,1DCNN ’s structure design and training parameters optimization are completed.Combined with the training mode of learning rate exponential decay,realize the method of accurately diagnosing fault categories relying on bearing vibration signals.Experiments show that the 1DCNN fault diagnosis model has strong feature extraction abilities,and the accuracy of fault diagnosis model is as high as 99.42%,and it is able to meet high demand of fault diagnosis.(2)Gabor filter has good resolution in time domain and frequency domain,which can reduce the network depth of 1DCNN.Therefore,a fault diagnosis model based on Gabor filter combined with one-dimensional convolution neural network(G-1DCNN)is proposed,that is,Gabor filter is used to enhance the feature extraction ability of 1DCNN without affecting the accuracy of fault diagnosis,and at the same time operation speed of the model is improved.The experiment shows that when the software and hardware conditions and training parameters remain the same,the feature extraction ability of the network is greatly enhanced by adding Gabor filter,which can still achieve a high fault diagnosis accuracy(99.72%)by reducing the network depth,and has a good fault diagnosis effect.