Research On Fault Diagnosis Method Based On Deep Learning And Image Processing

Motor rolling bearing is an important part of rotating machinery.Once the fault of motor bearing occurs,it is easy to cause serious mechanical accidents.Therefore,it is of great significance to study the fault diagnosis method of motor bearing.Because of the motor vibration signal extracted is non-stationary,the method of extracting features only based on time domain or frequency domain often fails to achieve ideal results under the operating conditions of complex working conditions.Moreover,the complex neural Network(Wavelet Package transform-depthwise Sparable Convolution-Residual Network)is commonly associated with large number of parameters.WPT-DSC-ResNet is proposed in this paper.WPT-DSC-ResNet is a Wavelet Package transform-depthwise Sparable convolution-residual Network,which can extract more useful features from the signals and reduce the training time of the model.Firstly,in the aspect of data preprocessing,this paper analyzes the influence of data preprocessing method on fault diagnosis effect from the aspects of time domain,frequency domain and time frequency domain.On this basis,the node energy decomposed by the signal in the subspace is used to reconstruct the two-dimensional image.High-dimensional images are easier to extract features and have better image classification effect.The preprocessed two-dimensional data has higher fault diagnosis accuracy than the one-dimensional original vibration signal.Image processing methods based on deep learning have been applied to image classification,target detection,image segmentation and other fields.The image classification network of WPT-DSC-ResNet method is improved on the basis of residual network(ResNet),and features of different scales are extracted by using the deep network layers of ResNet.At the same time,a lightweight deep neural network is constructed by combining the structure of Depthwise Sparable Convolution(DSConv)in Mobile Net with ResNet network.Through a case study of motor bearing data,it is verified that this method can reduce the time of model training without losing the accuracy of image classification.At the same time,aiming at the problem of data imbalance in fault diagnosis,this thesis improves the loss function in the network model,and sets up a comparative test to verify the effectiveness of WPT-DSC-ResNet proposed in this thesis for fault diagnosis.