| Rotating electrical machines are the most commonly used devices for converting electrical energy into mechanical energy.The survey shows that the motor bearing fault occupies a high proportion in the motor fault,and it is very necessary to detect the bearing fault.The traditional diagnosis method mainly relies on the extraction method of expert experience design features,which will occupy too many human resources in the case of increasingly complex bearing fault features.Motor bearing fault diagnosis based on deep learning can automatically learn complex feature information under supervised training to achieve effective diagnosis of motor bearing faults.However,in order to obtain a high accuracy of motor bearing fault diagnosis based on deep learning,a deep network is usually required,which requires a high memory of the equipment,and a small model cannot achieve good diagnostic accuracy;at the same time,the bearing fault data The collection of sets is sometimes difficult,and it is impossible to obtain a large number of samples for training in deep learning.In this paper,the convolutional neural network is used to diagnose bearing faults.The main research contents are as follows:(1)Aiming at the problem that when deep learning has high diagnostic accuracy in bearing fault diagnosis,the number of model parameters is usually large and occupies a lot of memory resources.A bearing fault diagnosis method based on lightweight deep convolutional neural network is proposed.First,the one-dimensional time series signal of the bearing is transformed into a two-dimensional time-frequency map by continuous wavelet transform to enhance the feature information;secondly,a deep convolutional neural network is built,and the residual structure is introduced to improve the network’s diagnosis accuracy for 16 types of bearing faults;Then,the depthwise separable convolution is fused,and the fully connected layer is replaced by global average pooling to construct a lightweight deep convolutional neural network,and the Adam optimizer is used to improve the convergence speed of the network.The experimental results show that the constructed lightweight deep convolutional neural network not only has higher accuracy than the classical classification network,but also occupies less device memory.(2)When the number of bearing fault samples is small,directly using the above-mentioned diagnosis method will lead to insufficient training and over-fitting phenomenon.An ECACGAN network method is proposed to augment the bearing fault data.First,the ECA mechanism was integrated into the ACGAN network to improve the quality of the images generated by the ACGAN network;then,a small number of bearing datasets collected in the laboratory were used to train multiple Gan networks,and the FID and IS indicators were used to verify the images generated by the ECACGAN network.Finally,the effectiveness of the generated samples is verified by the constructed lightweight deep convolutional neural network,Lenet5 and Alex Net,and it is further proved that the performance of the proposed lightweight deep convolutional neural network is better than the other two classical classification network.Starting from industrial needs,this paper conducts research based on convolutional neural networks and generative adversarial networks.Through the lightweight deep convolutional neural network and the improved generative adversarial network,a new idea is provided for the application of deep learning in bearing fault diagnosis,which has certain engineering application value and theoretical reference value. |
