Research On Rolling-element Bearing Fault Diagnosis Using Improved LeNet-5 Network

As one of the most vulnerable parts in mechanical equipment,rolling bearing’s working state directly affects the performance and production efficiency of the whole equipment,which will cause economic losses,or even disastrous consequences.Therefore,fault diagnosis of rolling bearings is essential to ensure the safe and stable operation of modern machinery and equipment.In view of this,a rolling bearing fault diagnosis method based on deep learning is proposed.In order to solve the problems of traditional LeNet-5 network in rolling bearing fault diagnosis,such as low recognition accuracy,slow convergence speed and weak generalization ability,a rolling bearing fault diagnosis method based on improved 2D LeNet-5 network was proposed.Firstly,the traditional LeNet-5 network is improved,the convolutional layer and pooling layer are designed reasonably,and the size and number of convolutional cores are carefully adjusted to improve the fault classification ability.Batch normalization is adopted after each convolution layer to accelerate the convergence speed.Dropout operation is introduced to enhance and improve the generalization ability after the full connection layer besides the last full connection layer.Then,the one-dimensional original vibration signals are transformed into two-dimensional gray-scale images,and histogram equalization is carried out to solve the problem that the local features are not obvious.Experimental results show that the proposed rolling bearing fault diagnosis method based on the improved 2D LeNet-5 network can realize multi-state recognition and classification of rolling bearings,and the average accuracy of fault identification can reach 99.25%.Compared with other rolling bearing fault diagnosis methods,the proposed rolling bearing fault diagnosis method achieves higher accuracy at lower computational cost.To further improve the effectiveness of fault diagnosis,on the basis of improved 2D LeNet-5 network,an end-to-end rolling-element bearing fault diagnosis method based on the improved 1D LeNet-5 network is proposed.Firstly,an improved 1D LeNet-5 network for rolling bearing fault diagnosis is designed,which has a similar structure to the improved2 D LeNet-5 network,including 5 convolution layers,5 pooling layers and3 full connection layers.One-dimensional convolution and pooling operations can be carried out directly on the original vibration signals without any data preprocessing.Then,the optimal honey source found by the artificial bee colony algorithm was used as the initial weight and bias value of the improved 1D LeNet-5 network,which further improved the convergence speed and classification accuracy of the diagnostic model,and enhanced the adaptability of the improved 1D LeNet-5 network.The experimental results show that the adaptive 1D LeNet-5 network has higher fault diagnosis accuracy and less training time in most cases,but the performance of the improved 2D LeNet-5 network is better than that of the adaptive 1D LeNet-5 network in the environment of small training samples and strong noise.